An automatic data augment method for remaining useful life prediction of aeroengines

The prediction of remaining service life in complex aviation engine systems is of great significance for airlines to develop maintenance plans for engines and reduce maintenance cost.However, the complex operating conditions of the engine and insufficient fault mode data limit the prediction accuracy. One direction to solve such problems is data augmentation, which aims to generate synthetic data from real datasets to expand training samples and improve the model's generalization ability.Admittedly, There are already many mature data augmentation methods, but the optimal data augmentation strategy for RUL prediction tasks varies in different situations. Confirming which data augmentation strategy is most suitable for the current remaining useful life prediction problem requires human experience or extensive parameter experiments.This work proposes an automatic data augmentation method(AdaRUL),Build an automatic search space and use reinforcement learning algorithms to search for the optimal strategy in the automatic search space to expand the sample dataset. The experiments conducted on the C-MAPSS public dataset provided by NASA demonstrate that AdaRUL has successfully generated high fidelity multivariate monitoring data. In addition, these generated data effectively support RUL prediction tasks and significantly improve the predictive ability of underlying deep learning models.