From Polynomial Fitting to Kernel Ridge Regression: A Generalized Difference Filter for Encoder Signal Analysis

Encoder signal analysis (ESA) provides a novel tool for condition monitoring of machinery. Nevertheless, how to enhance the incipient fault signatures using ESA, especially in noisy measurements, remains a challenging issue. In view of this limitation, a new method termed generalized difference filter (GDF) is proposed for weak feature enhancement of encoder signals. To better capture the fault transients, kernel ridge regression is first introduced for signal approximation in high-dimensional feature space. A stochastic error minimization scheme is then proposed to improve approximation accuracy in a data-driven manner. Finally, a fast algorithm for GDF is developed based on the Gaussian kernel. With this method, the denoising and differencing of encoder signals can be treated in a unified framework. The effectiveness of the proposed method is validated by both simulated studies and experimental data.