Pattern recognition in hydraulic backlash using neural network

An approach for estimating and classifying backlash clearance fault condition in hydraulic actuators is presented. Three networks (ADALINE network, a nonlinear neuron network, and multilayer perceptron network) are trained and applied to an experimental hydraulic system to identify the gap between an actuator pin and a load mass. The networks are trained on five clearance gaps of widths 1, 7, 12, 25, and 40 thousandths of an inch. They are tested on three clearance gaps of widths 10, 20, and 35 thousandths of an inch. The multilayer perceptron network performed very well in all testing. The other two networks did not perform well, except for small gaps.