Development of a Soft Sensor to Estimate Fermentation Rate Using Temperature Measurements During Ethanol Fermentation
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of the Witwatersrand, Johannesburg
Abstract
This dissertation presents the development of a soft sensor designed to estimate ethanol fermentation rates using only temperature-derived measurements. The sensor was constructed through a delta-based machine learning approach, tracking the rate of change rather than absolute values, with thermodynamic reasoning embedded into the feature engineering process. Ten fed-batch fermentation experiments were conducted using Saccharomyces cerevisiae in a 100 L stainless-steel bioreactor, with data collected on fermenter, jacket, and ambient temperatures, alongside specific gravity. Ethanol concentrations were estimated via a validated model based on the work of Cutaia et al. (2009) and cross-verified against HPLC readings, achieving a combined R2 of 0.75 and NRMSE of 0.17 across nine experiments. A Gaussian Process Regression (GPR) model, trained on four delta-based input features and smoothed via a 40-point moving average, emerged as the top performer. The final model reduced RMSE by 79.18%, from 5.62 g/L (ABS) to 1.17 g/L (DELTA) and achieved a maximum R2 of 0.81 during validation. Testing on three unseen experiments, including disturbed and feedback-controlled scenarios, demonstrated robust trend tracking, with the best case yielding an RMSE of 0.34 g/L and R2 of 0.50. This research confirms the viability of using thermal data to infer yeast activity and ethanol production in real- time, offering a low-cost, non-invasive alternative to traditional analytical methods for fermentation monitoring in industrial environments.
Description
A research report submitted in fulfillment of the requirements for the Master of Science in Engineering, in the Faculty of Engineering and the Built Environment, School of Chemical and Metallurgical Engineering, University of the Witwatersrand, Johannesburg, 2025
Citation
Naidoo, Adam Kyle. (2025). Development of a Soft Sensor to Estimate Fermentation Rate Using Temperature Measurements During Ethanol Fermentation [Master’s dissertation, University of the Witwatersrand, Johannesburg]. WIReDSpace. https://hdl.handle.net/10539/50264