Actuator Dynamics and Delay Compensation Using Neurocontrollers

This study compares the performances of a conventional feedforward controller and a neurocontroller in compensating the effects of the actuator dynamics and computational phase delay in a simple digital vibration control system. The model of this system is based on an earlier experimental study where a two-degree-of-freedom dynamic system was constructed in the laboratory. This system consisted of two electrohydraulic, position-controlled actuator-mass systems, mounted one on top of the other. Bode frequency domain plots were used to identify the governing parameters of the system. Based on the identified model of the experimental setup, a conventional feedforward controller and a neurocontroller are designed to compensate for the adverse effects of actuator dynamics and computational phase delay. The identified model is used to generate training information for the neurocontroller in a frequency domain of interest. To accomplish this endeavor, a neural network simulator is developed. This software uses a modified generalized delta rule with an adaptive momentum term for its learning mechanism and has a dynamic network topology capability. Through experiments and numerical simulations, it is shown that the neurocontroller is far more effective in compensating for actuator dynamics and time delay than the conventional feedforward controller. Factors contributing to the superior performance of the neurocontroller are identified and discussed.