Experiences of recursive identification applied to electrical machines

Recursive estimation is a well-proven tool for empirical modeling in the process industry. By contrast, its use in identifying common electrical systems such as transformers or synchronous and induction machines is far less documented. This paper reports on an attempt to apply this tool in the context of small rotating machines. Although recursive estimation is very attractive for its easy implementation, even on small computers, its practical application to continuous electrical devices following the so-called indirect discrete approach raises some problems related to the synchronization of the excitation and acquisition, as well as the proper choices of sampling interval, discrete-to-continuous operator, and noise model. Simulations are used to highlight the effect of a few specific factors but the paper really focuses on analyzing practical data collected over the years by the authors in their investigations into time-domain identification of laboratory-scale machines. Parameters estimation as such relies on PWM excitation with a random duty cycle, while deterministic PWM and step responses are used to assess the representativeness of the estimated models. Analysis techniques include the recursive least-squares (RLS) and recursive maximum likelihood (RML) methods, with usual enhancements such as UD factorization, forgetting factor and covariance resetting. In benchmarks, off-line subspace identification is also used for achieving very high accuracy in noise-free situations