A Data-Driven Robust Fault Detection Method for Traction Motors Based on Reduced-Order Residual Generators

Motors play an important role in industrial production. However, noises and disturbances are naturally generated when the motors are running, and they may mask the real fault signals and affect the accuracy of the fault detection. It is difficult to extract effective fault features in complex noise environment. Therefore, this paper proposes a data-driven approach for closed-loop systems with disturbances. It establishes a robust reduced-order observer to achieve the fault detection. Under the framework of robust stable kernel representation (SKR), subspace identification is utilized to establish the observer directly. The residual generator based on the iterative optimization Kalman filter is constructed by data-driven approaches. Therefore, the dimension of the observer is reduced to alleviate the computing load. This manner is different from traditional residual generators. This iterative optimization Kalman filter is used to reduce the dependence on prior knowledge of the noise covariance matrix. Therefore, it is appropriate for real industrial systems.Finally, a permanent magnet synchronous motor is used to verify the effectiveness of the proposed method.