Predictive Maintenance of Aircraft Engines: Machine Learning Approaches for Remaining Useful Life Estimation

In the realm of predictive maintenance for aircraft engines, ensuring precise and timely predictions of Remaining Useful Life (RUL) is paramount to enhancing safety, reducing costs, and minimizing downtime. While deep learning models dominate this field, their complexity, high computational demand, and lack of interpretability can hinder practical deployment in real-world scenarios. This study demonstrates the potential of machine learning models as an efficient and interpretable alternative for RUL prediction, leveraging the publicly available C-MAPSS dataset from NASA. Using C-MAPSS dataset, we integrated PostgreSQL for scalable data storage and Flask for interactive real-time visualization, creating a seamless pipeline from data ingestion to result interpretation. Preprocessing steps, including normalization, noise removal via box plots, and feature engineering, ensured a clean and reliable dataset. Among the models evaluated-XGBoost, Random Forest, SVM, KNN, and Linear Regression-XGBoost emerged as the best performer, achieving an RMSE of 23.8 and an R2 score of 0.67. These results nearly matches the accuracy typically associated with deep learning models while maintaining computational efficiency. Feature importance analysis through Explainable AI (XAI) further revealed critical insights into sensor contributions, enhancing interpretability and trust in the predictions. This work underscores the practicality and effectiveness of machine learning in predictive maintenance, offering a robust, transparent, and resource-efficient approach to RUL prediction. The findings highlight its potential to meet industry needs where simplicity, accuracy, and interpretability are critical for decision-making and operational efficiency.