ProphetJet: Predictive Maintenance Modelling Using LSTM, Random Forest, and XGBoosting to Forecast RUL Metrics of NASA Turbofan Jet Engines

Abstract

This project develops a predictive maintenance model for jet engines using the NASA C-MAPSS dataset. The model utilizes supervised learning to classify engine health states and predict Remaining Useful Life (RUL). Key techniques include data preprocessing, feature engineering, and machine learning algorithms optimized for time-series forecasting. Model performance is evaluated using RMSE, MAE, and overall loss between epoch gradients, with correlation matrices aiding feature selection. Future improvements include advanced deep learning techniques to enhance accuracy and adaptability, allowing machine owners to fine-tune the model with custom data for broader deployment. The model achieves a considerable accuracy of 87.4% with a 2% standard deviation. This approach enables proactive maintenance, reducing downtime and operational costs. See the project Github.