Robust harmonic detection, classification and compensation method for electric drives based on the sparse fast Fourier transform and the Mahalanobis distance

This work presents a method for the robust detection, classification and possible compensation of harmonics in electric drives during real-time operation, with the aim of providing a framework for monitoring and diagnostic without the need of additional hardware. The detection is performed with a sparse fast Fourier transform algorithm, for its lower computational effort when the signals are sparse (which, by definition, contain few meaningful spectral lines). The classification is performed on the results of the signal frequency analysis by means of the Mahalanobis distance concept, improving the robustness and noise rejection properties of the method. The compensation part relies on a family of regulators in parallel, each operating in the rotating reference frame of the harmonic to be cancelled. The theoretical background is followed by a discussion on the implementation and the interaction of the three blocks for a successful real-time operation. The system was tested in laboratory and proved to fulfil the requirements, by running in parallel to a vector control for synchronous machines. It was also found that the method is a useful tool to determine the presence of unknown harmonics in an electric drive system, thus potentially providing early warnings of unexpected failures.