A Lightweight Dual-Compression Fault Diagnosis Framework for High-Speed Train Bogie Bearing

The vibration monitoring data of high-speed train (HST) bogie bearings exhibit high redundancy and limited effective fault information, impacting diagnostic accuracy and speed. To address this, a lightweight dual-compression diagnostic framework (DCDF) is proposed, based on damage extraction and a squeeze self-attention network. For data compression, a novel damage extraction method is implemented using peak recognition based on wavelet coefficients (PRWC). This method automates the extraction of damaged segments that meet fault feature requirements from the complete signal by establishing thresholds based on average and maximum peak values, using peak strength in four directions for each data point as the criterion. This approach enables the model to be trained and diagnosed using shorter samples that contain richer information, thereby reducing computational overhead. For model compression, a lightweight diagnosis network based on the squeeze-and-excitation self-attention (SESAT) is constructed to reduce model parameters. The effectiveness of this framework is verified by comparison with other advanced lightweight networks on scaled-down and real HST bogie bearing datasets. Finally, a scientific analysis and recommendations are provided for the extraction rules of damage data.