Real-Time Implementation of Data-Driven Predictive Controller for an Artificial Muscle

Abstract

This study presents a position tracking control method, with reference to Data-Driven Predictive Controller (DDPC), for a Pneumatic Artificial Muscle (PAM) system. The design of predictive controller is created from the subspace identification matrices acquired by input/output data. The control scheme is entirely data-based without explicit use of a model in the control application that can rectify the nonlinearity and uncertainties of the PAM. Firstly, subspace matrices are developed employing the identification method as a predictor by using open-loop experiments. Secondly, the estimated subspace matrices are used to design the so-called DDPC. In this instance, the quadratic programming (QR) decomposition method is used to obtain the prediction matrices. Consequently, experiments are carried out by the PAM actuator with different testing and loading conditions. The real-time experimental results demonstrate the feasibility and efficiency of the suggested control approach for nonlinear systems.