Neural network-based PID compensation for nonlinear systems: ball-on-plate example

In this paper, a neural network (NN)-based feedback controller is proposed in order to compensate for the errors caused by using an approximated dynamic model in controller design. The controller consists of two subcontrollers working in parallel: base linear controller and NN-based PID compensator. The former can be any controller that is easily designed based on the system’s linearized or simplified model. The latter is based on a PID controller with adjustable gains and a neural network is used to update the PID gains during control process, aiming to compensate for the nonlinear effects ignored in the base controller. The performance of the proposed controller is demonstrated with a ball-on-plate system built for this study. Approximate feedback linearization is used to design the base controller in this work and the NN-based PID compensator is used in parallel. Simulation and experimental results that achieve better stabilization and trajectory tracking performance are provided and discussed.