Temporal periodicity modeling and spatial transformation for fault detection in metro train bogie transmission systems

Reliable condition monitoring of metro train bogie transmission systems is essential for operational safety and predictive maintenance. However, the multichannel signals collected from these systems are strongly affected by operating-condition variation, transient disturbances, and sensor noise, making stable fault discrimination difficult. This study introduces an innovative framework for diagnosing faults in metro train bogie transmission systems, utilizing temporal periodicity analysis and spatial transformation methodologies. The developed technique overcomes the shortcomings of traditional diagnostic methods, which are prone to challenges such as noise interference, incorrect identification of transient impacts, and limited generalizability under fluctuating operational conditions. Through the integration of Fast Fourier Transform (FFT), multi-scale convolutional feature aggregation, and spatial transformation processes, the model adeptly isolates periodic fault patterns from stochastic noise elements, facilitating accurate profiling of both intra-period and inter-period fluctuations in intricate electromechanical signals. The architecture dynamically extracts cross-operational invariant characteristics across various temporal dimensions, thereby substantially improving sensitivity to nascent weak fault indicators while minimizing false positives induced by environmental fluctuations or signal transmission distortions. Across four operating conditions and 20 independent repetitions per condition, the proposed method attained an overall clean-signal accuracy of 97.66% ± 5.38% and outperformed six conventional and recent multivariate time-series baselines; additional test-time experiments quantify its behavior under four additive-noise levels and four random channel-missing rates.