Real-time fuzzy logic control of a smart material actuated, robotic prosthetic hand with grip compensation

Abstract

In recent years, several attempts have been made towards the control of a robotic prosthetic-hand for lower-arm amputees. However, these attempts have often been limited as they do not consider the entire coupled system during the development and tuning of the control system, spanning electromyographic signal input to positional and force output of the robotic prosthetic hand. Considering this, a robotic prosthetichand model is developed encompassing a back-propagation artificial neural network electromyographic grasp classification model, electromyographic-force regression model, non-linear autoregressive with exogenous excitation shape memory alloy model, anthropomorphic hand mechanism and weighted Mamdani fuzzy logic control system. The performance of the control system was assessed based on the error in positional and force tracking of the hand in achieving the various grasps and forces as interpreted from the electromyographic signals. Overall, the error in position as well as force evaluated to less than 6% of the setpoint, thereby demonstrating the feasibility of using fuzzy logic in the control of a shape memory alloy actuated robotic prosthetic-hand. The controllers’ success and legitimacy were partly due to the very effective electromyographic grasp classifier which made use of extracted Hudgin’s features to yield a postprocessed classification rate of 100%. Additionally, the controllers’ ability to track force hinged on the developed electromyographic-force model yielding an acceptable MSE of 0.293N and R2 of 0.73. Though acceptable, the model yielded a less than expected correlation due to the scattering of electromyographic data points for any singular force recorded. Critical to the angular movement of the robotic prosthetic-hand and consequent position tracking was the highly non-linear shape memory alloy actuator. Characteristics such as pseudo-elasticity, shape memory effect, hysteresis, resistive heating, stress effects and convection were all found to be well captured in the closed-loop non-linear autoregressive with exogenous excitation model, resulting in an RMSE of 0.04. Apart from the aforementioned systems, the anthropomorphic hand itself was modelled with linkages and end-point trajectory mimicking realistic hand movement. The overall system was controlled by a weighted Mamdani fuzzy logic controller. The controller was iteratively tuned and was found to effectively control the robotic prosthetic hand in both movement and force. The findings and the developed models may be used to develop an embedded system for a robotic prosthetic-hand. Furthermore, the work may be extended to other fields of engineering including smart materials research, bioinformatics, and control systemsdevelopment.