Control of Electro-Hydraulic Poppet Valves via Online Learning and Estimation

Abstract

Online learning state trajectory control applied to Electro-Hydraulic Poppet Valves (EHPV) is considered herein. The control problem is to track a desired flow conductance coefficient KV for pressure or flow control applications. In general terms, the control methodology employed herein computes the input signal sent to the valve from the addition of three components. The first component comes from an experimentally approximated inverse input-output map of the system which gives a nominal input. The second component is computed through a neural network structure called the Nodal Link Perceptron Network that learns online the adjustment of this nominal map. The third component is an adaptive proportional feedback control input. This last component uses two system parameters known as the Jacobian and the Controllability parameter, which are estimated online via a recursive least squares algorithm with forgetting factor. The proposed controller is explored through experimental data on a pressure control application and the results are discussed.