Neural network modelling of fluid power control systems using internal state variables

Abstract

An approach is developed for identifying the behaviour of fluid power control systems using Artificial Neural Networks in conjunction with frequency-rich input excitation. Two different motor speed control systems are studied, the first considering the prediction of output torque using a mathematical model to identify the dynamic behaviour followed by predictions of the actual steady-state behaviour, the second considering the predictions of the output speed using direct experimental data to identify the dynamic behaviour. In both cases comparisons are made between the use of multi-sinusoidal and pseudo random binary input signals for network training and validation. A unique feature in both system studies is the use of internal state variables, pressure and flow rate, for network training. The results have implications for on-line identification of fluid power dynamic components with potential for adaptive control and fault diagnosis applications.