High bandwidth direct adaptive neurocontrol of induction motor current and speed using continual online random weight change training

This paper reports on direct adaptive neurocontrol of induction motors using the random weight change (RWC) training algorithm for continually online trained (COT) neural network (NN) ASICs. A previous practical implementation of a direct backpropagation (BP) COT NN induction motor (IM) current control scheme is used to demonstrate the sufficiency of local minimum tracking (LMT) for direct control. Subsequent work on the substitution of RWC-COT into this current control scheme is also summarised. New work is presented which shows that, while sufficient for direct control, LMT can cause instability in NN model based indirect control schemes. This new work also shows how the BP-COT NN controller of an indirect IM speed control scheme may be replaced by a RWC-COT NN in a direct control structure to effectively avoid LMT instability