Comparison of Sliding-Mode Control and Model-Predictive Control Algorithms for Pneumatic Actuators

Abstract

Pneumatic actuators are low-cost, clean, safe and provide a high power to weight ratio. In this thesis, the modeling and position control of pneumatic actuators is presented. Sliding-mode control and model-predictive control algorithms are compared. The actuator’s main components consist of a double acting pneumatic cylinder and four two-way on/off valves. A nonlinear system model was developed. Its parameters were estimated from experiments. A novel friction model was presented and shown to be superior to the classical friction model. The system model was validated by comparing simulation and experiment results. Three novel nonlinear control algorithms are designed and compared with two existing state-of-the-art sliding-mode control (SMC) algorithms. Two of the novel algorithms are modified versions of the existing SMC algorithms. The third is a discrete-valued model predictive control (DVMPC) algorithm. The designs and performance of the five control algorithms were compared. Simulations and experiments demonstrated that the two modified SMC algorithms reduced both the position tracking errors and the valve switching frequency. Reducing a valve’s switching frequency has the benefit of prolonging its life. In the experiments on a cylinder with high friction seals, the steady state errors reduced 42% to ±0.3 mm. The valve switching frequency was also reduced by 34%. The switching frequency was further reduced by 32%, without significantly affecting the tracking performance, by incorporating a 5 ms zero-order hold. The five algorithms were simulated and compared on a high friction cylinder and a low friction cylinder to demonstrate their generality. In the simulations, the position tracking performance with DVMPC was similar to the best SMC algorithm, and the valve switching frequency with DVMPC was 34% lower. The three novel controllers were shown to be robust to increased and decreased payload mass. The DVMPC calculation times demonstrated that future experimental implementation of DVMPC is possible.