ASOFNN-Enhanced Sliding Mode Control of Soft Actuators: Design and Experimental Validation

Abstract

This paper proposes an adaptive fast terminal sliding mode control (FTSMC) strategy for a self-designed soft actuator, in which a vision-based non-contact measurement method is employed to detect the end-point position of the soft actuator and convert it into the actuator’s bending angle. An Euler-Lagrange method is used to model the soft actuator, and a gray-box experiment is conducted to identify the system parameters. Furthermore, in order to counteract the time-varying nonlinearity and the susceptibility to disturbances in the soft actuator, an extended state observer (ESO) is incorporated into the adaptive FTSMC, in which an adaptive self-organizing fuzzy neural network (FNN) is applied to adjust the sliding function parameters adaptively, such that the transient performance and system’s steady-state error can be improved. Besides, statistical methods are used to dynamically adjust the number of fuzzy rules based on the spiking intensity of these rules, thus enhancing the performance of the FNN. The ESO parameters are optimized using the elite ant lion optimization algorithm and the optimal learning rates in the designed adaptive self-organizing FNN are determined via a grid searching algorithm. Finally, both simulations and physical experiments are carried out to illustrate the proposed control strategy for the soft actuators. Note to Practitioners—Soft actuators are widely employed in various applications such as industrial precision component grasping and fragile fruit sorting. In practical applications, the control of soft actuators is often compromised by inherent nonlinear dynamics and sensitivity to external disturbances, which result in an unsatisfactory control performance. In this paper, we propose an adaptive self-organizing fuzzy neural network (ASOFNN) to dynamically tune the sliding function parameters of FTSMC, thereby achieving superior control performance. As a result, the proposed control method deployed on the experiment platform ensures that the tracking error of the bending angle converges to a neighborhood of the origin within a finite time. Besides, we adopt a statistically-driven dynamic threshold instead of the conventional fixed-threshold method in self-organizing mechanism. When applied to practical systems with varying external disturbances and system states, the statistically-driven method is observed to result in more uniform and reasonable spiking intensities for each fuzzy rule, consequently leading to enhanced FNN performance.