An Stepwise Approach for Degenerate Fault Data Generation

Balancing data in fault diagnosis poses challenges due to the rarity of failures caused by different working conditions and equipment. Generative models encounter reliability issues when generating long time series data. This paper proposes an algorithm for generating fault degradation data in a more reliable and controllable manner. The algorithm adopts a step-by-step approach, initially creating a smoothed version of the samples and subsequently generating fault samples with increased randomness. The effectiveness of the algorithm is evaluated using the C-MAPSS dataset, demonstrating improved stability in the training process and enhanced depiction of data degradation states. The introduced randomness improves the generalization ability of the diagnostic modeling process. The results indicate increased fault diagnosis accuracy and improved generalization capability. This algorithm effectively addresses the limitations of existing methods and provides a valuable solution for generating fault diagnosis data in real-world scenarios.