Online Self Tuning of Parameters of a PID Controller that Uses a Radial Basis Neural Network

A proportional-integral-derivative (PID) controller is one of the most popular and commonly used controllers. Although this controller has been established as a control standard, still it has to cope with some difficulties. Tuning the parameters (proportional, integral and derivative gains) of a PID controller manually requires a large experience and can be a tedious task. In this work, we propose an optimization based approach to automatically tune these three parameters as the system is driven towards its desired behaviour. The parameters of the PID controller are tuned using a neural network (NN) with a radial basis (RB) activation function, while the parameters of the NN are optimized using a stochastic gradient descent (SGD) algorithm. This enables the system to learn online in realtime. Further, this method is tested in Simulink environment on a benchmark of the vibration suppression for a clamped-free flexible aluminum beam. The starting point for the controller design is the model of the beam obtained through the subspace model identification. Further on, using the NN the model update is performed along with the PID parameter optimization.