Spectral distribution decomposition and its application in gearbox fault diagnosis

Abstract

Existing signal decomposition algorithms suffer from mode mixing, decomposition instability, and poor noise resistance, hindering subsequent feature extraction and fault diagnosis. To address these issues, this paper proposes a spectral distribution decomposition (SDD) algorithm based on the spectral probability density function (SPDF) for gearbox fault diagnosis. The proposed SDD employs a spectral distribution strategy utilizing maximum broadcasting and Gaussian kernel convolution to transform raw spectra into the SPDF, which characterizes signal distribution properties. Integrating the SPDF with the expectation-maximization (EM) algorithm, SDD decomposes signals into multiple distribution mode functions (DMFs) with well-defined frequency boundaries and a residual signal. For gearbox fault diagnosis, an adaptive filtering strategy incorporating SDD is developed to extract fault-related frequency bands and diagnosis indicators. Comparative simulations with well-known decomposition techniques, together with experimental tests on the fault diagnosis of laboratory gearboxes and actual wind turbine gearboxes, demonstrate the effectiveness and superiority of the proposed approach.