OPTIMIZATION OF THE DC MOTOR MODEL COMPLEXITY LEVEL SELECTION FOR THE SYSTEM USING A NEURAL NETWORK

Abstract

A mathematical and computer model of the torque control system of a DC motor with independent excitation was built using the NARMA 2 neurocontroller. NARMA - a nonlinear moving average autoregressive model is one of the basic structures of a discrete and nonlinear model. The task was to conduct a comparative analysis of the operation of the neurocontroller-based control system using plant model of varying complexity. For the experiment, three models of the object were built with different degrees of detail, and therefore the reliability of the object. The full model includes an independently excited DC motor model that takes into account all the main parameters of the DC machine, internal friction parameters, excitation circuits, etc. The simplified model is a linear approximation of the DC motor by a second-order system that takes into account both mechanical and electrical time constants. The simplest model is a linear approximation of a DC motor by a first-order aperiodic link that takes into account only the mechanical time constant. The neurocontroller was trained using the above object models and an experiment was conducted to process the reference torque signal by the system. For each of the three training cases, the neurocontroller was set to 1000 epochs, since further training performance improvement is unnecessary due to the huge loss of computation time. The NARMA-L2 neurocontroller, which is also called feedback linearization control, was used. This controller can be implemented using a pre-identified NARMA-L2 object model. Neural networks that are trained on simplified object models do not take into account most of the electrical processes in a DC motor, especially the excitation winding is not modeled at all. The obtained results were compared with the use of a neurocontroller as a torque regulator. Further research in this direction involves the study of the necessary computing power for the microprocessor implementation of the neurocontroller.