Neural adaptive control of nonlinear MIMO electrohydraulic servosystem
Electrohydraulic servomechanisms are well known for their fast dynamic response, high power to inertia ratio, and control accuracy. If the system dynamics can be precisely described and the plant dynamics vary in the vicinity of the designed operation point, a fixed parameter controller may be designed using conventional control theory to acquire the desired output. However, for most industrial systems, it is very difficult to describe the system precisely. In addition, due to disturbances, variations of loads, and changing process dynamics, the system parameters may vary. Traditional linear control techniques based on small perturbation theory, which can deal with a system operated in the vicinity of the designed working state, may lead to a degradation in the performance of a system under varying parameter conditions. The neural network approach, using the parallel distributed processing concept with the capability of an ever-improving performance through a dynamic learning process, provides a powerful adaptation ability. Its implementation is thus quite feasible for the control of electrohydraulic servosystems. The major objective of the research undertaken in this thesis was to apply the neural network control architecture to a nonlinear multiple input-multiple output (MIMO) electrohydraulic servosystem to improve its position and force output performance. This objective was achieved through the use of a neural adaptive control scheme. A neuro-controller was implemented as a subsystem to control the real-time electrohydraulic system so as to track the desired signals defined by a reference model when subjected to system internal interactions and load variations. Experiments were conducted to illustrate the feasibility and benefits of the neural network approach in comparison with the traditional PID control strategies. The position and force outputs of the plant followed the reference model outputs successfully. The proposed control scheme forced the plant outputs to track those of the reference model simultaneously under changes of the load disturbances.
Doctor of Philosophy (Ph.D.)