PERFORMANCE ANALYSIS OF PLATE HEAT EXCHANGERS USED AS REFRIGERANT EVAPORATORS Jianchang Huang A thesis submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, in fulfillment of the requirements for the degree of Doctor of Philosophy. University of the Witwatersrand, Johannesburg August, 2010 i ABSTRACT In this study the heat transfer and frictional pressure drop performance characteristics of plate heat exchangers (PHE?s) used as refrigerant liquid over- feed evaporators were investigated. PHE?s have been gaining new applications in the refrigeration industry, especially as evaporators, during the last few decades, but the available information in the open literature for operation in this mode is rather limited. This study aims to extend the knowledge of PHE evaporator performance and to develop a model for use in evaluating heat transfer and pressure drop over as wide a range of operating conditions as possible. A laboratory experimental facility was constructed and the thermal-hydraulic characteristics of three middle-size industrial PHE?s were measured. The exchangers all had 24 plates of the same size but with different chevron angle combinations of 28?/28?, 28?/60?, and 60?/60?. Two sets of tests were carried out with the three units: single-phase performance tests with water, and evaporator performance tests with R134a and R507A, for which the exchangers operated as refrigerant liquid over-feed evaporators. The tests with water served to provide accurate water-side heat transfer information for the evaporator performance analysis which is the primary purpose of this study. In the evaporator performance tests, refrigerant flow boiling heat transfer and two-phase pressure drop data were obtained under steady conditions, over a range of heat flux from 1.9 to 6.9 kW/m2, refrigerant mass flux from 5.6 to 31.4 kg/(m2s), outlet vapour quality from 0.2 to 0/95, and saturation temperatures from 5.9 to 13.0 ?C. Additional field data of thermal performance were collected on an ammonia and a R12 PHE water chiller, operating as thermo-siphon evaporators at their design working conditions. All experimental data were reduced and analyzed to obtain the refrigerant-side heat transfer coefficients and frictional pressure drops in the PHE evaporators. The heat transfer results showed a strong dependence on heat flux and a weak dependence on mass flux, vapour fraction and the chevron angle. Along with the observations from the ammonia and R12 evaporators, it is concluded that the dominating heat transfer mechanism in this type of evaporator is nucleate boiling rather than forced convection. In contrast to the heat transfer characteristics, the refrigerant two-phase frictional pressure drop was found to be strongly influenced by mass flow rate, vapour fraction and also the chevron angle. An almost linear increase of the frictional pressure drop with the homogeneous two-phase kinetic energy per unit volume was observed for both refrigerants. ii Based on the experimental data, correlations were developed for predicting the refrigerant boiling heat transfer coefficient and two-phase frictional pressure drop in PHE liquid over-feed evaporators. Two correlations were developed for boiling heat transfer, one of these reflecting the h-q relationship in pool boiling, the other with all constants and exponents determined by regression analysis. The mean absolute errors are respectively 7.3% and 6.8% for these correlations. For two-phase frictional pressure drop, data were correlated using two established methods, namely the homogeneous and the Lockhart-Martinelli methods, with means absolute errors of 6.7% and 4.2%, respectively. The homogeneous model showed a slightly higher discrepancy with the experimental data but is likely to be more physically sound for PHE evaporators, and is much simpler to apply. Validation of these correlations with other data has been difficult due to the shortage of published information. For other refrigerants operating at comparable conditions, these correlations should serve as a guide, while more accurate design or evaluation may need to be based on further testing. The performance analysis carried out in this study was based on systematic experimental investigations and field tests on industrial PHE units. Correlations were developed covering a rather extensive range of flow parameters, plate geometry and various refrigerants. Such correlations have not been reported previously for PHE liquid over-feed evaporators. The results simplify the performance analysis of PHE evaporators and provide a reliable thermal- hydraulic model of PHE liquid over-feed evaporators, which can be used for system modeling of water-chilling machines employing this type of evaporator. iii ACKNOWLEDGEMENT It is difficult to express my gratitude to Professor Thomas J. Sheer, my supervisor, for his support both academically and financially, and for his enormous encouragement and patience during the course of this study. Thank you for your guidance and help, for so long. I wish to give my sincere thanks to Dr. Michael Bailey-McEwan, my co- supervisor, for his time and all the discussions. Your advice and help had always been valuable to me. I would also like to convey my gratitude to those who have helped me in any way to finish this work. To Chunyan Shen, for her encouragement and long-time support. To the Mechanical Lab staff, especially Mr. John D. Cooper and Mr. Albertus P. Moller, for their considerable assistance with the experimental apparatus. To Alexander Russell, for his excellent assistance during the water tests. To Envirocool refrigeration company, for suggestions and building up the experimental infrastructure. To Mr. Basil Chassoulas, of the School of Chemistry, for his advice on selecting temperature sensors. To my friends and office neighbors, for all the discussion and leisure talking. To the University of The Witwatersrand, Johannesburg, for granting me the Postgraduate Merit Award for three years, for the service and facility, and the time and memories I shared with. iv CONTENTS Abstract i Acknowledgement iii Contents iv Nomenclature viii CHAPTER 1 INTRODUCTION 1 1.1 Background 1 1.2 Objectives, Scope and Methodology 2 1.3 Structure of the Thesis 3 CHAPTER 2 LITERATURE REVIEW 5 2.1 Pool Boiling 5 2.1.1 Overview of Boiling 5 2.1.2 Pool Boiling 6 2.2 Flow Boiling 8 2.2.1 Flow Patterns 9 2.2.2 Flow Boiling in Microchannels 10 2.2.3 Flow Boiling Heat Transfer Correlations 11 2.2.4 Estimation of Boiling Heat Transfer Coefficient of Refrigerant Evaporators 24 2.3 Two-phase Pressure Drop 26 2.3.1 An Overview 26 2.3.2 The Homogeneous Flow Model 28 2.3.3 The Separated Flow Model 31 2.3.4 Two-phase Pressure Drop Correlations 32 2.4 Heat Exchanger Theory 39 2.4.1 Classification of Heat Exchangers 39 2.4.2 Basics of Thermal Characteristics 39 2.5 Plate Heat Exchangers 41 2.5.1 Types of Plate Heat Exchangers 41 2.5.2 Plate Geometry 43 2.5.3 Flow Arrangements 46 2.5.4 Hydraulic diameter, Reynolds Number, and Friction Factor 47 2.6 Single-phase Flow in PHE?s 50 v 2.6.1 Overall Thermal-hydraulic Performance 50 2.6.2 Single-phase Heat Transfer and Pressure Drop Correlations 53 2.6.3 Flow Distribution 63 2.6.4 Fouling 64 2.6.5 Numerical Simulations 65 2.7 Two-phase Flow in PHE?s 66 2.7.1 Adiabatic Two-phase Flow Characteristics 67 2.7.2 Flow Boiling Heat Transfer in PHE?s 70 2.7.3 Two-phase Pressure Drop 76 2.7.4 Heat Transfer and Pressure Drop Correlations 80 CHAPTER 3 EXPERIMENTAL INFRASTRUCTURE 87 3.1 Introduction 87 3.2 Brazed Plate Heat Exchangers 88 3.3 Water Test Arrangements 89 3.4 Refrigerant Evaporator Test Arrangements 90 3.4.1 Liquid Over-feed System 90 3.4.2 System Design 93 3.4.3 The Experimental Facility 96 3.5 Instrumentation and Measurement 99 3.5.1 Water Test System 99 3.5.2 Refrigerant Evaporator Test System 101 3.6 Data Acquisition System 104 3.7 Sensor Calibration 108 CHAPTER 4 SINGLE-PHASE PERFORMANCE TESTS WITH WATER 111 4.1 Introduction 111 4.2 Test Procedure 111 4.3 Data Reduction 113 4.3.1 Heat Transfer 113 4.3.2 Pressure Drop 116 4.4 Uncertainty Analysis 119 4.5 Results and Discussion 120 4.5.1 Heat Transfer Characteristics 120 4.5.2 Pressure Drop Characteristics 124 4.6 Conclusion 126 CHAPTER 5 FIELD TESTS ON LARGE-SCALE PHE?S 128 vi 5.1 Introduction 128 5.2 PHE Thermo-Siphon Evaporators 128 5.3 Measurements and Data Reduction 131 5.4 Results and Discussion 132 5.5 Conclusion 134 CHAPTER 6 EVAPORATOR PERFORMANCE TESTS USING R134a AND R507A 136 6.1 Introduction 136 6.2 Test Procedure 138 6.3 Data Reduction 139 6.3.1 Flow Boiling Heat Transfer 139 6.3.2 Two-phase Pressure Drop 141 6.4 Uncertainty Analysis 146 6.5 Results and Discussion 146 6.5.1 Heat Transfer Characteristics 147 6.5.2 Pressure Drop Characteristics 152 6.6 Conclusion 154 CHAPTER 7 DEVELOPMENT OF THERMAL-HYDRAULIC CORRELATIONS FOR PHE EVAPORATORS 155 7.1 Introduction 155 7.2 Flow Boiling Heat Transfer 155 7.3 Two-phase Frictional Pressure Drop 160 7.3.1 Correlation using the Homogeneous Model 160 7.3.2 Correlation using the Lockhart-Martinelli Method 165 7.4 Conclusion 169 CHAPTER 8 CONCLUSIONS AND RECOMMENDATIONS 171 8.1 Conclusions 171 8.2 Suggestions for Future Work 174 8.3 Suggestions on Experimental Facility Design 175 References 177 Appendix A Plate Geometrical Information 190 Appendix B Sensor Technical Data and Wire Connections 193 vii Appendix C Sensor Calibration 199 Appendix D Data Acquisition System Specifications 207 Appendix E Experiment Results: Water Tests 209 Appendix F Experimental Results: Refrigerant Evaporator Tests 212 Appendix G Sample Calculation: Refrigerant Evaporator Tests 224 Appendix H MATLAB Program Flow Chart: Water Heat Transfer Coefficient 229 Appendix I Uncertainty Analysis: General Theory, the ASME and ISO Method 232 Appendix J Uncertainty Analysis: Sensor Calibration 249 Appendix K Uncertainty Analysis: Water Tests 260 Appendix L Uncertainty Analysis: Refrigerant Evaporator Tests 270 Appendix M Heat Transfer Surface Partition for a DX Evaporator 279 Appendix N Operating Instructions for the Refrigerant Evaporator Test Facility 281 viii Nomenclature A heat transfer area m2 cA flow channel cross-section area m 2 chA single channel flow cross-section area m 2 pA heat transfer area per plate m 2 a acceleration m/s2 Bo boiling number, ( )fgBo /q G i= ? ? b plate corrugation depth m C Chisholm parameter m fc Darcy friction factor, ( )21f w 2/c u? ?= ? pc specific heat at constant pressure J/(kg?K) Co convection number, 0.50.8 g f 1Co x x ? ? ? ? ?? ? = ? ?? ?? ? ? ? ? D mass transfer diffusivity m2/s d diameter m 0d bubble departure diameter, 0.5 0 f g 20.0146 ( )d g ?? ? ? ? ? = ? ? ?? ?? ? m ed equivalent diameter m hd hydraulic diameter, h 4 VSd = m E enhancement factor for forced convection ? re percentage error, corr expe exper e ? ? ? ? = ? me mean absolute error, ( )corr expeexpem 1e n ? ?? ?= ?? ? F 1. force N 2. log mean temperature difference correction factor ? 3. suppression factor of nucleate boiling in Chen corr. ? f Fanning friction factor, f4f c= ? Fr Froude number, ( )2Fr /u gL= ? fFr Froude number assuming all mass flow as liquid ? ( )2 2f fFr /G gd?= G mass velocity kg/(m2?s) g gravitational constant m/s2 ix cg gravitational constant in English units (ft?lbm)/(s2?lbf) h 1. heat transfer coefficient W/(m2?K) 2. specific enthalpy kJ/kg tph two-phase flow boiling heat transfer coefficient W/(m2?K) nbh nucleate boiling heat transfer coefficient W/(m2?K) fh single-phase convective heat transfer coefficient W/(m2?K) fgi latent heat of vaporization J/kg j 1. phase superficial velocity, i i c/j Q A= m/s 2. Chilton-Colburn j-factor, 1/3 1/3 Nu Sh Re Pr Re Sc j = = ? ? ? Nuj heat transfer group, ( )Nu 1/61/3 wall Nu Pr / j ? ? = ? k thermal conductivity W/(m?K) ck mass transfer coefficient m/s L length m effL effective channel length m pL port center to port center channel length m M 1. molecular weight kg/mol 2. mass kg m? mass flow rate kg/s chN number of channels ? pN number of plates ? p, effN number of effective plates ? passN number of passes in a PHE ? RCn recirculation rate in liquid over-feed system ? Ntu heat exchanger number of transfer units ? Nu Nusselt number ? P pressure Pa critP critical pressure of fluid Pa rP reduced pressure, r crit/P P P= ? P? pressure difference Pa Pr Prandtl number, pPr /c k?= ? Q? total heat transfer rate W Q volumetric flow rate m3/s q heat flux W/m2 R thermal resistance m2?K/W fR fouling factor m 2 ?K/W x pR surface roughness m Re Reynolds number, Re /ud? ?= ? foRe liquid-only Reynolds number, fo fRe /Gd ?= ? goRe vapour-only Reynolds number, go gRe /Gd ?= ? ar channel aspect ratio, a p /r L w= ? wetS flow channel wetted surface area m 2 Sc Schmidt number, Sc / D?= ? Sh Sherwood number, cSh /k L D= ? ? T temperature ?C t time s LMT? log mean temperature difference ?C satT? excess temperature in boiling, sat wall satT T T? = ? ?C U over-all heat transfer coefficient W/(m2?K) u velocity m/s uncertainty ? V volume m3 V arrowrightnosp velocity vector m/s We Webber number, ( )2We /G d ??= ? w channel width m x vapour quality, g /x m m= ? ? ? mx mean vapour quality, ( )m in out0.5x x x= + ? X Martinelli parameter ? 2 fric, f fric, g dP dPX dz dz ? ? ? ? = ? ? ? ?? ? ? ? Y parameter used in the Chisholm (1973) correlation ? 2 fric, fo fric, go dP dPY dz dz ? ? ? ? = ? ? ? ?? ? ? ? z channel axial co-ordinate m Greek ? 1. thermal diffusivity, ( )p/ k c? ?= m2/s 2. void fraction, g c/A A? = ? ? plate corrugation angle degrees (measured from vertical direction) ? thickness m xi ? heat exchanger effectiveness ? ? 1. plate surface enlargement factor ? 2. general term for undefined parameter ? 2 fo? two-phase friction multiplier based on pressure gradient assuming total mass flow as liquid ? 2 f? two-phase friction multiplier based on pressure gradient assuming only liquid phase fraction flowing ? ? plate corrugation pitch m ? viscosity Pa?s ? two-phase mean viscosity Pa?s ? 90?- ? degrees ? kinematic viscosity, /? ? ?= m2/s ? density kg/m3 ? two-phase mean density kg/m3 ? surface tension N/m w? shear stress at the wall N/m 2 ? heat transfer enhancement factor, tp f/h h? = ? (introduced in Shah 1976 correlation) Subscripts 0 reference condition acce acceleration c 1. cross-section 2. convective ch channel corr correlation crit critical e equivalent eff effective elev elevation expe experiment f liquid phase fo all mass flow as liquid fg change of phase, liquid to vapour fric friction g gas/vapour phase h hydraulic i inlet m mean max maximum meas measurement min minimum nb nucleate boiling o outlet p plate port port r refrigerant sat saturation sub subcooling tp two-phase w 1. water, 2. wall wall wall 1 CHAPTER 1 INTRODUCTION 1.1 Background Plate heat exchangers (PHE?s) were first commercially introduced in the 1920?s to meet the hygienic demands of the dairy industry (Seligman, 1963, Carlson, 1992), while some patents existed as early as in the 1870?s in Germany (Clark, 1974). Design of this type of exchanger reached maturity in the 1960?s with the development of more effective plate geometries, assemblies, and improved gasket material, and the range of possible applications has widened considerably (Kakac S. and Liu H., 2002). PHE?s are nowadays widely used in a broad range of heating and cooling applications in food processing, chemical reaction processes, petroleum, pulp and paper, as well as in many water-chilling applications. Some basic features of PHE?s include high efficiency and compactness (i.e., high heat transfer capacity per unit volume compared to conventional, shell-and-tube heat exchangers), high flexibility for desired load and pressure drop, easy cleaning, and cost competitiveness. While PHE?s became popular for liquid-to-liquid heat transfer duties, their use in phase-changing applications was not common initially. Before the 1990?s such applications were mostly in the fields of concentrating liquid food and drying of chemicals (Usher, 1970, Kumar, 1993). Applications in refrigeration systems were rare, mainly because of concerns over refrigerant leakage, and also because of the pressure limits required, especially on condensation applications. In the last two decades, with the introduction of semi-welded and brazed PHE?s, this type of exchanger has been increasingly used in refrigeration systems, from domestic heat pumps to large ammonia installations for water-chilling duties. Heat exchangers, including PHE?s and other types, are designed and employed according to two criteria: heat transfer and pressure drop. Characteristics of the heat exchanger?s thermo-hydraulic performance are the primary interest of most investigations. The corrugated channels of PHE?s have probably the most complicated geometry of all flow ducts. As the number of applications has 2 increased, single-phase flows in PHE?s have been well investigated in the last few decades. Although a general theory about the influence of some geometric parameters is lacking, due to the complex geometry of the flow channel, a large number of correlations for heat transfer and pressure drop are available in the open literature. The number of published investigations on two-phase heat transfer and pressure drop in PHE?s is increasing but is not yet comprehensive. Due to the complexity of the two-phase flow process, a rigorous theoretical analysis is not feasible. The fundamental understanding of the flow boiling mechanism in this type of channel is rather limited. A few two-phase flow boiling heat transfer and frictional pressure drop correlations exist in the literature, obtained from water, R-22, R-134a, R-410a, and ammonia. The coverage of important flow boiling parameters, including the heat flux, mass flux and vapour fraction, is usually limited and unsystematic and the operating conditions vary. As a result, there is not a generally accepted calculation method for flow boiling heat transfer and two-phase pressure drop for PHE evaporators. Although accurate and complex formulations were developed by manufacturers, based on many years of research and systematic testing of specified plates and arrangements, these formulations are confidential and closely guarded. Vapour-compression refrigerating machines employing PHE evaporators have been used for some years in South African deep mining for water-chilling duties. For the requirement of large cooling loads (well over 1 MW) and also for thermal efficiency, liquid over-feed systems are usually considered rather than direct expansion (DX) systems. System modelling of the water-chilling machines has been difficult due to the shortage of reliable information to model the evaporator performance. Such a need motivated an extensive survey and new research on PHE evaporator performance analysis, to obtain a validated thermal-hydraulic model of PHE evaporators which can be used in the simulation of compete refrigeration systems. 1.2 Objectives, Scope and Methodology The purpose of the present study is to investigate the performance of plate heat exchangers used as refrigerant liquid over-feed evaporators. The objective is to extend the present knowledge of the thermo-hydraulic performance of PHE evaporators with consideration of flow conditions (mass flux, heat flux, vapour fraction) and channel geometries (chevron angles). This study aims to obtain a generally applicable model, which is able to provide design and operating guidelines for this type of evaporator, and to be used for the simulation of complete refrigeration systems. The specific objectives of this thesis are: 3 1. To carry out a thorough literature review on plate heat exchangers, their terminology, working principles, single-phase performance characteristics and most importantly the state of the art of evaluating the flow boiling heat transfer and two-phase frictional pressure drop in PHE channels, 2. To obtain single-phase water-water heat transfer and pressure drop data for a range of PHE?s. This serves to provide a quantitative understanding of the performance of PHE?s in single-phase applications, and to provide accurate water-side heat transfer coefficient information for the PHE evaporators. 3. To obtain flow boiling heat transfer and two-phase frictional pressure drop data, to cover various refrigerants, flow conditions (mass flow rate, heat flux, vapour quality), and channel geometries (chevron angles). 4. As the final and principal aim, to extend the present knowledge of the thermo-hydraulic performance of PHE liquid over-feed evaporators. It is the task of this study to develop general correlations which cover more than one refrigerant and must be able to predict the refrigerant flow boiling heart transfer and frictional pressure drop with reasonable accuracy. There are a few types of PHE?s available, but this study was limited to the chevron (herring-bone) corrugation type. Inlet flow conditioning (normally commercially available via flow distributors) was not considered 1. A laboratory experimental facility was to be constructed. The facility was to be designed as a liquid over-feed system. This type of system is more flexible for experimental purposes compared with dry expansion and thermo-siphon systems, giving a wide spectrum of possibilities for evaporator running conditions. The experimental equipment was to employ multiple PHE units covering different chevron angles which are typical of industrial applications, and various refrigerants needed to be used. Development of a flow boiling thermo-hydraulic model of the evaporators was to be carried out based on the experimental results, with the assistance of statistical analysis. 1.3 Structure of the Thesis This thesis contains eight chapters. Chapter 1 provides an introduction to plate heat exchanger applications in refrigeration, followed by the objectives, scope, motivation and methodology of the current study. Chapter 2 provides a critical 1 Flow distributors are designed to gain uniform flow distribution among channels. This kind of device is more commonly employed in DX evaporators. The Effect of distributors on the thermo-hydraulic performance of a PHE is hard to estimate, and for the sake of clarity and simplicity, distributors are not used in the present study. 4 review of the literature. This chapter includes a thorough examination of the established theories and methods for flow boiling heat transfer and two-phase pressure drop in general, an in-depth review of plate heat exchanger theory, terminology, working principles, and single-phase performance characteristics, with special attention then paid to flow boiling heat transfer and two-phase frictional pressure drop characteristics in PHE channels. Chapter 3 provides a detailed description of the experimental facility, including the plate heat exchangers, the water test facility and the refrigerant evaporator test facility, instrumentation and the data acquisition system. Chapter 4 describes the single- phase performance tests with water. Chapter 5 analyzes the results of previous field tests on large-scale PHE water-chillers employing ammonia and R12. Chapter 6 presents results of the evaporator performance tests using R134a and R507A. Chapter 7 discusses the results of the data analysis, and presents the development of thermal and hydraulic correlations for PHE liquid over-feed evaporators. Chapter 8 summarizes the overall results and gives suggestion for future work and experimental facility design. 5 CHAPTER 2 LITERATURE REVIEW 2.1 Pool Boiling 2.1.1 Overview of Boiling Boiling is a thermodynamic process occurring at a solid-liquid interface in which a fluid is heated by a surface with a temperature above the saturation temperature of the liquid which changes from liquid to vapour phase. Pool boiling is the simplest form of boiling with the heating surface being immersed in a large pool of stagnant liquid. If boiling occurs in a liquid that is in motion relative to the surface, it is called forced convective boiling, or flow boiling. In both types, depending on whether the main body of the liquid in the vicinity of the heated surface is at (or slightly above) or below the saturation temperature, either saturated boiling or subcooled boiling occurs, respectively. Boiling occurs only when the temperature of the solid surface Twall exceeds the liquid saturation temperature Tsat corresponding to its local pressure. Heat is transferred from the solid surface to the liquid, as can be expressed by Newton?s law of cooling: ( )wall sat satq h T T h T= ? = ? (2. 1) where ?Tsat = Twall - Tsat is called the excess temperature. Equation (2.1) applies to all types of phase-changing heat transfer, including pool boiling, flow boiling, and condensation. Generally, heat transfer coefficients associated with phase changing, including boiling and condensation, are larger compared with those associated with single-phase forced and natural convection. The process of boiling is characterized by the formation of bubbles, which grow and detach from the surface. Vapour bubble growth and dynamics depend, in a complicated manner, on the excess temperature, heating surface nature, and 6 liquid properties (Incropera and DeWitt, 1990). Boiling is a complex phenomenon that is difficult to model analytically. Despite many works on bubble dynamics, boiling nucleation, influence of free convection, etc., a reliable and accurate analytical theory is not available. 2.1.2 Pool Boiling Nucleate pool boiling is affected by various parameters. In many cases, an accurate quantitative description of those parameters that affect the heat transfer coefficient is very difficult. The three most investigated parameters include heat flux, saturation pressure, and the thermo-physical properties of the working fluid (Pioro et al., 2004). Despite the large number of articles on the subject, a theoretically consistent calculation method on pool boiling heat transfer does not exist (Gorenflo, 1999). The process of pool boiling is best described experimentally by the boiling curve, as available from most heat transfer text books. Briefly, four boiling regimes are found in pool boiling, with unique characteristics, those are natural convection, nucleate boiling, transition boiling, and stable film boiling. Of those, the nucleate boiling region is the primary interest of industry applications. Data from nucleate pool boiling experiments are usually fitted with expressions such as (Thome, 2003): n satq T? ? (2. 2) where the exponent n is of the order of 3. Collier (1983) pointed out that the value of n is dependent on the physical properties of the liquid and vapour and on the nucleation properties of the surface, normally in the range of 2 - 4, with a most probable value around 3 to 3.3. With n = 3, it can be obtained from Equations (2.1) and (2.2) that: 2/3h q? (2. 3) Literally there are hundreds of pool boiling correlations that have been proposed, and those can be classified into several groups according to the assumed dominating mechanism, for example, bubble agitation, reduced pressure, fluid physical properties, etc. Some of the well organized and widely quoted correlations include those of Rohsenow (1952), Stephan and Abdelsalam (1980), Cooper (1984), Gorenflo (1993). Thome (2003) presented a detailed examination of those correlations. Three of those correlations are employed in the present study for comparison purposes, and are given below. 7 Stephan and Abdelsalam (1980) Correlation 0.745 0.581 0.533 g f p, f0 0 f f sat f f 207 chd qd k k T k ? ? ? ? ? ? ? ? ? = ? ? ? ? ? ?? ? ? ?? ? (2. 4) where Tsat is in K, ( )0 f g0.0146 2 /d g? ? ? ?? ?= ?? ? is the bubble departure diameter an ? = 35? is the contact angle. For refrigerants, 3 r3 10 0.78P ?? ? ? , mean absolute error = 10.57%. Equation (2.4) is one of several correlations for classes of fluids including water, organics, refrigerants and cryogens, developed by the authors using statistical regression techniques. Cooper (1984) Correlation Cooper proposed the following reduced pressure expression for the nucleate pool boiling heat transfer coefficient: 10 p0.12 0.2log 0.55 0.5 0.67 nb r r55 ( log10 )Rh P P M q? ? ?= ? (2. 5) Note that this is a dimensional correlation in which: nbh , nucleate boiling heat transfer coefficient, W/(m2?K), q , heat flux, W/m2 rP , reduced pressure, r crit/P P P= , where critP is the critical pressure of the fluid, M , molecular weight of the fluid, pR , surface roughness parameter in ?m, pR =1 ?m for unspecified surface. Cooper?s correlation is for nucleate pool boiling on plane surfaces; for boiling on horizontal copper tubes, h is suggested to be multiplied by a factor of 1.7. The correlation is based on a large range of data, covering reduced pressures from 0.001 to 0.9, and molecular weights from 2 to 200. This correlation has probably the simplest form of all those available for nucleate boiling heat transfer. However, for its accuracy this correlation has gained popularity among many researchers. The very basic h-q relationship as suggested by Equation (2.3) is clearly evident in this correlation. Collier and Thome (1994) recommended this correlation for water, refrigerants, and organic fluids with poorly defined physical properties. Thome (2003) suggested this correlation for general use. 8 Gorenflo (1993) Correlation ( ) 0.133n p r 0 0 p, 0 Rh qF P h q R ? ?? ? = ? ?? ? ? ?? ? ? ? (2. 6) where h0 is the reference heat transfer coefficient, based on Pr = 0.1 and Rp,0 = 0.4 ?m, q0 = 20 kW/m2 is the reference heat flux, ( ) ( )0.27r r r r r 1.2 2.5 / 1F P P P P P= + + ? , and 0.3r0.9 0.3n P= ? for all fluids except for water and helium. The Gorenflo method is based on the concept of reduced pressure but also on a reference heat transfer coefficient at specified standard conditions. The reference value of h0 are given by the author for a selection of fluids. For fluids outside this selection, additional correlations determining the reference h0 are needed. 2.2 Flow Boiling Flow boiling, or forced convective boiling, can have two categories according to the surface geometry: internal flow and external flow. Internal flow boiling is associated with bubble formation at the inner surface of a heated channel through with a liquid flowing. Bubble growth and separation are believed to be strongly influenced by fluid motion, and hydrodynamic effects differ significantly from those corresponding to pool boiling (Incropera and DeWitt, 1990). The process of flow boiling is associated with the existence of various two-phase flow patterns and is difficult to model. In the literature, it is generally agreed that flow boiling heat transfer has two fundamental components: nucleate boiling and forced convection. The nucleate boiling part resembles the nucleate pool boiling heat transfer in which heat is transferred into bulk fluid by means of local bubble nucleation, growth and subsequent departure from the surface. The forced convection part is simply assumed as a representation of single-phase heat transfer. The total heat transfer coefficient is calculated by combining the two, with weighting. The combined effects of the two components are not well understood, many hypotheses have been proposed, based on which correlations have been developed. Usually the forced convection part is believed to be enhanced, because the fluid velocity, which is the dominating parameter in single phase convective heat transfer, is higher in two-phase flow than in single-phase 9 liquid flow given that the channel cross-section is partially occupied by gas (vapour) phase. Other the other hand, the nucleate boiling part is suppressed due to the fact that forced convection effect reduces the thermal boundary layer thickness on the heated surface, and thus suppresses bubble nucleation. 2.2.1 Flow Patterns Various flow structures are observed in flow boiling, and those are defined as two-phase flow patterns with identifying characteristics. Flow patterns of vertical and horizontal flow generally differ, as flow in horizontal pipes is influenced by the effect of gravity, which acts to stratify the liquid to the bottom and the gas/vapour to the top of the channel. Refrigerant evaporators usually have vertically orientated channels, where two-phase liquid and vapour refrigerant flows in a co-current upward manner. Four flow patterns are usually identified for vertical up-flows in conventional tubes: bubble, slug (plug), churn, and annular (McQuillan and Whalley, 1985). Figure 2.1 gives the flow patterns in an uniformly heated vertical tube with a low heat flux, with associated wall and fluid temperature variations and heat transfer regions. Churn flow is not present in Figure 2.1, but it can be identified as a highly disturbed flow pattern between slug and annular flow. Flow patterns in horizontal flow will be somewhat different from those shown in this figure. Taitel and Dukler (1976) specified five flow regimes in horizontal two-phase flow, those are smooth stratified, wavy stratified, intermittent (slug and plug), annular and dispersed bubble. It should be pointed out that a two-phase flow pattern is a subjective observation, and there is no general assessing method which identifies and describes flow patterns precisely. One simple method of representing flow pattern transitions that occur at particular local conditions is by the form of a two dimensional flow pattern map. Respective flow patterns are represented as areas on a graph, with the co- ordinates being most often chosen as superficial phase velocities, jg and jf, or other generalized parameters containing these velocities. Coordinates of flow patterns can be arbitrarily chosen, as pointed out by Taitel et al. (1980). Because there was little theoretical basis for selection of coordinates, their generality and accuracy are limited. Nevertheless, flow pattern maps have long been widely used in the industry, the most recognized ones are probably that proposed by Hewitt and Roberts for vertical flow, and that of Baker for horizontal flow. The channel geometry of a plate heat exchanger is very different from a straight circular tube; flow patterns identified in conventional tubes may not be entirely applicable. For example, due to the highly three-dimensional flow directions, 10 annular flow is not likely to be present. Investigations on flow patterns in plate heat exchangers have been very few, and limited to adiabatic air-water flows, as will be discussed in Section 2.7.1. 2.2.2 Flow Boiling in Microchannels Channel diameter is an important parameter which has an effect on flow boiling heat transfer. As the channel diameter decreases, the bubble departure diameter reaches that of the flow channel; some macro-scale thermal and fluid phenomena may be suppressed, while others may be enhanced, for example, surface tension forces may become important. Another important difference in two-phase microchannels flows, compared with macro-channel flows, is that the liquid flow is laminar at most conditions, with typical values of Reynolds number from 100 to 4000 (Thome, 2004a, Peters and Kandlikar, 2007). Knadlikar (2004) showed Figure 2.1: Flow patterns with associated heat transfer regions in uniformly heated tube with a low heat flux (Collier and Thome, 1994). 11 that the flow boiling phenomena in micro-channels are very similar to the nucleate boiling phenomena, with the periodic dry-out and rewetting of the wall having the same characteristics. Thome (2004) concluded that nucleate boiling controls evaporation in microchannels. Flow boiling heat transfer coefficients are largely dependent on heat flux and saturation pressure while they depend only slightly on mass velocity and vapor quality. It seems clear from the majority of experimental results, as summarized by Bertsch et al. (2008), that nucleate boiling plays a major role in flow boiling in microchannels, while the convective boiling mechanism is diminished. It should be mentioned that the classification of micro- and conventional channels is not very well agreed in the literature. Many experimental investigations on ?micro-? or ?small? channels, as termed by different authors, have covered the channel equivalent diameter range from 0.2 to 3 mm, and this range might probably be taken as a rough assessment. Plate heat exchanger channels usually have hydraulic diameters from 3 to 10 mm, and are in the midrange between micro- and conventional size channels. 2.2.3 Flow Boiling Heat Transfer Correlations To date there is no general theory witch can predict boiling heat transfer coefficients with satisfactory accuracy, due to the deficiencies in the understanding of the complex thermo-hydraulic processes involved. The prediction of heat transfer remains essentially empirical. Numerous correlations have been developed, many of those being inconsistent with others, and should be reviewed critically. For the purpose of the current study, only those most known and verified correlations which are applicable for vertical conduits and for pure substances are introduced in this section. It is highly important to note that all those correlations give local values for the predicted flow boiling heat transfer coefficients. The flow boiling heat transfer coefficient is defined by Equation (2.1). In general there are three types of models existing in the literature for evaluation of flow boiling heat transfer. All types utilize the assumption that two distinct mechanisms: nucleate boiling and forced convection, have contributions to the overall flow boiling heat transfer. These are: 1. The superposition models assuming that the flow boiling coefficient is an additive of nucleate boiling and forced convection contributions, with weighting factors for each. The Chen 1966 correlation is a typical representative of this category of modelling. This type of correlation has the form 12 tp nb fh S h E h= ? + ? (2. 7) where htp is the two-phase flow boiling heat transfer coefficient, hnb is the nucleate boiling heat transfer coefficient, and hf is the single-phase liquid convective heat transfer coefficient. S and E are suppression and enhancement factors, respectively. 2. Models that use the larger of the two components, hnb and hf, as the two- phase flow boiling heat transfer coefficient, with hnb calculated by modified correlations. The Shah (1982) correlation is one example of this type. 3. The asymptotic models that combine the two components using a power- law type of equation: ( )1/nn ntp nb fh h h= + (2. 8) The mathematical behavior of the asymptotic model is that at n =1, the model is a superposition model, for large values of n, htp tends to the larger of the two components. The Steiner and Taborek (1992) correlation is a representative of this type. Chen (1966) Correlation Based on 665 data points for water and some hydrocarbons, in the conditions of vertical, saturated flow boiling, Chen (1966) proposed his later widely adopted correlation. It is postulated that two basic mechanisms take part in the heat transfer process for saturated flow boiling: the macro-convective mechanism of single-phase liquid forced convection and micro-convective mechanism associated with bubble nucleation and growth. The flow boiling coefficient is expressed in Equation (2.7), with E being denoted by F in the original paper. In this method, the component of forced convection is calculated by the Dittus- Boelter equation: ( ) ( )0.8 0.4 ff f f0.023 Re Pr kh d= (2. 9) where f fRe (1 ) /G x d ?= ? , is the Reynolds number for liquid fraction. hnb is calculated using the Forster and Zuber analysis of nucleate boiling correlation, modified to account for the reduced average superheat in the thermal boundary layer for bubble nucleation at the wall: 13 0.79 0.45 0.49 0.25 f p,f f c 0.24 0.75 nb sat sat0.5 0.29 0.24 0.24 f fg g 0.00122 k c g h T P i ? ? ? ? ? ? = ? ?? ?? ?? ? (2. 10) where sat wall satT T T? = ? is the excess temperature, satP? is the difference in saturated vapour pressure corresponding to satT? , sat sat w sat sat( ) ( )P P T P T? = ? . It is important to note that Equation (2.10) is derived from non-dimensional equations (Forster and Zuber?s correlation for pool boiling), as such, gc is needed only with English units. As already noted, the effect of forced convection is postulated to be enhanced due to the fact that liquid phase fraction flow velocity is increased, while the nucleate boiling part is suppressed because a sharper temperature gradient at the wall is created by the forced flow, which has an adverse effect on bubble nucleation. The enlargement factor, F, is always greater than unity. The author argued that F is strictly a flow parameter, and therefore can be assumed as a function of the Martinelli parameter Xtt1. S was estimated as a function of F and Ref. F and S were both given graphically in the original paper. Collier (1983) later proposed the following curve fitting equations for these two parameters: tt 0.736 tt tt 1 1 for 0.1 1 1 2.35 0.213 for 0.1 X F X X ? ??? = ? ? ?? + >? ?? ? ?? (2. 11) ( )1.176 1.25f 1 1 2.53 10 Re S F? = + ? (2. 12) The Chen correlation is based on 6 different data sources of over 600 data points. The correlation is for vertical flow only, and valid for pressures in the range of 1- 35 bar, vapour quality 0-0.7, heat flux 6-2400 kW/m2, liquid inlet velocity 0.06- 4.5 m/s. One shortcoming of Chen?s correlation is that a rather large number of physical properties are required, and since the wall temperature is not known beforehand, an iteration procedure is needed in calculation. Chen?s correlation is among the oldest yet most widely used correlations for saturated flow boiling heat transfer in vertical channels. The correlation showed 1 0.10.50.9 g f tt f g 1 xX x ? ? ? ? ? ?? ? ?? ? = ? ?? ?? ? ? ?? ? ? ? ? ? for straight pipes. For more details refer to Section 2.3.3. 14 excellent agreement with the data analyzed by Chen, however, some later researchers have reported that this correlation did not agree well with their experimental data, such as Kandlikar (1990) and Gungor and Winterton (1986). Especially, large deviations with refrigerant data have been observed (Liu and Winterton, 1990, Kandlikar and Nariai, 1999). Based on the form of Equation (2.7), large numbers of correlations have been proposed since Chen?s correlation. Many modifications and analytical examinations of the Chen correlation can be found in the literature in the effort of expanding the correlation to more fluids and operating parameters, for example, Bennett and Chen (1980) and Edelstein et al. (1984). Shah (1976, 1982) Correlation Shah in 1976 proposed a general correlation in the form of a graphical chart for the estimation of heat transfer coefficients during saturated boiling at sub-critical heat flux in tubes and annuli, both vertical and horizontal. Equations representing the chart correlation were presented in his 1982 paper. Similar to Chen, he included the two distinct mechanisms: nucleate boiling and forced convection. However, instead of adding the two contributions, this method chooses the larger as the final result. No explicit nucleate boiling term is used in this method, while the two mechanisms are both attributed to a heat transfer enlargement factor ?. Three flow boiling regions are defined: a nucleate boiling dominant region at low vapour quality x, a convective dominant region corresponding to high x, and in between a bubble suppression region where both mechanisms are significant. In the development of this chart correlation four dimensionless parameters are employed, defined by: tp f h h ? = (2. 13) 0.50.8 g f 1Co x x ? ? ? ? ?? ? = ? ?? ?? ? ? ? (2. 14) fg Bo q Gi = (2. 15) 2 f 2 f Fr G gd? = (2. 16) Co is the convection number, first introduced by Shah in this correlation. Co is proposed to be a replacement of the Martinelli parameter Xtt, since the viscosity 15 ratio was found to have no significant influence. The other three dimensionless parameters have been used in prior correlations. ? is the heat transfer enhancement factor, Bo is the boiling number and Frf is the Froude number assuming all mass flow as liquid. In this method the single-phase liquid heat transfer is calculated, again, by the Dittus-Boelter equation with only the liquid fraction considered. The chart form correlation is very convenient to apply. For the equations representing the chart, the method is shown as follows. First a dimensionless parameter N is defined: f 0.3 f f Co vertical, horizontal with Fr 0.4 0.38Fr Co horizontal with Fr 0.4 N ? >? = ? 1.0, the heat transfer enhancements are calculated by: 0.5 4 nb 0.5 4 230 Bo Bo 0.3 10 1 46 Bo Bo 0.3 10 ? ? ? ? > ?? = ? + < ??? (2. 19) cb 0.8 1.8 N ? = (2. 20) The final heat transfer enhancement is the larger of the two: ( )nb cbmax ,? ? ?= (2. 21) For N ? 1.0, flow boiling heat transfer is dominated by the boiling suppression region and the convective boiling region. The following equations apply: 0.1 0.15 0.5 2.74 bs 0.5 2.74 Bo 0.1 1.0 Bo 0.1 N N F e N F e N ? ? ? ? < ?? = ? physically indicates that momentum diffusivity is greater than the thermal diffusivity, resulting a flatter velocity profile than the temperature profile across the flow. In flow boiling a sharpened temperature profile at the vicinity of the wall is expected, compared with that of single-phase flow heat transfer, because the bulk of the flow is at, or very close to, saturation temperature. 20 where the subscripts NBD and CBD refer to the nucleate boiling dominant and the convective boiling dominant regions. The correlation for vertical tubes and horizontal tubes with Froude number, fFr 0.04> , is ( ) ( ) tp,NBD 0.2 0.7 fl0.8 f tp,CBD 0.9 0.7 fl0.8 f 0.6683Co 1058.0Bo 1 1.136Co 667.2Bo 1 h F x h h F x h ? ? ? = +? ???? = +? ?? (2. 35) The single-phase liquid-only heat transfer coefficient fh is calculated by the Dittus-Boelter equation, with the Reynolds number computed by the liquid fraction only, i.e., f fRe (1 ) /Gd x ?= ? . This treatment is the same as used by Shah. All three dimensionless groups, Co, Frfo, and Bo are the same as given in the Shah correlation. Values for the fluid dependent parameter Ffl were reported for 10 fluids including water and some common refrigerants, and later for R134a in another paper (Kandlikar, 1991), those values are shown in Table 2.1. This later paper further reported that all values are applicable to copper tubes only, for stainless steel tubes Ffl =1. The Kandlikar correlation is based on 5246 data points for water and some refrigerants from 24 sources. Introduction of the fluid dependent parameter flF in the nucleate boiling term can be viewed as an important aspect of the Kandlikar correlation. It is clear that this correlation has been developed under the scheme of the Shah correlation; however, not much improvement has been shown in either accuracy or physical basis. The correlation was later extended to microchannels and minichannels by Kandlikar and Steinke (2003) by introducing the use of laminar flow equations, good agreement with experimental data were reported. Table 2.1: Fluid dependent parameter flF in the Kandlikar (1990) correlation Fluid flF Fluid flF water 1 R113 1.30 R11 1.30 R114 1.24 R12 1.50 R134a 1.63 R13B1 1.31 R152a 1.10 R22 2.20 21 Steiner and Taborek (1992) Correlation Steiner and Taborek (1992) proposed a new general correlation for boiling in vertical tubes. The correlation has an open-ended feature, i.e., parts of it might be exchanged to fit the problem at hand without invalidating the final results. The local two-phase flow boiling coefficient is obtained from an asymptotic equation: ( )1/nn ntp nb fh h h= + (2. 36) The exponent n is to be assigned on empirical grounds to the value that gives the best fit of the data set. n =3 was obtained by the authors on evaluating their large and diversified database. Liu and Winterton (1991) used this type of model in their correlation with n =2. This so called ?asymptotic model? is accepted as the most logical model (Darabi et al., 1995). As n approaches infinity, it represents the case of the ?greater of the two components?, similar to the correlations of Shah (1982) and Kandlikar (1990). For n =1, it represents the superposition model, as that of Chen (1966) and Gungor and Winterton (1986). The local flow boiling heat transfer coefficient is given: ( ) ( ) 1/333tp nb,0 nb fo tph h F h F? ?= +? ?? ? (2. 37) where hnb,0 is the local ?normalized? nucleate pool boiling coefficient at standard conditions of heat flux q0 and reference reduced pressure Pr. Fnb is the nucleate boiling correction factor that compensates for the differences between pool and flow boiling. hfo is the local liquid phase forced convection coefficient based on the total mass flow as liquid. Ftp is the two-phase multiplier that accounts for the enhancement of convection in a two-phase flow, assumed as a function of vapour quality x and the liquid/vapour density ratio ?f/ ?g. For hnb,0, the standard conditions refer to reduced pressure Pr =0.1, mean surface roughness Rp =1 ?m, and the heat flux with specified values for four types of fluid classes (inorganic, hydrocarbons and refrigerants, cryogens, Helium I). hnb,0 can be obtained from experimental data or a correlation of the user?s choice, for example, the Cooper (1984) pool boiling correlation. The nucleate boiling correction factor nbF is given by: 22 ( ) 0.133nf 0.4 p nb pf 0 0 p,0 Rq dF F F M q d R ? ? ?? ? ? ? = ? ?? ? ? ? ? ?? ? ? ? ? ? (2. 38) where, 0.45 3.7 pf r r7 r 1.72.816 3.4 1 F P P P ? ? = + + ?? ? ?? ? r r 1.75 1.105 0.8 0.1 for all fluids except cryogens 0.7 0.13 for cryogens P P e nf e ? ?? = ? ??? ( ) ( ) 5 20.377 0.199ln 2.8427 10F M M M?= + + ? The authors excluded the notion of Fnb as a ?nucleate boiling suppression factor?, which has been conventionally used for the correction factor for hnb, since no effect of mass velocity and vapour quality on flow nucleate boiling had been observed by the authors, over wide ranges of heat flux and pressure. It is therefore concluded that a nucleate boiling ?suppression? does not occur in flow boiling. In Equation (2.38), Fpf is a pressure correction factor valid for Pr <0.95. F(M) is a correction factor accounting for the effect of molecular weight. The reference tube diameter d0 is equal to 0.01 m, and reference surface roughness Rp,0 is 1 ?m which is about the average for commercial tubes. hfo is calculated with the total mass flow as liquid. The two-phase multiplier Ftp is calculated as: ( ) ( ) ( ) ( ) 1.10.35 1.5 0.6 f g 0.52.20.35 1.5 0.010.6 f tp g 20.67 0.7go 0.01 f fo g 1 1.9 0 0.6 1 1.9 1 0.6 1 1 8 1 x x x F x x x x h x x h ? ? ? ? ? ? ? ? ? ? ? ?? ?? ? ? ? + ? ?? ?? ? ?? ?? ?? ???? ?? ? ?? ?? ?? ?= ? + ?? ? ?? ?? ?? ?? ?? ?? ? ? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?+ + ? ? ?? ?? ? ? ?? ?? ?? ?? ?? ?? ?? ?? ? ? ??????? (2. 39) The terms 0.01x and ( ) 0.011 x? were included to arrive at the proper limits at 0x = and 1x = . The Steiner and Taborek (1992) correlation is based on a very extensive data base of 12,607 data points, including 10262 for water and the remaining for four refrigerants (R11, R12, R22 and R113), seven hydrocarbons, three cryogens and 23 ammonia. Besides, much effort has been placed on the physical basis of the model respecting the established principles of pool and convective boiling, rather than relying on statistical development of correlations. The correlation is well structured with logical basis and currently regarded as the most accurate vertical tube boiling correlation for pure fluids (Thome, 2003, Ghiaasiaan, 2007). A Summary It is difficult to predict the boiling coefficient accurately because of the complexities of the mechanisms. The large number of flow boiling correlations available in the literature reveals the fact that the subject is under extensive investigation yet is not well understood. To date a fundamental understanding of the heat transfer mechanism in flow boiling, or any boiling, is still in need. Since the success of Chen?s 1966 correlation, it has been widely accepted that two heat transfer mechanisms apply to flow boiling ? nucleate boiling and forced convection. Various approaches have been proposed in combining these two contributions, resulting in three types of modelling which essentially cover all correlations available. All types of correlations were developed empirically, with important functions or constants defined by experimental results. With the accumulation of experimental data in the subject field, later researchers are able to use larger and larger data bases aiding the development of their correlations. Still, there are deficiencies of data available which place limitations on the accuracy of correlations. For example, as pointed out by Gungor and Winterton (1987), there are at least two such limitations, one concerns the difficulty of measuring the sometimes very small temperature differences in boiling heat transfer, another limitation is that the importance of the heat transfer surface condition has not been accounted for adequately; with the lack of this information there is little justification for attempting a complicated correlation procedure. Chisholm (1983) noted that prediction of two-phase flow parameters may carry an uncertainty of 50%, due to the large number of variables encountered. Comparisons of their correlations are often made by authors with others against the data base from which the new correlation is based. This kind of comparison has always favored the newly developed one. It is always hard to say why one of the approaches should be preferred to the others, since they all possess merits as well as shortcomings. The Chen correlation remains classic and always a popular objective for modification and comparison. The Shah (1982) correlation and Steiner and Taborek (1992) correlation can be regarded as the most accurate ones that are currently available. All correlations have different underlying postulations which could not be fully evaluated with the current knowledge of the 24 subject (it should be remembered that even the knowledge of single-phase flow is essentially empirical.). 2.2.4 Estimation of Boiling Heat Transfer Coefficient of Refrigerant Evaporators As already noted, most, if not all, general flow boiling correlations available provide methods to estimate local heat transfer coefficients of the liquid. A number of those correlations are recommended by ASHRAE (2005 ASHRAE Handbook: Fundamentals) to estimate local heat transfer coefficients of refrigerants during evaporation, including the Gungor and Winterton (1986), Kandlikar (1999), Shah (1982), and the Chen (1966) correlation. For refrigerant boiling inside tubes, the heat transfer coefficient changes progressively as the refrigerant flows through the tube, mainly as a function of the mass flow rate and vapour quality. It is difficult to develop a single relation to describe the heat transfer performance for evaporation in a tube over the full quality range. Some Figure 2.2: Heat transfer coefficient vs. vapour quality for partial evaporation. (ASHRAE Handbook 1997: Fundamentals) 25 correlations on average heat transfer coefficients for specified refrigerants evaporating in fixed size tubes can be found, for example, Figure 2.2 shows the graphical presentation of the boiling heat transfer coefficient of R22 in a horizontal tube with an outside diameter of 19.6 mm when the quality of refrigerant is varied. However, in general the prediction of heat transfer coefficient is difficult even for a given refrigerant in a certain size tube with specified flow rate (Stoecker, 1998). In the industry, evaporator performance is usually determined by actual testing of the evaporator (Dossat, 1991). It is a common practice for manufacturers and designers of evaporators to provide data of heat transfer coefficients for certain types of evaporators under ranges of operating conditions, and gives the users and application engineers the convenience of selection. Despite the difficulty of predicting accurately the refrigerant boiling heat transfer coefficient, the trend is clearly known. Figure 2.3 shows the typical refrigerant boiling heat transfer coefficient variation along a DX evaporator tube. The changes in the heat transfer coefficient are associated with changes of flow pattern as a function of the vapour quality and flow velocity that change along the tube. Refrigerant enters the evaporator with a small fraction of flashing vapour as a result of the sudden pressure drop across the expansion valve. Along the tube more refrigerant evaporates progressively and the velocity increases. At the entering section, bubbles and plugs of vapour flow along with the liquid; further along the tube, the flow becomes annular with high-velocity vapour rushing Figure 2.3: Heat transfer coefficient as refrigerant boils in an evaporator tube (Stoecker, 1998) 26 through the center with liquid clinging to the inside surface of the tube. Still further the flow converts to a mist of non-equilibrium mixture of liquid droplets and superheated vapour (sometimes called dry-out because the surface is no longer wetted by liquid) till all liquid finally evaporates. For DX evaporator tubes, annular flow occurs in most of the tube length (ASHRAE Handbook 2005: Fundamentals). The distribution of heat transfer coefficients as a function of vapour quality helps to understand why flooded and liquid overfeed evaporators have some heat transfer advantages. Single-phase vapour(gas) convection heat transfer is associated with low heat transfer rates. In a DX tube the heat transfer coefficient tends to drop sharply after dry-out and keeps low in the length of super-heating. In flooded and liquid over-feed evaporators the refrigerant side surface area is always wetted by the liquid refrigerant, which results in a higher surface heat transfer coefficient. 2.3 Two-phase Pressure Drop 2.3.1 An Overview The total pressure drop for two-phase flow in tubes consists of three components: pressure drops due to friction, acceleration (change in momentum), and elevation (gravitation): total fric acce elev dP dP dP dP dz dz dz dz ? ? ? ? ? ? ? ? = + +? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? (2. 40) The acceleration and gravitation pressure gradient components are due to vapour generation (or degeneration) and non-horizontal orientation of the flow channel. Traditionally, the two most important modelling methods of two-phase gas-liquid flow are the ?homogeneous? model and the ?separated flow? model. Both methods provide evaluations of the acceleration and gravitational pressure gradient components on a theoretical basis, while the frictional pressure gradient part is usually evaluated empirically, and is the focus of investigations on two- phase pressure drops. It is convenient and a conventional practice to relate the two-phase frictional pressure gradient to that for the flow of the liquid phase (or gas phase) flowing alone in the channel. Four single-phase pressure gradients are thus defined, for the liquid phase: 27 fric, f dP dz ? ?? ?? ? : liquid fraction assumed to flow alone in the whole channel fric, fo dP dz ? ?? ?? ? : all mass flow assumed as liquid For the gas phase, subscripts of ?g? and ?go? are to be used. With the single- phase pressure gradients used as reference values, two-phase multipliers are defined accordingly: 2 f fric fric,f 2 fo fric fric,fo ( ) ( ) dP dP a dz dz dP dP b dz dz ? ? ? ? ? ? ? = ? ? ? ?? ? ? ? ??? ? ? ? ?? = ? ? ? ?? ? ? ? ?? (2. 41) The single-phase pressure gradients are to be calculated from standard single- phase correlations, for instance, for the liquid phase pressure gradient: ( )2212 f fric,f f 21 2 fo fric,fo f 1 ( ) ( ) G xdP f a dz d GdP f b dz d ? ? ? ?? ? ? = ?? ? ?? ??? ? ?? ? = ?? ?? ? ?? (2. 42) where ff and ffo are Darcy friction factors, obtained from standard single-phase equations or charts with respective Reynolds numbers. For example, when the Blasius equation ( 1/40.3164 / Ref = ) is used, ( ) f f0.25 f f fo fo0.25 fo f 10.3164 , Re ( ) Re 0.3164 , Re ( ) Re G x df a Gdf b ? ? ? ? = =???? = =?? (2. 43) The treatment of the two-phase pressure gradient by a single-phase flow assumption and a two-phase multiplier reduces the analytical task to the determination of the two-phase multiplier ? fo 2 or ? f 2 with independent flow variables, and has proved successful. Many classical works are found in this category, such as the Lockhart-Martinelli (1949) method, that of Chisholm (1973) and Freidel (1979). It shall be noted that the expression of local pressure gradient in terms of ? fo 2 is more convenient for evaporating and condensing problems, 28 where the vapour quality x varies along the tube length and an integration of the local pressure gradients is needed. This is because (dP/dz)fric, fo is a constant, not a function of the vapour quality, which makes the integration far easier (or indeed possible, for some cases). For this reason, correlations that have been designed for such duties purposely utilize the ? fo 2 expression, as seen in correlations of Martinelli-Nelson (1948), and that of Chisholm (1973) which will be later introduced. 2.3.2 The Homogeneous Flow Model The homogeneous model considers the two phases to flow as a single phase possessing mean fluid properties. Those properties basically include the mean density and mean viscosity. Figure 2.4 shows the momentum conservation in a homogeneous model. The model has been in use in various forms in adiabatic two-phase flow and refrigeration systems for a considerable time (Collier and Thome, 1994). This method has also found its application in two-phase pressure drop calculation in PHE channels, though the treatment for some parameters varies. Two important assumptions in a homogeneous model are: 1. equal vapor and liquid velocities: f gu u u= = , 2. thermodynamic equilibrium between the phases. Components of pressure gradients can be derived from the momentum conservation equation: fric elev- - -A dP F F m du? = ?? (2. 44) The individual forces are: Figure 2.4: Conservation of momentum in a homogeneous model 29 c w fric w wet w c h h elev c acce c 4 4 ( ) sin ( ) ( ) AF S dz dz A dz a d d F g A dz b duF mdu G A dz c dz ? ? ? ? ? ? = = = ???? = ???? = = ??? ? (2. 45) The pressure gradients are calculated accordingly: fric w fric c grav elev c acce acce c 4 ( ) sin ( ) ( ) FdP a dz A dz d FdP g b dz A dz FdP duG c dz A dz dz ? ? ? ? ? ? = =? ?? ? ? ? ? = =? ?? ? ? ? ? = =? ?? ? (2. 46) For the frictional pressure gradient, w? is expressed in terms of a two-phase friction factor: ( ) 221w tp tp21 14 8 Gf u f? ? ? = = ? (2. 47) Now the two-phase frictional pressure gradient can be expressed as: 2 w tp fric 4 2 dP Gf dz d d ? ? ? ? ? = =? ?? ? (2. 48) The two-phase friction factor ftp can be calculated by any single-phase friction factor equations with the Reynolds number determined using mean fluid properties or, alternatively, ftp can be determined directly from measured two- phase pressure drops, the latter usually has higher accuracy for the specified testing conditions. The homogeneous model assumes the two-phase flow as single-phase, and there are three properties to be defined, namely the two-phase mean velocity, density and viscosity. The mean density ? is defined by the basic conservation of mass: g f g f 1 1 m m x xQ Q Q? ? ? = = = ?+ + ? ? (2. 49) 30 The mean velocity u is determined accordingly: c m u A? = ? (2. 50) The two-phase mean viscosity can be evaluated in many ways, while the following conditions must be satisfied: f g 0, 1, x x ? ? ? ? = =??? = =?? (2. 51) Three definitions are summarized by Collier and Thome (1994), as given by Equation (2.52). Of those, the most commonly used definition of ? is probably that proposed by McAdams et al, which has the form similar to the definition of ? . Another definition of ? often used in the design of water-tube boilers is f? ?= (Chisholm, 1983), but this definition dose not meet the condition specified by Equation (2.51). Still other definitions of mean two-phase viscosity are possible. Collier and Thome (1994) argued that the failure of establishing an accepted definition is that the dependence of the friction factor on viscosity is small. 1 g f g f g f g f 1McAdams et al., 1942: ( ) Cichitti et al., 1960: (1 ) ( ) Dukler et al., 1964: x 1 ( ) x x a x x b ( x) c ? ? ? ? ? ? ? ?? ? ? ? ?? ? ? = +? ?? ?? ? = + ? ? ? = + ?? ?? ?? ? (2. 52) Once the mean fluid properties defined, pressure gradients of the three components can then be integrated stepwise to obtain the pressure drops, provided a number of simplifications are made. Those include assuming ftp, ?f, ?g as constant and x changes linearly over the channel length. The final form of the integration of the three components can be given as: ( ) ( ) ( ) ( ) f g f g z=L 2 2 i o f fric tp tp f gz=0 z=L 2 acce o i o i g fz=0 z=L of g elev f g o iz=0 i 1 1 ( ) 2 2 2 1 1( ) ( ) 1 1sin sin ln ( ) - 1 1 x xG GP f dz f L a d d P Gdu G u u G x x b xLgP g dz c x x x ? ? ? ? ? ? ? ? ? ? ? ? ?? ? ? ? ? ? ?? ?+? ?? = ? ? = ? ? ? + ?? ?? ?? ? ?? ?? ?? ???? ?? = = ? = ? ??? + ? ?? = = ? ? ? + ?? ? ? ? ???? (2. 53) 31 The homogeneous model provides a simple method for computing acceleration and gravitational components of pressure drop and is still a very common method of evaluating two-phase frictional pressure gradients. Up until the 1940s, this method, with mean viscosity taken as that of liquid, was used exclusively in designing water-tube boilers with good agreements, but later research indicated this was because of the high mass velocity in such devices (Chisholm, 1983). Different possible treatments of some parameters, for example ftp and ? as shown in the foregoing, result in some variances of this method, but in general the homogenous model tends to under-predict the frictional pressure drop. It is also generally agreed that this method gives reasonable results at high reduced pressure Pr and high mass flux (at more than 2000 kg/(m2 ? s)) for tube flow (Chisholm, 1983, Thome, 2004b). This is probably because at such conditions, the flow is in the regime of dispersed (mist) flow, for which the basic assumptions for the homogeneous model are closely met. 2.3.3 The Separated Flow Model The separated flow model considers the phases to be artificially segregated into two streams, one of liquid and one of vapour. The most famous and certainly most widely adopted approach in this type of modelling is the Lockhart- Martinelli (1949) method. Basic assumptions of the separated model are: 1. constant but not necessarily equal velocities for vapour and liquid phases, 2. thermodynamic equilibrium between the phases. Again as for the homogeneous flow model, the frictional pressure gradient can be expressed in terms of the single-phase pressure gradient with a two-phase multiplier: 2 2 fo f fric fric, fo fric, f dP dP dP dz dz dz ? ?? ? ? ? ? ?? = ? ? = ? ?? ? ? ? ? ?? ? ? ? ? ? (2. 54) For the acceleration pressure gradient consider a control volume (CV). The inertial force applied on a CV is equal to the time rate of change of momentum of the CV, i.e., the net momentum flow rate across the CV, which is expressed as: ( )g g f f acce ( ) d m u m uD MVF Ma dz Dt dz + = = = ? arrowrightnosp ? ? (2. 55) 32 Considering acce acce cF = -dP A? , the following is obtained: ( ) ( ) 22 2 acce g f 1 1 xdP d x - = G + dz dz ?? ? ? ? ? ?? ? ? ?? ? ? ? ?? ? ? ? (2. 56) The gravitational pressure gradient is obtained from the gravitational force acting on the CV: ( )elev elev c g g f fsinF dP A g A A dz? ? ?? = ? ? = + (2. 57) and ( )g f elev sin 1dP g dz ? ? ? ? ?? ? ? ?= + ?? ? ? ?? ? (2. 58) In common with the homogeneous model, a stepwise integration of the pressure gradient components requires simplifying assumptions. The result of integration is given by Equation (2.59), assuming ffo, ?f, and ?g being constant and a linear change of x over the channel length L: ( ) ( ) ( ) o i o i o i 21 x 22 fric fo fo x f o i x22 2 acce g f x x elev g f x o i ( ) 1 ( ) 1 sin 1 ( ) G LP f dx a d x x xxP G + b LP g dx c x x ? ? ?? ? ? ? ? ? ? ? ? ?? = ? ?? ??? ? ? ?? ? ??? =? ? ? ?? ? ??? ? ??? = ? + ?? ?? ?? ? ? (2. 59) As is seen the separated model does not supply simple solutions for the acceleration and gravitational components of pressure drop. Additional information is needed for the void fraction ?, which brings extra difficulties for the calculation. From considerations of the flow patterns it is expected that this model would be most valid for the annular flow pattern. 2.3.4 Two-phase Pressure Drop Correlations Numerous correlations on two-phase frictional pressure drop are available in the open literature. It is not in the scope of the present study to cover them all; only some of the commonly used and well-cited correlations are compiled and reviewed. 33 Lockhart-Martinelli (1949) Correlation Lockhart and Martinelli (1949) presented their classic paper on frictional pressure gradients of two-phase, two-component air-liquid flow. In this method it is assumed that the two phases flow separately without interaction, and pressure drops of the two phases are equal regardless of flow patterns. Four two-phase flow patterns were defined; those are combinations of the liquid and gas phases being individually turbulent or laminar. A new parameter X, which was later denoted as the Lockhart-Martinelli parameter (also sometimes called the Martinelli parameter) is defined as: 2 fric, f fric, g dP dPX dz dz ? ? ? ? = ? ? ? ?? ? ? ? (2. 60) For conventional tubes X2 is readily calculated with single-phase pressure drop correlations. For example, for the turbulent-turbulent flow region: ( ) ( ) 0.22 1.81 f f f2 g2 f tt 21 g fg g g2 / 1 / f j d xX xf j d ? ?? ? ?? ? ? ? ?? ? = = ? ?? ? ? ?? ? ? ? ?? ? (2. 61) where ( )f f c/fj m A?= ? and ( )g g c/gj m A?= ? are superficial velocities of the two phases. f is the single-phase Darcy friction factor given by: 0.20.184/ Ref = . It is important to recognize that Equation (2.61) has been quoted widely, sometimes without acknowledging that it was obtained for circular tube flow only. The Martinelli parameter can be expressed in a more general form. This is simply done by defining the Darcy friction factor by f = a / Re n. The Martinelli parameter is then expressed as: ( ) ( ) n2 2-n1 f f f2 g2 f tt 21 g fg g g2 / 1 / f j d xX xf j d ? ?? ? ?? ? ? ? ?? ? = = ? ?? ? ? ?? ? ? ? ?? ? (2. 62) Equation (2.62) is applicable for any type of channel as long as the friction factor can be expressed as f = a / Re n. This is the case for PHE channels. The core of a separated model is to find the two-phase multiplier ? f 2 , which has been defined in Equation (2.54), and is represented by: 34 2 f fric fric, f dP dP dz dz ? ? ? ? ?= ? ? ? ?? ? ? ? (2. 63) The two-phase frictional pressure drop can also be expressed in terms of a two- phase multiplier based on the gaseous phase, ? g 2 , given by: 2 g fric fric, g dP dP dz dz ? ? ? ? ?= ? ? ? ?? ? ? ? (2. 64) The authors successfully correlated ? f 2 and ? g 2 as a unique function of X 2, and the result was presented originally in a graphical form. Chisholm (1967) later conducted a theoretical analysis and developed an equation form of the correlation: 2 f 2 11 C X X ? = + + (2. 65) where the constant C, later known as the Chisholm parameter, has the following values: Flow pattern liquid gas C tt turbulent turbulent 20 vt viscous turbulent 12 tv turbulent viscous 10 vv viscous viscous 5 C = 20 applies for most cases of interest in internal flow in HVAC and refrigeration systems. Citations of the Chisholm parameter C in the literature are numerous; many fail to mention that the initial values are restricted to mixtures with gas-liquid density ratios close to air-water. The Chisholm treatment of the Lockhart-Martinelli correlation greatly eases its application in non-regular flow channels and flow conditions other than those tested in the original paper, as C can be readily modified to fit experimental data. This approach is also found in two-phase pressure drop investigations in PHE channels, as is introduced in Section 2.7.3. The Lockhart-Martinelli correlation was developed for two-phase, two- component adiabatic flow of air and liquids including benzene, kerosene, water and various oils in horizontal pipes with diameters from 1.49 to 25.8 mm and at low pressures close to atmosphere. The correlation has also proved to be quite successful when applied to other fluids and for tubes of larger diameters (Chhabra and Richardson, 1999). The Lockhart-Martinelli correlation is the best known separated flow model, it is very simple to use and has long been the most 35 widely applied correlation for two-phase frictional pressure drop calculations and comparisons. Chisholm (1973) Correlation Chisholm (1973) proposed an extensive empirical method applicable to both local pressure gradients and overall evaporating pressure drop calculations. In this method, the liquid-only two-phase multiplier is given as: ( ) ( ) ( )( )2 n /22 n /22 2 2 n fo 1 1 1Y B x x x? ?? ?? ?= + ? ? ? +? ? (2. 66) where the parameter Y is determined from 2 fric,fo fric,go dP dPY dz dz ? ? ? ? = ? ? ? ?? ? ? ? (2. 67) and n is the exponent of the Reynolds number in the friction factor expression (n = 0.25 for the Blasius equation). B is given by B Y G , kg/(m2? s) 4.8 ? 500 2400 / G 500-1900 0.555 / G ? 9.5 ? 1900 ( )0.5520 / YG ? 600 21/ Y 9.5-28 ? 600 ( )2 0.515,000 / Y G ? 28 The correlation can be extended to evaporating flows. The author argued that where the change in pressure along a tube is sufficiently small in relation to the absolute pressure, Y can be assumed constant, and it is then possible to integrate Equation (2.66), provided the vapour quality x is assumed to vary linearly along the tube length L. The Chisholm correlation is essentially a transformation of some previous correlations including the Lockhart-Martinelli correlation, in forms of ? fo 2 instead of ? f 2 , with modifications by data from several sources. This method has the advantage of more convenient applications for both local and phase-changing flows, and is quite simple to use for engineering calculations. 36 Friedel (1979) Correlation Friedel (1979) developed a two-phase frictional pressure drop correlation using a large data base. For vertical up-flow and horizontal flow, the two-phase multiplier is given as: 2 fo 0.045 0.035 3.23 Fr We F HE? ?= + (2. 68) where ( )2 go2 f g fo 1 f E x x f ? ? = ? + ? ? ( )0.2240.78 1F x x= ? 0.91 0.19 0.7 g gf g f f 1H ? ?? ? ? ? ? ? ? ? ? ? = ?? ? ? ? ? ?? ? ? ? ? ?? ? 2 2Fr G gd? = , Froude number 2 We G d ?? = , Weber number In the above equations, f go and f fo are friction factors calculated by standard single-phase correlations assuming all mass flow as vapour and as liquid, respectively. ? is the mean two-phase density given by Equation (2.49). The Friedel correlation is regarded as one of the most accurate, and is recommended for ?f / ?g <1000, which covers most fluids and operating conditions (Collier and Thome, 1994). A comparison of the correlation with a data bank of 25,000 data points (Hewitt, 1983) showed 40-50 % standard deviation, which is regarded as common for two-phase flow correlations. M?ller-Steinhagen and Heck (1986) Correlation A very simple and purely empirical correlation was proposed by M?ller- Steinhagen and Heck (1986). They noted that the two-phase frictional pressure drop increases with x almost linearly up to a maximum at x ? 0.85 and then falls (this is similar to the local flow boiling heat transfer coefficient, see Figure 2.3). 37 The pressure drop at x=0.5 is nearly always, regardless of mass velocity, identical to that at x=1, which accounts for single-phase gas flow. Based on this information, they proposed: ( )1/3 3 fric 1dP x Bx dz ? ? = ? ? +? ?? ? (2. 69) where ( )2A B A x? = + ? , 2 fo fric,fo f2 dP GA f dz d? ? ? = = ?? ?? ? , 2 go fric,go g2 dP GB f dz d? ? ? = = ?? ?? ? , fof , gof are calculated with standard single-phase correlations. This method is applicable for 0 1x? ? . Equation (2.69) has the advantage of being easily integrated for evaporating liquids, with the assumption of linear change of x over the tube length. A comparison was done by the authors of their correlation with fourteen previous correlations against a data bank of 9300 data points covering a variety of fluids and flow conditions. The correlation performed well with 41.9% relative error, with the best correlation giving 32.6% (a very complicated correlation by Bandel, 1973, as given in the authors? paper). The Lockhart-Martinelli correlation gave a relative error of 62.8%. It was concluded by the authors that prediction of frictional pressure drop for two-phase flow in pipes was not satisfactory with the current knowledge, and average deviations of more than 30% should be expected. Tribbe and M?ller-Steinhagen (2000) and Ould Didi et al. (2002) have found this correlation worked quite well and is at least as good as others for air/water, air/oil, and several refrigerants. Other Correlations Some other well known two-phase frictional pressure drop correlations include those of Dukler et al. (1964), Baroczy (1964) and Gr?nnerud (1972), among others. Assessment of those correlations can be found in various review articles, for example, Ould Didi et al. (2002). 38 A Summary Two-phase flow pressure drop depends on a large number of independent parameters including channel geometry, mass and volume fractions of the individual phases, fluid properties, pressure, mass flux, orientation of the channel (having an influence on the flow patterns), one or two components in the system, adiabatic or with phase-change, etc. Large number of correlations are reported in the literature, to cater for the needs of diverse applications. For improved accuracy, the parameter ranges need to be determined for the problem at hand, and correlations chosen accordingly. Due to the complexity of the problem, mean deviations of as much as 30% are common using these correlations; calculations for individual flow conditions can easily deviate 50% or more from experimental data (ASHRAE 2005: Fundamentals). For conditions outside the range of the original data from which these correlations were derived, deviations of more than 100% may be seen. It is then important to realize that the predictions made using these models must be treated with reservation and used only as estimates. It is seen that a common strategy in both two-phase heat transfer and pressure drop modelling is to begin with a single-phase model and determine an appropriate two-phase multiplier to account for the two-phase effects. This ?multiplier concept? approach goes back at least as far as the classic work of Lockhart and Martinelli (1949), and the majority of two-phase frictional pressure drop correlations are found to be falling into this category. As previously mentioned, of the three components of two-phase pressure gradients, the frictional pressure gradient has been the topic of almost all empirical investigations in the subject. The other two parts are evaluated either by the homogeneous or the separated flow method, with the latter requiring additional information about the void fraction. Investigations were most often carried out on horizontal-adiabatic flow wherein the other two components vanish, but this leaves the effect of flow orientation unattended on application of the correlations obtained. Results of frictional pressure gradient have always been presented with the vapour quality as a primary parameter. To obtain the overall pressure drop in evaporating flow, integration is needed which requires a x-z (vapour quality vs. tube length coordinate) relationship to be determined. Usually a linear change of x over the tube length is assumed which is only strictly satisfied with constant heat flux along the tube but nevertheless has been adopted widely for all other conditions. Examples of this treatment are seen in the correlation of Chisholm (1973) and that of M?ller-Steinhagen and Heck (1986). Integration of complex correlations can be carried out numerically, if an explicit mathematical expression is not obtainable. 39 2.4 Heat Exchanger Theory A heat exchanger is a heat transfer device that exchanges heat between two (or more) process fluids. Refrigerant evaporators are types of heat exchangers where one of the two fluids undergoes phase-changing. 2.4.1 Classification of Heat Exchangers Heat exchangers can be classified in many ways. In terms of surface compactness, there are conventional exchangers and compact exchangers. Conventional exchangers largely refer to shell-and-tube exchangers, compact heat exchangers have many sub-categories but all are characterized by a large heat transfer surface area per unit volume, sometimes termed as surface area density. A gas-liquid exchanger is referred to as a compact heat exchanger if it has a surface area density greater than about 700 m2/m3 or a hydraulic diameter dh < 6 mm. A liquid/two-phase fluid heat exchanger is referred to as a compact heat exchanger if the surface area density on any one fluid side is greater than about 400 m2/m3 (Shah and Sekulic, 2003). Bergles et al. (2003) recommended hydraulic diameter dh of 1 to 6 mm for compact heat exchangers and dh > 6 mm for conventional heat exchangers. A typical process industry shell-and-tube exchanger with plain tubes has a surface area density of less than 100 m2/m3 on one fluid side, and two to three times greater than that with finned tubing. Plate-fin, tube-fin, and rotary regenerators are examples of compact heat exchangers for gas flow on one or both fluid sides. Plate heat exchangers (gasketed, welded, brazed) are examples of compact heat exchangers for liquid flows. A typical plate heat exchanger has about twice the average heat transfer coefficient h on one fluid side or the average overall heat transfer coefficient U compared with a shell-and-tube exchanger for water/water applications (Shah and Sekulic, 2003). 2.4.2 Basics of Thermal Characteristics For any type of exchanger (shell-and-tube, plate, etc), any kind of flow configuration (co-current, counter-current, cross flow, etc), and regardless of type of phase-changing (evaporation or condensation), the basic energy equation is in the form: mQ U A T= ? ? ?? (2. 70) 40 where ?Tm is the mean temperature difference between the two fluids. It can be derived theoretically that for counter-current, co-current, or when temperature of one fluid does not change (the situation approximately met when the fluid undergoes phase-changing, for example, evaporation and condensation): m LMT T? = ? (2. 71) ?TLM is the familiar log mean temperature difference and defined as: max min LM max min ln T TT T T ? ? ?? = ? ? (2. 72) where the temperature differences are defined in Figure 2.5 for counter-current flow, but also apply to other types of flow arrangements. For heat exchangers that do not operate in purely co-current or counter-current flow configuration, a correction factor, F, is multiplied by the value of ?TLM that would be obtained by assuming a counter-current flow configuration, and the mean temperature difference between the two fluids becomes: m LMT T F? = ? ? (2. 73) and the energy equation becomes: LMQ U A T F= ? ? ? ?? (2. 74) Analytical relations for F, and their graphical representations, for the most Figure 2.5: Temperature profile in a counter-current heat exchanger 41 common configurations are available from most heat exchanger design handbooks. For pure co- and counter-current heat exchangers and also for evaporators and condensers, F = 1. For any other kind of heat exchangers, F < 1. For PHE?s with both fluids taking the same number of passes, F is usually in excess of 0.95 (Cooper, 1974 a). It is important to realize that in applying Equations (2.70) and (2.74), the overall heat transfer coefficient U has to be taken as the average value along the heat exchanger surface. U can be calculated by: f, 1 f, 2 wall1 2 1 1 1 U R R h h k ?= ? ?+ + + +? ?? ? (2. 75) where h1 and h2 are individual convective heat transfer coefficients and Rf are fouling resistances. Equation (2.75) assumes equal heat transfer areas on both sides, this is the case for plate heat exchangers. For cases where the heat transfer areas are not the same on the two sides, for example the tubular exchangers, a simple correction is taken. The method of using the log mean temperature difference in heat exchangers analysis and design is usually referred to as the LMTD method. Another widely used method is the ?-Ntu method (not employed in this study), where ? is the heat exchanger effectiveness and Ntu is number of transfer units. Details of the two methods can be found in any heat exchanger design handbooks (for example, Shah and Sekulic, 2003, Kuppan, 2000). It shall suffice to mention here that both methods employ the concept of an average value of U along the heat exchanger channels, and for any calculation duties, both methods can be used to obtain equivalent results, though they are each more suitable for certain situations. 2.5 Plate Heat Exchangers 2.5.1 Types of Plate Heat Exchangers Several types of PHE?s are available both for liquid-liquid duties and in refrigeration systems, these being: 1. conventional gasketed plate-and-frame 2. semi-welded plate and frame 42 3. brazed plate heat exchangers (BPHE?s, also known as compact brazed exchangers, CBE?s). A conventional plate-and-frame heat exchanger consists of a series of thin corrugated metal plates (typically stainless steel, Titanium, or other alloys) between which a number of channels are formed. Channels are sealed with gaskets and corner rings of elastomer materials, most commonly nitrile (Katzel, 2000). The plates and gaskets are held together in a frame by a pressing arrangement. Two fluids flow through alternate inter-plate channels and heat transfer takes place across the plate. The fluids can pass through the heat exchanger either in co-current or counter-current flow, and can be in single-pass or multi-pass arrangement. The basic principle of a single-pass, counter-current flow in a PHE is shown schematically in Figure 2.6. The chevron angle is reversed on adjacent plates so that when plates are assembled together, the corrugations provide numerous contact points. The corrugation on the plates generally perform three functions: they increase the effective surface, they give rigidity and strength, and they promote turbulence (Clark, 1974, Troupe et al., 1960). Normally identical plates are used in one single exchanger pack; however, mixed plates may also be used where there is a demand for balancing heat transfer load and available pressure drop. The design pressure limit of this type of PHE is around 10 bar, and operating temperatures are limited by the availability of suitable gasket materials (Kakac S. and Liu H., 2002). Generally, they are not suited for pressures exceeding 25 bar and temperatures over 250 ?C (Raju and Chand, 1980). The semi-welded PHE is similar to a gasketed plate and frame except the two adjacent plates are welded, usually by laser welding. The welded pair is called a plate cassette, and is sealed at the ports by two ring gaskets (and is thus not gasket-free). Semi-welded channels offer better sealing quality than the frame- Figure 2.6: Basic flow principle of a PHE in single-pass counter- current flow arrangement 43 and-plate type, and are designed for high pressure and for aggressive media applications. For the fluid flowing in the welded cassette, temperature limits range from -50 up to 350 ?C and operating pressures can be from full vacuum to 40 bar (Reppich, 1999). Brazed PHE?s (BPHE?s) have similar plate geometry to the former two types, but gaskets are completely removed and the whole pack is non-opening. In such a unit, plates are brazed together with copper or nickel alloy and the exchanger is completely sealed and leak-tight. Without the limitations brought by the gasket, which has long been a weak point in the design and applications of PHE?s, brazed PHE?s have typical design temperatures of -160 to 190 ?C, and design pressure up to 30 bar (Goldfinch, 1994, Fijas, 1997). This manufacturing solution has greatly increased the exchangers? mechanical resistance, but also eliminated the flexibility and ease of cleaning features available on gasketed PHE?s. Nevertheless, the ability to withstand higher pressures and the elimination of sealing problems made the brazed PHE?s interesting and applicable for the refrigeration industry, to be used both as evaporators and condensers. 2.5.2 Plate Geometry Technically, plate corrugations can be of many types, for example, chevron (herringbone), washboard (lateral), wavy-groove, zig-zag. However, over the years, the chevron wave pattern has proved to be the most successful and popular design, and is offered in rather similar shapes by the majority of manufacturers (Carlson, 1992, Martin, 1996). The heat exchangers investigated in this study are of the chevron type, consequently the introduction of this particular type of corrugation pattern is the task of this section. Chevron Corrugations Geometrical features of the chevron type of corrugation are given in Figure 2.7. A single plate comprises four corner ports and the corrugated area. The corrugated pattern has a chevron angle ?, evaluated as the angle between the corrugation trough to the vertical axis, or in some reports to the horizontal axis. In a PHE unit plates are installed with the apex of the chevron pointing in opposite directions. The chevron design brings four main effects, it 1. increases flow turbulence level, 2. increases the effective heat-transfer area (typically by a factor of 1.1-1.25), 3. increases stiffness of the plate pack, 4. induces turbulence and high wall shear force which reduces fouling. 44 Figure 2.7: Corrugation features of the chevron type plate The chevron angle of commercially available plates varies between the extremes of about 25 to 65?, and is perhaps the most important geometrical parameter of PHE?s relating to thermal and hydraulic performance (Kakac and Liu, 2002). Conventional plates have approximately sinusoidal profiles. Corrugations that are asymmetrical in profile are not common and found to be less efficient (Focke, 1985). The surface enlargement factor, ?, is another important parameter, which is defined as the ratio of actual heat transfer area to the projected area. Most commercial plates have enlargement factors usually in a relatively small range of 1.1-1.25 (Kumar, 1984). ? can be calculated approximately, for a sinusoidal corrugation profile, from a three-point integration formula (Martin, 1996): 2 21 (1 1 4 1 ) 6 2 XX? ? + + + + (2. 76) where X = ?b/? is a dimensionless parameter, b is two times the amplitude of the sinusoidal wave (b = mean flow channel gap) and ? is the wave pitch. For a strict sinusoidal wave, ? =1.22. The channel aspect ratio is defined as ra = Lp/w. Common industrial plates usually have this ratio around 2, and it is not likely to be much lower than 1.8 (Cooper and Usher, 1983). The heat transfer area of a single plate is calculated by: effA L w ?= ? ? (2. 77) 45 where effL is the effective channel length and w the channel width. Flow Channel Geometry A flow channel is formed by two adjacent plates, sealed with gaskets, welding or brazing at the edges of plates and ports. The plates are usually identical but positioned with the chevron angle reversed, and the geometry of the flow channel is identical on both sides of the plates. Connecting points are formed at the crests of both plates where they meet. The channel flow area is calculated by: chA w b= ? (2. 78) Also commonly seen are units with channels which are formed by plates of different chevron angles, this is referred to as the mixed-angle arrangement. From the manufacturer?s side, the purpose of using mixed plate channels is to achieve the required thermal duty while also using the available pressure drops, and at the same time avoid expenses on pressing tools for a large number of plate types (Raju and Chand, 1980). This solution makes the design work highly flexible, and has long gained popularity in applications as least as early as in the 1970?s (Clark, 1974). One of the three units that has been tested in the present study has the mixed angle arrangement, and the other two use identical plates with high and low chevron angles, respectively. These arrangements are shown schematically in Figure 2.8. Focke (1983) argued that for practical purposes two different plate chevron angles suffice: ?=45? at which the ratio of heat transfer/ pressure drop is at its Figure 2.8: Channel of combined plates 46 highest, and ? = 72? which results in the maximum heat transfer coefficient value. However, manufacturers usually opt for angles approximately 30? and 60?, presumably to cover low heat transfer rate duties. 2.5.3 Flow Arrangements The flow arrangement in a PHE can be very flexible. There are basically four types of configurations that can be used, as will be shown below, which are related to flow distribution inside or between channels. 1) Inside Single Channel In a single channel, the flow of one fluid can be either diagonal or parallel, depending on the sealing arrangement on the ports. A diagonal arrangement refers to the flow that enters and leaves the channel at diagonally opposite ports, whereas in the parallel (or vertical, same-side) arrangement the flow enters and leaves on the same side (Kumar et al., 1994). As an example, the gasket set as shown in Figure 2.7 would give a parallel flow if a channel is formed by two such plates. 2) Pass Pass refers to a group of channels in which the flow is in the same direction. The two streams in a PHE unit can have different pass arrangements, single or multiple, with the latter consisting of passes connected in series. Figure 2.9-a shows a 2 pass / 1 pass configuration where one fluid flows in a single pass manner and the other in two. 3) U and Z Type Arrangement in Single-pass Flows The so-called U and Z type arrangements are found in single-pass flows. For any one fluid stream if the inlet and outlet ports are on the same side of the exchanger unit, it is called a U type arrangement, otherwise it is a Z type, as shown Figure 2.9: Flow arrangement of multi-channel PHE?s 47 schematically in Figure 2.9-b. The U type permits easy cleaning or repair on the exchanger without disturbing external piping, while the Z type offers more uniform flow distribution (Kakac and Liu, 2002). As is clear in the illustration, the Z type arrangement assures that liquid streams passing through different channels will cover the same flow distances, while in a U type arrangement channels further away from the entrance have longer manifold passages and may thus cause greater pressure drops. 4) Relative Flows between Two Fluids The two fluids can be organized in either co-current or counter-current flow. For single-phase applications, counter-current arrangement is a more favorable choice and gives higher exchanger effectiveness than the co-current flow. This can be shown by a theoretical Ntu? ? (Effectiveness - Number of transfer units) analysis on pure co- and counter- current flows, which is very well represented by PHE channels. 2.5.4 Hydraulic diameter, Reynolds Number, and Friction Factor Equivalent vs. Hydraulic Diameter The corrugated channel found in PHE?s is a unique and complex channel; definition of the characteristic diameter of this type of channel has not been agreed on in the research community. Two definitions are found in the open literature, the first one is often called the "equivalent diameter" which is twice the corrugation depth: e 2d b= (2. 79) A second and possibly more common definition is often named as the ?hydraulic diameter?, defined in the traditional way for hydraulic diameter of non-circular tubes (four times the volume divided by flow channel wetted surface area) as: p h wet p 44 2 2 L wbV bd S L w ? ?= = = (2. 80) where ? is the surface enlargement factor. The hydraulic diameter method of definition seems more physically sound since the surface enlargement factor does change the actual channel cross-section area for a given value of corrugation depth. Furthermore, introducing the enlargement factor to the hydraulic diameter might be a simple and effective way of taking 48 into account the influence of ? on the PHE performance. In the present study, the hydraulic diameter definition is used uniquely as the characteristic diameter for the corrugated channels. For clarity, these two definitions will be distinguished in this study by their subscripts, e for equivalent and h for hydraulic. It is very important to note that when correlations are quoted the definitions used in the original texts should be mentioned, to avoid confusion or misunderstanding. Reynolds Number and Other Non-dimensional Groups Non-dimensional groups are defined with characteristic lengths and velocities. For the chevron type of corrugated channel the characteristic diameter can be either de or dh, as mentioned earlier, but the characteristic velocity is always the nominal flow velocity, or bulk velocity, defined by: ch Q u A = (2. 81) It shall be noted that the nominal flow velocity obtained is not the true velocity of the fluid, due to the effect of the corrugation. Depending on the particular corrugation pattern, true velocities may be higher by a factor of up to about four (Marriott, 1971). Now, the Reynolds number is defined: h Re u d? ? = (2. 82) which can also be expressed as hRe /G d ?= ? , where ch ch/ ( )G m A N= ? is the mass velocity in the vertical direction. With the characteristic length and velocity defined, other non-dimensional groups are now determined accordingly, for example, the Nusselt number is given by: h Nu h d k = (2. 83) Friction Factor Definition of the friction factor can be sometimes confusing, more so when friction factors are mixed freely with different definitions of characteristic diameters and channel lengths, as is the case in the open literature. The two most commonly used definitions of friction factors are the well-known Darcy and the Fanning types, denoted in this study as f and cf, respectively, and defined as: 49 Table 2.2: Definitions and conversions of friction factors Friction factor: Characteristic diameter h 2 /d b ?= e 2d b= 21 2 / / P Lf u d? ? = hf ef f 21 2 1 / 4 / P L c u d? ? = ? f, hc f, ec Conversion: h e /f f ?= h f, h4 f c= e f, e4 f c= 21 2 / / P Lf u d? ? = (2. 84) f 2 / 2 / P L c u d? ? = (2. 85) In the above definitions, d is de or dh, whichever is chosen by the investigator. u is the nominal velocity, given by Equation (2.81). L stands for the characteristic length of the flow channel. To make the issue more complicated, there are also two different definitions of L to be found in the literature. In most situations, fortunately, L has been used as the length between upper and lower port centres, denoted as Lp (see Figure 2.7), while in others L can be chosen as the developed flow length defined as the ratio of the actual heat transfer area to the flow width, given by L = Ap / w. No reasonable explanations are available for this definition. It is of interest to note that the developed flow length still does not represent the actual fluid flow distance in the channel, since flows are highly three dimensional and directed by corrugation angles, and not along routes normal to the corrugated waves. In this study, the friction factor used is the Darcy type definition f, based on the hydraulic diameter dh and the projected channel length Lp between the upper and lower ports. The two friction factors have the simple relationship of f = 4cf in any case, regardless the geometry of flow channel. Care must be taken to correctly interpret correlations from different authors, as the free choice of two types of characteristic diameters combined with two types of friction factors results in four possible ways to express the friction factor, and all these four are found in the open literature. For clarity, definitions and conversions between the four 50 expressions are given in Table 2.2. 2.6 Single-phase Flow in PHE?s Over the years, investigations into the single-phase thermo-hydraulic performance of PHE?s have become rather extensive. Technical reports on their performance assessment, construction development, and application recommendations are found as early as in the 1960?s (for example, Troupe et al, 1960, Dummett, 1964). This is partially due to the PHE?s widening applications from the food industry to the chemical industry. There are currently well over 30 correlations available which could be considered as practicable for single-phase duties, mainly for water-water, some for non-Newtonian fluids in laminar flows. PHE?s are generally considered not suited for gaseous fluid duties because of high pressure drops, and not for highly viscous fluids due to problems in flow distribution (Kandlikar and Shah, 1989, Raju and Chand, 1980). 2.6.1 Overall Thermal-hydraulic Performance PHE?s are usually much more thermally efficient than their shell-and-tube counterparts, particularly for liquid/liquid duties. Film coefficients can be two to four times those for tubular units of the same duty, at the same or even lower pressure drops (Bond, 1981). At normal working ranges the overall heat transfer coefficient U can be expected to be 2300-5800 W/( m2 ? K), depending on plate corrugation and flow conditions (Raju and Chand, 1980). The highest U value that could be achieved by a PHE was reported up to 8500 W/(m2 ? K), making it capable of working with film coefficients three to five times higher than tubular or spiral-plate designs (Carlson, 1992). The augmented heat transfer performance of a PHE is due to several enhancement mechanisms, which directly result from the complex plate surface characteristics. These surface effects include disruption and reattachment of boundary layers, swirling motion of the fluids, continuous change in flow directions and velocity, combining to promote early transition to turbulence and produce exceptionally high film coefficients of heat transfer. The analogy between heat and momentum transfer would indicate that as the heat transfer is improved there is a higher pressure drop penalty as well. In fact, it is widely noticed that increase of the friction factor is far quicker than that of the Nusselt number with increasing chevron angles. Thonon et al. (1995) reported that at the same Reynolds numbers, when compared with smooth channels the 51 enhancement of heat transfer is up to 6 times but at the same time the friction factor can be increased by a factor of one hundred. However, in practice for the same heat transfer area or at the same cooling/heating duties, pressure drops in a PHE are usually lower (Troupe et al., 1960), due to the fact that the flow velocities in a PHE unit are usually very low, and the flow channel length is always much shorter than that in tube-type exchangers. Of all the dimensional parameters that describe a PHE channel, the chevron angle ? is probably the most important. Generally both heat transfer and pressure drop increase with higher chevron angles. Mixed plate channels are commonly evaluated with the averaged chevron angle (Bond, 1981, Heavner et al., 1993), though a rigorous examination of this treatment is not found in published works. The surface enlargement factor ? might have similar effects, as deeper corrugations increase the effective surface area as well as promote greater swirl and mixing, resulting in higher heat transfer rate and pressure drop. A comparison of experimental data from several sources by Mulley and Monglik (1999) has confirmed this. When the hydraulic diameter dh =2b/? is used, the effect of ? can be regarded as included in that of the hydraulic diameter; in other cases, this effect is usually ignored since an accurate quantitative evaluation of ? is difficult to obtain. Another parameter not usually attended to is the channel aspect ratio. While its effect on the exchanger thermal-hydraulic performance is apparent, this parameter has hardly become a consideration in any available correlations. High aspect ratio (a ?narrow? channel) at the same heat transfer area will increase the flow velocity for given heat transfer area and thus increase pressure drops; low aspect ratio will bring problems of flow distribution inside individual channels and so will reduce the exchanger efficiency. It is also noticed that in the open literature geometrical parameters are rarely all given in detail. It is widely recognized in the research community that turbulence is attained in PHE?s at much lower Reynolds numbers than in circular tubes, however, the values of the critical Reynolds number, Recrit, are reported differently from various sources from 10 to 500 (Raju and Chand, 1980, Carlson, 1992, Reppich, 1999, etc.). This may partially be due to the particular plate configurations tested. Cooper and Usher (1983) pointed out that it is difficult to predict the flow pattern in a particular PHE unit without testing it. However, it seems quite safe to conclude, from those reported values, that all types of PHE will be in the turbulent flow region at Re > 500, and in the laminar region at Re < 10. In most cases the transient region is in the Re range of 10-150. The early transition of flow patterns in PHE?s might be explained by two main reasons: 1. Corrugation features. Corrugations break down the stagnant insulating film at the plate surface and trigger turbulence (Raju and Chand, 1980). It is 52 obvious that a higher chevron angle and deeper corrugation will promote higher levels of disturbance and earlier transition. 2. Actual flow velocity. In the corrugated channel, the actual flow would most likely follow the corrugations rather than flow in the vertical direction (Focke et al, 1985). As a result, the actual flow velocity is much higher than the mean value in the vertical direction, which is used to calculate the Reynolds number. PHE's could be treated as pure co-current or counter-current heat exchangers in principle if end effects are neglected. For a shell-and-tube exchanger this can hardly be the case, due to cross flows resulting from baffles. For a two-channel PHE, pure counter-current flow may be assumed; for multiple channel units, a correction factor of the LMTD (Log Mean Temperature Difference) is sometimes recommended. Buonopane et al. (1963) initially addressed the issue and reported that for lateral-corrugated plates, with packs in a 1 pass/ 1 pass counter-current arrangement, an average correction factor of 0.95 should be applied to the LMTD. Marriott (1971) later presented a chart for approximate values of the correction factor as functions of the Ntu (Number of transfer units) for various pass arrangements, which confirmed Buonopane et al?s correction factor value. Kandlikar and Shah (1989) also tabulated correction factor values based on a more refined numerical analysis. Usher (1970) pointed out that corrections are needed when the flow departs from the two-channel equal flow condition, to account for flow ratio, number of passes and end effects of passes, and these can only be determined empirically. Pressure Drop Components Pressure drop in a PHE consists of three contributions: (1) frictional pressure drop within the core (plate passages), (2) pressure drop due to elevation change, (3) pressure drop in inlet and outlet manifolds (ports). Of these, the frictional pressure drop is of main interest and has been addressed in various empirical correlations in the open literature. Elevation pressure drop is calculated in a straightforward way by P=?gh. For the manifold pressure drop, an accurate evaluation method is not available. A very widely cited equation by Shah and Focke (1988) gives an estimation of the manifold pressure drop, without a reference, as: 2 manifolds pass inlet 1.5 2 GP N ? ? ? ? = ? ?? ?? ? (2.86) The manifold pressure drop is usually considered much smaller than the other components. Generally the manifold pressure drop is lower than 10% of the 53 overall pressure drop, but can be as high as 30% or higher in certain designs (Shah and Sekulic, 2003). Bond (1981) pointed out that different port designs, producing variation in manifold roughness, could result in the overall pressure drop to be two to three times higher than the passage pressure drop. 2.6.2 Single-phase Heat Transfer and Pressure Drop Correlations A considerable amount of research has been conducted to determine heat transfer and flow friction characteristics for the PHE geometry. There are well over 30 correlations available in the open literature, most of which are for a specific plate geometry. Those correlations are generally developed for turbulent water-water duties, with a few for laminar flows of Newtonian and non-Newtonian fluids. Most of them are in forms of the familiar Sieder-Tate type, with modified constants and exponents. Some later correlations express coefficients or exponents of the Reynolds number as functions of the chevron angle and the enlargement factor to cover wider applications. The exponent of Re in most of the water-water correlations is close to 2/3, for all chevron angles. This value was also confirmed by Stasiek et al. (1996) in their tests on turbulent air flow (Re = 400-3500) through air pre-heaters which have very similar geometries to those of PHE channels. It seems clear that correlation forms which have been successfully applied in tube flow are also applicable to PHE?s, while a more general one would require considerations of geometrical effects, especially the chevron angle. Edwards et al. (1974) of the APV Company 1 tested various non-Newtonian fluids including glucose solutions, lubricating oil and poly-acrylamide solutions in one PHE unit with a chevron angle of 60? and very high aspect ratio (?8). It was found that laminar flow conditions exist up to a Reynolds number of 10, and a very gradual transition to turbulent flow follows, in contrast to pipe flows where the onset of turbulent flow is rapid. The authors had, interestingly, noticed the similarity of channel geometry between a PHE and that of a packed bed of near-spherical particles, in that flow channels are both featured with series of expansions and constrictions. Certain correlations for packed beds were then compared with pressure drop and heat transfer data obtained from the PHE unit and very good agreement was obtained. The authors pointed out that the same measure of agreement would not be expected for plate exchangers having markedly different corrugation patterns. While the analogy of these two types of exchangers remains interesting, verifications of this approach are not found in other works. 1 APV Company Ltd is an UK based major PHE manufacturer. 54 Based on extensive test data, Bond (1981) of the APV Company reported PHE heat transfer and pressure drop correlations in graphic form, for ? = 25?- 60?, ? = 1.14 fixed, and Re = 0.1-10,000. This correlation is reproduced in Figure 2.10, with its basic form expressed in Equation (2.87). The heat transfer relationship in the Bond correlation is of typical Sieder-Tate type, with modified constants accounting for the chevron angle. For each chevron angle the flow is divided into three regions. It is seen that the higher the chevron angle the lower the critical Reynolds number. For ?=60?, laminar flow occurs below Re=10. Pressure drops determined from the plots refer to passage pressure drops excluding passage entry and exit losses. The author stated that diverse designs of ports, producing variation in port manifold roughness, could result in the overall pressure drop two to three times the passage pressure drop. Kumar (1984) reported that the accuracy of the Bond correlation is about ?20% except in laminar flow regions, for which greater inaccuracies may occur on account of influence of fluid viscosity. Bond?s correlation is one of the earliest with a wide applicability, and remains the only one that is based on manufacturer data covering more than one plate model. This correlation has also become a favorite comparison basis for many later investigations. For the convenience of computation, constants of the correlation were derived in this study from the original graph, and listed in Table 2.3. Figure 2.10: Bond (1981) correlation for chevron-type PHE?s ? (= 90 ??ringoperator ) is the angle between corrugation trough to horizontal direction, f refers to the Fanning friction factor, 2 /d b ?= 55 m Nu 11/6 1/3 wall 2 f p Nu Re Pr Re j c c c ? ? ? = =? ? ??? ? ?? ? ??? =?? (2.87) 1 2 Re Re 25 20 0.543 0.33 50 25 1 20 500 0.323 0.50 50 500 3.13 0.47 500 0.090 0.71 500 0.62 0.21 30 20 0.543 0.33 20 400 0.300 0.53 400 0.104 0.71 c m c p? ? ? ? ? ? ? ? ? ? ? 40 25 1 40 400 3.60 0.47 400 0.742 0.21 40 20 0.618 0.34 20 33 1 20 300 0.284 0.60 20 300 11.1 0.63 400 0.133 0.73 300 0.78 0.16 45 20 0.725 0.35 15 48 1 10 100 0.420 0.59 15 3 100 0.290 0.67 ? ? ? ? ? ? ? ? ? ? ? ? ? ? 00 18.6 0.65 300 1.30 0.19 60 10 0.725 0.35 10 50 1 10 100 0.375 0.64 10 100 19.0 0.58 100 0.309 0.68 100 3.1 0.19 ? ? ? ? ? ? ? ? h 2 /d b ?= , 1.14? = Table 2.3: Constants in the Bond (1981) correlation (Equation. 2.87) Table 2.4: Constants in the Heavner et al. (1993) correlation (Equation. 2.87) avg 1 2 45 / 90 67.5 0.278 0.683 1.715 0.0838 23 / 90 56.5 0.308 0.667 1.645 0.1353 45 / 45 45 0.195 0.692 0.810 0.1405 23 / 45 34 0.118 0.720 0.649 0.1555 23 / 23 23 0.089 0.718 0.571 0.1814 c m c p? ? Re 400 10,000= ? , h 2 /d b ?= 56 Heavner et al. (1993) of the APV Company later published test results on five units with different chevron-angle combinations: 23?/23?, 23?/45?, 45?/45?, 23?/90? and 45?/90?. Each unit had 12 channels, other geometrical information was not provided due to commercial considerations. All tests were in turbulent regions with Reynolds number in the range of 400-10000. A correlation was given in the same form as the Bond correlation (Equation (2.87)), with new constants given by Table 2.4. In this correlation, the chevron angle of a mixed- plate channel is evaluated by the average of the two plates. The authors compared their data with the Bond (1981) correlation curves, though chevron angles covered in the two investigations were not identical, and rough agreements were obtained. It is noticed that the Reynolds number index for the heat transfer coefficient is around the value of 0.7 but dose not show much relevance to the chevron angle, which is consistent with the Bond correlation for turbulent flows. For the friction factor it is noticed that the Bond curve has a Re exponent of 0.2, remaining constant for all chevron angle plates, while Heavner et al?s data have shown this exponent to decrease with increasing chevron angle. This relationship can be explained, as tried by the authors, by considering the corrugated channel as a rough tube, whose roughness increases with increasing chevron angle. At ? = 0? the channel is a collection of smooth vertical tubes which would have a Re exponent of 0.25, as suggested by the Blasius equation (cf = 0.79/ Re1/4), on the other hand, at ? = 90? the exponent reaches 0, as for very rough tubes the friction factor is no longer a function of the Reynolds number (which is also indicated by the familiar Moody chart). Wanniarachchi et al. (1995) developed a set of more generalized correlations based on the curve correlation of Bond (1981) and data reported by Heavner et al (1993). The correlations are in a third-order asymptotic form, which is given by Equation (2.88). The authors explained that the choice of this form is due to the fact that there was no suitable information to define the critical Reynolds numbers. Mathematically, an asymptotic equation always approaches the larger of the two components and assures a smooth transition. Edwards et al. (1974)?s observation of very gradual transition from laminar to turbulent flow in PHE channels, as opposite to a rapid one in conventional pipes, might have at least partially justified this choice of correlation form. ( ) ( ) 1/33 3 Nu Nu, l Nu, t 1/33 3 f f, l f, t j j j c c c ? = +??? = +? (2.88) where Nu, l 0.455 0.339 3.65 Re j ? = , 0.646 0.00111 Nu, t 1.142 12.6 Rej ? ? + ? = 57 f, l 1.026 1774 Re c ? = , 3 5 2f, t 1.08 4.23 10 2.23 10 46.6 Re c ? ?? ? ?? ? ? ? ? = , 90? ?= ?ringoperator Valid for 23 62, Re=1-10000? = ? Crozier et al (1964) suggested from investigation on Non-Newtonian flows that, for laminar flow, the L?v?que approximation, initially developed for heat transfer calculations of laminar flow in short pipes, can be used with good results, whereas a clear description of the method was not given. The idea was later developed independently and more successfully by Martin (1996) into a semi- theoretical correlation for turbulent heat transfer and pressure drop in PHE?s. The L?v?que approximation, shown by Equation (2.89), is a purely analytical equation for hydrodynamically developed and thermally developing laminar flows in circular pipes of short lengths. This equation can be found in many heat transfer handbooks (for instance, Bejan and Kraus, 2003, Baehr and Stephan, 2006). Martin proposed a generalized form of this equation, as given by Equation (2.90), based on which his heat transfer correlation for PHE channels was developed. For a circular tube, with f ? Re = 64, the generalized L?v?que equation (Equation (2.90)) reduces to the conventional one (Equation 2.89). The author argued that though the L?v?que equation had not been previously applied to turbulent flow, there is no reason to restrict its application to the laminar range. A further analysis on this equation with its applications for the prediction of heat and mass transfer rates from pressure drop can be found in a later paper by the same author (Martin, 2002). 1/3 Nu 1.615 Re Pr d L ? ? = ? ?? ?? ? (2.89) 1/3 2 1/3 Nu 0.404 Re Pr df L ? ? = ? ?? ?? ? (2.90) To use Equation (2.90), the friction factor has to be determined first. Martin carried out an analysis on the frictional pressure drop based on the observation (Focke et al., 1985) that there are two kinds of flows coexisting in a corrugated channel: crossing flow following the furrows of adjacent plates and longitudinal flow between two vertical rows of contact points. To obtain an overall friction factor, effects of the two are combined tentatively using a reciprocal square root type of equation, with each part containing constants to be determined by experimental data. The expression for the heat transfer group takes the form of Equation (2.90), with the characteristic length being the distance between two crossings, i.e., L = ? /sin(2?), and the leading constant and the exponent for f ? Re2 58 further refined by experimental data. The final expression of the Martin correlation is: ( ) ( ) 0 1,0 2 q Nu q 1 cos 1 cos tan sin / cos [ Re sin(2 )] af b c f a f j c f b ? ? ? ? ? ? ?? = +? ? + ? + ??? = ?? (2.91) where ( ) ( ) ( ) ( )q 0 1,0 2 0.289 10 h , , 3.8, 0.18, 0.36 , 0.122, 0.374 64 597Re 2000 3.85 Re 2000Re Re , 1 39Re 2000 Re 2000(1.8log Re 1.5) Re for Re 400 10,000, 0 80 , 2 / a b c c q f f d d b? ? = = ? ?< + 1000. The conventional Sieder-Tate form (Equation (2.87)) was again used, with the slight difference of the viscosity ratio exponent being 0.14 instead of 0.17. In the effort to obtain a general correlation that covers all the three units, the lead coefficients and Re exponents were described as functions of the chevron angle ?. The effect of the enlargement factor ? was included by using a third-order polynomial correlated with more data from other sources. The final correlation shows: ( ) m Nu 11.141/3 wall 2 f p Nu Re Pr / Re j c c c ? ? ? = =???? =?? (2.92) where ( ) ( ) ( ) ( ) 5 2 2 3 1 2 2 3 2 0.2668 0.006967 7.244 10 20.78 50.94 41.16 10.51 0.728 0.0543sin( / 45 3.7) 2.917 0.1277 0.002016 5.474 19.02 18.93 5.341 0.2 0.0577sin( / 45 2.1) c m c p ? ? ? ? ? pi? ? ? ? ? ? pi? ? = ? + ? ? ? + ? = + + = ? + ? ? + ? = + + ? in degrees, / 45 3.7pi? + and / 45 2.1pi? + are in radians. For Re 1000, 30 60, 1 1.5? ?? = ? = ? , e 2d d b= = The agreements for Nusselt number and friction factor with the experimental data were claimed as 10% and 5%, respectively. However, the accuracy of this correlation needs verification by more tests, since it was based on only three different PHE?s. Moreover, the treatment of the effect of the enlargement factor in Muley and Manglik?s correlation can be arguable. The third order polynomial expressions accounting for ?, in both the Nusselt number and the friction factor equations, were based on a very limited number of data points (less than 10) with relatively big scatters. A simple calculation would indicate that in the range of ?=1.1-1.25, which is typical for industrial plates (Kumar, 1984), the correlation gives an increment of 52% for Nu and 143% for f. Even though reports on the effect of ? have not been quantitatively conclusive in the literature, those figures are likely to be exaggerated. It can be shown that when the enlargement factor is smaller than its tested value, i.e., 1.29, the Muley and Manglik correlation tends to give continuously lower values of Nu and f than those predicted by most other correlations. Focke and his coworkers conducted a series of experimental studies on PHE?s including flow visualizations (Focke and Knibbe, 1986), optimum PHE surface 60 patterns selection (Focke, 1986) and performance assessment of asymmetrically corrugated PHE plates (Focke, 1985). Based on mass transfer and pressure drop data, Focke et al. (1985) reported a correlation for heat transfer and friction factor. The heat transfer rate is evaluated not by direct measurement data but by heat- mass transfer analogy theory, with the Colburn j-factor which states: 1/2 1/2Sh/ (Re Sc ) Nu/ (Re Pr )j = ? = ? 1. The testing unit consisted of a single channel machined in sinusoidal corrugation shape with an enlargement factor of 1.464, which is considerably higher than most commercially available industrial plates. Care should be taken when using this correlation as the heat transfer data are based on the projected and not the developed transfer surface area. This is a rather unusual treatment, but explanations were not given in the paper. The correlation is found to give exceptionally high Nu when compared with others, while the friction factor is fairly similar. Some other correlations are available, though less cited and usually lack generality. A total of 28 correlations were summarized by Ayub (2003) in his review article 2, which brings a good and brief introduction. It is recommended here, however, that for the details and applicability of a particular correlation, reference should be made to the original source, as the obscure terminology and mixed use of various parameters often causes not only confusion but also erroneous quotations. It might be worthwhile, for the sake of a straightforward perception, to compare the introduced correlations against each other. Figure 2.11 shows the calculation curves of Nu and f against the Reynolds number in the range of Re=100-10000, at ? = 45? and ? = 1.14, from six correlations as introduced in this section. Valid ranges of those correlations are given in Table 2.5. The chevron angle of 45? is covered by all correlations while the enlargement factor of 1.14 is not, this value is chosen for a fairest possible comparison basis for all. Disagreements between those correlations are clear, being relatively greater at low Reynolds numbers. The Nu predicted by Focke et al?s correlation is significantly different, which could possibly be explained by their heat-mass transfer analogy treatment of experimental data, also with the tested channel?s large enlargement factor. Muley and Manglik?s correlation gives a Nu number lower than all the others, possibly because of an exaggerated effect of the enlargement factor. 1 The familiar Chilton-Colburn analogy states, as seen in many heat transfer textbooks: 1/3 1/3 Nu Sh Re Pr Re Sc j = = ? ? It is noticed that Focke et al. (1985) has chosen the exponent of the Pr number as 1/2, instead of 1/3, no explanation of this treatment was given. 2 Reproduction of those correlations in Ayub?s paper was not without mistakes. Besides typographical errors, one apparent cause has been, for certain correlations, the author?s incorrect interpretation of the various characteristic length and friction factors used by different authors. The use of particular correlations is thus recommended to refer to the original papers. 61 Figure 2.11: A comparison of six correlations Calculation based on ? =45?, ? =1.14, at Pr =6.13 (water at 1bar, 25?C) f is Darcy friction factor, characteristic diameter is 2 /d b ?= Table 2.5: Some single-phase PHE correlation with their validity ranges Correlation Re ? , degrees ? ch PassN ? PL w? (mm) Description Bond, 1981 0.1-10,000 25, 30, 40, 45, 60 1.14 ? ? Focke et al., 1983 90- 56,000 0, 30, 45, 60, 72, 80, 90 1.464 Single channel 440?100 Heat transfer data based on mass transfer measurement Heavner et al., 1993 400- 10,000 23/23, 23/45, 45/45, 23/90, 45/90 ? 12?1/12?1 ? PHE typical of modern designs Wanniarachchi et al., 1995 1-10,000 23-67.5 ? ? ? Based on Bond and Heavner et al. data Martin, 1996 All range 0-80 All range ? ? Heavily based on Heavner et al data Muley and Manglik, 1999 >1,000 30/30, 30/60, 60/60 1-1.5 12?1/12?1 392?163 ? =1.29 for the tested units 62 A Summary In contrast with the established knowledge and standards for heat transfer and pressure drop in straight pipes and conventional heat exchangers, a generally accepted theory, or correlation, for PHE design and performance assessment has not yet become available. This is certainly attributed to the highly complex nature of the PHE channel, in terms of its corrugation-featured geometry, also in terms of the almost infinite combinations of many other parameters including the aspect ratio, channel depth, port size, flow distribution, plus many types of optional flow arrangements. Plate corrugation is not an industry standard, there are over 60 different corrugation patterns from different manufacturers (Cooper and Usher, 1983). Although it is true that the chevron-type corrugation is dominant, geometries of those plates from different manufacturers are not identical, a typical example is the types of patented corrugation features around the entrance region aiming at more uniform flow distributions. It is believed that the only accurate correlations are those produced by manufacturers, based on many years of research and extensive tests of specified plates and arrangements but those are usually kept confidential (Cooper, 1974a, Reppich, 1999). Such correlations, once available, are still to be regarded as case studies with restricted applications for specified conditions. With all this in consideration, the task of achieving a general solution seems hopeless. On the other hand, however, a generally applicable correlation could have to focus only on important and common features while omitting minor and irregular ones, at the expense of lower accuracy. From the most widely accepted correlations, including all that have been reviewed in this section, it is seen that chevron angle is (of course) a priority, enlargement factor is an option, other geometrical parameters and the effect of any flow arrangements are usually not considered. Of those correlations, there is currently no suitable information one can rely on to judge which one is more accurate than others, and it is usually difficult to explain the disagreements between them, as those disagreements may have resulted from, among others, geometry differences, incorrect treatment in the data reduction, and measurement errors. It is not unusual to find calculations using some of those correlations to differ by more than ?30% or even over 100% especially at lower Reynolds numbers. As such, correlations shall be chosen according to the plate geometry and used as rough estimates only. For rigorous assessments and for situations where high accuracy is required, it is always preferable to conduct individual performance tests. 63 2.6.3 Flow Distribution The concept of flow distribution, in the case of plate heat exchangers, refers to that between channels and thus applies only to multi-channel units. Although the flow distribution inside an individual channel can be another important technical issue, it is rarely addressed in the open literature. Thermo-hydraulic theories and correlations for PHE?s largely work on the assumption of uniform flow distribution between channels, this is however not the case in practice. It is generally accepted that flow maldistribution can increase the overall pressure drop and decrease the thermal performance. There is, however, no generally accepted quantitative determination method accounting for this effect. Flow distribution is determined by inlet and exit manifold pressure profiles, for which two factors can have influences: frictional force and momentum change. Frictional force always causes pressure drops, the momentum change may however have two different effects: in the inlet manifold the deceleration of fluid, due to outflow (into channels), results in actually a pressure rise, whereas in the exit manifold the fluid confluence gives pressure drop in the flow direction. The net effect of the two mechanisms depends on many factors, most importantly channel pressure drop characteristics which determines the flow rate of the fluid leaving the manifold and entering the channels, fluid flow rate, and port size which determines the frictional losses in the manifolds. Most industrial PHE units are arranged in single-pass U or Z type flows; effects of friction loss and momentum change on the pressure profile in these two arrangements are schematically illustrated in Figure 2.12. A vertical line between the two pressure profile lines represents the pressure drop. Even distribution of pressure drops means even distribution of flow, since all channels are identical. Figure 2.12: Manifold pressure and channel flow distribution 64 As introduced earlier (Section 2.5.3), the Z-type arrangement is normally expected to give more uniform flow distributions than the U-type (Kakac and Liu, 2002). This was confirmed by Wilkinson (1974) in his test on units of six channels with pressure measurement on each individual channel. The result also indicated that in the U-type arrangement the liquid stream always ?favors? channels closer to the inlet nozzle, more so at higher volume flow rates. The author further pointed out that this holds true also for large units. Bassiouny and Martin (1984 a, 1984 b) carried out a theoretical analysis on the flow distribution and pressure drop in PHE?s, based on the assumption that the friction loss in the manifolds is negligible. The direct result from this assumption is that pressure always rises in the inlet manifold along the flow direction, and consequently the Z-type arrangement always suffers more severe maldistribution than the U-type. This is because, with such an assumption, pressure profiles in the Z-type arrangement always tend to diverge, while in a U-type they tend to remain relatively parallel (see Figure 2.12). While this assumption remains itself highly arguable, Rao and Das (2004) concluded using this same model that maldistribution is more severe in the Z-type compared with the U type. They conducted tests of overall pressure drop with the two arrangements, but no obvious difference of the overall pressure drop at the two conditions could be seen. 2.6.4 Fouling Predicting the fouling resistance accurately is essential for any rigorous performance analysis or design of heat exchangers. For refrigeration applications as in the current study, the refrigerant flows in a closed loop and is usually assumed as clean, information of fouling is needed for the water side. In practice, scale or deposits on heat exchanger surfaces cause a decrease in thermal performance during operation due to reduced overall heat transfer coefficient. The effect of fouling is evaluated by a fouling factor, Rf, which is defined as the thermal resistance across the fouling substance. Fouling is found to be considerably less in PHE?s than in other types of exchangers. The main reason for this is the high degrees of turbulence which maintain solids in suspension and removes foulant (Marriott, 1971). Rf can be generally expected to be about 10- 20% of that in a tubular exchanger, from reported data available in the open literature (Cooper, 1974a, Marriott, 1971, Stromblad, 1989, Panchal and Rabas, 1999). It is evident but worth stressing that an over-estimated value of the fouling factor will have a much greater effect on the overall coefficient of a high- efficiency exchanger such as a PHE than on that of an exchanger having a low overall thermal efficiency. In the current study typical values of fouling factors 65 for plate heat exchangers as quoted from two sources, along with those of the tubular exchangers for the convenience of comparison, are given in Table 2.6. The fouling factors of tubular exchangers are quoted from the TEMA (Tubular Exchanger Manufacturers Association) standard as provided by Panchal and Rabas (1999). There are currently no standard values of fouling factors available for PHE?s, in contrast to those for tubular exchangers. Considering the great variety of corrugation patterns, it is reasonable to assume a geometry-related fouling factor applicable to a specific exchanger. Under the circumstance where this information is lacking, values provided in Table 2.6 can be used as a general and best estimation. 2.6.5 Numerical Simulations Numerical simulation, or in other terms, computational fluid dynamics (CFD), has been the subject of recent interest for many heat exchanger analysts, due to the vast computational resources available and development of numerical solution techniques particularly in the last thirty years (Nithiarasu, 2005). Numerical simulation of any heat transfer problem essentially involves the solution of the Navier-Stokes equations with the continuity and energy equations, by commercial or specially developed software. Simplifications are necessary, of mathematical expressions and geometrical conditions of the problem. Table 2.6: Recommended fouling resistances of PHE?s vs. TEMA values Process fluid fR - PHE ?10-3 m2K/W fR - TEMA ?10-3 m2K/W Marriott, 1971 Panchal and Rabas, 1999 Panchal and Rabas, 1999 Water soft 0.018 0.018 0.18-0.35 hard 0.043 - - Cooling tower water 0.034 0.044 0.18-0.35 Sea water 0.026 0.026 0.18-0.35 River water 0.043 0.044 0.35-0.53 Lubricant oil 0.017-0.043 0.053 0.36 Organic solvents 0.009-0.026 0.018-0.053 0.36 Steam 0.009 0.009 0.18 Process fluids, general 0.009-0.052 - - 66 There are a number of CFD simulations on the flow pattern or heat transfer in plate heat exchangers, e.g., Trifonov (1998), Mehrabian and Poulter (2000), Kanaris et al. (2004), Pelletier et al. (2005). The majority of those concern heat transfer and pressure drop in a single unitary cell or a simplified single channel with only one corrugated plate, due to the highly complex channel geometry. No numerical simulation of a complete PHE channel has been found. CFD modeling has the advantage, among others, of being able to show local behaviour and interactions, such as of temperature and velocity vector profiles, usually in graphical interfaces. On the other hand, however, those simulated results are difficult to validate, especially for flows in confined and limited spaces such as a PHE. Validations and engineering applications of the CFD simulation method in plate heat exchangers have so far been limited, though it is known that some manufacturers have used this method in their product development process for years. 2.7 Two-phase Flow in PHE?s A brief history review shows that PHE?s were not normally considered as suitable for refrigerant evaporators and condensers before the 1990?s. Concerns included leakage from gasket materials, pressure limits (especially for condensers), freezing risk, and high pressure drops of generated, or in the case of a condenser, entering, vapour in both the manifolds and plate passages. Early applications of PHE?s in two-phase flow duties commonly involved heating process fluids using condensing steam (Cooper, 1974b, Kumar, 1983) with usually moderate pressure on the steam side, one typical example of this kind is pasteurizing milk. PHE?s were also used as evaporators for liquid food concentration and sea water desalination (Kumar, 1993). Even though some of the major manufacturers had started development programs on PHE?s as refrigerant evaporators and condensers since the middle of 1970?s, as initiated by the first energy crisis (Str?mblad, 1989), applications of this type of exchanger in refrigeration systems only became popular after the invention of semi-welded PHE?s and brazed PHE?s in the early 1990?s. Meanwhile, the ozone-depletion issue in the mid-1980s also became a factor promoting the use of PHE?s in refrigeration systems because their compactness would result in a low refrigerant charge (Ayub, 2003). In the last two decades, with their inherent advantage of high efficiency and compactness, it is believed that PHE?s have enjoyed a rapidly increasing market in the refrigeration industry with applications including liquid chillers, indirect systems, residential heat pumps, and absorption chillers (Mencke et al, 2005). 67 Investigations on two-phase flow in PHE?s are increasing but not are yet extensive. Due to the complex nature of two-phase flow, fundamental understandings of the flow boiling mechanism in this unique type of channel (actually, in any type of channel) is rather limited. In contrast to the thermal- hydraulic process in single-phase flow, flow boiling is controlled by some additional parameters, most importantly heat flux, vapour quality and local flow regimes. Of the few empirical correlations available in the open literature, coverage of those important parameters is usually limited and unsystematic and the operating conditions vary. As a result, there is not a generally accepted calculation method for two-phase heat transfer and pressure drop in PHE channels. Some manufacturer-published information can help, but this is either too general or restricted to a few aspects of certain products (Ayub, 2003). 2.7.1 Adiabatic Two-phase Flow Characteristics Investigations on adiabatic flows have mainly focused on pressure drop characteristics and sometimes on flow visualization. The air-water system is most often used. It has been a common practice to model a liquid with its vapour by water and air for two-phase flows in conventional pipes. This strategy normally helps to ease the experimental arrangements. It is widely agreed that PHE?s are not suitable for gas applications due to high pressure drops. At the same time, the topic is rarely addressed in any depth in the open literature. With the lack of information on gaseous fluid pressure drop in PHE channels, one question may rise: could the friction factor correlations obtained from liquid (water) flow be applicable to gaseous fluids (air)? This question becomes important when two-phase pressure drop calculation is concerned, in adiabatic or phase-changing flows. Most, if not all, empirical methods employing the separated flow concept require pressure drop determination of both the liquid and gaseous phases when they flowing alone in the channel. One would naturally assume that the same correlation should apply for both phases. This is the case for conventional tubular channels, and has been adopted by many investigators for PHE channels without questioning. Experimental results from a few reports may have supported this assumption. In a test of air flow in corrugated channels by Gaiser and Kottke (1989), the friction factors at Re = 2000 were compared with Focke et al.(1985)?s data using water. Good agreement was obtained with deviations smaller than 20% for all six chevron angles in the range of ? = 20?-77.5?. Laboratory machined plates were used in both investigations where ? = 1.57 and 1.46, respectively. In another test by Stasiek et al.(1996) on air flow in an industrial power plant regenerator, which 68 has corrugated passages very similar to those of a PHE, the friction factor at Re = 500 - 4000 and fixed ? =30? would give close agreement to that of Focke et al., with deviations smaller than 10% for most data points. Reports on air flow pressure drops in industrial PHE units were not found. On the other hand, in an effort of calculating PHE condenser pressure drop, Kumar (1983) obtained different correlations for liquid and vapour phase: 0.2 f 0.3 g / Re liquid / (Re ) vapour f A f B ? ? =?? = ??? (2.93) where A and B are geometry-dependent constants, unique for each plate. Kumar?s expression for liquid friction factor was in agreement with the correlation suggested by Bond (1981). More specific investigations on gas pressure drop in PHE?s were not found. Nevertheless, information available may have indicated that a single correlation could be applied to both phases without serious error, and most investigators have used this treatment. Some fundamental issues need to be considered in adiabatic flow pressure drop: 1. the dependence of two-phase pressure drop on mass flux G and mixture quality x, 2. applicability of correlations from conventional channels for PHE?s, 3. influence of geometrical parameters, particularly chevron angle. Creissig and M?ller-Steinhagen (1992) measured port-to-port pressure drops of air/water flow in an Alfa-Laval commercial PHE with ? = 60?, in the range of Re=4000-64000. Considerable pressure drops were observed at such high flow Reynolds numbers, single phase air pressure drop was 90 kPa/m at Re = 4000 and 300 kPa/m at Re = 8000, for instance. It was found that the pressure drop increases almost linearly with the mixture quality x, which differs from tubular flow where maximum pressure drops are normally observed at x = 0.8 - 0.9. Four correlations previously established for conventional channels were tried to predict the data, three of which are based on separated flow concepts including the Lockhart-Martinelli (1949) correlation and that of M?ller-Steinhagen and Heck (1986). All of those tend to over-predict the pressure drop data but the Dukler (1964) correlation, which is based on the homogeneous model, gave reasonable agreement, being better at high mixture qualities. The authors argued, reasonably, that the complicated flow geometry in a PHE would keep the two phases well mixed, therefore heterogeneous flow patterns which are often seen in tubular flow (most commonly annular flow), are unlikely to occur. Winkelmann et al. (1999) experimentally investigated air-water down-flow in a corrugated channel with ? = 27?. At two mass fluxes of 50 and 100 kg/(m2 ? s) (air and water combined), it was observed that the frictional pressure drop 69 increases almost linearly with the mixture quality x. This observation agrees with that of Creissig and M?ller-Steinhagen (1992). The authors correlated their data by the Lockhart-Martinelli method using the original Chisholm type of equation (see Equation (2.65)), an average value of C = 6 was obtained for all data points covering x = 0.05 to 0.9. The reported accuracy of the calculation is within 25% of the measured values of two-phase pressure drop. Tribbe and M?ller-Steinhagen (2001a, 2001b) investigated air-water flow in typical industrial PHE?s in greater depth. Measurements of pressure drop in the corrugated section, using thin needles interested through the gasket, were taken on three chevron angle combinations of 30?/30?, 30?/60? and 60?/60?, under constant mass flux conditions in the range of 50 to 600 kg/(m2 ? s). It was found that regardless of channel geometry, the measured total pressure drop tended toward a linear function of mixture quality x except at very low qualities. They found that, upon applying the Lockhart-Martinelli method, the Chisholm parameter C varies considerably as a function of the Martinelli parameter X, and they recommended not to use a unique value for C 1. Claesson and Simanic (2003) experimentally investigated pressure drop of R134a adiabatic flow inside a BPHE channel. Plate geometry information was not given, but it was reported that in the range of x = 0.1-0.7 and G = 18-48 kg/(m2 ? s), the frictional pressure drop is almost linear with the quality x. The pressure drops were high, for example, 100 kPa at G = 48 kg/(m2 ? s) and x = 0.3. A general correlation was not obtained due to the small number of data points, but efforts were made to correlate the two-phase multiplier f? to the Martinelli parameter X and a power law type of ? f ?X relationship was suggested. A constant C approach of the Chisholm method would not succeed because of large variations. Vlasogiannis et al. (2002) investigated air/water mixture down-flow in an Alfa- Laval P-01 PHE with 6 channels and a chevron angle of 60?. Four flow regimes were identified based on visual observations through a Plexiglas plate, embossed with the same corrugation and placed as the last plate of the exchanger pack. At water superficial velocity jf lower than 0.025 m/s, liquid mainly moves at the bottom of corrugation furrows leaving a continuous gas phase flowing in the major part of the channel space (regime A). At jf > 0.1 m/s and low jg, dispersed flow was observed where small bubbles flow inside a liquid continuous stream (regime B). At other superficial velocities of both phases the flow was interpreted 1 Upon using the Lockhart-Martinelli approach, the f X? ? relationship could be given in graphical form (as in the original paper, and Kumar(1984) for PHE condensation), by the Chisholm C constant, or in other forms. The authors proposed a new empirical correlation for the f X? ? relationship. Though they suggested not to use the C constant method, as an unique value of C would result in large errors, a close examination shows that when a C value of 4.00 is used the maximum deviation from their suggested correlation is -23.7%, not so large a value for two-phase pressure drop predictions. 70 as combinations of the two. In another flow visualization investigation by Shiomi et al. (2004), only dispersed flow was observed in the range of jf =0.01-0.56 and jg, = 0.14-1.2 m/s, on three PHE?s of 30?/30?, 30?/60? and 60?/60? chevron angles. Flow visualization investigations had often used transparent materials as a cover plate, and it is noticed that pictures are always vague and blurry because of metal reflection, a problem not suffered by investigations on pipe flows. This affects flow pattern identifications which are already heavily associated with subjective judgments. 2.7.2 Flow Boiling Heat Transfer in PHE?s Flow boiling heat transfer is a very complex process and the physics of this process is still not well understood. Practical applications require quantitative prediction of the flow boiling coefficient. Since it is not possible at present to solve boiling heat transfer problems by analytical or numerical solutions of the conservation laws, the coefficient is always obtained through experimental studies and expressed by empirical correlations. The flow boiling heat transfer coefficient is a complex function of transfer surface finish, fluid thermo-physical properties but most importantly two-phase flow parameters, including primarily mass flux, heat flux, vapour quality, and system pressure. Conventional treatments involve the combination of two contributions: nucleate boiling and forced convection (for more details refer to Section 2.2.3). For circular pipes, these two are usually treated as additive or in other forms. For PHE channels, while the same two components are considered to present, their contributions are not clearly known. It is also generally agreed that two-phase heat transfer and pressure drop are flow pattern dependent, but most conventional models have ignored this effect. Flow pattern related investigations on two-phase flow in PHE channels are very scarce, and there is no heat transfer associated modelling currently available. Overall Performance and Basic Features Panchal et al.(1983) investigated the overall performance of two industrial PHE?s used as refrigerant evaporators. A series of tests were conducted with ammonia and R22 at various exit vapour qualities, chevron angles, heat flux, and mass flow rates. In general, ammonia showed better performance than R22 with higher overall heat transfer coefficients and lower pressure drops at same heat flux and exit qualities. The two major conclusions were: 71 1. nucleate boiling is a major mode of heat transfer, but forced-convection will be more important for the upper region. Results showed that the overall heat transfer coefficient is not substantially influenced by the refrigerant flow rate and relatively insensitive to the chevron angle, both indicating a nucleate boiling dominant process. 2. the controlling heat transfer resistance is on the refrigerant side. As a result, increasing the water side flow rate (and thus hw) did not significantly increase the overall heat transfer coefficient. A test was conducted on one exchanger unit before and after a nucleate boiling promoting surface was coated on the heat transfer surface; the overall heat transfer coefficient was increased by a factor of 2 or more at the same heat fluxes. Since only overall performance was tested, water and refrigerant side heat transfer coefficients were not individually determined, and the relative contribution of nucleate boiling and forced convection could not be determined because local measurements were not made. Nevertheless, the results are important in academic value as well as for engineering practice. It might be inferred from Panchal et al?s observations that PHE evaporators will not benefit from high angle plates, which only marginally increase the heat transfer rates but significantly increase water-side pressure drop. In a later report Panchal and Rabas (1993) confirmed that nucleate boiling is the dominant evaporation mechanism up to the upper part of a PHE evaporator. As such, a nucleate boiling promoting coating on the refrigerant side can be applied but is only necessary to the lower half of the plates. More data were reported on individual film coefficients of both fluid sides of a 60? chevron angle unit in this latter report, and it was shown that the porous surface coating increased the ammonia side heat transfer coefficient from 5700 to 29000 W/(m2 ? K), with the water side coefficient remains as high as 15000 W/(m2 ? K) and resulted in an overall heat transfer coefficient increase from 3500 to 8100 W/(m2 ? K). Engelhorn and Reinhart (1990) conducted a series of tests on a R22 direct expansion (DX) PHE evaporator with and without an inlet flow distributor. The influence of vapour exit conditions was also investigated, the overall heat transfer values U for superheated refrigerant at the exit were approximately 25% lower than those of the saturated state. Evaporator performance was considerably improved after the installation of a flow distributor. Experimental results showed that the refrigerant heat transfer coefficient h r had a strong dependence on heat flux q, and no significant influence of mass flow rate or evaporating pressure was observed. Data points of h r were explained quite well by methods for nucleate pool boiling. Kumar (1993) reported two basic conclusions for evaporation of water where the heat source is condensing steam: 1. overall heat transfer coefficient U is invariant 72 with water mass flow rate G, and 2. U is significantly lower at low evaporating pressures P. Boccardi et al. (2000) tested three refrigerants (R134a, R407C and R410A) and it was found that R410A had the best overall thermal performance under the same working conditions, while the other two had similar performances. More recently Thonon (2008) reviewed experimental work on propane evaporation in BPHE?s, it was concluded that the refrigerant side heat transfer coefficient for hydrocarbons is usually lower than for R22 due to their specific physical properties. For example, the higher latent heat of propane implies a lower flow rate. Propane heat transfer coefficients of 1700-2750 W/(m2 ? K) were reported in normal working ranges of heat flux of q = 6-10 kW/m2. Dominant Flow Boiling Process in PHE: Forced Convection or Nucleate Boiling? Flow boiling heat transfer in pipe flow is established to be the combination of two basic mechanisms: forced convection and nucleate boiling, which well represent the term ?flow boiling?. Conventionally, the nucleate boiling contribution is often calculated from pool boiling correlations, sometimes with a suppression factor introduced, while the forced convection contribution is calculated from single-phase correlations. For compact heat exchangers including PHE?s, it is generally admitted that the two basic mechanisms occur during flow boiling, but no general predictive method is available (Thonon et al., 1997). There are four most important working parameters to be considered in flow boiling in any channel geometries: evaporating pressure P, heat flux q, local vapour quality x, and the mass velocity G. Pool boiling heat transfer is a strong function of P and q, plus surface roughness Rp, which is a fixed feature for a certain channel. In general, nucleate boiling heat transfer coefficients increase with increasing P, q, and Rp. On the other hand, single-phase forced convection heat transfer coefficients are mainly a function of flow velocity, which is directly relevant to the mass velocity G and local vapour quality x; P and q have only marginal effects via fluid thermal-physical properties. The influence of those parameters on the overall heat transfer performance indicates the contribution of relative heat transfer mechanisms. Thonon et al. (1997) concluded from experiments on R22 and R134a vaporization in PHE?s that the heat transfer coefficients could be either in a nucleate or a convective boiling regime. The transition between these two basic mechanisms depends on flow characteristics and also on channel geometry. It was concluded that for PHE?s running as DX evaporators, the dominant boiling mechanism is forced convection. The authors suggested a transition criterion, Bo ? X tt: 73 3 tt 3 tt Bo 0.15 10 Nucleate boiling dominant Bo 0.15 10 Forced convection dominant X X ? ? ? > ? ? < ? (2.94) where tt f g/X P P= ? ? In the above equation Bo ? X tt is a product of the boiling number and the Martinelli parameter, this criterion takes into account the operating parameters but not the geometrical effects. Ayub (2003) pointed out that in PHE flooded evaporators it is possible that the majority of the heat transfer is through forced convection rather than local nucleate boiling, due to the narrow complex passages, but nucleate boiling probably plays its role in the lower sections. Stromblad (1989) argued that high shear and homogeneous two-phase flow characteristics in corrugated channels result in rapid suppression of nucleate boiling, and as a consequence, in the entire channels the boiling process takes place by two-phase convective heat transfer. While both statements are technically reasonable, later experimental observations of various authors suggested otherwise. Panchal et al. (1983) observed that in an ammonia liquid-overfeed system the overall heat transfer coefficient is not changed significantly by increasing the ammonia flow rate. Engelhorn and Reinhart (1990) found that data points in a R22 evaporator showed strong dependence on q. Kumar (1993)?s test on water flow boiling showed that U is invariant with water flow rate G, Panchal et al. (1983) and Uehara et al. (1997) both obtained very high overall heat transfer coefficients applying nucleate boiling-promoting material on the refrigerant side. All those observations seem to suggest a nucleate boiling dominant process in PHE evaporators. Claesson and Palm (1999) experimentally determined the evaporating and superheating regions of a brazed PHE R22 DX evaporator using thermochromic liquid crystal (TLC, a material which changes colour with temperature) techniques. In the range of q = 6-15 kW/m2 with superheating of 1-10?C the evaporating region heat transfer coefficient was only a function of the boiling heat flux and not the mass flux, which led to the authors? conclusion that the boiling process in this type of heat exchanger is mainly in the nucleation boiling region. From the foregoing, it seems clear that nucleate boiling will be a dominant boiling process in a liquid-overfeed PHE where a liquid-deficient region is not likely to occur. For DX systems two-phase forced-convection evaporation may become important for the upper part. There are two factors which are unique in a PHE evaporator and need some consideration. Firstly the flow velocity in a PHE evaporator is usually significantly lower than that in pipe flow at comparable mass flow rates (cooling loads), due to the wide flow area, and this effect favours local nucleate boiling. Secondly, most, if not all, PHE evaporators are arranged in 74 single-pass configuration for the refrigerant side and the flow length is very short. Consequently, when dry-out occurs, even for a short length of the channel, it will occupy a considerable portion of the channel and bring a severe effect on the overall heat transfer coefficient. The latter feature was seen in a number of investigations. Panchal et al. (1983) observed that at in a R22 evaporator, at comparable heat flux and both water and refrigerant side mass velocities, the overall heat transfer coefficient dropped 44% when the exit quality changed from 0.63 to 0.98. Claesson and Palm?s test (1999) showed that when the exit superheat changed from 1?C to 10?C, the evaporating area dropped sharply from 90% of the total surface to 20%. The same authors (Claesson and Palm, 2002) in a later paper pointed out that when there are two regions present ? boiling and superheating regions, those two need to be treated individually in predicting the evaporator performance. Sterner and Sunden (2006) and Jokar et al.(2006) divided the PHE DX evaporator into two regions in their experimental analysis. The basic method of determining the individual areas of these two regions, which was also outlined by all groups of authors, is given in Appendix M. However, Claesson and Palm (2002) had pointed out that this method tends to underestimate the area of the superheating region when compared with measured areas using the TLC (Thermochromic Liquid Crystals) technique. Other accurate calculation methods for determining the areas of these two regions have not been available. Sterner and Sunden (2006) argued that for small channels of dh = 2.5-3 mm the channel dimension plays a critical role in determining heat transfer mechanisms, and conflicting data interpretations and conclusions could be drawn depending on whether the channel size is greater or less than some critical space dimension. This is a reasonable statement, but may not be the main cause of the issue. Most industrial plates are in a rather small range of corrugation depth and hydraulic diameters dh are usually well beyond 3 mm; in cases where laboratory machined plates are used, those are often of similar or scaled-up corrugations. Plam and Claesson (2006) in another paper noticed the complexity of flow boiling and disagreement of different authors, they stated that the difference in opinion seems not to be explained by differences in the geometric parameters or in the flow conditions. They further reported that for DX evaporators there is a dominating influence of heat flux and only a minor influence of mass flux and vapour fraction, but this does not necessarily mean that nucleation is an active process under all conditions; thin film evaporation might be an important heat transfer mechanism which is also a heat flux dominating process. It is evident that local flow structure has the strongest influence on the local boiling mechanism. It is also possible that the effects of heat flux and mass flux on the heat transfer coefficient only become significant when those parameters are greater than certain magnitudes. The choice of the mass flux in a PHE evaporator is flexible even at comparable heat fluxes using the same refrigerant 75 because of the free choice of the number of channels. For DX evaporators there is a direct connection between heat flux and mass flux, which makes it very difficult to distinguish between the two contributions. The question of which mechanism is dominating in a PHE evaporator has not gained an agreed answer and thus remains open. Results from various investigators have led to different, and sometimes conflicting conclusions. There is no in-depth and systematic investigation devoted to this topic in the open literature. All correlations available are based on statistical analysis without regard to the physics of the boiling process that occurs in this type of channel. Heat Transfer Enhancement in PHE Evaporators It is a well known fact that liquid over-feed, or flooded, evaporators are more thermally effective than direct expansion (DX) evaporators. This also applies to other types of evaporators. One of the reasons for this improvement is certainly due to the fact, as is also known for pipe flow, that dry-out occurs at the upper section of the transfer surface of a DX evaporator. The situation is more severe for a high performance evaporator, bearing in mind that in normal one-pass working conditions the flow channel is already short. One other reason, not usually addressed, might be attributed to the inlet flow distribution. A DX evaporator has liquid-vapour mixtures at the inlet, due to liquid flashing at the thermal expansion device. Hewitt et al. (1993) conducted an excellent analysis on the liquid/vapour interface in the manifold. They demonstrated that from the inlet port along the flow direction, vapour would rise first and then the liquid. As a result, a certain number of channels would contain vapour and the remainder pure liquid, and this effectively leads to the ?under-utilizing? of the heat exchanger. This analysis gives a reasonable perspective of the two-phase maldistribution at the PHE evaporator inlet, while its effect on the overall heat transfer rate depends on two things: number of plates and inlet vapour quality. In a test by Kedzierski (1995) on a DX evaporator of 36 plates with R22 as the working fluid, the effect of flow maldistribution was investigated by feeding both saturated and subcooled entering refrigerants; in the latter case the amount of subcooling was set to offset pressure drops that would occur and cause flashing, however, no significant difference in the heat transfer performance of the evaporator was observed. Some authors (Engelhorn and Reinhart, 1990, Sterner and Sunden, 2006) have reported heat transfer improvement by the installation of an inlet refrigerant distributor1, all concerning DX systems. This is justifiably so according to the 1 A distributor basically consists of a tube with small bores, one per channel. But there are various patented designs. Installation of a distributor is a convenient option usually offered by manufacturers. 76 analysis above, and it could possibly be said that a distributor would not be as effective for a liquid-overfeed evaporator since this type of evaporator does not suffer from liquid/vapour maldistribution in the first place. Stromblad (1989) reported that the largest liquid-overfeed evaporators have about 300 refrigerant channels (600 plates for the exchanger pack), while for DX systems the number of channels should not be excessive (maximum numbers not given). Promotion of heat transfer on the refrigerant side has proved effective in increasing the PHE evaporator overall heat transfer rate. This is because PHE evaporators usually have the major thermal resistance on the refrigerant side. Panchal et al. (1983) obtained an increase of the overall heat transfer coefficient by a factor of 2 after application on the refrigerant side (ammonia) of a porous surface coating specially designed for promoting nucleate boiling. Uehara et al. (1997) tested a R22 shell-and-plate evaporator (plate geometries very similar to that of a PHE) where the refrigerant side was coated with aluminium powder, very high overall heat transfer coefficients of 3200-4800 W/(m2 ? K) were observed. Longo et al. (2004) obtained improved R22 heat transfer coefficient by 30% - 40% using a ?cross-grooved plate surface?. It is established that the water side heat transfer coefficient h w can be several times greater than that of the refrigerant side, h r, as Gray (1984) stated: PHE evaporators usually confer very high heat transfer coefficients on the liquid side and good coefficients on the boiling side. According to various sources (Marriott, 1971, Raju and Chand, 1980, Panchal et al., 1983, Stromblad M., 1989, Engelhorn and Reinhart, 1990), at normal conditions, values of h w = 3000 - 12000 W/(m2 ? K) for water, and h r = 1000 - 3000 W/(m2 ? K) for R22 and 2000 - 5000 W/(m2 ? K) for ammonia can be expected. 2.7.3 Two-phase Pressure Drop Two-phase pressure drop in a PHE unit has many components. Under common conditions pressure measurements are taken at the inlet and exit of the unit, and the total pressure drop (measured value) includes four parts: pressure drop due to the channel friction ?Pfric, pressure drop at the ports and manifolds ?Pport, pressure drop due to momentum change (acceleration) ?Pacce and pressure drop due to gravity (elevation) ?Pelev. Of those the frictional pressure drop is by far the major part and almost the sole interest of all investigations. Evaluation methods of those components are reviewed in this section. 77 Frictional Pressure Drop For the calculation of two-phase frictional pressure drop in PHE channels, many authors have chosen the Lockhart-Martinelli method. Others have used the homogeneous model treatment. The Lockhart-Martinelli method has been reviewed in detail in Section 2.3.3; this method requires determination of the ? f- X, i.e., two-phase multiplier and Martinelli parameter, relationship. This relationship can be given in many ways, for example, it can be in graphic forms (as in the original paper), or by the Chisholm method with a C constant (Equation (2.65)), or other equation forms as found suitable for the experimental results. The Chisholm method is widely employed; however, different values of the Chisholm constant C have been reported with large variations. Cooper (1974b) of the APV company might be the first in the literature who proposed the Lockhart-Martinelli method in PHE two-phase flow applications. The author reported that for steam condensation the use of the Lockhart- Martinelli formulation gave the best fit of the data among a number of correlations tried; ? g was used instead of ? f and the ? g ?X relationship was obtained by curve fits but not given in the paper. Kumar (1983) of the same company later reported more details in applying this method for steam condensation pressure drop calculations. A figure form of the ? f ?X and also ? g ? X relationship was presented, which yields non-fixed values of the C parameter (C ? 5.4 at X = 1) if the Chisholm method is to be used. It is important to note that Kumar had used different correlations for the liquid and vapour phase pressure drops in his method. The Martinelli parameter X is determined from its original definition, as given by Equation (2.60), which gives different values than if determined by the widely recognized equation for circular tubes, Equation (2.61). Thonon et al.(1995) reported that for two-phase pressure drop in corrugated channels the original Chisholm equation can be used with C=8 to obtain good estimation of the two-phase flow multiplier ? f. The Martinelli parameter X was however evaluated by the circular pipe flow Equation (2.61) and no information on the channel geometry was given. Winkelmann et al. (1999) later reported C = 6 for air-water adiabatic flow in a single corrugated channel with chevron angle of 27?. Asano et al.(2004) reported C = 2.73 for air-water adiabatic flow in a single channel with chevron angle of 60?. Tribbe and Muller-Steinhagen (2001) observed that in their air-water adiabatic flow experiments the data obtained could not be represented by the Chisholm equation with a unique value of C, and a form of ? f ?X correlation was suggested. Claesson (2005b) used the Chisholm method for calculation of refrigerant pressure drop in DX evaporators, the Chisholm parameter was fitted from the experimental data in the form C = f (Refo, X). Palm and Claesson (2006) later reported that a single value of C=4.67 78 correlated the data reasonably well. Sterner and Sunden (2006) applied the Chisholm parameter for their ammonia DX evaporators, C was found as a function of the liquid-only Reynolds number, i.e., C = f (Refo). It is seen that many investigators have favoured using the Lockhart-Martinelli approach in calculating two-phase frictional pressure drop in PHE channels, especially for adiabatic two-phase flows. However the ? f ?X relationship, or value of the Chisholm parameter C where used, could not be agreed upon. The disagreement is certainly relevant to the variety of channel geometries tested. Another possible reason, among others, could be due to the difficulty of accurately predicting the single-phase gaseous pressure gradient. Confirmation is still needed whether correlations obtained from single-phase water are applicable to refrigerant liquid and vapours. It is important to note that where the Lockhart- Martinelli method is used for phase-changing calculations, such as evaporation and condensation, the channel length must be split into several zones; in each zone the individual liquid and vapour component pressure drops are to be calculated using pertinent correlations. This is because the method is for the adiabatic flow condition only. With modern computation facilities available, a stepwise integration can be carried out numerically by dividing the flow length into a great number of subsections, say 100 or more. Other methods have also been proposed. In fact most proposed frictional factor correlations from experiments on PHE evaporators have been based on homogeneous theory, with the friction factor being a function of local mean vapour quality and liquid-only Reynolds number. This is seen in a list of empirical correlations available, as given in Table 2.7 in the following section. Yan and Lin (1999) first proposed a simple two-phase friction factor correlation (Equation 2.98), followed by Hsieh and Lin (2002, 2003) and Han et al. (2003) with similar forms (see Equations (2.101), (2.103) and (2.105)). Ayub (2003) presented a very simple correlation for two-phase Fanning friction factor which is a function of vapour-only Reynolds number and chevron angle (Equation 2.107). Jassim et al. (2005) observed from R134a adiabatic flow in a single chevron-type channel with ?=60? that a linear relationship between pressure drop ?P and homogeneous kinetic energy G2/(2?m) exists in the two-phase flow at a given quality in the whole range of x = 0-1. Longo and Gasparella (2007) also reported the same linear relationship between ?P and G2/(2?m), from their tests on R134a, R410A and R236fa evaporation in a BPHE with ?=65?. If the observations from those two groups of authors could be validated, calculation of two-phase frictional pressure drop in PHE channels will be greatly simplified by using the homogeneous model with a relatively fixed friction factor. There is no published information available for the effect of flow maldistribution on two-phase pressure drops, since most two-phase investigations were carried out in just a few channels and no severe flow maldistributions were present. One 79 would assume that the same type of pressure-profile-driving flow distribution may also occur for two-phase flow. In the case of an evaporator, vapour flows at much higher velocity than that of liquid in the exit manifold; the pressure profile should be changing more steeply than in the case of single-phase flow, hence the flow maldistribution would be expected to be more severe. This can not, however, be taken as conclusive without verifications from experimental work. Other Pressure Drop Components There are basically three other pressure drop components: acceleration pressure drop ?Pacce, elevation pressure drop ?Pelev and port pressure drop ?Pport. The first two components can be readily calculated by the homogenous model as given by Equation (2.53.b) and (2.53.c), or alternatively they can be calculated by the separated model as given by Equation (2.59.b) and (2.59.c), but in the latter the void fraction ? needs to be determined using additional empirical correlations. The port pressure drop is usually evaluated by the single-phase correlation, Equation (2.86), assuming homogeneous flow. This equation gives: ( )21port 2 m1.5P u?? = (2.95) where m indicates that density and port velocity are to be evaluated by homogeneous mean values. Equation (2.95) was suggested first by Yan and Lin (1999) and followed by a number of other investigators. Ayub (2003) proposed another simple correlation: ( )21port 2 g1.1P u?? = (2.96) Equation (2.96) is a conversion from the original one which was in English units. The equation accounts for pressure drop in both inlet and outlet ports. Sometimes the calculation of overall pressure drop is further complicated when a flow distributor is present, whose effect on pressure drop is very difficult to evaluate without relevant information from the manufacturer. As it is seen that some of the pressure drop components are evaluated by empirical methods, the accuracy and thus their effect on the final frictional pressure drop is hard to evaluate. Fortunately, it has been found that the frictional pressure drop component is by far the largest part, usually being more than 90% (Yan and Lin, 1999, Han et al., 2003, Lango and Gasparella, 2007) of the measured overall pressure drop. 80 2.7.4 Heat Transfer and Pressure Drop Correlations A handful of correlations are available in the open literature, for various flow and plate geometry conditions. None of these have been verified for a wide range of applications. Some of the correlations are reviewed in this section; for clarity, details of these correlations reviewed are tabulated in Table 2.7. Yan and Lin (1999) investigated local R134a flow boiling heat transfer and pressure drop in a two-channel PHE with a chevron angle of 60?, in the quality range of x = 0.1-0.8 with quality change of ?x = 0.09-0.18. They observed that at low values of vapour quality (x < 0.45), mass flux had a small influence on hr, while beyond that, hr increased rapidly with x. Heat flux had very little effect on the overall heat transfer performance. The authors proposed two-phase refrigerant Nu and f correlations, which are among the earliest published and one of a few based on local vapour qualities. However, difficulties in applying those correlations have been reported. It is noticed that the heat transfer correlation (Equation (2.97)) shows very poor agreement with the authors? own data (given in graphic form in the paper) with the predicted value falling far below the data points. This problem was also reported by Donowski and Kandlikar (2000). There might be typographical errors in the published version, or perhaps direction on the proper use of their correlation is needed from the authors. More recently Garcia-Cascales et al. (2007) used this correlation in a comparison work and the predicted heat transfer coefficient needed to be multiplied by a factor of 8 to match it to the experimental results obtained. Donowski and Kandlikar (2000) used Yan and Lin?s data and proposed an improved correlation. The new correlation, given by Equation (2.99), is based on the form of the Kandlikar correlation (1991) for augmented tubes but exponents for Bo, Co and (1-x) are re-evaluated from best-fits to the experimental data. The original Kandlikar correlation is given by Equation (2.35). Good agreement was reported for the improved correlation with a mean error of 16%. Hsieh and Lin (2002), following Yan and Lin and on the same experimental apparatus, carried out an investigation on R410A saturated flow boiling heat transfer and pressure drop. They observed that both hr and ?P increase almost linearly with heat flux q, and that mass flux G had a significant effect on hr only at higher q. Empirical correlations were provided for Nu and f, given by Equations (2.100) and (2.101). In their test the R410a inlet quality was always preset to be saturated liquid, while outlet qualities were not given. This makes their observations less valuable and the reported correlation is lacking a comparison basis to others, since the tested conditions were neither as a whole at standard conditions for refrigerant evaporators (DX or liquid-overfeed) nor with locally specified x or range of ?x. The same authors (Hsieh and Lin, 2003) in a 81 later report presented R410a evaporation performance data based on local mean vapour qualities, at xm = 0.1-0.8 and ?x = 0.126-0.337, and a new set of correlations was presented, shown by Equations (2.102) and (2.103). The new correlation for hr, however, was noticed to be a complete adoption of the Gungor and Winterton 1986 correlation (see Equations (2.24) through (2.27)) for tube flow without any modification. The single-phase liquid heat transfer coefficient hf and the Martinelli parameter Xtt, as required by the method, were both evaluated using smooth circular tube flow equations. Acceptable agreement with data was claimed by the authors but this treatment remains certainly questionable, and no explanation of its physical basis was given. Han et al. (2003) performed experiments on three BPHE?s of different chevron angles with two refrigerants, R22 and R410A. Evaporation heat transfer coefficients and pressure drops were measured with varying P, q, G and local x. Heat transfer coefficients were observed to increase with increasing vapour quality in all units. Empirical correlations were obtained which are given by Equations (2.104) and (2.105). Han et al.?s correlation is another one that is based on local mean vapour qualities, but the determination of this local quality was not clearly given. A tentative calculation, for example, for the testing condition of R410A at G = 27 kg/(m2 ? s) and q = 5.5 kW/m2, showed that ?x is 0.5, this brings the question whether the result can be regarded as ?local?, it is also confusing as the range of data points covered xm = 0.2-0.95 at such a ?x (xm ? 0.5 ?x will be out of the range of x = 0-1, at extreme points). Explanations of those questions could not be found in the paper. Ayub (2003) presented heat transfer coefficient and two-phase friction factor correlations, based on a decade of design and field experience, and data collected from a number of ammonia and R22 installations. The correlation for heat transfer is independent of heat flux, but a leading constant takes different values for flooded or DX systems. The correlations were originally in English units, which were converted into Metric units and given by Equations (2.106) and (2.107). Ayub?s correlations are so far the most universal for PHE evaporators, they cover the two most widely used refrigerants, two types of systems (DX and liquid overfeed), and any type of commercially available plates with any chevron angles. The correlation for heat transfer is however of dimensional form, which is easy to use by field engineers but dimensionally inconsistent. Sterner and Sunden (2006) investigated ammonia heat transfer and pressure drop on five commercial PHE?s with different geometrical configurations. Three sets of correlations were obtained using linear regression analysis for the five evaporator units, those are given by Equations (2.108) and (2.109). All tests were performed under realistic DX conditions where the ammonia inlet quality is in the range of 0.05-0.1 and outlets were at superheats of 2-10 ?C. The correlations gave reasonable accuracy in the range over which they were developed, however 82 any application outside those conditions should be done with caution. They are not likely to be suitable for liquid-overfeed and any systems employing halocarbon (HCFC, CFC) refrigerants, the reasons being: 1. The mass flux G has a high value of exponent of 1.05-1.41. In a liquid- overfeed system where the recirculation rate (and thus G) is high, this correlation will give unreasonably high values of the heat transfer coefficient. Also, under comparable cooling loads, CFC refrigerant mass flux will be 5- 10 times higher than that of ammonia, which will result in a 5-26 times higher heat transfer coefficient by this correlation. 2. The parameter 1/x-1 (in the Co number) has an exponent of 1.44-2.24. In DX systems, the value of x is between 0.53-0.55 (corresponding to inlet vapour quality of 0.05 and 0.1, as in Sterner and Sunden's (2006) paper), while in liquid overfeed systems x is usually around 0.25-0.35 (corresponding to outlet quality of 0.5 to 0.7). It can be shown that this again will bring the heat transfer coefficient up to 6 times higher. Jokar et al. (2006) carried out a dimensional analysis on two-phase flow in corrugated channels and presented empirical correlations based on data collected from a PHE DX evaporator using R134a. The authors reported that in correlating heat transfer coefficient with flow parameters the Chen type of correlation, i.e., the summation of the two terms of forced convection and nucleate boiling, did not work well. For pressure drops, the Lockhart-Martinelli model was found not suitable, while the homogeneous model best described the experimental data. Empirical correlations were proposed for Nu and two-phase cf as given by Equations (2.110) and (2.111). The authors reported average deviations of 31% for the Nu correlation and 46% for cf, which are higher than in most other reported correlations. Again, the form of Jokar et al?s correlation for heat transfer coefficient indicated that it is not suitable for liquid-overfeed systems, as the term 1/x has an exponent of 2 on hr. Moreover, the vapour quality x had to be evaluated by the inlet and outlet mean (0.4 and 1 in their test) since no local measurements were taken. This type of treatment makes the effect of local vapour quality more obscure than clear, and at times when a large exponent is assigned to the mean vapour quality by means of regression analysis, as in this correlation, a change of xm due to change of working conditions will result in much greater or even unreasonable changes of the heat transfer coefficient calculated. The authors made a comparison between their data with the Yan and Lin (1999) correlation, and (surprisingly) it showed consistency in trend and magnitudes. It has been reported earlier in this section that Yan and Lin?s correlation, at least in its equation form as given in the original paper, has major errors and deviates significantly from their own data. The agreement of Jokar et al?s data with this correlation is confusing, which makes the application of this correlation questionable. 83 Palm and Claesson (2006) recommended that for flow boiling heat transfer the Cooper (1984, Equation 2.5) pool boiling correlation is to be used with a correction factor of 1.5. It was concluded that the flow boiling in a DX PHE evaporator is a heat flux dominant process; as such, the Cooper correlation showed good accuracy for a number of refrigerants, PHE?s and pressure level, and is very simple to use. For two-phase frictional pressure drop calculations, the Lockhart-Martinelli method was suggested with a fixed Chisholm parameter C set to 3. Longo and Gasparella (2007) experimentally investigated heat transfer and pressure drop of R134a, R410A and R236fa in a small BPHE. As outlet vapour quality was a controlling parameter, there is a direct connection between heat flux and mass flux and their contributions could not be separated. The heat transfer coefficients showed great sensitivity to heat flux (and corresponding mass flux), weak sensitivity to evaporating pressure, and dropped markedly when outlet superheat was increased. The authors concluded that dry-out occurs in the upper part of the evaporator and the decrease of the heat transfer coefficient is due to the increase in the superheating portion of the heat transfer surface. R410A showed heat transfer coefficients 40-50% higher than R134a and 50-60% higher than R236fa under the same operating conditions. No correlation was proposed for the flow boiling heat transfer coefficient, but experimental results were compared with two well known correlations for nucleate boiling: the Cooper (1984, Equation 2.5) and Gorenflo (1993, Equation 2.6) correlations. It was found that both correlations were able to predict R134a and R410A data with acceptable agreement, with the Cooper correlation performing better (approximately ?20% deviation), whereas both under-predict R236fa data. The authors assumed that this is due to the fact that R134a and R410A evaporating data were controlled by nucleate boiling, whereas R236fa data is affected by both nucleate boiling and forced convection. This assumption was justified by applying the criterion suggested by Thonon et al (Equation 2.94). Two-phase pressure drop was well predicted by the homogeneous model, but an explicit correlation for friction factor was not given. Garc?a-Cascales et al. (2007) compared six correlations with experimental results obtained for R22 and R290 from two PHE evaporators. Those correlations included that of Yan and Lin (1999, with hr multiplied by 8), Hsieh and Lin (2002), Han et al.(2003), the Cooper (1984) pool boiling correlation, and two others traditionally applied to fin and tube heat exchangers but adapted by the authors to plate heat exchangers. Good results were obtained with the Cooper correlation which led to the conclusion that that nucleate boiling plays an important role in the test cases. 84 A Summary It is noticed that most correlations available were developed by the investigators to represent their own data, covering usually limited plate geometry and working conditions. A few authors made comparisons of their correlations with those of others, but none have gained wide validation to any practical extent. All correlations are based heavily or entirely on statistical analysis, which makes their application beyond their tested range highly risky. As demonstrated for some, doing so can brought unreasonable results. To date a general method or correlation for predicting reliably the flow boiling heat transfer and pressure drop in PHE evaporators has not become available. Predictions by different correlations can give very different results at specified conditions. In terms of generality, Ayub?s correlation (2003) is the most promising one, which covers two of the most common refrigerants, two evaporator types, any plate chevron angle, over a wide range of practical conditions. The Donowski and Kandlikar (1999) correlation also remains promising. This correlation works with local heat transfer coefficients which makes it preferable for wider applications. It is based on the form of one that has gained some popularity for round and augmented tubes. The good agreement of this correlation with data might indicate that compact heat exchangers could be treated as augmented tubes, with introduced augmentation factors accounting for surface enhancements. This correlation was developed for a chevron angle of 60? only and different sets of constants might be needed for other plate geometries; moreover, the employment of a fluid-dependent parameter in this correlation may restrict its use with certain refrigerants. Table 2.7: PHE two-phase heat transfer and pressure drop correlations Investigator Correlation Details Yan and Lin, 1999 0.5 0.3 0.33 r eq fo eq fNu 1.926 Re Re Bo Pr ? = ? ? ? ? (2.97) for eq2,000 Re 10,000< < 5 eq1.109 0.5 eq fo f, tp 0.04557 eq eq0.5 fo 6.947 10 Re 6,000 Re Re 31.21 Re Re 6,000 Re c ? ? Sterner and Sunden, 2006 32 4cc c 1 foNu Re Ja Coc= ? ? ? (2.108) 6 5 c foRe cC = (2.109) fo fRe /Gd ?= , fo fRe /Gd ?= , p, f in f fg g Ja c T h ? ? ? ? = ? , e 2d d b= = in w,in r,inT T T? = ? (count-current), 0.50.8 g f 1Co x x ? ? ? ? ?? ? = ? ?? ? ? ?? ? ? ? C is the Chisholm parameter Ammonia/water, or ethylene glycol-water, 5 PHE?s, 1 with distributor, ?=59? for 4, 65? for 1 2b=3.6, 5, 5.6 mm Nch=63-99 DX, overall, xin=0.05-0.1, xout=1, ?Tsup=2-10?C, q not given, Gr=0.5-9.5 kg/(m2? s) Tsat=-3-6?C Jokar et al., 2006 -0.1 2 0.5 0.1 r fo f 2 f p,f sat -0.5 21.12 f fg f f 2 f f g Nu 0.603Re Pr 1 G c T i G xG ? ? ? ? ? ? ? ? ? ? = ? ?? ??? ? ? ? ? ?? ? ?? ? ? ?? ? ? ?? ? ?? ? ? ?? ? (2.110) 4 f, r 1.35 fo x fo 13.521 10 Re for 70 Re 420 c C = ? ? ? (2.111) e 2d d b= = xC same in Yan and Lin (1999) correlation R134a/ glycol-water, ?=60?, 2b=4 mm, Np >30 DX, overall, xin?0.4, xout=1, P= 200-600 kPa, q and Gr not clearly given. 87 CHAPTER 3 EXPERIMENTAL INFRASTRUCTURE 3.1 Introduction The experimental infrastructure in this study has two systems: 1. the water test system, where three brazed plate heat exchanger (BPHE) units operate as single-phase water-water heat exchangers, 2. the refrigeration test system, where the same BPHE?s operate as refrigerant evaporators, with water as process liquid. Using the two systems, two sets of experiments were carried out: the water test and the refrigeration system test. The water test serves to provide heat transfer information on the water side which is to be used for the refrigerant evaporator analysis. The water test system was designed and built in such a way that both thermal characteristics and pressure drops of water can be tested. Pressure drop data on single-phase water are not necessary for evaporator analysis; however, the information helps to understand the hydraulic performance of the corrugated channels in general. The refrigeration test arrangement was built as a liquid over-feed system; experimental facilities were designed in order to study the detailed global performance of the brazed plate heat exchangers and to generate data which can be used to evaluate the developed model. Two circuits are included in this system: the water flowing on one side of the corrugated channel and the refrigerant flowing on the other. The two circuits were designed to provide stable controlled conditions in the desired range of temperatures, fluid flows, and thermal capacities. Measuring instrumentation was selected and installed to quantify all physical variables necessary for determining the exchangers? thermal capacity and hydraulic resistances, with high accuracy. A data acquisition system with PC based data logging was developed. Detailed descriptions of the measuring instruments are presented in the following sections. 88 3.2 Brazed Plate Heat Exchangers Three BPHE units of the herring-bone type were tested in this study. All plates are made of stainless steel AISI 316. The three units have different chevron angle combination: 28?/28?, 28?/60?, and 60?/60?. Plate surface geometry is characterized by the chevron angle ? , corrugation depth b , and the surface enlargement factor ? (ratio of actual corrugated surface area to the projected area of the plate). The enhanced heat transfer is directly related to these features, which provide increased effective heat transfer area, disruption and reattachment of boundary layers, and swirl flow generation. Figure 3.1 shows the geometry of a single plate. All important geometric parameters of the tested plate heat exchanger units were doubly confirmed, in addition to the manufacturer-supplied information. Refer to Appendix A for details of this information on the dimensions and how they were confirmed. Some important geometrical and operational parameters are summarized in Table 3.1. For the fluid on any one side, the BPHE unit is manufactured in a single-pass U-type arrangement (inlet and outlet at the same side of the exchanger unit), and the ports are positioned in a parallel (as opposed to diagonal) manner. Figure 3.1: Single plate geometry 89 Table 3.1: Summary of three BPHE units? geometrical and operating parameters Number of plates: p 24N = Number of effective plates: p,eff p 2 22N N= ? = Number of channels: ch p 1 23N N= ? = , Side 1 (Refrigerant): 12, Side 2 (water): 11 Volume per channel: ch 0.201V = l Heat transfer area per plate: p 0.095A = m 2 Design pressure: 30 bar Design temperature: -196 to 200 ?C Port to port channel length: p 519L = mm Effective channel length: eff 466L = mm Width of flow channel: 180w = mm Plate port diameter: port 53d = mm Corrugation depth: 2b = mm Plate thickness: 0.4? = mm Chevron angle: 28?/28?, 28?/60?, 60?/60? Corrugation wavelength: 8.1? = mm Enlargement factor: 1.14? = Single channel flow area: 4 ch 3.6 10A wb ? = = ? m 2 Equivalent diameter: e 2 2 2 4d b= = ? = mm Hydraulic diameter: h 2 / 3.51d b ?= = mm Total heat transfer area: p p,eff 0.095 22 2.09A A N= = ? = m 2 3.3 Water Test Arrangements The water test system consists of one BPHE, two water tanks, water pump and pipe lines, and the instrumentation. The system was designed in such a way that one BPHE is tested at one time. This arrangement brings the inconvenience of replacing the BPHE?s when more than one unit is tested, but simplifies the system design. Counter-current arrangement is implemented for the two water streams. Hot water is circulated from the water tank to the PHE and flows upwards. Cold water is pumped from the water sump to the PHE and flows downwards. The experimental apparatus is shown schematically in Figure 3.2. The hot and cold water streams were thus arranged so that the cold water flows in 90 the same direction both in this single-phase performance test and later in the refrigerant evaporator tests. In the evaporator tests refrigerant flows upwards and water (from the water sump, as used here for the cold side) downwards, forming a counter-current flow arrangement. Because the water side heat transfer coefficient of the evaporator is to be evaluated by a correlation based on the current cold side data, it is preferable to keep the cold water flow in the same flow direction in both set of tests, to reduce experimental errors on the water side for the evaporator test. The two water tanks have volumes of 1.3 m3 and are well insulated to ensure that water is supplied at constant temperature during the test. It was observed that during any single test (about 45 minutes) the water temperature change was less than 0.5 ?C. Two centrifugal pumps were used to circulate the two water streams. Hot water supply was obtained by using four immersed heaters with a capacity of 2 kW each; water temperature is raised to the desired value, around 55 ?C, before each test run. PVC pipes with an inside diameter of 28 mm are used for all water piping. 3.4 Refrigerant Evaporator Test Arrangements 3.4.1 Liquid Over-feed System The evaporator feeding method was designed to be the liquid over-feed type, with liquid refrigerant being circulated by a liquid pump. Liquid over-feed systems Figure 3.2: Schematic diagram of water test facility layout 91 refer to those in which ?excess liquid is forced, either mechanically or by gas pressure, through organized flow evaporators, separated from the evaporator, and returned to the evaporators? (ASHRAE 2006). Figure 3.3 shows a typical liquid over-feed system. There are generally two types of evaporator feeding methods applicable to PHE evaporators: direct expansion (or dry expansion, DX) and liquid over-feed. From an experimental point of view, the liquid overfeed system is preferable to a DX system, in that some important parameters are more freely adjustable and independent of the cooling load, these include: refrigerant flow rate rm? (feeding rate), inlet and outlet vapour quality inx and outx , and evaporating pressure. The refrigerant flow rate is controlled by a liquid pump and can be in a wide range, the evaporating pressure will depend on the surge drum pressure. Inlet vapour quality is also controllable provided a pre-heater is installed upstream of the evaporator (this arrangement is not employed in the present study). On the other hand, in a DX evaporator these parameters are always coupled and not individually adjustable, because certain degrees of superheat required at the exit of the evaporator will determine the operation of the expansion valve, resulting in a simultaneous change of refrigerant flow rate, evaporating pressure, and inlet vapour quality. From an industrial point of view, almost all large units invariably use over-feed systems where PHE evaporators are involved. Halocarbon refrigerants are not commonly used in large over-feed systems, the reasons for this can be: Figure 3.3: Typical liquid over-feed system 92 1. oil returning problems; 2. high density of halocarbon refrigerants requires high pump power input; 3. low latent heat requires more refrigerant to be pumped for the same cooling duties. Ammonia is usually the choice for large-scale liquid overfeed installations for the above reasons. In the current study, however, halocarbon refrigerant was used in a liquid overfeed system, but the focus of the investigation was on the flow boiling mechanism rather than on any specified feeding method (to this aim, it would also be desirable to test other refrigerants in further experiments). It might be worthwhile to make the concept clear that liquid overfeed systems are those in which a greater rate of liquid refrigerant is delivered to the evaporator than evaporated. At the exit of the evaporator a mixture of liquid and vapour flows out. Liquid overfeed systems are characterized by having a low-pressure receiver, also sometimes called a surge drum, from which the liquid refrigerant is circulated through the evaporator either mechanically or by thermo-siphon. What is normally called a flooded evaporator is a shell-and-tube exchanger where chilled process liquid flows in the tubes which are submerged in boiling refrigerant in the shell. Vapour refrigerant is discharged from the top of the evaporator and directed to the compressor. The difference between an overfeed and a flooded evaporator is that in the latter only vapour leaves the evaporator, instead of a two-phase mixture, and for this purpose it is necessary to design a dropout area inside or install an eliminator at the top of the evaporator, to separate the liquid refrigerant from the vapour. A liquid overfeed evaporator can be either a shell-and-tube exchanger or a PHE, while a flooded evaporator is always a shell-and-tube exchanger with refrigerant boiling on the shell side. It is also important to note that sometimes the terms ?flooded? and ?liquid overfeed? are mixed freely and may refer to the same evaporator, possibly because the liquid overfeed evaporator operates also in a flooded ?manner?. Thermo-siphon evaporators are another example in the liquid overfeed category, these types of evaporators operate by means of natural density differences in the refrigerant liquid and vapour and a liquid pump is not used. Thermo-siphon systems are more commonly seen in large-sized industrial installations. Field tests were carried out on two large-scale thermo-siphon units using ammonia and refrigerant R12, which are described in Chapter 5. In a liquid overfeed system, the overfeed extent is evaluated by the recirculation rate RCn , which is defined as the mass ratio of liquid entering to the amount of liquid vaporized in the evaporator : RC g, out out/ 1/n m m x= =? ? (3.1) 93 Typical values of recirculation rate for conventional evaporators are 2 to 7 (ASHRAE 2006), while for PHE liquid over-feed evaporators the common value of recirculation rate is 1.2-1.5 (Stromblad, 1989, Hanssen, 1997), corresponding to outlet vapour quality of 0.67-0.83. The current system was designed for this range of recirculation rate. With fine adjustment of the system operating parameters, i.e., water and refrigerant flow rates and the surge drum pressure, it is also possible to obtain fully vaporized or superheated outlet conditions, where the liquid-overfeed system will run as a dry system and the recirculation rate is 1 in this case. 3.4.2 System Design Feeding Method The experimental facility serves for experimental investigation of the refrigerant flow boiling mechanism in corrugated channels of the plate heat exchangers under various flow conditions. To this aim, the equipment had to be able to control, in a certain range, some of the important operating parameters; those basically include flow rate, outlet vapour quality, evaporating pressure, and the cooling load. The evaporator feeding method was chosen to be the liquid overfeed type; as mentioned previously, this feeding method gives the system the flexibility that a DX system does not offer. Choosing the PHE Type Choosing the PHE type is another task in the system design. There are mainly three types of PHE?s applicable to refrigeration duties, namely frame-and-plate, semi-welded plate pack, and brazed PHE?s (BPHE?s). The frame-and-plate type of PHE is widely used for liquid-liquid duties and its applications in refrigeration duties are not commonly seen nowadays, mainly due to the concerns of refrigerant leakage. With chosen gasket material compatible with the refrigerant applied, this type of PHE can be used as refrigerant evaporators. The frame-and- plate type exchanger has many virtues when used as testing equipment. The plates can be replaced to obtain different channel geometries; the number of flow channels can be changed by adding or removing plates when there is a need of cooling load change; needle-type pressure tappings can be inserted through the gasket at certain locations if local pressure is to be measured. Moreover, plates for this type of PHE usually have the widest product spectrum from manufacturers, compared with the other two. However, when used in a refrigeration system, the evaporator requires high levels of sealing quality; new 94 gasket sets need to be used if new plates are to be tested. This type of exchanger is not considered in the present study. Semi-welded PHE?s are refrigerant leakage-proof and can work under higher pressure and temperature limits. For a testing rig, one shortcoming for semi- welded units is that the plate geometries to choose from are usually limited. Mixed-plate channels are possible to obtain from manufacturers when special orders are made, but this will depend on the available plate geometrical patterns. BPHE?s are widely used in refrigeration systems as evaporators and condensers, where the plates are brazed together as an integral pack which makes it not possible to change channel geometry without using new sets of PHE?s. Some BPHE?s are equipped with flow distributors, and because the effect of distributors on flow distribution, pressure drop, and heat transfer are not well understood, the installation of a distributor may not be desirable for experimental purposes. For this study, three sets of BPHE units with different chevron angles were used, flow distributors were not fitted. Refrigerants The system used in this study was initially designed for refrigerant R134a, which is gaining an increasing application worldwide in the refrigeration industry. R507A was used for additional tests with new lubricant oil. R410A was a consideration but not used due to its high working pressure. Oil Return Oil return is a critical design parameter in all refrigerant systems; it is more so in a liquid overfeed system using halocarbon refrigerants. Liquid overfeed systems typically use a surge drum to feed the evaporator, either by gravity or by a liquid pump. No matter how efficient the oil separator is, some lubricant oil will always find its way into the low pressure side of the system and will inevitably accumulate in the surge drum. The approach to returning oil in an ammonia system is relatively simple because ammonia and mineral oil are not miscible (to be accurate, miscible to a very limited extent) and because the oil has a greater density than liquid ammonia, when oil passes on to the surge drum it separates from the refrigerant and settles to the bottom. An oil sump installed at the bottom of the surge drum can then collect the oil and a drain line on the sump will direct the oil back to the compressor. In halocarbon systems, however, the refrigerant and lubricant oil are at least partially miscible, the miscibility depends on refrigerant type, temperature and pressure. For partially miscible oil/refrigerant solutions, for instance mineral oil/R22, the two substances are miscible above a certain temperature, below that temperature there is a separation, and because the 95 oil-rich layer is less dense than the halocarbon-rich layer, it rises to the top of the liquid mass. For mutually soluble pairs such as Polyol Ester (POE)/R134a, the two substances are completely soluble. For these reasons, oil return must be handled differently than in an ammonia system (Stoecker, 1998). In the current R134a system, a flow of liquid refrigerant and its dissolved oil is extracted from an outlet of the surge drum (it could also be on a point of the feeding line to the evaporator), and directed to the suction line. Because there is liquid refrigerant in this mixture that should not be permitted to return to the compressor, a liquid/extraction heat exchanger is located on the liquid line connecting the compressor discharge and the surge drum. This method will not return all the oil in the surge drum, but the amount of oil in the surge drum will be maintained at a certain level in any stable running condition. Figure 3.4 shows the schematic of the oil return system. The amount of extraction is controlled by a metering valve and should be kept small. To ensure that a suitable return flow is taken and at the same time no liquid refrigerant is passing to the compressor, some additional components are required: Upstream of the heat exchanger: ? A liquid solenoid valve, actuated by the compressor, to prevent flow of liquid refrigerant to the suction line when the compressor is idle; Figure 3.4: Schematic of the oil returning arrangement 96 ? A metering valve, to control the extraction flow rate. Downstream of the heat exchanger: ? A sight glass in which no liquid should be observed. Also very important is that where a riser is needed in the suction line, the sizing of the riser is to be carefully designed to make sure that a high vapour flow velocity is obtained sufficient to carry the oil back to the compressor, as in the case of a direct expansion system. 3.4.3 The Experimental Facility The refrigeration test facility consists principally of two circuits: the water circuit and the refrigerant circuit. Each circuit has independent control. The test facility has been designed to give controllable and stable working conditions for the refrigeration cycle, and to provide desired evaporating pressure and liquid refrigerant recirculation rates RCn for the evaporator. The system layout is given schematically in Figure 3.5, and the equipment shown in Figure 3.6. Water Circuit The water circuit circulates water at the environment temperature (between 17 and 25 ?C, depending on season) from the water sump through the brazed plate heat exchangers and back to the water sump, at a position away from the point of water suction. This is to ensure that the supply water temperature will not be affected by the retuning cooled water. The water sump has a large volume of clean water collection (more than 40 m3) and constant feeding temperature during the testing period is assured. Water flows downwards in the exchanger to form a counter-current flow arrangement. Water flow rate was mutually controlled by a gate valve on the feeding line and a ball valve on the by-pass line.1 To reduce water side fouling on the exchangers? surface, two fine-mesh filters were fitted at the pump suction and pump outlet. 1 A by-passing arrangement is more preferable in flow rate control than a single regulating valve fitted in the main line. Centrifugal pumps can run stably in a certain range of flow rates, below that, vibration and flow fluctuation usually occur. A by-passing arrangement can effectively reduce the flow rate in the main line without much effect on that of the pump; this can be very useful if stable and small flow rates are desired from a relatively large capacity pump. 97 Figure 3.5: System layout of the refrigeration system test facility 98 Refrigerant Circuit The refrigerant circuit consists of the BPHE evaporators, the condensing unit with reciprocating compressor and air-cooled condenser, the surge drum, a high pressure receiver, the oil returning system, and other accessories necessary for system running. The three brazed plate heat exchangers were organized in a parallel manner; each unit can be conveniently isolated from the system by shutting down two isolating valves positioned in the inlet and outlet pipe lines. Refrigerant liquid enters the evaporator at the bottom and flows upwards, being heated by the process water when flowing through the channel. The refrigerant flow rate was controlled by a by-pass arrangement. A by-pass line is installed at a point between the refrigerant liquid pump and the flow meter; the line joins the refrigerant returning line from evaporator and returns to the surge drum. Refrigerant flow rate in the evaporator feeding line is thus controlled by a regulating valve fitted on the by-pass line. A back pressure regulator is installed at the surge drum outlet on the compressor suction line; this regulator controls the surge drum pressure, which determines the evaporating pressure in the evaporator. Figure 3.6: The experimental equipment of the refrigerant evaporator tests 99 Cooling Load At design conditions, the system has a cooling capacity of 13.5 kW minimum and 24 kW maximum, in the range of evaporating temperature of 1-15 ?C and condensing temperature of 35-45 ?C. The refrigerant flow rate is in the range 3.9- 7.6 l/min, giving recirculation rates of 1-1.4 at the given cooling loads. For a specific evaporator with fixed heat transfer area, cooling load is determined by the mean temperature difference LMT? between the two fluids and the overall heat transfer coefficient U . Water inlet temperature is not a controllable quantity, however, on the refrigerant side, the evaporating temperature can be controlled in a range by adjusting the surge drum pressure, via a back pressure regulator installed at the surge drum outlet on the compressor suction line. The overall heat transfer coefficient U depends largely on the water and refrigerant side heat transfer coefficient components, and can be controlled to a certain extent by changing the water flow rate, an operation conveniently available. 3.5 Instrumentation and Measurement The instrumentation used in the experimental facility was selected and applied for the measuring of all the variables necessary for determining the performance of the heat exchangers. The quantities measured, in both water and refrigeration system tests, basically included temperatures, pressures, and flow rates. There are 14 sensors that were employed in the study, of those 7 were used for the water test, 9 for the refrigeration test, some of which were used for both tests. All instruments have analogue outputs which are logged onto a PC via the data acquisition system. A description of the measuring instruments is presented in this section, organized separately for the water test system and the refrigeration test system. For detailed technical data and wire connections, Appendix B refers. 3.5.1 Water Test System Temperature T Temperatures are measured at four points, i.e., at inlets and outlets of both hot and cold water streams, approximated 200 mm from the PHE ports. Measurements were taken by sensors located at pertinent points with LM35DZ semi-conductor sensors, which are integrated circuit sensors that give electrical 100 voltage output proportional to the temperature in Celsius. The sensors are each powered by a 9V battery and give a 0-1 V signal corresponding to 0-100 ?C temperature spectrum, via a three wire system. The LM35DZ sensor is a tiny semi-conductor with a plastic tip. To use this sensor for the measurement of water temperature, a probe needed to be made which encloses the sensor in a copper sheath and thus can be inserted into the water pipe. The temperature sensor probe is shown schematically in Figure 3.7. Volumetric flow rate Q Volumetric flow rates of water on both sides are measured with two RS 257-026 turbine flowmeters. These flow meters have a rotor which spins when liquid passes through the flowmeter. The speed of the rotor is directly proportional to the flow velocity. As the rotor spins, the stainless steel blade tips pass a magnetic field and an AC voltage is induced in a coil, which is converted into an output 4- 20 mA current signal proportional to the flow. The two flowmeters have the capacity of 0-100 l/min and were, for the present experimentation, calibrated in the range of 0-60 l/min. Many piping configurations and fittings generate disturbances with unknown characteristics, for this reason, flow conditioning, i.e., minimum length of straight pipe runs before and after the meter, is required for certain types of flowmeters. This information is not available from the manufacturer?s data sheet, therefore as a general rule, a minimum length of 20 diameters of straight run piping (Feener, 1999) was applied. Figure 3.7: Temperature probe structure 101 Differential Pressure ?P Differential pressure is the pressure drop between the PHE inlet and outlet, on the cold water side in the test. The pressure drop is measured by a Rosemount 3051 CD differential pressure transmitter. The transmitter has a working range of 0-80 kPa and was calibrated in the same range using water. Figure 3.8 shows the layout of the measurement arrangement. It is important to note that the measured pressure drop has many components, and is not the frictional pressure drop coreP? in the corrugated channel, which is of the interest of the test. Evaluations of other components and an extraction procedure are required to obtain coreP? , as will be given in Chapter 4. 3.5.2 Refrigerant Evaporator Test System Temperature T Temperatures are measured at four points: inlets and outlets of both refrigerant and water. Four resistance temperature detectors (RTD?s) are used for the measurements. All the RTD sensors use a VDC 24V power supply and give standard 4-20 mA current signals. On the water pipe line, two RTD?s are Figure 3.8: Layout of the pressure drop measurement in water test 102 mounted using pressure couplings. On the refrigerant pipe line, the two sensors are mounted in well insulated T-connections where sealed copper pipes are brazed into a Tee and the sensors are inserted into the pipes. All sensors were extended approximately to the pipe center lines. Schematics of the mounting arrangement are shown in Figures 3.9 and 3.10. Refrigerant Absolute Pressure P Static pressures at the refrigerant inlet and outlet are measured by two WIKA pressure transmitters. The two sensors are mounted in well insulated T- connections on the pipe line. Also on the compressor suction line a pressure gauge is fitted on the back pressure regulator, for monitoring of the surge drum pressure. Both the WIKA sensors and the pressure gauge give readings of the ?gauge pressure?, which is the absolute pressure less the local atmosphere pressure. The WIKA sensors use a VDC 24V power supply and give 4-20 mA current signals. Refrigerant Differential Pressure P? The pressure drop through the evaporator is measured with a differential pressure transmitter 3051 CD, from Rosemount Inc. Careful consideration is needed for this measurement. The inlet refrigerant is liquid and the outlet refrigerant is a liquid and vapour mixture. The transmitter can not simply be installed with vertical connection between the inlet and outlet pipe lines because the connection could be partially filled with liquid and vapour. The unknown liquid column in the pressure connection can exert static pressure and thus has an effect on the measured result. For heavy refrigerants such as R134a this could be of the order Figure 3.9: Water line RTD mounting Figure 3.10: Refrigerant line RTD mounting 103 of the measurement itself. To solve this problem, all sensor connection tubes must be horizontally oriented, and the evaporator outlet port and the sensor low- side connection must be at the same height. The arrangement in the current facility is shown schematically in Figure 3.11. This arrangement eliminates the influence of refrigerant fluids in the connecting tubes on the pressure measurement. Water Volumetric Flowrate wQ Volumetric flow rate of water on the process liquid sides is measured with one RS 257-026 turbine flowmeter, which is one of the two flowmeters which were used for the water test. The flowmeter was positioned at the water pipe line exit. The same flow conditioning was applied for this flowmeter, namely a minimum length of 20 diameters of straight run piping provided on both sides. Refrigerant Volumetric Flow rate rQ The flow rate of the liquid refrigerant is measured with a Trimec MP 15S multipulse flowmeter, from Trimec Industries (Australia). The flow meter utilizes the oscillating-piston principle for measuring the flow rate. The passage of liquid causes a piston to oscillate smoothly in a circular motion within the round measuring chamber. Each piston cycle displaces a known volume of liquid Figure 3.11: Connection of the differential pressure transmitter 104 from the inlet to the outlet port, while small high energy magnets in the piston activate a reed switch and a solid state sensor (Hall effect sensor), which provides pulse outputs for remote flow monitoring. The flowmeter has a flow rate range of 0.2-10 l/min, the calibration was done in the range of 0-8.5 l/min for the current study. This flowmeter is insensitive to viscosity changes and does not require flow conditioning. The manufacturer?s instruction states that the flowmeter does not create pulses in the flow, however, pulsation of the water stream was observed during the calibration (refer to Appendix C), more apparent at relatively smaller flow rates. The device uses a 5-24V VDC power supply and gives a variable frequency signal. The frequency signal is converted to a current signal via a converter (Model 1100P, IQ Instruments cc, South Africa), and then sent to the data acquisition system. 3.6 Data Acquisition System Sensors and Data Acquisition It is generally accepted in the field of measurement that sensors which give electrical signals are termed as ?transducers?, while transducers which give 4-20 mA signal ranges are called ?transmitters?. All sensors used in this study have electrical analog signal outputs. The four temperature sensors used in the water test give voltage signals in the range of 0-1 V. All other sensors are standard transmitters and give current output signals of 4-20 mA. Outputs from the sensors need to be converted into a meaningful quantity, for example, the output from the turbine flowmeter, which is a current signal, needs to be converted into volumetric flow rate. In most cases, measuring transmitters are linearized devices; the conversion rate is given in the manufacturer?s instruction, along with its accuracy specifications. For higher accuracy, a calibration process needs to be carried out with a properly designed procedure. The data acquisition system used in the present study is the National Instruments? NI PCI-6224 DAQ system along with the operating software LabVIEW (version 8.0). Most data acquisition systems read voltage signals only, including the NI PCI-6224 system. Therefore, it is necessary to make an arrangement in electrical connection which can give voltage signals. Because current output is the same at any point in the circuit, it can be converted to a voltage output for measurement purposes at any point in the circuit by adding a load resistor in series. The basic 105 circuit for measuring the output of a transmitter is then to connect a power supply, a precision load resistor and the transmitter in series. The analogue signal is measured indirectly by connecting a voltmeter (built-in with the data acquisition system) across the load resistor which produces a voltage drop and is proportional to the 4 to 20 mA current loop. This is shown in Figure 3.12. The load resistor brings extra uncertainty to the measurement accuracy. However, a calibration procedure can compensate for this uncertainty if the measurement channel, including the sensor and all connections, is calibrated as one integrated unit. General Features of the Data Acquisition System Used The NI PCI-6224 data acquisition system is one of the M series products from National Instruments. Voltage signals are the only analogue signals that can be read by the device. Three types of analogue input ground-reference settings are supported, namely: Differential mode, Referenced single-ended mode, Non- referenced single-ended mode. Depending on the ground-reference mode, 16 or 32 channels in total are supported (16 channels for differential mode, 32 for other modes). The Differential mode reduces noise pickup and should be used whenever possible; this was used in these tests. It is recommended by NI that the M-Series device should be self-calibrated after installation and whenever the ambient temperature changes. Self-calibration should be performed after the device has warmed up for the recommended time period (15 minutes for NI 6224). This function measures the onboard reference voltage of the device and adjusts the self-calibration constants to account for any errors caused by short-term fluctuations in the environment. In the process of self-calibration all external signals need to be disconnected. 106 Multichannel Scanning Considerations In multichannel scanning applications, accuracy is affected by settling time. Settling time refers to the time the device takes to amplify the input signal to the desired stability before it is sampled. Several factors can increase the settling time which decrease the accuracy of measurements. To ensure fast settling times, the following should be considered (in order of importance): ? Use low impedance sources Large source impedances increase the settling time. To ensure fast settling times, the signal sources should have an impedance of less than 1k?. If the source impedance is high, the scan rate can be decreased to allow more time to settle. The NI 6224 has a default settling time of 14 ?s, but this was manually adjusted to 1000 ?s in the current study. For sensors that have a current signal output, the signal source is actually a 47 ohm resistor and will thus not cause a problem. ? Use short high-quality cabling Short high-quality cables can minimize several effects including crosstalk, transmission line effects, and noise. It is recommended by NI to use individually shielded, twisted-pair wires that are 2m long or less to connect analogue signals to the device. Figure 3.12: Connection loop of transmitters with current 107 ? Carefully choose the channel scanning order a. Avoid switching from a large to a small input range. Switching from a channel with a large input range to a channel with a small input range can greatly increase the settling time. In the current study, all channels have an input range of 1 V. b. Minimize the voltage step between adjacent channels. When scanning between channels that have the same input range, the settling time increases with the voltage step between channels. If the expected input range of the signals is known, then similar expected ranges should be scanned together in groups. ? Avoid scanning faster than necessary Scanning slower gives the system more time to settle to a more accurate level. There are two cases to consider: a. Averaging large number of samples can increase the accuracy by reducing noise effects but it also decreases the required settling time. b. If the time relationship between channels is not critical, it is preferable to scan the same channel multiple times and scan less frequently. Technical Specifications and Analogue Input Circuitry For more detailed information on technical specifications and analogue input circuitry of the data acquisition system, refer to Appendix D. Scanning Settings and Channel Assignments The data acquisition system was set to have a sampling rate of 100 Hz for each channel, and the samples were filtered by a built-in third order Butterworth low- pass filter, the filtered signals were then further averaged over every 50 sample. Finally two readings per second were obtained. This setting was maintained throughout all the calibrations and tests. Channel assignments for the water and refrigeration system tests are given in Tables 3.2 and 3.3, respectively. The channel-sensor pairs must be kept unchanged for all calibration and measurement procedures. Table 3.2: Channel assignment in water test Physical channel Sensor Measuring quantity 108 AI 0 Temperature sensor No.1 Hot inlet AI 1 Temperature sensor No.2 Hot outlet AI 2 Temperature sensor No.3 Cold outlet AI 3 Temperature sensor No.4 Cold inlet AI 4 Water flowmeter No.1 Hot water stream (at outlet) AI 5 Water flowmeter No.2 Cold water stream (at outlet) AI 6 Diff. pressure transmitter Cold water (between outlet and inlet) Table 3.3: Channel assignment in refrigeration system test Physical channel Sensor Measuring quantity AI 16 RTD No.1 Water outlet AI 17 RTD No.2 Water inlet AI 18 RTD No.3 Refrigerant inlet AI 19 RTD No.4 Refrigerant outlet AI 20 Refrigerant flowmeter Refrigerant flow rate AI 21 Pressure transmitter No.1 Refrigerant outlet AI 22 Pressure transmitter No.2 Refrigerant inlet AI 4 Water flowmeter No.1 Water flow rate AI 6 Diff. pressure transmitter Refrigerant (between outlet and inlet) 3.7 Sensor Calibration 109 Calibration Procedure and Range All 14 sensors were carefully calibrated before the experimentation. A best fitting curve was obtained for each sensor and used in the measurements. As a summary, Table 3.4 gives a list of all the sensors with their manufacturer-specified and calibration ranges. Refer to Appendix C for details of the calibration procedures and evaluation of the fitting curves. Uncertainty analysis for sensor calibration is given in Appendix J. It is very important to note that the sensors and data acquisition channels had to be calibrated together, and to be used together for any measurements afterwards. This is because an individual channel brings its own error. A data acquisition channel includes the wire connections, load resistance, and the data acquisition system physical channel with all its settings. Channel-sensor pairs should be kept unchanged, for example, if channel No.1 is used for sensor No.1 in the calibration, then the same pair should remain for any further measurements. Calibration helps to improve the sensors? accuracy, in addition to that provided by the manufacturer?s product specifications. Although sensors with factory calibration certificate can be used straight away, it must be noted that the accuracy level degrades after a period of time, normally 1 year after the calibration is taken. This same time period applies to calibrations carried out by the experimenter. Table 3.4: Summary of sensor calibration range Sensor Measuring quantity Manufacturer-specified working range Calibration range RS flowmeter 1 Q , l/min 0-100 0-65.0 RS flowmeter 2 Q , l/min 0-100 0-65.1 LM35DZ 1 T , ?C LM35DZ 2 T , ?C LM35DZ 3 T , ?C LM35DZ 4 T , ?C 0-100 0-63.3 RTD 1 T , ?C 0-100 RTD 2 T , ?C 0-100 RTD 3 T , ?C 0-200 RTD 4 T , ?C 0-200 0-59.5 110 WIKA 1 P , Pa WIKA 2 P , Pa 0-1000k 0-840k Rosemount 1151 CD P? , Pa 0-80k 0-73.3k Trimec flowmeter Q , l/min 0.2-10 0-8.5 Calibration Uncertainty For the sensor calibration, and also in the following experimental measurements, the uncertainties were calculated according to the ISO (ISO/1995) method. For clarity, only a summary of the analysis results is presented here, as given by Table 3.5. Extended details of the uncertainty analysis for the sensor calibration are given in Appendix J. sensor Maximum uncertainty LM35DZ temperature sensors (4 in total) ? 0.08 ?C RTD?s (4 in total) ? 0.08 ?C RS 257-026 Turbine Flow Meters (2 in total) ? 0.65 % Trimec MP15S Flowmeter ? 0.29 % Rosemount 3051 CD Diff. Pres. Transmitter ? 1.9 % WIKA Static Pressure Transmitters (2 in total) ? 0.72 % Table 3.5: Summary of sensor calibration uncertainty 111 CHAPTER 4 SINGLE-PHASE PERFORMANCE TESTS WITH WATER 4.1 Introduction The experimental results on the three BPHE units operating as water-water heat exchangers are reported in this chapter. Correlations for Nu and isothermal friction factor f are developed, based on the experimental data. The obtained correlations are compared with seven other published correlations, and discussions on the thermal and hydraulic performance of PHE?s are given. The tests with water have three main purposes: 1. to obtain an overall understanding of the performance of PHE?s operating in single-phase water-water applications; 2. to obtain accurate water side heat transfer coefficient information. This was needed for the subsequent performance analysis of the same units when used as water-cooling evaporators; 3. to obtain single-phase water pressure drop data, for the evaporator pressure drop analysis when the separated model (the Lockhart-Martinelli method) is used. The Reynolds number range in these tests was relatively small, from 280 to 1100. As discussed in Section 2.6.1, single-phase flows in this range can be considered to fall in the turbulent flow region for PHE?s. 4.2 Test Procedure The water test facility and arrangements have been described in detail in Section 3.3. BPHE units were configured in a single-pass U-type (inlet and outlet at the same side of the exchanger units), counter-current arrangement. Experiments on 112 heat transfer and pressure drop were carried out separately. In each set of experiments, tests were carried out on the three BPHE units individually, following the same procedure. Measurements were taken in a range of flow rates, as summarized in Table 4.1. For tests on the heat transfer performance, hot water flow was maintained at fixed flow rates for each unit, and was kept at high Reynolds numbers to ensure turbulent flow. Cold water was supplied in a range of flow rates. Hot water was maintained at approximately 55 ?C for all tests, and the cold water was supplied at around 18 ?C from the water sump. It is to be noticed that hot water achieves a higher Reynolds number than the cold water at the same flow rate, due to the temperature-dependent characteristics of water viscosity. For both the hot and cold water streams, measurements were taken for volumetric flow rates and inlet and outlet temperatures. All data were taken under steady-state conditions, this was confirmed by: 1. Stable fluid outlet temperatures. Variation of outlet temperatures of the two water streams were within 0.1 ?C during a 2 minute time period before each measurement was taken. 2. Close energy balance. The heating and cooling load Q? calculated from the hot and cold sides did not differ by more than 2%. For tests on the pressure drop performance, adiabatic flow conditions were maintained with only cold water flowing through the exchangers at the environment temperature (around 20?). A differential pressure transmitter was positioned to measure the pressure drop across the PHE?s, at various volumetric flow rates. Again all data were taken under steady-state conditions. Table 4.1: Summary of water-water performance tests of the three PHE units Test PHE Chevron angle No. of data Hot side Cold side Re Tin , ?C Re Tin , ?C 1 28?/28? 22 510-560 55.0-56.0 280-1100 18.4-19.3 2 28?/60? 18 450-510 55.0-55.5 330-1110 18.4-19.2 Heat transfer 3 60?/60? 16 430-460 56.0-56.5 350-1030 18.3-19.1 4 28?/28? 17 ? ? 280-1010 18.0-22.0 5 28?/60? 16 ? ? 220-990 19.0 Pressure drop 6 60?/60? 18 ? ? 210-840 18.0 113 4.3 Data Reduction Experimental data for the heat transfer coefficient were processed by a MATLAB code, using data reduction methods described in this section. The program flowchart is given in Appendix H. Water properties were evaluated according to a NIST Standard Reference Database (Lemmon et al., 2005). All fluid properties are calculated at the bulk mean temperature: ( )1mean in out2T T T= + (4. 1) 4.3.1 Heat Transfer The primary measurements consist of six parameters, namely the volumetric flow rates and the inlet and outlet temperatures of both streams. Heat loads (total heat transfer rates) on the hot and cold side are calculated by: out hot hot p in out cold cold p in Q m c dT Q m c dT ? = ???? =? ? ? ? ? ? ? (4. 2) Under ideally steady conditions and provided that the measurements are absolutely accurate, there is an energy balance between the two streams: hot coldQ Q Q= =? ? ? (4. 3) In the real experiment, due to flow conditions not reaching equilibrium and measurement errors, there will always be some discrepancy between the two. Either the arithmetical average of the two or the one which has a smaller measurement uncertainty can be used as the heat load of the exchanger. In the present study, the hot side has stable flow rates during one test run which led to a smaller measurement uncertainty and is thus chosen. For all measurement points the discrepancy of the heat load does not exceed 2%. The overall heat transfer coefficient is determined by LM qU T = ? (4. 4) where /q Q A= ? is the heat flux. ?TLM is the log-mean-temperature-difference defined by Equation (2.72). 114 The film coefficient h is not a direct measurement quantity, and must be extracted from the overall heat transfer coefficient, through a proper reduction procedure . (Although a direct method is possible to find the film coefficient, this will involve the measurement of wall temperatures, a condition not conveniently available in most situations.) The overall heat transfer coefficient can be expressed in terms of thermal resistances by: wall fouling, hot fouling, cold hot cold 1 1 1 R R R U h h = + + + + (4. 5) Where hhot and hcold are the film coefficients on the hot and cold side, respectively. Rwall is the wall thermal resistance determined by: wall wall R k ?? ? = ? ?? ? (4. 6) Fouling resistances can be found from reference books or published information. Information from two selected sources is given in Section 2.6.4. Since brand new PHE units were used in the experiment, the fouling resistances were assumed to be negligible. To determine hhot and hcold from Equation (4.5), some assumptions must be made, especially for the expression form of the film coefficient. A new method is proposed here which has not been found in the open literature, as described below. Step 1: Determine the form of the equation for heat transfer coefficient h. It is customary to express the Nusselt number in the following form, as seen in much of the literature: 0.17 m 0.33 w Nu Re Prc ? ? ? ? = ? ?? ? (4. 7) which can be rewritten as: 0.17 m 0.33 w Re Pr kh c d ? ? ? ? ? ? = ? ? ? ?? ?? ? (4. 8) Step 2: Assume that the same form of expression for film coefficient applies on both sides. This requires that during the test, the two fluid streams shall be at sufficiently high Reynolds numbers and thus are both in the turbulent region. The trend of the Nu-Re relationship curve changes 115 when flows change from laminar to turbulent regions1, and so does the correlation equation. In the current experiment, this change was observed at Reynolds number around 360 for the low-angle PHE unit, but not for the mixed and high-angle PHE units at similar Re numbers. For this reason, only data obtained at Reynolds number beyond 360 are used for the low-angle PHE, and the developed correlation applies to turbulent flow conditions only. Step 3: Assume a value of m in Equation (4.7), and find the wall temperature to evaluate ?w. The heat flux q is expressed as: wall, hot wall, cold hot hot wall, hot cold wall, cold cold wall ( ) ( )T Tq h T T h T T R ? = = ? = ? (4. 9) Using Equation (4.9), an iteration program is carried out to find wall, hotT and wall, coldT : 1. set initial value of wall, coldT as ( ) 0wall, coldT , for each measurement point of flow rate 2. ( ) 0wall, hot wall, cold wall qT T R = + 3. find iteration value of wall,cT = ( )1wall, coldT , where ( )1 hotwall, cold cold hot wall, hot cold ( ) hT T T T h = + ? where hot cold/h h is obtained by Equation (4.8), the constant 1c cancelled in the ratio. 4. If ( ) ( )1 0wall, cold wall, cold 0.1e T T= ? < , iteration finished, otherwise ( ) ( ) ( )0 0 11wall, cold wall, cold wall, cold2T T T? ?= +? ?? ? , and repeat 2-4. Step 4: Use linear regression to find the leading constant c. To make the regression possible, some rearrangements are needed. The film coefficient is rewritten as: 0.17 m 0.33 0 w Re Prh kh c d ? ? ? ? ? ? = = ? ? ? ?? ?? ? (4. 10) 1 The Nu Re? curve is generally flatter in the laminar region and steeper in the turbulent region, according to the current experimental observations and correlations from open literature. This is the same trend as shown in the better-known circular tube heat transfer mechanisms, where for the two extreme cases of constant wall temperature and constant heat flux, laminar flows have constant Nusselt number, though with different values in each case. 116 note that h0 is now obtained with the assumed m. Substitute Equation (4.10) into equation (4.5) and rearrange: wall 0, hot 0, cold 1 1 1 1R U c h h ? ?? ? ? = +? ?? ? ? ?? ? ? ? (4. 11) Let 0, hot 0, cold 1 1 x h h ? ? = +? ?? ?? ? and wall 1y R U ? ? = ?? ?? ? , use linear regression (least square fitting) to find the constant for y = ax. The data size for x and y is the number of experimental measurement points. Now c is obtained by a = 1/ c. Step 5: Determine m. Every pre-assumed m value would have resulted in a corresponding c, and therefore solely determines the fitting curve by the regression as shown in step 4. The best value for m is the one from which the resulted fitting curve best fits the experimental data. Generally the goodness of a fitting is evaluated by two parameters, the Root Mean Square Error (RMSE) and the Coefficient of Multiple Determination (usually called R-square). In the present method, the best m is determined by trial-and-error: values from 0.01 to 2 with a step of 0.01 were tried, the one which has the smallest RMSE was chosen. 4.3.2 Pressure Drop Contributions of the Total Pressure Drop For single-phase flow, the total pressure drop across a plate heat exchanger consists of three contributions: 1. pressure drop associated with the inlet and outlet manifolds and ports, ?Pport , 2. friction pressure drop within the core, ?Pcore, 3. pressure drop due to the static water head (elevation change), ?Pelev. Summing all contributions, the total pressure drop across a plate heat exchanger is given by: PHE core port elevP P P P? = ? + ? + ? (4. 12) 117 The elevation pressure drop (or rise, depending on flow direction) is due to the static water head, and is calculated by: elevP g h?? = ? (4. 13) where ?h is the water head, measured as the vertical distance between the upper and lower port centre line. The pressure drop in the manifolds and ports is calculated by an empirical equation (Shah and Sekulic, 2003): 2 port port 1.5 2 G P ? ? ?? ?? = ? ?? ? (4. 14) The frictional pressure drop within the core is the purpose of the investigation, and is usually by far the major part of the total pressure loss. The pressure drop measurement arrangement is given in Figure 3.8. The measured value of pressure drop is not directly the total pressure drop across a PHE but has extra contributions, as can by expressed by: measured PHE pipe fittings elevP P P P P? = ? + ? + ? ? ? (4. 15) Note that in this equation the term of the elevation pressure drop has a negative value; this is because the static pressure due to the water head in the transmitter connection tube will counteract that in the PHE. Combining Equation (4.12) and (4.15), we have: measured core port pipe fittingsP P P P P? = ? + ? + ? + ? (4. 16) The dimensions of the pipeline and the fittings are given in Figure 4.1. The pressure loss along the inlet and outlet pipelines, ?Ppipe, is calculated by 2 pipe pipe 1 2 LP f u d ?? ?? = ? ?? ?? ? (4. 17) where f can be obtained from the Moody Chart, or any fluid mechanics text book. For convenience of numerical calculation, the well-known Blasius equation is used which states: 0.25 5 0.3164 Re for smooth tube, 30000.01 1/6 m 1/3cold cold 0,cold cold cold wall,cold i 1/6 m 1/3hot hot 0,hot hot hot wall,hot i ( ) Re Pr ( ) Re Pr kh i dh kh i dh ? ? ? ? ? ?? ?? ?= ? ? ?? ?? ?? ?? ? ? ?? ?? ?= ? ? ?? ?? ?? ?? ? if 100count > Return Yes 0,hot cold hot 0,cold i 2' ( ) ( 1) 2 ( ) 2' ( ) 1 h T i T T T h e T i T i count count ? ? = + ?? ?? ?? ? = ? = + No 2 ( ) 40 0.1 0 T i e count = = = wall1 ( ) 2 (i) ( )T i T q i R= + ? [ ]122 ( ) 2 ( ) 2'T i T i T= + If iteration > 100 times, Stop and look for a better iterative formula. wall,hot wall,cold ( ,:) 1 ( ,:) 2 T j T T j T = = 0Cold 0Hot wall 1 1 1 x h h y R U = + = ? find the best fitting curve for: y ax= ( ) ( ) ( ) i i 2 i 2 i i 2 i ( ) 1 ( ) 1 x y a x SSE ax y SST y y SSERMSE j n SSER - square j SST ? = = ? = ? = ? = ? ? ? ? ? 1( )C j a = Least square regression at fixed m, to find the leading constant c in: ? ? ? ? = ? ? ?? ? 1/6 m 1/3 wall Nu Re Prc ( ) ( ) wall,cold wall,hot ( ) 2 ( ) ( ) 1( ) i T i i T i ? ? ? ? = = 231 Variables list (names are case-sensitive) Variable Name in Matlab Description Variable Name in Matlab Description A AHeatTransfer Total heat transfer area coldq qCold c,coldA AcCold hotq qHot Heat load, 2 p1 t t mq c dt A = ?? c,hotA AcHot Flow cross-section area wallR RWall Wall thermal resistance pc cp Specific heat Re Re Reynolds number hd dh Hydraulic diameter cold,inT TColdIn wallk kWall Wall thermal conductivity cold,outT TColdOut LMTD LMTD Log mean temperature difference hot,inT THotIn coldm? massFlowCold hot,outT THotOut Temperatures of hot and cold streams at inlet and outlet hotm? massFlowHot Mass flow rate ? beta Chevron angle Nu Nu Nusselt number wall? wallThickness Wall thickness Pr Pr Prandtal number ? fai Enlargement factor coldQ voluFlowCold ? miu viscosity hotQ voluFlowHot Volumetric flow rates ? rho density q q Heat load, ( )1 hot cold2q q q= + for i from 1 to 201 by 1 if ( ) min( )RMSE i RMSE= Yes No bestM i= ( ) ( ) wall,cold wall,hot 2 1 T T ? ? ? ? = = wall,hot wall,cold 1 ( ,:) 2 ( ,:) T T bestM T T bestM = = ( ) ( ) m M bestM c C bestM = = 1/6 m 1/3cold cold cold cold cold wall,cold 1/6 m 1/3hot hot hot hot hot wall,hot Re Pr Re Pr kh dh kh dh ? ? ? ? ? ? = ? ? ?? ?? ? ? ? = ? ? ?? ?? ? cold cold cold hot hot hot Nu Nu dhh k dhh k = ? = End 232 Appendix I Uncertainty Analysis: General Theory, the ASME and ISO Methods I.1 Introduction I.2 Basic Mathematics of Uncertainty Calculation I.3 Traditional Classifications of Errors: Random and Systematic Errors I.4 The ASME Method I.4.1 Types of Errors I.4.2 Precision Error, P I.4.3 Bias Error, B I.4.4 Combining Errors I.4.5 Uncertainty of a Parameter I.4.6 Uncertainty of a Result I.5 The ISO Method (GUM Method) I.5.1 Basis of GUM I.5.2 Terms Used in GUM I.5.3 Type A Evaluation of Standard Uncertainty, uA I.5.4 Type B Evaluation of Standard Uncertainty, uB I.5.5 Combined Standard Uncertainty I.5.6 Expanded Uncertainty, U Additional References I.1 Introduction All measurements have errors, due to measurement procedures, instrumental inaccuracy, and environmental influences. By definition, errors are the differences between the measurements and the true values. Even the most carefully calibrated instruments will have errors. In practice, the true values of measured quantities are rarely known, thus, the best one can do is to obtain the ?best estimate? of the true value, along with the estimate of the error. This is the task of uncertainty analysis. Uncertainty analysis is imperative when using data directly or in calibrating or validating simulation methods (Stem et. al., 1999). The fact is that until today, there is not a uniform and unambiguous method of uncertainty analysis, though some general mathematics applies to all possible methods. There are two widely accepted methods from two organizations: the ASME method and the ISO method. The main difference between the two methods is the way in which the random and systematic uncertainty estimates are handled (though the terms ?random? and ?systematic? are not recommended in the ISO method). Moffat (1982, 1985, and 1988) also proposed a method but his approach is not consistent with either of those standards. Other approaches 233 include those of NIST and AIAA, as briefly introduced by Coleman and Steele (1995). The ASME method was briefly described by Abernethy et al. (1985) and the updated revision can be found in the standard: ANSI/ASME PTC 19.1, 2005. ISO published the document ?Guide to the Expression of Uncertainty in Measurement? in 1993, revised in 1995 and usually referred to as the ?GUM? method. This method was developed by a joint committee of seven international organizations (BIPM, IEC, IFCC, ISO, IUPAC, IUPAP and OIML) and is gaining wider applications. Steele et al. (1994) made a comparison between the two methods applying the Monte Carlo techniques in a series of example experiments. It was shown that the ISO model is more consistent in providing an uncertainty interval. Coleman and Steele (1995) represented (but not with the same terminology) a discussion of the assumptions and approximations used in the method and also a brief overview of uncertainty analysis. Also found in Kirkup ?s web article, the ISO method is well summarized and represented. Another good literature source is the UKAS (United Kingdom Accreditation Service) document on measurement uncertainty (Publication reference: M3003, 2nd edition, 2007). The ISO method of uncertainty analysis is followed in this study. I.2 Basic Mathematics of Uncertainty Calculation Kline and McClintock (1953) showed (as described in the article by Moffat, 1988) that the uncertainty in a computed result could be estimated with good accuracy using a root-sum-square (RSS) combination of the effects of each of the individual inputs. This treatment is shown below. The result of the experiment is assumed to be calculated from a set of measurements using a data interpretation program: 1 2 n ( , , . . . , )R R X X X= (I. 1) The effect of an error in one measurement on the error of the calculated result is: ix i i RR X X ? ??= ? (I. 2) 234 The term i R X ? ? is often called the sensitivity factor (or sensitivity coefficient) and denoted as ?i. When several independent variables are used in the function R, the terms are combined by a root-sum-square method: 2 i 1 i n i RR X X ? ? ? ? ?? = ? ??? ?? (I. 3) This single equation is the basic equation of all types of uncertainty analysis. Equation (I. 3) applies when: 1. Each measurement (Xi) is independent, 2. Each measurement fits a Gaussian distribution, It would be straightforward in applying Equation (I. 3) to the following two basic conditions: 1. for 1 2R X X= + , ( ) ( )2 21 2R X X? ? ?= + (I. 4) 2. for 1 2R X X= , ( ) ( ) 2 2 2 2 1 2 2 1 1 2 1 2 X XR X X X X R X X ? ?? ? ? ? ? ? ?= ? + ? = ? +? ? ? ?? ? ? ? (I. 5) Note that Equation (I. 5) applies also to 1 2/R X X= , this can be proved by: 1 2 1 2 2 2 2 2 2 2 1 1 1 2 1 22 2 2 2 1 2 1 , , 1 XR R X X X X X X X XR X X X X X X X ? ?? ? ? ? ? = = ? ? ? ? ? ? ? ? ? ? ? ? = + = +? ? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? In general, when an equation describing the results is a pure ?product form?, then the relative uncertainty can be found directly. That is, if a b m 1 2 nR X X X= ? (I. 6) then 2 2 2 1 2 n 1 2 n X X XR a b m R X X X ? ? ?? ? ? ? ? ? ? = + + +? ? ? ? ? ?? ? ? ? ? ?? (I. 7) 235 This can be proved by: b m a-1 2 n 1 1 1 1 1 1 1 R aX X aX R X X XR X R a X X ?? ? = ? = ? ? ? ?? = ? ?? ? ? ? apply the same principle to other terms, substitute into Equation (I. 3), and the result is Equation (I. 6). There is no particular difficulty in the evaluation of Equation (I. 3), provided that each iX? is known. The central problems are the assignment of the proper value to the uncertainty of each input item and the selection of which items to be included in the analysis. This is where the various methods differ from each other. It is important to recognize that Equation (I. 3) applies regardless of different methods used for uncertainty analysis. In most situations, the overall uncertainty is dominated by only a few of its terms. Terms that are smaller than the largest term by a factor of 3 or more can usually be ignored. This is due to the RSS combination: small terms have very small effects (Moffat, 1988). I.3 Traditional Classifications of Errors: Random and Systematic Errors Errors are traditionally classified as random or systematic. Random errors (also called precision errors) follow a Normal distribution (Gaussian distribution), and so can be determined by statistical tools. Systematic errors (also called fixed errors, bias errors) are fixed in repeated trials (or readings) at a measurement point, and cause the measured value to be consistently above or consistently below the true value. Briefly, an instrument?s bias error is the difference between the average and true values, the precision error is the scatter about the bias. In other words, if only random errors exist then the mean of an infinite number of readings will be the true value. On the other hand, if only systematic errors exist then there is only one reading (no scatter and all readings are one) and that reading is away from the true value by the systematic error (so it is called ?bias?). Precision can be 236 improved only by selecting another measuring device, but bias error, if known, can be eliminated by correction. A particular device can be considered precise but biased, or biased but precise. I.4 The ASME Method I.4.1 Types of Errors At a single measurement point, the total error has two components, bias error (B) and precision errors (P). The total error is then expressed as: B P? = + (I. 8) Both the bias and precision are presumed to represent statistical properties of a Gaussian distributed data set. I.4.2 Precision Error, P The precision error P can be determined by statistical tools, from a number of repeated measurements. In the ASME method of uncertainty analysis, it is assumed that measurements without bias error will represent a Gaussian distribution around the true value of the parameter. Thus, the true value is the population mean. However, because the population has an infinite number of measurements, the population mean can only be estimated by a sample mean, and this brings the application of the Student?s t-distribution. The Student?s t-distribution (or t-distribution) is a probability distribution usually used to estimate the mean of a Normal population when the sample size is small. For a sample of size n drawn from a Normal population with mean ? and standard deviation ?, the quality / x t s n ?? = (I. 9) has a t distribution with n -1 degrees of freedom, where the x is the sample mean, s is the sample standard deviation and is called the ?precision index?: i 1 1 n i x x n = = ? (I. 10) 237 ( ) 2i 1 1 1 n i s x x n = = ? ? ? (I. 11) Since the true value of the variable is the population mean ?, the error between the measured value and the true value, defined as the precision error P, is: sP t n = (I. 12) The student?s t value is a function of the degrees of freedom of the data set, and can be found in statistics and probability reference books. For a sample size of n, the number of degrees of freedom is 1v n= ? . t value is always associated with a confidence level, it is a common practice to choose a t value where the confidence level is 95% in the two-tailed t-table. t is usually set equal to 2 for large samples, as a general rule, when 30n > . I.4.3 Bias Error, B The bias error B is constant during a given set. In repeated measurements of a given set, each measurement has the same bias. There is no statistical equation to calculate the bias error; instead, it must be estimated from the best available information (Moffat, 1982). The estimation of the bias error is not an easy task since the true value is not known. Calibration information and comparison of measurements by different methods can help but in general the estimate of bias must be based on judgment and is often subjective and hard to quantify or defend rigorously. Usually one must rely on the engineering judgment of instrumentation and measurement engineers to provide an upper limit on the bias. The bias errors include those which are known and can be calibrated out, those which are negligible and can be ignored and those which are estimated and included in the uncertainty analysis. An example of a bias error is the sensor uncertainty as specified by the sensor manufacturer. In some cases an instrument will be calibrated against a known standard, and the precision of the calibration (being a random error in the calibration procedure) can be used as an estimate of the bias error. 238 I.4.4 Combining Errors In the measurement of a parameter, errors can arise from many sources; these basically include calibration errors, data acquisition errors, and data reduction errors. For each of the sources there will be bias and precision components. To obtain the precision of each measured value (like temperature, pressure, or flow rate), the root sum square (RSS) method is used to combine the two components from the k sources of error. ( )2 2 21 2 K. . .P P P P= + + + (I. 13) ( )2 2 21 2 K. . .B B B B= + + + (I. 14) I.4.5 Uncertainty of a Parameter Now the precision error and bias error for a measured parameter are obtained. To express the total error for that parameter using a single number U, the two components need to be combined. The expression for the measured value, in the form of an interval within which the true value of the parameter is expected to lie, is: x U? (I. 15) Two methods can be used to combine the bias error with the precision error, addition of the two components (ADD) and combination by root sum square (RSS). It remained an argument as which method should be used till the late 1970s. The NBS (National Bureau of Standards, USA) suggested in late 1980 that either method should be accepted if: 1. the bias and precision components are propagated separately from the measurements to the final test result, 2. the method of combination is clearly stated. (Abernethy et al., 1985) It is not possible to define a rigorous confidence level for either method because the bias is an upper limit based on judgment which has unknown characteristics. It is usually recommended the ADD method has a 99% confidence level and the RSS has 95%. Thus 239 ADD 2 2 RSS with 99% confidence level with 95% confidence level U B P U B P = + = + (I. 16) I.4.6 Uncertainty of a Result A result is derived by measured parameters through the functional relationship, and errors of independent variables propagate into the result. The bias and precision error are kept separate until the last step of computing the uncertainty of a result. The precision and bias errors of a result are calculated separately by Equation (I. 3). Thus if the result is a function of independent parameters: 1 2 n( , ,. . . , )R R X X X= , the precision error of a result is given by: i 2 R 1 i n X i RP P X ? ? ?? = ? ??? ?? (I. 17) and the bias error of a result is given by: i 2 R 1 i n X i RB B X ? ? ?? = ? ??? ?? (I. 18) Again the total uncertainty of a result is given by the two models of ADD and RSS. I.5 The ISO Method (GUM Method) The ISO method is generally referred to as the GUM method, as an abbreviation of the standards (Guide to the Expression of Uncertainty in Measurement) that describes this method, and this term will be used in the following. I.5.1 Basis of GUM 1. Uncertainties are categorized in two ways according to how they are estimated: Type A: evaluated by the statistical analysis of a series of measurements, 240 Type B: evaluated by other means. The traditional classification of uncertainties, ?random? and ?systematic? errors is not recommended by GUM. 2. Combined uncertainty of the two types is calculated by taking the square root of the sum of the variances due to Type A and Type B uncertainties. 3. In situations where it may be necessary to quote a ?confidence level? within which the true value of the measurement is likely to lie, a coverage factor k shall be stated. I.5.2 Terms Used in GUM Standard uncertainty, u Uncertainty of the result of a measurement expressed as a standard deviation. Combined standard uncertainty, uc The positive root of the sum of the variances of the standard uncertainties of all contributing quantities (whether these be Type A or Type B uncertainties). Expanded uncertainty, U Quantity defining an interval which is expected to encompass a confidence level (usually 95%) Coverage factor, k This is used to obtain the expanded uncertainty, U, given the combined standard uncertainty, uc. Specifically, cU k u= ? (I. 19) For an expanded uncertainty where the confidence level is 95%, k usually lies between 2 and 3. I.5.3 Type A Evaluation of Standard Uncertainty, uA It is assumed that repeated measurements of a quantity X will represent a Gaussian distribution. The best estimate of the true value is found by the arithmetic mean of the measurements from n measurements: i 1 n i X X n = = ? (I. 20) 241 To show the variability of the measured values of the quantity X, we use the sample standard deviation: ( ) 2i 1 1 n i X X s n = ? = ? ? (I. 21) The measurement of the variability of the means (if an experiment consisting of measurements is repeated many times, then the distribution of means has its own standard deviation) is represented by the standard deviation of the means (often called standard error of the mean) as ( )s X : ( ) ss X n = (I. 22) In GUM ( )s X is called by a new term, standard uncertainty, and represented by the symbol, u. Thus, if X is the best estimate of X, then the standard uncertainty is written as ( )u X . I.5.4 Type B Evaluation of Standard Uncertainty, uB Where the best estimate of an input quantity is not determined by repeated measurements, we cannot use the type A evaluation methods to establish the standard uncertainty. In such a situation to evaluate the uncertainty we must use: 1. previous measurement data, 2. experience with or general knowledge of the behavior and properties of relevant materials and instruments, 3. calibration certificates, manufacturers? specifications, 4. data tables from handbooks. An uncertainty determined in this manner is referred to as a ?Type B? uncertainty. The proper use of available information for a Type B evaluation of standard uncertainty calls for insight based on experience and general knowledge, and is a skill that can be learned with experience. It should be recognized that a Type B evaluation of standard uncertainty can be as reliable as a Type A evaluation, especially when a Type A evaluation is based on a comparatively small number of statistically independent observations. Care must be taken to ensure that the uncertainty, as given in the manufactures? specifications, calibration certificates, and so on, is correctly interpreted. 242 ? If (from manufacturer?s specification, calibration certificate, handbook, or other sources) the quoted uncertainty is stated to be a multiple of a standard deviation, the standard uncertainty i( )u X is simply the standard deviation (the quoted value divided by the multiplier). ? If the quoted uncertainty is stated to have a confidence level of 90, 95, or 99%, one may assume a normal distribution of the quoted uncertainty, unless otherwise indicated, and recover the standard uncertainty of iX by dividing the quoted uncertainty by a appropriate factor. As for a normal distribution, the factor is 1.64, 1.96, and 2.58 for the above three levels of confidence. ? If the quoted uncertainty is stated as ?there is a x% chance that the quantity iX lies within the interval of a- to a+?, one may also assume a normal distribution of the uncertainty of iX and recover the standard uncertainty (standard deviation for that distribution). ? In cases where only the upper and lower limits are given, and there is no specific knowledge about the possible values of iX within the interval, one Figure I.1: Rectangular probability distribution Figure I.2 : Triangular probability distribution 243 can only assume that it is equally probable for iX to lie anywhere within it. This is a rectangular distribution. Denoting the upper limit as a + and the lower limit as a ? , and every value between those limits is equally probable to be the true value, as shown in Figure I.1. In a rectangular distribution the best estimate of 2? is given by: 2 2 2 3 a s ?= = (I. 23) where a is the half width of the distribution. The standard deviation, which is called the Type B standard uncertainty, u, of the rectangular distribution is: 3 a u s= = (I. 24) ? In cases when it is more realistic to expect that values near the bounds are less likely than those near the midpoint, a symmetric trapezoidal distribution having equal sloping sides (isosceles trapezoid) is assumed. Set the trapezoidal base width a+ ? a- = 2a, and a top of width 2a? where 0???1. As ? ?1 this trapezoidal distribution approaches the rectangular distribution (Figure I.1), while ? = 0 it is triangular distribution (Figure I.2). At this distribution, the expectation (best estimation) of iX is 12 ( )a a? ++ and the standard uncertainty is: 21 6 u s a ?+ = = (I. 25) ? If the measuring instrument has a resolution of x? , a rectangular distribution is assumed where the true value x can lie with equal probability anywhere in the interval / 2x x?? to / 2x x?+ . And thus the standard uncertainty is 0 .29 2 3 x u x ? ?= = . (Gum F.2.2.1) I.5.5 Combined Standard Uncertainty Combine type A uncertainty with Type B uncertainty for the same variable 244 This is the common situation where a quantity is measured using a calibrated sensor. The Type A uncertainty is obtained by repeat measurements and Type B uncertainty is evaluated by a manufacturer?s specification or calibration certificate. (GUM examples in 4.3.7 and 5.1.5). In this case, The GUM method treats the Type B uncertainty as providing bounds to an additive correction to the measured quantity X and thus the quantity is expressed as: X X X= + ? . The Type A uncertainty of X is expressed as ( )u X and Type B ( )u X? , respectively. Since the quantity X is now treated as the addition of two terms, the standard uncertainty of X is the combined standard uncertainty of Type A and Type B uncertainty (Equation I. 4): 2 2 c ( ) ( ) ( )u X u X u X= + ? (I. 26) Or, more generally, if the contributions to c ( )u y of the Type A and Type B standard uncertainties alone are denoted, respectively, by c, A ( )u y and c, B ( )u y , the combined standard uncertainty of y can be expressed as (GUM G.4.1, note 3): 2 2 c c, A c, B( ) ( ) ( )u y u y u y= + (I. 27) Combined standard uncertainty of independent input quantities If quantity Y depends on independent quantities (i.e., not related) 1X , 2X , ?, NX in the form: N( 1, 2, , )Y f X X X= ? (I. 28) The standard uncertainty of y, where y is the best estimate of the quantity Y (take note of the terminology) and thus the result of the measurement, is obtained by appropriately combining the standard uncertainties of the input estimates 1x , 2x , ?, Nx . The combined standard uncertainty of the estimate y is denoted by c ( )u y and given by: 2 c i 1 i ( ) ( ) N i f u y u x x = ? ?? = ? ??? ?? (I. 29) Note that this equation is the same as Equation (I. 3) (Equation 10 in GUM), and it can be deduced from this equation: 245 1. for 1 2Y X X= + , 2 2 c c 1 c 2( ) ( ) ( )u y u x u x= + (I.30) 2. for 1 2Y X X= , or 1 2/Y X X= 2 2 c c 1 c 2 1 2 ( ) ( ) ( )u y u x u x y x x ? ? ? ? = +? ? ? ?? ? ? ? (I.31) I.5.6 Expanded Uncertainty, U Expanded uncertainty U and combined standard uncertainty uc The standard uncertainty, uc, is the basic measure of uncertainty in the GUM method. However, there are situations in which it is preferable to quote an interval of confidence p (usually 95%) within which the true value will lie. Specifically, we can write: Y y U= ? (I. 32) where Y is the true value of the quantity, y is the best estimate of Y, and U is the expanded uncertainty. The equation is also expressed as y U Y y U? ? ? ? . The expanded uncertainty, U, is related to the combined standard uncertainty, c ( )u y , by the equation: c ( )U k u y= ? (I. 33) k is referred to as the coverage factor. In general, k is in the range 2 to 3. It should be recognized that multiplying c ( )u y by a constant provides no new information but presents the previously available information in a different form. It should also be recognized that in most cases the level of confidence p is rather uncertain, not only because the limited knowledge of the probability distribution characterized by y and c ( )u y , but also because of the uncertainty itself. (GUM, 6.2.3) Coverage factor in type A evaluation of uncertainty 246 If uc is determined through a Type A evaluation of uncertainty, it is usual to assume that the t distribution may be applied when determining the coverage factor, and c ( ) y Y t u y ? = . The Student?s t-distribution (or t-distribution) is a probability distribution usually used to estimate the mean of a Normal population when the sample size is small. For a sample of size n drawn from a Normal population with population mean ? and standard deviation ?, the quality / X t s n ?? = (I. 34) has a t distribution with n - 1 degrees of freedom. In the equation X is the sample mean, s is the sample deviation defined by: ( ) 2i1 1s x xn= ?? ? (I. 35) The student t value is a function of the degrees of freedom of the data set, for a sample size of n , the degrees of freedom is 1v n= ? . At the level of confidence of 95% (i.e., the probability that the true value lies within a specified interval is 0.95), Table I.1 gives the coverage factor for various degrees of freedom. For values of ? >10, k tends towards a value close to 2 and experimenters often use 2 as the coverage factor when the level of confidence required is 95%. For level of confidence of other values, for example, 99%, the t-table can be consulted to find out the k value with corresponding degrees of freedom. Specially, for ? >10 and p = 95%, k = 3. In the situation of 1 2 N( , , , )Y f X X X= ? where the estimate for each of 1 2 N, , ,X X X? is obtained from a different number of repeated observations, to obtain the coverage factor for c ( )u y , the effective degrees of freedom, effv , shall be calculated using the Welch-Satterthwaite formula: Table I.1: Coverage factors in Type A evaluation of uncertainty at level of confidence of 95% 247 4 c eff 4 i i ( ) ( )N i i u y u y ? ? = = ? (I. 36) where i? is the degrees of freedom of iX , and i c i i ( ) ( )fu y u x x ? = ? . Coverage factor in Type B evaluation of uncertainty If uc is determined through a Type B evaluation based on an assumed probability distribution, the k value is not evaluated by the t-distribution. For example, when ( )u x is obtained from a rectangular probability distribution of assumed half-width of a, ( )=a/ 3u x is viewed as a constant with no uncertainty (i.e., the uncertainty has no uncertainty for itself). The expanded uncertainty of a 95% level of confidence is simply 0.95 a, or 1 .65k = using the equation of ( )U ku x= . In the situation of N( 1, 2, , )Y f X X X= ? where i( )u x are obtained from both Type A and Type B evaluations, the probability distribution of the function ( ) c/ ( )t y Y u y= ? can still be approximated by the t-distribution. the Welch- Satterthwaite formula also applies, and the degrees of freedom for a standard uncertainty which is determined by a Type B evaluation is assigned by: 21 2 u u ? ? ?= ? ??? ? (I. 37) where u? is the standard deviation of the standard uncertainty, and 0u? = for rectangular distribution. At 0u? = it follows that ? = ? , in such a case, any term in the Welch-Satterthwaite formula with an infinite number of degrees of freedom is zero. Additional References Abernethy R.B., R.P. Benedict, R.B. Dowdell, ASME measurement uncertainty, Journal of Fluids Engineering, vol.107, pp.161-164, 1985 BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML, Guide to the Expression of Uncertainty in Measurement, ISO/1995, ISBN: 92-67-10188-9, First Edition 1993 Coleman H.W. and Steele W.G., Engineering application of experimental uncertainty analysis, AIAA Journal, vol.33, no.10, pp.1888-1896, 1995 248 Kline S.J., The purpose of uncertainty analysis, Journal of Fluid Engineering, vol.107, pp.153- 160, 1985. Kirkup L., Calculating and expressing uncertainty in measurement, Downloaded from: http://www.science.uts.edu.au/physics/uncertainty.pdf Moffat R.J., Contributions to the theory of single sample uncertainty analysis, Journal of fluids engineering, vol.104, pp.250-260, 1982 Moffat R.J., Using uncertainty analysis in the planning of an experiment, Journal of Fluids Engineering, vol.107, pp.173-178, 1985. Moffat R.J., Describing the uncertainties in experimental reports, Experimental Thermal and Fluid Science, vol.1, pp.3-17, 1988. Steele W.G., R.A. Ferguson, R.P. Taylor, H.W. Coleman, Comparison of ANSI/ASME and ISO models for calculation of uncertainty, ISA Transactions, vol.33, pp.339-352, 1994 Stem F., Muste M., Beninati M-L, and Eichinger W.E., Summary of experimental uncertainty assessment methodology with example, IIHR Technical Report No.406, Iowa Institute of Hydraulic Research, College of Engineering, The University of Iowa, 1999. UKAS, The expression of Uncertainty and Confidence in Measurement, UKAS Publication reference: M3003, 2nd edition, 2007. 249 Appendix J Uncertainty Analysis: Sensor Calibration The accuracy of a sensor can be obtained through either the manufacturer?s specification or a calibration procedure, and will contribute to the uncertainty of a measurement conducted by this sensor. There are two standards of uncertainty analysis one can follow: the ASME method and the ISO method. In this study, the ISO method of uncertainty analysis is used. J.1 Calculation Method J.2 Temperature sensors J.2.1 Mercury Thermometer J.2.2 LM35DZ Sensors J.2.3 Resistance Temperature Detectors (RTD?s) J.3 RS 257-026 Turbine Flow Meters J.4 Trimec MP15S Flowmeter J.5 Rosemount 3051 CD Differential Pressure Transmitter J.6 WIKA Static Pressure Transmitters J.1 Calculation Method For a sensor which measures the quantity x, the uncertainty (accuracy) of the sensor is obtained by both Type A and Type B evaluation of standard uncertainty. Type A evaluation has two components: ? A1( ) s u x n = , where s is the sample standard deviation, and n is the sample size ? A2 ( ) RMSE of calibrat ion curve fi t t ingu x = 1 Type B evaluation evaluates the standard uncertainty of the actual value of the quantity being measured, and is denoted by ( )Bu x . This means the actual flow rate for a flowmeter, the actual temperature for a temperature sensor, for example. 1 RMSE = Root Mean Squared Error. This error evaluates the goodness of the curve fitting. RMSE is calculated by ( )2i i'RMSE y y n p ? = ? ? , where iy is the original respond to ix , i'y is the response from the fitting curve ' ( )y f x= . n is number of data points, p is the number of unknown coefficients, called the degrees of freedom. 250 In most cases, the quantity of interest is a function of many other quantities, for example, in the water flowmeter calibration, the actual volume flow rate is calculated by : / ( )Q M t ?= ? , where M is the total mass of collected water. To be able to evaluate the uncertainty, a probability distribution must be assumed for each of the component quantities. The sensor?s accuracy, or the combined standard uncertainty, c ( )u x , is obtained by: 2 2 2 2 2 c A B A1 A2 B( ) ( ) ( ) ( ) ( ) ( )u x u x u x u x u x u x= + = + + (J. 1) J.2 Temperature sensors J.2.1 Mercury Thermometer Appendix C gives details of the mercury thermometer calibration, it is not to be repeated here. From the result of the calibration, the correction equation, as given by Equation (C.1), is reproduced as: actual reading1.004T T= ? (J. 2) Now assume that the water not being distilled and errors due to the lab barometer will bring an uncertainty of ? 0.1 ?C to the actual boiling temperature, also, the resolution of the thermometer will bring another error of ?0.029 ?C 1, thus at 94.3 ?C the mercury thermometer has a standard uncertainty of ? 0.104?C 2. This final error distributes linearly along the range of 0 to 94.3 ?C, for the thermometer itself, which is 0.0011 ?C per degree. At each reading of the thermometer, considering the resolution error of the thermometer (0.029 ?C), the standard uncertainty is thus: 0.11( ) Reding(corrected) 0.029 100 u T = ? + ?C (J. 3) 1 For sensor resolution of x? , under rectangular distribution, the standard deviation is 0.29 x? . 2 2 20.1 0.029 0.104+ = 251 which can be represented as: thermo min thermo max ( ) 0 .029 C, at reading=0 C ( ) 0.139 C, at reading=100 C u T u T = = ringoperator ringoperator ringoperator ringoperator (J. 4) J.2.2 LM35DZ Sensors Type A evaluation of standard uncertainty The sample size, standard deviation, and calibration RMSE for the four temperature sensors are: Sensors T1 T2 T3 T4 No. Sam, n 60 60 60 60 Max. Sam. Std. Dev., V 8.4?10-5 9.3?10-5 8.3?10-5 9.2?10-5 Max. Sam. Std. Dev. 1, ?C 0.0084 0.0093 0.0083 0.0092 Calibration curve fitting RMSE, ?C 0.0289 0.0297 0.0271 0.0282 The uncertainty of the four temperature sensors can now be calculated as: Sensor T1 T2 T3 T4 A,1 s u n = 0.001 0.001 0.001 0.001 A,2 RMSEu = 0.029 0.030 0.027 0.028 2 2 A A,1 A,2u u u= + 0.029 0.030 0.027 0.028 Type B evaluation of standard uncertainty This is the mercury thermometer uncertainty: B 0.0011 0.029 Cu T= + ringoperator Combined standard uncertainty 1 Converted from the reading Std. Dev. The reading standard deviation at each point is (in volts): Sensor T1 T2 T3 T4 Std. Dev. 8.4?10-5 9.3?10-5 8.3?10-5 9.2?10-5 The calibration equation for each sensor indicates that 1 V reading is compounding to 100 ?C. 252 It can be obtained from the above calculation that for all the four sensors, the standard uncertainty can be approximated as: ( ) A B 0.52 2 A B ( ) 0.03 C ( ) 0.0011 0.03 ( ) u T u T T u T u u = = ? + = + ringoperator The calibration of the four RTD?s has been in the range of about 0 to 60 ?C, similar to that for the LM35?s. The maximum uncertainty is also the same: ( )( ) [ ]( ) 0.522c max 0.03 0.0011 60 0.03 0.07524 Cu T = + ? + = ringoperator J.2.3 Resistance Temperature Detectors (RTD?s) Type A evaluation of standard uncertainty The sample size, standard deviation, and calibration RMSE for the four RTD?s are: Sensors RTD 1 RTD 2 RTD 3 RTD 4 No. Sam, n 60 60 60 60 Max. Sam. Std. Dev., V 10?10-5 10?10-5 7.3?10-5 9.9?10-5 Max. Sam. Std. Dev., ?C 0.013 0.013 0.027 0.027 Calibration curve fitting RMSE, ?C 0.026 0.020 0.033 0.029 The uncertainty of the four temperature sensors can now be calculated as: Sensor RTD 1 RTD 2 RTD 3 RTD 4 A,1 s u n = 0.0017 0.0017 0.0035 0.0035 A,2 RMSEu = 0.029 0.030 0.027 0.028 2 2 A A,1 A,2u u u= + 0.026 0.020 0.033 0.029 Type B evaluation of standard uncertainty This is the mercury thermometer uncertainty: B 0.0011 0.029 Cu T= + ringoperator 253 Combined standard uncertainty It can be obtained from the above calculation that for all the four sensors, the standard uncertainty can be approximated as: A B 2 2 A B ( ) 0.03 C ( ) 0.0011 0.03 ( ) u T u T T u T u u = = ? + = + ringoperator where T = 0 ~ 100 ?C, is the measured value. The result shows that the maximum uncertainty is with the maxim measured temperatures. In the current calibration the maximum temperature was around 60 ?C, the maximum uncertainty is then: ( )( ) [ ]( ) 0.522c max 0.03 0.0011 60 0.03 0.07524 Cu T = + ? + = ringoperator J.3 RS 257-026 Turbine Flow Meters Type A evaluation of standard uncertainty At maximum and minimum calibrated flow rates, The sample size, standard deviation, and calibration RMSE for flowmeter One and Two are: Flowmeter 1 Flowmeter 2 Flow rates minQ Q= maxQ Q= minQ Q= maxQ Q= No. Sam., n 564 136 532 136 Sam. Std. Dev 1, l/s 0.00084 0.0058 0.0014 0.0039 Calibration curve fitting RMSE, l/s 0.00151 0.00391 A,1 0.0058( ) 0.0005 l/ s , at maximum 136 s u Q n = = = A,2 0.0015 l / s , Flowmeter 1( ) RMSE 0.0039 l/ s , Flowmeter 2 u Q ?= = ?? 2 2 A A,1 A,2 0.0015 l / s , Flowmeter 1( ) 0 .0039 l/ s , Flowmeter 2 u Q u u ?= + = ?? 1 Converted from the reading Std. Dev., which is in V. 254 Type B evaluation of standard uncertainty The flow rate is calculated by: / ( )Q M t ?= ? , where M is the total mass of collected water. 1. The weight scale has a resolution of 0.1 kg, a rectangular distribution of the true value is assumed, for which it is equally probable for the true value X to lie anywhere within the band [ 0 .05 kg, 0 .05 kgX X? + ]. 2. Same distribution assumption is made for the time counted which is assigned with an uncertainty limit of 1s, and the density with an uncertainty limit of 1 kg/m3. Now the uncertainty of components for actual flow rates of flowmeter One and Two can be determined as: Value of iX Quantity iX unit Error limits, ix? Standard uncertainty, i i( ) 0.29u x x?= Flowmeter One Flowmeter Two M kg 0.1 0.029 min 61.25M = , max 62.35M = min 61.2M = , max 62.2M = t s 1 0.29 min 56.5t = , max 264.7t = min 56.5t = , max 278.3t = ? kg/m3 1 0.29 998.7 998.7 ( ) ( ) ( ) ( ) 22 22 2 2 B 2 2 ( ) 0.029 0.29 0.29 62 / 998.7 10 0.003 0.22 , in l/ s u M u t uu Q Q M t M Q Q Q ? ? ? ? ? ?? ? ? ? = + + = + +? ?? ? ? ?? ?? ? ? ?? ? = + (where ( ) 62/ 0.062 / 1000 M Q QQ? ? = ) Combined standard uncertainty The combined standard uncertainties are: Flowmeter Flowmeter 1 Flowmeter 2 A ( )u Q , l/s 0.0015 0.0039 B ( )u Q , l/s 2 210 0.003 0.22 , in l / sQ Q Q?? + c ( )u Q , l/s 2 2A Bu u+ 255 This result shows that the maximum uncertainty is with the maxim measured flow rates. In the current calibration the maximum flow was around 1.2 l/s. the maximum percentage uncertainty is then: ( ) 0.52 2 2 2 c max 0.0039 1.2 10 0.003 0.22 1.2 1.2 0.006522 u Q Q ?? ?? ?+ ? ? + ?? ?? ? ? ?? ? =? ?? ? = J.4 Trimec MP15S Flowmeter Type A evaluation of standard uncertainty At maximum and minimum calibrated flow rates, The sample size, standard deviation, and calibration RMSE for the Trimec MP15S flowmeter are: Flow rates minQ Q= maxQ Q= No. Sam., n 644 218 Sam. Std. Dev, l/min 0.0073 0.034 Calibration curve fitting RMSE, l/min 0.0172 A,1 0.034( ) 0 .0023 l /min , a t maximum 218 s u Q n = = = A,2 ( ) RMSE 0.0172 l /minu Q = = 2 2 A A,1 A,2( ) 0.0174 l /minu Q u u= + = Type B evaluation of standard uncertainty Following the same procedure as for the 257-026 Turbine Flow Meters, the flow rate is calculated by: / ( )Q M t ?= ? , where M is the total mass of collected water. 1. The weight scale has a resolution of 0.05 kg, assuming rectangular distribution of the weights, the true value X to lie anywhere within the band [ 0.025 kg, 0 .025 kgX X? + ]. 2. Same distribution assumption is made for the time counted which is assigned with an uncertainty limit of 1s, and the density with an uncertainty limit of 1 kg/m3. 256 Now the uncertainty of components for actual flow rates of the flowmeter is : Quantity iX unit Error limits, ix? Standard uncertainty, i i( ) 0.29u x x?= Value of iX M kg 0.05 0.015 min 21.1M = , max 20.9M = t s 1 0.29 min 140.9t = , max 582.7t = ? kg/m3 1 0.29 998.7 ( ) ( ) ( ) ( ) 22 22 2 2 B 2 2 ( ) 0 .015 0.29 0.29 21 / 998.7 10 0.006 0.00052 , in l /min u M u t uu Q Q M t M Q Q Q ? ? ? ? ? ?? ? ? ? = + + = + +? ?? ? ? ?? ?? ? ? ?? ? = + Combined standard uncertainty The combined standard uncertainties are: A ( )u Q , l/min 0.0174 B ( )u Q , l/min 2 210 0.006 0.00052 , in l /minQ Q Q?? + c ( )u Q , l/min 2 2A Bu u+ This result shows that the maximum uncertainty is with the maxim measured flow rates. In the current calibration the maximum flow was around 8.5 l/min. the maximum percentage uncertainty is then: ( ) 0.52 2 2 2 c max 0.0174 8.5 10 0.006 0.00052 8.5 8.5 0.002924 u Q Q ?? ?? ?+ ? ? + ?? ?? ? ? ?? ? =? ?? ? = J.5 Rosemount 3051 CD Differential Pressure Transmitter Type A evaluation of standard uncertainty The sample size, standard deviation, and calibration RMSE for the transmitter are: Water head 0 7.5 mh = ? 257 No. Sam., n 60 Max. Sam. Std. Dev., Pa 6.72 Calibration curve fitting RMSE, Pa 18.6 A,1 6.72( ) 0.87 Pa, at maximum 60 s u DP n = = = A,2 ( ) RMSE 18.6 Pau DP = = 2 2 A A,1 A,2( ) 18.6 Pau DP u u= + = Type B evaluation of standard uncertainty Differential pressure is calculated by: DP gh?= . 1. The Type measure has a resolution of 1 mm, a rectangular distribution of the true value is assumed, for which it is equally probable for the true value X to lie anywhere within the band [ 0.5 mm, 0 .5 mmX X? + ]. 2. Same assumption is made for the gravity which is assigned with an uncertainty limit of 0.01 m/s2, and the density with an uncertainty limit of 1 kg/m3. The uncertainty of components for actual differential pressure are: Quantity iX unit Error limits, ix? Std. uncertainty, i i( ) 0.29u x x?= Value of iX ? kg/m3 1 0.29 998.2 g m/s2 0.01 0.0029 9.79 h m 0.001 0.00029 min 0mh = , max 7.5mh = 2 22 2 2 2 B 2 8 ( ) ( ) ( ) ( ) 0 .29 0.0029 0.00029 998.2 9.79 / 2.917.1 10 DP in Pa u DP u u g u h DP g h DP g DP ? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? = + + = + + ? ?? ? ? ?? ?? ? ? ?? ? ? ? ? ?? ? ? ? ? ? ? ? = ? + ? ?? ? Combined standard uncertainty At any reading of DP where 30 73.3 10 PaDP? ? ? : 258 ( )Au DP , Pa 18.6 ( )Bu DP , Pa 2 8 2.917.1 10 , in PaDP DP DP ? ? ? ? ? + ? ?? ? c ( )u DP 2 2A Bu u+ The application of this device in the current study is in the range of 1-30 kPa in the current study. For this range: ( ) ( ) 0.52 22 8 c 3 DP=30kPa 18.6 30000 17.1 10 2.9 / 30000 30 10 0.00075 u DP DP ? ? ?? ?+ ? ? +? ?? ?? ? ? ?? ? =? ? ?? ? = ( ) ( ) 0.52 22 8 c DP=1kPa 18.6 1000 17.1 10 2.9 /1000 1000 0.0188 u DP DP ? ? ?? ?+ ? ? +? ?? ?? ? ? ?? ? =? ?? ? = J.6 WIKA Static Pressure Transmitters Type A evaluation of standard uncertainty The sample size, standard deviation, and calibration RMSE for the transmitter are: Weight pressure 0 - 840 kPa 0 - 840 kPa No. Sam., n 60 60 Max. Sam. Std. Dev., kPa 0.57 0.149 Calibration curve fitting RMSE, kPa 0.33 0.34 A,1 0.57( ) 0.074 kPa, at maximum 60 s u DP n = = = A,2 ( ) RMSE 0.33 kPau DP = = 2 2 A A,1 A,2( ) 0.34 kPau DP u u= + = Type B evaluation of standard uncertainty Differential pressure is calculated by: /P Mg A= . 259 1. The weights and the calibrator were in good maintained condition, it is safe to evaluated the relative uncertainty of the weights and the calibrator to be ? 0.5%. 2. Assuming rectangular distribution for g with an uncertainty limit of 1 kg/m3. The uncertainty of components for actual differential pressure are: Quantity iX unit Error limits, ix? Std. uncertainty, i i( ) 0.29u x x?= Value of iX M kg - - - A m2 - - - g m/s2 0.01 0.0029 9.79 22 2 2 2 2B ( ) ( ) ( ) ( ) 0 .00290.005 0.005 9.79 0.0071 u P u M u A u g P M A g ? ?? ? ? ? ? ? = + + = + + ? ?? ?? ? ? ?? ? ? ? ? ?? ? = Combined standard uncertainty At any reading of DP where 0 840 kPaDP? ? : ( )Au P , kPa 0.34 ( )Bu P , kPa 0.0071P ? c ( )u DP 2 2A Bu u+ The application of this device in the current study is around 300 kPa for R134a tests and 800 kPa for R507 tests. For this range: ( ) ( )( ) 0.522c P=300 0.34 300 0.0071 0.0072 300 u P P + ?? ? = =? ?? ? ( ) ( )( ) 0.522c P=800 0.34 800 0.0071 0.0071 800 u P P + ?? ? = =? ?? ? 260 Appendix K Uncertainty Analysis: Water Tests K.1 Calculation Method K.2 Experimental Results Mass Flow Rate, m? Log Mean Temperature Difference, ?TLM (LMTD) Differential pressure, ?P Heat Flux, q Overall Heat Transfer Coefficient, U Heat Transfer Coefficient, hc Reynolds Number, Re Nusselt Number, Nu Friction Factor, f K.1 Calculation Method For any measured quantity x, the standard uncertainty ( )u x is calculated by: 2 2 A B( ) ( ) ( )u x u x u x= + (K. 1) where A ( ) s u x n = , is the Type A evaluation, s is sample standard deviation, and n is sample size, B ( )u x is the Type B evaluation, taken as the sensor calibration uncertainty, or quoted uncertainty from the manufacturer. K.2 Experimental Results Mass Flow Rate, m? Mass flow rate of the water streams is calculated by: m Q?=? . The standard uncertainty of m? is calculated as: Hot stream (Flowmeter 1) Cold stream (Flowmeter 2) ( )u m? , kg/s ( ) /u m m? ? , % ( )u m? , kg/s ( ) /u m m? ? , % unit No. of data points min max min max min max min max Low angle 19 0.002 0.002 0.43 0.43 0.004 0.007 0.60 1.13 261 Mix angle 18 0.002 0.002 0.44 0.45 0.004 0.007 0.60 1.40 High angle 16 0.002 0.002 0.46 0.46 0.004 0.006 0.61 1.36 ( )u m? values are rounded to 0.001 kg/s Sample Calculation As an example, the calculation for first point in Low angle unit (19 points for this test) is shown below. The measurements are given as: Water stream Reading, l/s Std. Dev., l/s No. Sample Hot, Flowmeter 1 0.396 0.0006 40 Cold, Flowmeter 2 1.154 0.0013 40 For hot water: 5 A 0.0006( ) 9.5 10 l/ s 40 40 s u Q ?= = = ? 2 2 2 2 B ( ) 0.0015 10 0.003 0.22 =0.0017 l/ su Q Q Q?? ?= + +? ?? ? 2 2 c A B( ) =0.0017 l/su Q u u= + To calculate ( )u m? , Assume rectangular distribution of uncertainty band of 1 kg/m3 for density ? and so ( ) 30.29 kg/mu ? = : 2 2 2 2 c ( ) ( ) ( ) 0.0017 0.29 = 0.0043 0.396 1000 u m u Q u m Q ? ? ? ? ? ? ? ? ? ? = + + =? ? ? ? ? ? ? ?? ? ? ?? ?? ? ? ? c ( ) 0.396 1000 0.0042 0.0017 kg/su m = ? ? =? For cold water: A 0.0013( ) 0.00021 l / s 40 u Q = = 2 2 2 2 B ( ) 0.0039 10 0.003 0.22 0.0074 l/ su Q Q Q?? ?= + + =? ?? ? 2 2 c A B( ) =0.0074 l/su Q u u= + 2 2 c ( ) 0 .0074 0.29 = 0.0064 1.154 1000 u m m ? ? ? ? + =? ? ? ?? ? ? ? ? ? c ( ) 1.154 1000 0.0064 0.0074 kg/su m = ? ? =? 262 Log Mean Temperature Difference, ?TLM (LMTD) For clarity, the equation for log mean temperature difference is rewritten as: 1 2 1 2 LMTD ln( ) x xy x x ? = = (K. 2) where 1 1 h,in c,outx T T T= ? = ? , 2 2 h,out c, inx T T T= ? = ? . The combined uncertainty for LMT? is: 2 2 c 1 2 1 2 ( ) ( ) ( )y yu y u x u x x x ? ? ? ?? ? = +? ? ? ?? ?? ? ? ? (K. 3) where the partial differentials are: 1 2 1 2 11 1 2 2 1( ) 1 ln( ) ln( ) x x xy xx x x x ? ? ? = ? ? ? ?? ?? ? , 1 2 2 2 12 1 2 2 1( ) 1 ln( ) ln( ) x x xy xx x x x ? ?? ? ? ? ? ? ? = ? ?? ?? ? ?? ?? ?? ?? ?? ? Results for the current water test are: ( )u LMTD , ?C (round to 0.01) ( )u LMTD LMTD , % unit No. of data points min max min max Low angle 19 0.08 0.08 0.4 0.5 Mix angle 18 0.08 0.09 0.7 0.8 High angle 16 0.08 0.14 0.9 1.6 Sample Calculation As an example, the calculation for first point in Low angle unit test (19 points for this test) is given below. The measurement readings are: Sensor h,inT h,outT c,outT c, inT Reading, ?C 55.88 26.85 27.91 18.48 Std. Dev., ?C 0.008 0.009 0.06 0.005 263 Sample size 40 40 40 40 Calculation: 1 1 h,in c,out 27.97x T T T= ? = ? = , 2 2 h,out c, in 10.17x T T T= ? = ? = , ?C [ ] 1 2 1 2 2 11 1 2 2 1 1( ) 17.81 1 27.97 0.367 ln(2.75) ln(2.75)ln( ) ln( ) x x xy xx x x x ? ? ?? = ? = ? = ? ? ?? ?? ? [ ] 1 2 2 2 2 12 1 2 2 1 1( ) 17.81 1 10.17 0.722 ln(2.75) ln(2.75)ln( ) ln( ) x x xy xx x x x ? ?? ? ? ? ?? ? ? = ? ? = ? + =? ?? ? ?? ?? ?? ?? ?? ? For each sensor, from: B ( ) 0.04 0.001 reading Cu T = + ? ringoperator ( ) ( ) ( )h,in h,out c,out c, in( ) 0 .096, 0.069, 0.068, 0.058u T u T u T u T= = = = , ?C Uncertainty of individual temperatures: sensor h,inT h,outT c,outT c, inT A s u n = , ?C 0.001 0.001 0.01 0.001 B 0.04 0.001 readingu = + ? *, ?C 0.096 0.069 0.068 0.058 2 2 c Au u u= + B , ?C 0.096 0.069 0.069 0.058 * Bu is the sensor calibration uncertainty Uncertainty of temperature differences: 2 2 2 2 1 h, in c,out( ) ( ) ( ) 0 .096 0.068 0.118 Cu x u T u T= + = + = ringoperator 2 2 2 2 2 h,out c, in( ) ( ) ( ) 0 .069 0.058 0.090 Cu x u T u T= + = + = ringoperator Result: ( ) ( ) 2 2 2 2 c 1 2 1 2 ( ) ( ) ( ) 0 .367 0.118 0.722 0.090 0.078 C y y u y u x u x x x ? ? ? ?? ? = + = ? + ?? ? ? ?? ?? ? ? ? = ringoperator c ( ) 0 .078 0.0044 0.4% 17.8 / ln 2.75 u y y = = = 264 Differential pressure, ?P Result: ( ) , Pa u P? (Round to 1) ( ) / , % u P P? ? unit No. of data points No. of samples at each measuring point min max min max Low angle 17 40 19 22 0.2 1.9 Mix angle 16 40 19 27 0.2 2.4 High angle 18 40 19 33 0.1 1.0 Sample calculation For first point in Low angle unit test. Measurement readings are: Reading, Pa Std. Dev., Pa No. Sample 9475 49 40 For hot water: A 49( ) 7.7 Pa 40 s u DP n = = = 2 2 2 8 B 2.9( ) 18.6 17.1 10 19.2 Pau DP DP DP ? ? ?? ?? ?= + ? ? + =? ?? ?? ?? ? 2 2 c A B( ) =20.7 Pau DP u u= + c ( ) 20.7 0.0022 9475 u DP DP = = Heat Flux, q Heat flux is calculated by: 2 p1 1 t t q m c dT A = ?? (K. 4) For the calculation of the standard uncertainty, the equation is modified as: p 1q mc T A = ?? (K. 5) 265 Heat load can be calculated by either hot water side of the cold water side. A rectangular distribution is assumed for heat transfer area A with an uncertainty band of 3%, also a rectangular distribution is assumed for specific heat with an uncertainty band of 1%. Result: Hot stream Cold stream ( ) /u q q , % ( ) /u q q , % unit No. of data points min max min max Low angle 19 1.1 1.2 1.3 1.5 Mix angle 18 1.1 1.2 1.3 1.7 High angle 16 1.1 1.1 1.2 1.7 Sample Calculation For the first point in Low angle unit test, readings are: Quantity h,inT h,outT c ,outT c , inT hotm? coldm? unit ?C ?C ?C ?C kg/s kg/s Value, ix 55.88 28.65 27.91 18.48 0.396 1.154 i( )u x 0.10 0.07 0.07 0.06 - - i i( ) /u x x , % - - - - 0.43 0.64 Uncertainty of heat transfer area, ( ) 0 .29 0.03 0.0087u A A A A ? = = Uncertainty of specific heat, p p p p ( ) 0 .29 0.01 0.0029 u c c c c ? = = For hot stream: 2 2( ) 0 .1 0.07 0.0045 55.88 28.65 u t t ? + = = ? ? 22 2 2 2 2 2 2 p 2 p ( )( ) ( ) ( ) ( ) 0.45 0.43 0.87 0.29 100 0.011 u cu q u t u m u A q t m A c ? ?? + + +? ? ? ? ? ? = + + + =? ?? ? ? ? ? ??? ? ? ? ? ? ? ?? ? = ? ? For cold stream: 266 2 2( ) 0.07 0.06 0.0098 27.91 18.48 u t t ? + = = ? ? 2 2 2 2 2 ( ) 0.98 0.64 0.87 0.29 0.015 100 u q q + + + = = Overall Heat Transfer Coefficient, U U is calculated by qU LMTD = (K. 6) where q is calculated from the hot stream (lower uncertainty compared with the cold side). The standard uncertainty is: 22 LM LM ( )( ) ( ) u Tu U u q U q T ? ?? ? ? = + ? ?? ? ?? ? ? ? (K. 7) Result: ( ) /u U U , % unit No. of data points min max Low angle 19 1.2 1.3 Mix angle 18 1.3 1.4 High angle 16 1.4 1.9 Heat Transfer Coefficient, hc The film coefficient is obtained from: wall h c 1 1 1R U h h ? = + (K. 8) where ( )wall wall/R k?= . Only the cold side water stream is of interest, based on which the final correlation 267 wall Nu Re,Pr,f ? ? ? ?? ? = ? ?? ?? ?? ?? ? (K. 9) is drawn. Uncertainty of U and thus 1/U can be obtained from previous sections. The uncertainty of the wall resistance, Rwall, is calculated based on the assumptions that: 1. a rectangular distribution of wall thickness with an uncertainty band of 3%, and so: ( ) 0 .29 0.03 0.0087u ? ?? ? ? = = 2. a rectangular distribution of wall thermal conductivity with an uncertainty band of 1%, and so: ( ) 0 .29 0.01 0.0029u k k k k ? = = The relative standard uncertainty of wall resistance is thus: 2 2 2 2wall wall ( ) ( ) ( ) 0.0087 0.0029 0.0092u R u u k R k ? ? ? ? ? ? = + = + =? ? ? ?? ? ? ? The uncertainties of the film coefficients of both sides are coupled, to find the uncertainty of c( )u h some assumptions must be made. In the extreme case where c( ) 0u h = the uncertainty of cold side film coefficient will have its maximum value. From this assumption, it can be calculated: ( ) wallc c wall 1( ) 1 u Ru h U h R U ? = ? Result: c c( ) /u h h , % unit No. of data points min max Low angle 19 1.2 1.3 Mix angle 18 1.3 1.5 High angle 16 1.6 2.1 Reynolds Number, Re Reynolds number is calculated by: 268 h h c Re ud d Q A ? ? ? ? = = ? (K. 10) Assuming rectangular distribution for h c, , ,d A? ? with uncertainty bands of 0.1%, 1%, 3%, 3%, respectively, and it is obtained that: h h h h c c c c ( ) 0.29 0.001 0.00029 ( ) 0.29 0.01 0.0029 ( ) 0.29 0.03 0.0087 ( ) 0.29 0.03 0.0087 u u u d d d d u A A A A ? ? ? ? ? ? ? ? ? = = ? = = ? = = ? = = And: 22 2 22 2 ch h c ( )( )(Re) ( ) ( ) ( ) ( )0.00016 Re u Au du u u u Q u Q d A Q Q ? ? ? ? ? ?? ? ? ? ? ?? ? ? ? = + + + + = +? ?? ? ? ? ? ?? ? ? ?? ? ? ? ? ? ? ?? ? ? ? Result: c c(Re ) / Reu , % unit No. of data points min max Low angle 19 1.4 1.7 Mix angle 18 1.4 1.9 High angle 16 1.4 1.9 Nusselt Number, Nu Nusselt number is calculated by: hNu hd k = (K. 11) Assuming rectangular distribution for k, dh with uncertainty bands of 1%, 3%, respectively, it is obtained: h h h h ( ) 0.29 0.03( ) 0.29 0.01 0.0029, 0.0087u d du k k k k d d ?? = = = = 22 2 2 h h ( )(Nu) ( ) ( ) ( )0.00008 Nu u du u k u h u h k d h h ? ?? ? ? ? ? ? = + + = +? ?? ? ? ? ? ?? ? ? ? ? ?? ? 269 Result: (Nu) / Nuu , % unit No. of data points min max L angle 19 1.5 1.6 M angle 18 1.6 1.7 H angle 16 1.8 2.2 Friction Factor, f The Darcy friction factor is calculated by: 2 ch 2 21 h2 2/ / AdDP L DPf u d Q L? ?= = ? ? (K. 12) Assuming rectangular distribution for h c, , ,d L A? with uncertainty bands of 0.1%, 3%, 3%, 3% respectively, it is obtained: h h h h c c c c ( ) 0.29 0.001 0.00029 ( ) 0.29 0.03 0.0087 ( ) 0.29 0.03 0.0087 ( ) 0.29 0.03 0.0087 u u d d d d u L L L L u A A A A ? ? ? ? ? = = ? = = ? = = ? = = Use 1.4% as the relative uncertainty for Q : 22 22 2 2 2 ch h c ( )( )( ) ( ) ( ) ( ) ( ) ( )0.00042u Au du f u u L u Q u DP u DPf d L A Q DP DP ? ? ? ?? ? ? ?? ? ? ? ? ? ? ? = + + + + + = +? ?? ? ? ?? ? ? ? ? ? ? ?? ? ? ? ? ?? ? ? ?? ? ? ? Result: ( ) /u f f , % unit No. of data points min max L angle 17 2.1 3.1 M angle 16 2.1 3.1 H angle 18 2.1 2.3 270 Appendix L Uncertainty Analysis: Refrigerant Evaporator Tests L.1 Calculation Method L.2 Experimental Results Water Mass Flow Rate, w m? Heat Flux, q Log Mean Temperature Difference, ?TLM (LMTD) Overall Heat Transfer Coefficient, U Refrigerant Heat Transfer Coefficient, hr Refrigerant Mass Flow Rate rm? Outlet Vapour Quality, x Total Pressure Drop, ?Ptot Frictional Pressure Drop, ?Pfric Friction Factor, f L.1 Calculation Method For any measured quantity x, the standard uncertainty ( )u x is calculated by: 2 2 A B( ) ( ) ( )u x u x u x= + (L. 1) where A ( ) s u x n = , is the Type A evaluation, s is sample standard deviation, and n is sample size, B ( )u x is the Type B evaluation, taken as the sensor calibration uncertainty, or quoted uncertainty from the manufacturer. L.2 Experimental Results Water Mass Flow Rate, w m? Mass flow rate of the water streams is calculated by: ( )w wm Q?=? . The standard uncertainty of m? is calculated as: unit R134a R507 271 ( ) /u m m? ? , % ( ) /u m m? ? , % No. data min max No. data min max Low angle 58 0.38 0.43 22 0.42 0.42 Mix angle 58 0.38 0.42 18 0.42 0.42 High angle 59 0.39 0.42 20 0.42 0.42 Sample Calculation A sample calculation is given here for the first reading of the Low angle unit with R134a. The measurements are: Reading, l/s Std. Dev., l/s No. Sample 0.7977 0.000972 60 Result: 4 A 0.000972( ) 1.255 10 l / s 60 s u Q n ? = = = ? 2 2 2 2 B ( ) 0 .0015 10 0.003 0.22 =0.0034 l/ su Q Q Q?? ?= + +? ?? ? 2 2 c A B( ) =0.0034 l/su Q u u= + To calculate ( )u m? , Assume rectangular distribution of uncertainty band of 1 kg/m3 for density ? and so ( ) 30.29 kg/mu ? = : 2 2 2 2 c ( ) ( ) ( ) 0 .0034 0.29 = 0.0043 0.7977 1000 u m u Q u m Q ? ? ? ? ? ? ? ? ? ? = + + =? ? ? ? ? ? ? ?? ? ? ?? ?? ? ? ? Heat Flux, q Heat flux is calculated from: ( )p w1q mc TA= ?? (L. 2) A rectangular distribution is assumed for heat transfer area A with an uncertainty band of 3%, also a rectangular distribution is assumed for specific heat with an uncertainty band of 1%. Result: unit R134a R507 272 ( ) /u q q , % ( ) /u q q , % No. data min max No. data min max Low angle 58 1.8 4.3 22 2.2 3.4 Mix angle 58 1.8 4.3 18 2.1 3.4 High angle 59 1.7 4.3 20 2.1 3.5 Sample Calculation A sample calculation is given here for the first reading of the Low angle unit with R134a. The measurements are: Quantity w,inT w,outT wm? unit ?C ?C kg/s Value, ix 16.49 12.51 0.798 i( )u x 0.056 0.053 - i i( ) /u x x , % - - 0.42 Uncertainty of heat transfer area, ( ) 0 .29 0.03 0.0087u A A A A ? = = Uncertainty of specific heat, p p p p ( ) 0 .29 0.01 0.0029 u c c c c ? = = 2 2( ) 0 .056 0.053 0.01937 16.49 12.51 u t t ? + = = ? ? 22 2 2 2 2 2 2 p 2 p ( )( ) ( ) ( ) ( ) 1.94 0.42 0.87 0.29 100 0.0219 u cu q u t u m u A q t m A c ? ?? + + +? ? ? ? ? ? = + + + =? ?? ? ? ? ? ??? ? ? ? ? ? ? ?? ? = ? ? Log Mean Temperature Difference, ?TLM (LMTD) For clarity, the equation for log mean temperature difference is rewritten as: 1 2 1 2 LMTD ln( ) x xy x x ? = = (L. 3) 273 where 1 maxx T= ? , 2 minx T= ? . The combined uncertainty for LMT? is: 2 2 c 1 2 1 2 ( ) ( ) ( )y yu y u x u x x x ? ? ? ?? ? = +? ? ? ?? ?? ? ? ? (L. 4) where the partial differentials are: 1 2 1 2 11 1 2 2 1( ) 1 ln( ) ln( ) x x xy xx x x x ? ? ? = ? ? ? ?? ?? ? , 1 2 2 2 12 1 2 2 1( ) 1 ln( ) ln( ) x x xy xx x x x ? ?? ? ? ? ? ? ? = ? ?? ?? ? ?? ?? ?? ?? ?? ? Results: R134a R507 LM LM( ) /u T T? ? , % LM LM( ) /u T T? ? , % unit No. data min max No. data min max Low angle 58 0.8 1.9 22 0.8 1.2 Mix angle 58 0.9 2.3 18 1.0 1.9 High angle 59 1.2 2.9 20 1.2 2.3 Sample Calculation A sample calculation is given here for the first reading of the Low angle unit with R134a. Measurements: Sensor w, outT w, inT r , inT r , outT Reading, ?C 10.85 14.66 6.29 6.25 Std. Dev., ?C 0.002258 0.001306 0.03164 0.004135 Uncertainty of individual temperatures (sample size = 60): Sensor w, outT w, inT r , inT r , outT A s u n = , ?C 0.0003 0.0002 0.0041 0.0005 B 0.04 0.001 readingu = + ? *, ?C 0.0508 0.0547 0.0463 0.0462 2 2 c Au u u= + B , ?C 0.0508 0.0547 0.0465 0.0463 * Bu is the sensor calibration uncertainty 274 Uncertainty of temperature differences: 2 2 2 2 1 w, in r, out( ) ( ) ( ) 0.0547 0.0463 0.072 Cu x u T u T= + = + = ringoperator 2 2 2 2 2 w, out r, in( ) ( ) ( ) 0.0508 0.0465 0.069 Cu x u T u T= + = + = ringoperator Calculation: 1 max w, in r, out 8.41x T T T= ? = ? = , 2 min w, out r, in 4.56x T T T= ? = ? = , ?C [ ] 1 2 1 2 2 11 1 2 2 1 1( ) 3.851 1 8.41 0.41 ln(1.84) ln(1.84)ln( ) ln( ) x x xy xx x x x ? ? ?? = ? = ? = ? ? ?? ?? ? [ ] 1 2 2 2 2 12 1 2 2 1 1( ) 3.851 1 4.56 0.63 ln(1.84) ln(1.84)ln( ) ln( ) x x xy xx x x x ? ?? ? ? ? ?? ? ? = ? ? = ? + =? ?? ? ?? ?? ?? ?? ?? ? Result: ( ) ( ) 2 2 2 2 c 1 2 1 2 ( ) ( ) ( ) 0.41 0 .072 0.63 0.069 0.053 C y y u y u x u x x x ? ? ? ?? ? = + = ? + ?? ? ? ?? ?? ? ? ? = ringoperator c ( ) 0 .053 0.0084 0.84% 3.85 / ln1.84 u y y = = = Overall Heat Transfer Coefficient, U U is calculated by LMTD qU = (L. 5) The standard uncertainty of U is: 22 LM LM ( )( ) ( ) u Tu U u q U q T ? ?? ? ? = + ? ?? ? ?? ? ? ? (L. 6) Results: unit R134a R507 275 ( ) /u U U , % ( ) /u U U , % No. data min max No. data min max Low angle 58 2.0 4.7 22 2.4 3.6 Mix angle 58 2.1 4.9 18 2.3 3.9 High angle 59 2.2 5.1 20 2.4 4.2 Refrigerant Heat Transfer Coefficient, hr The film coefficient is obtained from: wall f w r 1 1 1 R R U h h = + + + (L. 7) where ( )wall wal l/R k?= , Rf is the fouling factor. Uncertainty of U and thus 1/U can be obtained from previous sections. The uncertainty of the wall resistance, wallR , was obtained in the water test as: 2 2 2 2wall wall ( ) ( ) ( ) 0.0087 0.0029 0.0092u R u u k R k ? ? ? ? ? ? = + = + =? ? ? ?? ? ? ? As a safe estimation, the percentage uncertainty of the fouling factor is set as 20%. Uncertainties of the water side heat transfer had been previously obtained in the water test as: w w( ) /u h h , % unit No. data points min max Low angle 19 1.2 1.3 Mix angle 18 1.3 1.5 High angle 16 1.6 2.1 The maximum values for w w( ) /u h h are to be used in this analysis. Uncertainties of the refrigerant side heat transfer coefficient can now be calculated: ( ) wall fr w r wall f w 1 1( ) 1 1 u R R u h U h h R R U h ? ? ? = ? ? ? (L. 8) Result: 276 R134a R507 r r( ) /u h h , % r r( ) /u h h , % unit No. data No. data min max Low angle 58 4.2 7.6 22 5.0 6.5 Mix angle 58 3.8 7.0 18 5.0 7.9 High angle 59 3.7 6.7 20 4.9 8.6 Refrigerant Mass Flow Rate rm? The R134a mass flow rate is calculated by: ( )r rm Q?=? . Assume a rectangular distribution of uncertainty band of 1 kg/m3 for density ? and so ( ) 30.29 kg/mu ? = . The standard uncertainty of m? is calculated as: R134a R507 ( ) /u m m? ? , % ( ) /u m m? ? , % unit No. data min max No. data min max Low angle 58 0.3 1.8 22 0.3 0.5 Mix angle 58 0.4 1.9 18 0.3 0.6 High angle 59 0.4 2.0 20 0.3 0.9 Outlet Vapour Quality, x The refrigerant outlet vapour quality is calculated from: r fgQ m x i= ? ? ?? ? (L. 9) where the inlet vapour quality is assumed as 0. Assume a rectangular distribution of uncertainty band of 1 kJ/kg for enthalpy of vaporization and so ( )fg 0.29 kJ/kgu i = . The results are: R134a R507 unit No. data out out( ) /u x x , % No. data out out( ) /u x x , % 277 min max min max Low angle 58 1.7 4.6 22 2.0 3.3 Mix angle 58 1.6 4.6 18 2.0 3.3 High angle 59 0.3 4.5 20 1.9 3.4 Total Pressure Drop, ?Ptot The total pressure drop is a direct measurement. The results are: R134a R507 to tP? , Pa to tP? , Pa unit No. data min max No. data min max Low angle 58 20 43 22 20 24 Mix angle 58 21 32 18 20 26 High angle 59 20 50 20 20 74 Frictional Pressure Drop, ?Pfric The frictional pressure drop is calculated by the measured pressure drop less the pressure drop through pipeline and all fittings, given by: fric measurement pipelineP P P? = ? ? ? (L. 10) Assume the calculation of pipelineP? has uncertainty of 20%. The results for the frictional pressure drop are: R134a R507 fr ic fr ic ( )u P P ? ? , % fr ic fr ic ( )u P P ? ? , % unit No. data min max No. data min max Low angle 58 7 12.4 22 9.5 12.4 Mix angle 58 5.1 12.1 18 10.8 12.5 High angle 59 5.0 11.6 20 8.2 12.8 278 Friction Factor, f The Darcy friction factor is calculated by: 2 ch 2 21 h2 2/ / AdDP L DPf u d Q L? ?= = ? ? (L. 11) Assuming rectangular distribution for h c, , ,d L A? with uncertainty bands of 0.1%, 3%, 3%, 3% respectively, it is obtained: h h h h c c c c ( ) 0.29 0.001 0.00029 ( ) 0.29 0.03 0.0087 ( ) 0.29 0.03 0.0087 ( ) 0.29 0.03 0.0087 u u d d d d u L L L L u A A A A ? ? ? ? ? = = ? = = ? = = ? = = Results: R134a R507 ( ) /u f f , % ( ) /u f f , % unit No. data min max No. data min max Low angle 58 7.3 12.5 22 9.7 12.5 Mix angle 58 5.6 12.2 18 10.9 12.6 High angle 59 5.4 11.7 20 8.4 12.9 279 Appendix M Heat Transfer Surface Partition for A DX Evaporator The heat transfer surface of the evaporator is divided into two regions: evaporating and superheating. To be able to calculate individual area of these two regions, single-phase heat transfer correlation must be known for refrigerant gas at the superheating region. Now consider only the most basic conditions: 1. no non-equilibrium between the two-phases, at x=1, flow enters the region of superheating, 2. no flow maldistribution, channel-average values of flow rate is taken for analysis, 3. saturation temperature is constant in the evaporating region, this will ignore the influence of pressure drop. the temperature profile in such an evaporator is schematically shown by Figure M.1 (Classon and Palm, 2002, Sterner and Sunden, 2006). The heat transfer rate superheating the vapour (single-phase) is: ( )sup r r, sup p, r r, o r, iQ m h c T T= ? = ?? ? (M. 1) where ?hr,sup is the refrigerant enthalpy change due to superheating. The temperature of water at the dividing point, Tw,b, can now be determined from: ( )sup w p, w w, i w, bQ m c T T= ?? ? (M. 2) The superheating area, supA , can now be determined from: Figure M.1:Temperature profile of a DX PHE evaporator 280 sup w r, gwall 1 1 1 U h k h ?? ? = + +? ?? ? (M. 3) sup sup sup lm, sup Q U A T = ? ? (M. 4) where lm, supT? is the log mean temperature difference at the superheating region. The total heat transfer area is usually known, and the evaporating area tpA can be obtained from: tot sup tpA A A= + (M. 5) 281 Appendix N Operating Instructions for the Refrigerant Evaporator Test Facility The experimental facility used in this study was designed and built as a liquid over-feed system. Refrigerant evaporator tests were carried out on three middle- size industrial brazed plate heat exchangers. The three units were installed in parallel, and the test was carried out individually for each unit. The detailed operating instruction of the facility, for each evaporator test, are given as following. i. Before the refrigeration system is started, the compressor should be checked for any refrigerant accumulation in the crank case. Liquid refrigerant in the crank case will boil upon starting the compressor as a result of sudden pressure drop, and the lubricant oil in the crank case will be brought into compressor cylinders by refrigerant foams and further into the system. A convenient method to avoid this situation is to install a heater inside the crack case, and to allow the oil to be warmed for some time, say 30 minutes, before starting the system. As a heater is not available for the current equipment, a mini flow torch was used. Before starting the compressor, the blow torch was placed on the underside of the compressor just below the sight glass of the crank case. If the fluid as seen from the sight glass begins to boil then there is refrigerant in the compressor. Allow the fluid to boil until the boiling appears to have stooped completely. Once this has occurred then there should be only oil left in the crank case. ii. Start the water pump, open the inlet and outlet valves for the evaporator to be opened while also making sure the water line valves for other evaporators are closed. iii. Start the compressor of the refrigeration system, open the inlet and outlet valves for the evaporator while also making sure the refrigerant line valves for other evaporators are closed. iv. Switch the data acquisition system, launch the measurement software, and carried out a self-calibration of the software. 282 v. Adjust the water flow rate at a fixed value at approximately 0.8 kg/s, and maintain at this flow rate for all tests. This is done by operating the bypass valve and the main line valve. vi. Fully open the refrigerant bypass valve to achieve the maximum possible flow rate of the refrigerant, allow the system to run 40 minutes to stabilize. vii. Once the system is stabilized, record the first reading by the data acquisition system. the data recording takes 30 seconds, as the sample rate is 2, therefore 60 data samples are taken, for each measured quantity. viii. Reduce the refrigerant flow rate by adjusting the bypass valve, to preset values. Allow the system to run for approximately 10 minutes to stabilize, once the system is stabilized, another reading is to be taken. ix. Follow the previous step until the flow rate is reduced to the lowest possible of the system. x. Once all the readings are taken for the first cooling load, reduce the system cooling load, repeat steps vi ? ix. Three methods were used for controlling the cooling load: (1) by the surge drum back pressure regulator, (2) by the surge drum isolating valve, and (3) by damping the condenser surface. Of these the method (3) proved the most effective. Dampers made of cardboard strips were used, those strips were placed to over the condenser surface and the air flow rate across the condenser was controlled by adjusting the number of dampers. xi. After all readings are saved, shut down the system and close all inlet and outlet refrigerant valves. xii. Steps i -xi are to be repeated for other evaporators.