A Digital Twin-Enabled Deep Learning Framework for Remaining Useful Life Prediction of Turbofan Engines

Accurate prediction of the Remaining Useful Life (RUL) of industrial machinery is essential for developing intelligent predictive maintenance and digital twin systems. This study proposes a Long Short-Term Memory (LSTM) neural network model to estimate the RUL of turbofan engines by analyzing multivariate time-series sensor data obtained from the NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset. The model was designed to capture temporal dependencies within the sensor readings in order to learn the complex patterns of degradation that occur over time. Four datasets, namely FD001, FD002, FD003, and FD004, were examined, and the FD001 dataset was selected as the baseline because it represents a single operational condition with a clearly defined degradation trend. The trained LSTM model achieved a Mean Absolute Error (MAE) of 38.533 and a Root Mean Square Error (RMSE) of 51.069, showing that it can closely follow the actual degradation trajectory with a high degree of accuracy. Correlation analysis identified several key sensors, including sensor_7, sensor_12, sensor_20, and sensor_21, as the most influential variables for predicting RUL. The findings indicate that deep learning models can effectively represent mechanical degradation and can be integrated into digital twin frameworks to enable real-time health monitoring, proactive maintenance scheduling, and data-driven decision-making in industrial environments.