Machine parameter estimation as a pattern recognition problem

On-line techniques for parameter estimation face practical restrictions and innumerable concerns about the possibility of undesirable effects. To overcome this barrier, an on-line parameter estimation technique that is both conceptually clean and highly nonintrusive is presented in this paper. By reformulating the parameter estimation problem as a pattern recognition problem, a practical solution has been achieved by using the ability of neural networks to recognize patterns from on-line data. Studies on a salient-pole micro-alternator and a turbo-alternator illustrate the effectiveness of the proposed technique.