A Transformer Framework for Remaining Useful Life Prediction Using Sensor Attention and Operational Context Embeddings

This work presents a novel and efficient approach for predicting the Remaining Useful Life (RUL) of turbofan aircraft engines using NASA's C-MAPSS dataset. A two-phase methodology is introduced: baseline modeling with classical regressors (Ridge, Random Forest, XGBoost), followed by advanced deep learning techniques employing Long Short-Term Memory (LSTM) networks and a custom transformer-based architecture. The proposed transformer model integrates dynamic operatingcondition embeddings and a sensor-specific attention mechanism inspired by squeeze-and-excitation (SE) networks. This design enhances the interpretability while capturing complex multivariate temporal dependencies. Comparative experiments across multiple configurations d emonstrated s uperior p erformance in t erms of RMSE and R², outperforming both traditional models and recent state-of-the-art deep learning approaches. The proposed method is computationally efficient and can be generalized across diverse degradation patterns. These findings r einforce t he r ole of lean transformer architecture as a scalable, interpretable, and effective tool for predictive maintenance in aerospace applications.