A multi-scale adaptive dual forecast model applied to remaining useful life prediction of air turbine starter bearing

The air turbine starter (ATS) is a critical device for starting aircraft engines, in which rolling bearings play a core role. Predicting the remaining useful life (RUL) of rolling bearings is challenged by complex dependencies. Feature extraction, as a crucial step for RUL prediction, reveals the operational state of the bearings. To comprehensively extract the operational characteristics of rolling bearings, this paper introduces the multi-scale adaptive dual forecast (MADF) model. First, the model employs a multi-scale adaptive router to dynamically select optimal patch scales. Subsequently, these scales are used to segment the data into different temporal slices. For each scale, the model applies a dual forecast block to capture global and local temporal dependencies. Finally, a multi-scale aggregator integrates information across different bearing scales, further refining the multi-scale modeling process to predict the RUL. The experimental results demonstrate that the proposed method achieves reductions in mean absolute error (MAE) by 15.76%, 24.79%, 33.08%, 39.83%, and 49.15%; root mean square error (RMSE) by 16.01%, 29.64%, 29.97%, 38.72%, and 47.92%; and score by 21.01%, 34.52%, 41.29%, 43.61%, and 47.52% compared to the multi-resolution time-series transformer (MTST), patch time-series transformer (PatchTST), TimesNet, Scaleformer, and convolutional neural network (CNN) models, respectively. The effectiveness of the proposed method is validated through experimental data from the RUL prediction of civil aircraft-bearing components, showcasing its ability to achieve high-precision predictions of bearing RUL.