Neural networks have been applied to a wide spectrum of applications in the last decade, including the fluid power area. The popularity of neural networks can be attributed, in part, to their ability to deal with nonlinear systems. Their use as a controller, however, poses many interesting problems, especially when trying to apply them to practical systems. In hydraulics, PID controllers have been used with some degree of success; however, when it comes to highly nonlinear systems, this established method encounters some difficulties.

This paper considers the applied problem of a hydraulic servovalve controlling a actuator with nonlinear friction characteristics. A neural net controller is pre-trained to the performance of a special PID controller which has been tuned to a specific waveform. The plant is a model of a hydraulic servovalve and a linear actuator with slip-stick friction characteristics. The neural net controller replaces the PID controller and shows superior performance for waveforms that it was not trained for. The addition of a “kicker” signal reduced the distortion of velocity (a consequence of nonlinear friction) when the velocity was near zero. It is concluded that in certain applications, a neural net controller does show potential for use in nonlinear systems, but many other issues such as stability, reliability, and adaptability must be addressed in the future.

This content is only available via PDF.
You do not currently have access to this content.