Adaptive Control of Hydraulic Systems with MML Inferred RBF Networks

In this paper the problem of adaptively controlling a hydraulic system with uncertainties is considered. An adaptive controller is derived to control actuator force with unknown valve flow coefficients and fluid parameters. This is subsequently cascaded into a position controller which uses RBF networks to compensate for the effects of friction in the system. In contrast to conventional adaptive controllers, the controller is augmented with a further layer that adaptively determines the optimal architecture for the RBF networks using the Minimum Message Length costing criterion. This provides an automated method of determining when it is no longer advantageous to increase the network size. Stability results are presented, and simulation demonstrates the ability of the MML criterion to determine when a suitable fit has been achieved.