A Normalised Gradient Descent PI Controller For Speed Servomechanism

Performance of digitally controlled speed servomechanisms is very important for the overall performance of modern motion control systems. Proportional and integral controller is simple, yet with clear physical interpretation. This controller, even when optimally tuned, due to fixed values of parameters might lead to a poor performance of the overall system. One way to resolve this problem is to use gradient descent optimisation of controller parameters, with simple update rules for a real time operation, and with intuitive graphical interpretation. However, gradient descent algorithms have slow convergence and decreased performance when operate in nonstationary and/or nonlinear environment. Learning rate normalisation gives time varying learning rate and minimised a posterior output error. Due to these facts, normalised gradient descent proportional and integral controller for digitally controlled speed servomechanism is presented in this paper. The recommended values of the algorithm parameters, based on convergence analysis, are proposed. Results of the experiments support the analysis.