PBMT: A Novel Transformer-Based Model for Accurate RUL Prediction in Industrial Systems

In recent years, the prediction of remaining useful life (RUL) has become a research hotspot about prognostics and health management (PHM). Highly accurate RUL prediction is the basis for improving the reliability of mechanical system and reducing the probability of unexpected accidents. Although deep learning-based methods for RUL prediction have made significant progress in applications, the existing models still suffer from high computational complexity and insensitivity to the dependencies of each time step when dealing with long sequence data. To solve these problems, this paper proposes a new model that combines the time-step encoder with the ProbSparse Self-Attention mechanism, replacing the encoder in the traditional Transformer. This approach not only reduces the computational complexity of the model when dealing with long sequence data but also maintains a high prediction accuracy. It has been validated using NASA's C-MAPSS and N-CMAPSS datasets, and the results demonstrate the method's effectiveness and accuracy.