Neural-network-based adaptive sliding mode control with reinforcement learning for robust pneumatic valve regulation

Abstract

Inherent model uncertainties in pneumatic servo systems as well as issues such as dead zone hysteresis caused by gas compressibility and friction pose notable challenges in the design of high-performance and high-precision regulators. Therefore, this study proposes a neural-network-based adaptive sliding mode control algorithm to address critical engineering problems encountered during the operation of pneumatic regulating valve systems such as external disturbances, dead zones, and time-varying mass. To eliminate dead zone errors, an error compensation auxiliary signal and sliding mode structure are initially designed, and the Lyapunov method is then employed to theoretically demonstrate that the proposed algorithm can stabilize a pneumatic system within a finite time. Subsequently, reinforcement learning is utilized to optimize the key parameters of the proposed method in real-time, enabling the adaptive control of the system in response to environmental disturbances, thereby enhancing system robustness. Finally, results show that the proposed controller effectively reduces dead zone hysteresis and suppresses vibrations caused by environmental disturbances during the regulation of a pneumatic valve, thereby considerably improving control performance.