Research on Gear Fault Diagnosis Based on Feature Mode Decomposition and CNN-CSNet

Gearboxes, as the core components of rotating machinery power transmission, are crucial for ensuring equipment operational safety through fault diagnosis. To improve the accuracy of gearbox fault diagnosis, this paper proposes a method based on Feature Mode Decomposition (FMD) and Markov Transition Field (MTF), combined with attention mechanisms in Convolutional Neural Networks (CNN) to address the insufficient feature extraction in conventional CNNs. The method adaptively extracts the 5 most fault-sensitive modal components from raw signals using FMD, then converts the 1D time series signals into 2D state transition images through MTF to effectively capture state transition patterns during fault evolution. Finally, the processed data are input into the CNN-CSNet fault diagnosis network for feature extraction and classification. Experiments show that this method exhibits superior performance and higher accuracy in feature extraction and fault diagnosis compared to other popular methods, thus verifying its feasibility.