Data-Driven Electrical Machines Structural Model Using the Vibration Synthesis Method

Abstract

This paper proposes a data-driven structural modeling approach for electrical machine air-born vibration that can be used both in the design stage for fast optimization routines and in system-level simulations. This method is able to capture both in-plane and out-of-plane modal behavior along with local effects, such as tooth bending. The model relies on fitting unitary-force shape responses to transfer function models using a zero-pole representation. These frequency response functions are obtained from applying the vibration synthesis process to a series of finite element stator models that have material properties updated using experimental modal analysis. The obtained zeros and poles are used to construct three look-up table models with their accuracy varying between the three of them with the number of samples used in the data-generation process. The accuracy of these models is validated at the unitary force shape response level using four stator finite element model samples. Also, at the run-up vibration spectrogram level, the models are validated, with good results, using radial and tangential force excitation obtained from an updated electromagnetic finite element model of an interior permanent magnet synchronous machine designed for electric power steering. The computational cost compared to the standard finite-element vibration-synthesis method is shown.