Multi-task learning boosted predictions of the remaining useful life of aero-engines under scenarios of working-condition shift

The aero-engine is a typical equipment operating under variable working conditions. Changes in the working conditions of an aero-engine can cause data distribution divergence, making the remaining useful life (RUL) prediction task more challenging. Previous domain adaptation (DA) approaches have the limitation on the prerequisite of data availability in the target domain when handling the domain discrepancy and arranging data alignment. The target working condition is more likely to be unseen, resulting in the unavailability of the corresponding condition monitoring data of this working scenario. This study presents the research topic: the RUL prediction of aero-engines under working-condition shift scenarios in the absence of target domain data. To this end, we propose a multi-task learning-boosted method (MTLTrans) for the cross-domain RUL prediction of aero-engines. The MTLTrans is built upon the Transformer backbone in a hierarchical sharing style with two auxiliary prognostics tasks, i.e., state of health (SOH) assessment and performance degradation (PD) prediction. The trade-off learning of these three tasks facilitates producing reliable RUL prediction results robust against the data shift. Experiments on 12 cross-domain scenarios have shown that the proposed method significantly outperforms state-of-the-art methods, with an improvement of 18.83% of the root mean square error (RMSE).