A Data-Driven PID Controller For Flexible Joint Manipulator Using Normalized Simultaneous Perturbation Stochastic Approximation

• Main Authors: Mustapha, Nik Mohd Zaitul Akmal, Suid, Mohd Helmi, Raja Ismail, Raja Mohd Taufika, Nasir, Ahmad Nor Kasruddin, Ahmad, Mohd Ashraf, Mohd Tumari, Mohd Zaidi, Jui, Julakha Jahan
• Format: Article
• Language: English
• Published: Penerbit Universiti Teknikal Malaysia Melaka 2019
• Online Access: http://eprints.utem.edu.my/id/eprint/25191/2/SPSA%20FLEXIBLE%20JOINT.PDF; http://eprints.utem.edu.my/id/eprint/25191/; https://jamt.utem.edu.my/jamt/article/view/5514/3789

This paper presents a data-driven PID controller based on Normalized Simultaneous Perturbation Stochastic Approximation (SPSA). Initially, an unstable convergence of conventional SPSA is illustrated, which motivate us to introduce its improved version. The unstable convergence always happened in the data-driven controller tuning, when the closed-loop control system became unstable. In the case of flexible joint manipulator, it will exhibit unstable tip angular position with high magnitude of vibration. Here, the conventional SPSA is modified by introducing a normalized gradient approximation to update the design variable. To be more specific, each measurement of the cost function from the perturbations is normalized to the maximum cost function measurement at the current iteration. As a result, this improvement is expected to avoid the updated control parameter from producing an unstable control performance. The effectiveness of the normalized SPSA is tested to the data-driven PID control scheme of a flexible joint plant. The simulation result shows that the data-driven controller tuning using the normalized SPSA is able to provide a stable convergence with 76.68 % improvement in average cost function. Moreover, it also exhibits lower average and best values for both norms of error and input performances as compared to the existing modified SPSA.