In many areas, neural network technology has made a successful transition from theory to practical application, primarily due to the advances that have been made in computer technology and digital signal processing. Research at the University of Saskatchewan over the past few years has focused on applying neural network technology to fluid power systems. This paper will examine four projects that have been initiated by the authors and their graduate students which use neural networks for purposes of open loop pattern following, multiple input - multiple output control, indirect measurement of actuator displacement, and hydraulic component identification. A brief introduction to static and dynamic neural networks is given. Descriptions of the individual project objectives, the experimental implementation of neural networks to achieve these objectives, and some typical experimental results are considered. The results of these studies indicate that the use of neural networks for hydraulic systems shows significant potential for a variety of applications in the fluid power industry and for its customers in the off-highway vehicle business.
A unique contribution of the paper is a discussion of the philosophy of this research approach by an (with respect to the University of Saskatchewan) “at arms length” industrial supplier of custom designed electronic sensors and electrohydraulic control systems for OEM (Phoenix International). This company has been attracted to the neural network approach because these networks do not demand system linearity, perform well in noisy data environments, and can be quickly implemented in rather elegant real-time control solutions through the use of microcontrollers and digital signal processors. A discussion on the potential and conversion of this, and other similar type of research, to commercially viable products is forwarded.