Multiscale feature extraction from the perspective of graph for hob fault diagnosis using spectral graph wavelet transform combined with improved random forest

The hob forms key component in gear hobbing machine and its health condition directly affects the reliability and safety of entire machine. This paper proposes a multiscale feature extraction scheme based on spectral graph wavelet transform combined with improved random forest, forming a novel hob fault diagnosis technique and realizing multi-scale analysis of vibration signals from the perspective of graph. Firstly, the vibration signal samples are transformed into path graphs, which contain the vertices information and similarity information between connected vertices, enriching the input information. Then, the path graphs are preprocessed by spectral graph wavelet transform at five-level decomposition for feature extraction. Finally, the random forest improved by adaptive beetle antennae search is utilized to identify the hob fault. Two groups of experimental results indicate that the proposed method has high effectiveness and robustness, achieving all the identification accuracy greater than 90% under multiple operating conditions and various environmental noises.