Model-Based Current Filter Design for Control of Vibrations due to Electric Motor Faults

Excessive vibration caused by motor faults poses significant challenges to system performance and structural stability. This research presents a model-based approach to designing motor current signature filters for effective fault detection and vibration control. The proposed method trains convolution filters exclusively on modeled fault data, leveraging the hypothesis that local minima in the loss function during training on modeled data correspond to those on measured data. Stochastic gradient descent is employed to optimize filter weights and biases, while the minimum redundancy maximum relevance (mRMR) method identifies the most effective filters. Validation on measured data demonstrates the ability of the trained filters to accurately detect and classify various motor states, including healthy conditions, stator open-circuit faults, bearing outer-race faults, and mechanical looseness. By enabling precise fault detection, the designed filters facilitate targeted vibration control, reducing fault-induced vibrations and enhancing system reliability. This model-based approach offers a cost-effective and efficient solution for fault diagnosis and vibration management, advancing the field of noise and vibration engineering.