Estimating Parameters of a Proportional Solenoid Valve using Neural Networks
Rosa, Arlen M.
Many control schemes, simple or sophisticated, utilize high performance electro-hydraulic components as an interface to mechanical hardware. Failure of an electro-hydraulic component in these applications may impact safety, maintenance schedules and/or productivity of the overall operation. A condition monitoring scheme that would measure the performance or health of critical electro-hydraulic components would be desirable in addressing the above mentioned concerns. A neural network approach in estimating on-line proportional solenoid valve parameters is the focus of this study. A neural network algorithm is well suited for this type of parameter estimation in that it is capable of learning and best fitting a solution to given data. It is also capable of handling inherent noise in sensor data and is simple enough to allow for non-complex processing. Some of the parameters that affect the performance of the main stage valve spool which may be used in a condition monitoring scheme include orifice area gradient, spring rates, spring pre-compressions and spool friction. This thesis considers the use of individual neuron structures to estimate each valve parameter. Each neuron trains to a specific valve parameter with its output being the estimated parameter's main spool force contribution. The individual neurons act and train independently; however, their outputs are integrated to compute the main spool differential pressure. This differential pressure calculation is compared to the measured value and the difference is used to train each neural structure. The neural algorithm was found to be quite capable of estimating simulated valve parameters. When applied to experimental data, the neural algorithm estimated comparable main spool valve parameters to results from other studies conducted at the University of Saskatchewan. In all, it was determined that the neural algorithm was capable of estimating accurate and repeatable valve parameters, which reveals the feasibility of the parameter estimation procedure.