Linear speed-up parallel implementation of continually online trained neural networks for identification and control of fast processes [induction motor control]

This paper summarises progress to date and presents the latest findings and results of ongoing work on high speed ASIC and parallel DSP implementation of continually online trained (COT) artificial neural networks (ANNs) for system identification and control, with particular reference to induction motor current control. Practical current responses produced by a low speed single Inmos T800 transputer implementation of the current loop demonstrate the practical success of this particular application of COT ANNs and the need for high speed implementation of COT. Two new methods of high speed COT ANN implementation on parallel hardware are described and evaluated by means of practical results obtained using two T800s. Several issues regarding high speed parallel implementation of COT ANNs on several Texas Instruments TMS320C40 DSPs are also described