Diesel Engine Intelligent Fault Diagnosis Based on Synchro Extracting Enhanced Generalized S-Transform and Improved Swin Transformer

To address the issue where irrelevant information interferes with subtle fault features, making it difficult to clearly define the boundaries between different fault types and thereby affecting the accuracy of fault diagnosis, an intelligent fault diagnosis method for diesel engines is proposed based on the Synchro Extracting Enhanced Generalized S-Transform (SEEGST) and an improved Swin Transformer. By processing the raw signals using the SEEGST method, the acquired vibration signals are converted into time-frequency spectra, comprehensively revealing the feature information contained in the raw signals; An improved Swin Transformer model is proposed, which inserts a local enhancement perception module before each cascaded Swin Transformer block and replaces the Window-Based Multi-Head Self-Attention (W-MSA) and the Shifted W-MSA (SW-MSA) modules with a k-Nearest Neighbor (k-NN) attention mechanism; and replaces the Multi-Layer Perceptron (MLP) modules in the Swin Transformer blocks at the output stages of each Swin Transformer stage with a multi-scale channel attention mechanism module to address issues such as slow feature fusion inference and insufficient interaction of contextual feature information in existing models; By extracting and training time-frequency image features using the proposed model, we achieve diesel engine fault diagnosis. Case studies and comparative analyses show that the proposed method can accurately and intelligently locate diesel engine faults.