Performance Evaluation of Neural Network and RBF-Based Controllers for Pneumatic Artificial Muscles

This research addresses the significant control challenges inherent in the Pneumatic Artificial Muscle (PAM) systems such as high nonlinearities, hysteresis and parametric uncertainties which conventional control methods struggle to mitigate effectively. The primary objective was to design and comparatively evaluate two advanced adaptive control strategies namely the Neural-Second-Order Sliding Mode Control (Neural-SOSMC) and the Radial Basis Function-based Second-Order Sliding Mode Control (RBF-SOSMC). Their performance was benchmarked against a conventional SOSMC algorithm using a comprehensive MATLAB/Simulink model across two distinct scenarios. The results unequivocally demonstrate the superior performance of the adaptive controllers. In the first scenario the Neural-SOSMC controller exhibited a remarkably fast rise time of 0.2306 s and a minimal steady-state error of 0.0093, while the RBF-SOSMC achieved a slightly higher rise time of 0.3509 s with an overshoot of 54.0516%. In the second scenario the RBF-SOSMC demonstrated the fastest transient response with a rise time of 0.3832 s and an overshoot of 54.0877%, whereas the Neural-SOSMC achieved the lowest steady-state error of 0.0046. The comparative analysis reveals that while the RBF-SOSMC excels in minimizing transient response time, the Neural-SOSMC provides a more robust and precise solution for steady-state tracking. The study concludes that intelligent control methods like Neural-SOSMC and RBF-SOSMC are highly effective in overcoming the inherent complexities of PAM systems, providing robust and precise solutions for advanced robotic applications.