Design, DSP implementation, and performance of artificial-intelligence-based speed estimators for electromechanical drives

Abstract

The paper discusses the design and DSP implementation of artificial-intelligence-based (AIB) speed estimators for control applications in electromechanical drives. The design and performance of AIB estimators based on feedforward and recursive artificial neural networks (ANNs), associative memory networks (AMNs) and neuro-fuzzy networks (NFNs) are compared and discussed. Emphasis is placed on the development of minimal configuration estimators with a view to reducing DSP requirements. It is shown that it is an advantage of the AIB approach to estimator design that neither a conventional drive model nor a knowledge of any drive parameters are required and that an estimate of rotor speed can be obtained using only measurements of supply voltages and/or currents. The DSP system used is based on the Texas Instruments TMS320C31 mounted in a host PC. Results are presented for the real-time application to the speed control of a small DC drive and the estimators are shown to provide a sufficiently accurate speed estimate resulting in stable, robust, speed control. The DSP requirements and performances of each of the estimator forms are presented and compared and it is shown that the overheads imposed by implementation of these estimators is small.