TI-Former: End-to-End Useful Life Prediction Model Based on Transformer-Informer

In recent years, the advent of Transformer and Informer models has catalyzed a paradigm shift in the domain of Remaining Useful Life (RUL) prediction, showcasing unparalleled predictive proficiency. However, despite their prowess, each model is encumbered with intrinsic limitations. To mitigate these deficiencies, the present study introduces the Transformer-Informer (TI-former), an innovative end-to-end methodology for the prediction of Remaining Useful Life (RUL). The TI-former ingeniously amalgamates the Transformer's exemplary global feature extraction capabilities with the Informer's adeptness in modeling long-range dependencies. This potent synergy aims to equal or surpass the RUL prediction accuracy of the CNN-Informer, while necessitating reduced computational resources, thereby rendering the TI-former more accessible to researchers and practitioners with limited computational infrastructure. Rigorous empirical assessment on two authentic datasets substantiates that the TI-former outperforms, manifesting marginal superiority over the CNN-Informer in accuracy metrics, and a marked improvement in computational efficiency and GPU resource utilization.