Integrated on-line recognition and control for adaptive hysteresis compensation in pneumatic muscle actuators

Pneumatic Muscle Actuators (PMAs) present a persistent challenge in high-precision motion control due to their pronounced hysteresis nonlinearity. While conventional Proportional-Integral-Derivative (PID) controllers based on offline-identified hysteresis models are widely used, they fail to capture the time-varying and load-dependent nature of PMA hysteresis, leading to significant modeling and tracking errors. To overcome this limitation, this paper introduces a novel integrated control architecture that, for the first time, tightly couples real-time hysteresis identification with adaptive PID regulation. The proposed method retains the simplicity of the classical Prandtl–Ishlinskii (PI) hysteresis model but employs an online parameter estimator-specifically the Forgetting-Factor Recursive Least Squares (FFRLS) algorithm—to continuously update the model under varying loads and frequencies. Furthermore, three distinct adaptive PID controllers are developed and experimentally compared, whose gains are adjusted online via a Back-Propagation (BP) neural network, a Radial Basis Function (RBF) neural network, or a fuzzy-logic system. Extensive trajectory-tracking tests demonstrate that: the FFRLS-based online identification effectively adapts to load changes and achieves the highest modeling accuracy among the tested online methods; all three adaptive PID variants significantly outperform the conventional PID controller; the BP-based adaptive PID controller delivers the best overall performance-requiring no manual tuning, exhibiting superior robustness against both load and frequency disturbances, and reducing tracking error by more than 40% compared to the classical PID approach. Thus, this work establishes a practical, readily implementable control solution that exploits the synergy between online hysteresis compensation and adaptive PID tuning, offering a robust framework for precision motion control of PMA-driven systems in real-world applications.