Life prediction and health assessment of aero-engine gas path using digital twin and deep learning

This study proposes a system for health monitoring and remaining useful life (RUL) prediction of aviation engine gas path that integrates digital twins and deep learning. Addressing the limitations of existing data-driven methods in feature extraction capabilities and dynamic interaction mechanisms, this study designs an improved Transformer model (TMR). By incorporating learnable position encoding, serial multi-head attention mechanisms, and an optimized multi-layer perceptron (MLP) structure, the model significantly enhances its ability to model time-series data and improve prediction accuracy. Based on this, a multi-level virtual-reality interaction mechanism was constructed to achieve real-time linkage between the physical space, health assessment module, and maintenance decision-making scenarios, enhancing the integration capabilities of state perception, information transmission, and collaborative decision-making. At the same time, a three-dimensional visualization interaction framework for digital twin systems was developed to improve the system’s observability and interpretability. Experiments were conducted using the NASA CMAPSS dataset, and the results showed that TMR outperformed mainstream models such as LSTM, CNN, and traditional Transformers in terms of RMSE and Score metrics, particularly in complex multi-condition scenarios. Ablation experiments further validated the critical contributions of each component to the overall performance. Finally, this study constructed a digital twin platform prototype, achieving a complete closed-loop process from data collection to condition monitoring, performance prediction, and decision support, providing a high-precision, real-time solution for aviation engine health management with promising engineering application prospects.