Remaining Useful Life Prediction based on Multisource Domain Transfer and Unsupervised Alignment

Transfer learning enhances remaining useful life (RUL) predictions by addressing data scarcity and operational challenges. Nonetheless, when a significant disparity in degradation data distribution exists between source and target domains, single-source domain transfer learning risks misleading or negative transfer. Multisource domain transfer learning partially addresses these issues. However, it ignores substantial discrepancies in feature-label correlations, which would impair the RUL prediction accuracy. Thus, we propose to develop a multisource domain unsupervised adaptive learning method, which is powered by a temporal convolutional network. Using a multilinear conditioning strategy, we combine degradation data and subregion labels to construct input characteristics for the domain discriminator. Additionally, we design a feature extractor that produces label-related features, invariant across domains, effectively enhancing prediction precision. We evaluate our method using the publicly available C-MAPSS degradation dataset with a case study and ablation experiments.