A Data Augmentation Boosted Dual Informer Framework for the Performance Degradation Prediction of Aero-Engines

Abstract

The prediction of performance degradation for the aero-engine is crucial to its health management, but the handling of the dynamic spatio-temporal dependence between condition monitoring (CM) data of multiple sensors and the status of performance degradation is non-trivial. Most previous prediction models of performance degradation treat different health stages equally in the training process, although the data in the initial degradation stage is relatively sparse and more important for training a decent prediction model. To bridge this gap, we propose a data-augmentation-boosted dual Informer framework (DARWIN) for predicting the performance degradation of aero-engines. First, we present a degradation time-series data augmentation model based on Informer to increase the amount of degradation data, making it possible to emphasize the importance of data in the initial degradation stage in the following prediction stage. Second, we design a padding strategy for the run-to-failure (RtF) data so as to preserve the integrity of the degradation context comprehensively. Third, we invoke another Informer model for predicting the performance degradation in which a generative decoder is implemented to get predictions in one forward process instead of the recursive manner for fast computation speed and avoid error accumulation. On the benchmark C-MAPSS datasets, DARWIN yields a 27% accuracy improvement compared with the state-of-the-art method, Informer, in the performance degradation prediction of aero-engines. Furthermore, we demonstrate the feasibility of DARWIN on a fleet of eight turbofan engines under real flight conditions, thereby confirming its applicability.