Naive Bayes Predictive Model for Failure Detection in CODLAG Propulsion Systems

Abstract

This study implements a Naive Bayes Classifier algorithm to detect failures in gas turbines operating within a CODLAG (Combined Diesel-Electric and Gas) propulsion system. The complexity of hybrid propulsion systems necessitates reliable data-driven monitoring methods to support early anomaly detection and predictive maintenance. An open-access dataset from Kaggle was utilized as the source of gas turbine operational data, with five key parameters (GTn, T48, ṁf, P1, and P2) selected due to their strong correlation with turbine thermodynamic performance. Following data preprocessing and an 80:20 train–test split, the model was trained to classify operating conditions into Normal and Faulty states. The evaluation results demonstrate an accuracy of 86.89%, accompanied by high precision and recall values, indicating the model’s capability to identify anomalies with minimal misclassification. Furthermore, the Receiver Operating Characteristic (ROC) curve yields an Area Under the Curve (AUC) of 0.96, reflecting strong discriminative performance. These findings confirm that the Naive Bayes approach is computationally efficient and suitable for real-time implementation within shipboard Condition-Based Monitoring (CBM) systems, thereby enhancing the reliability and operational efficiency of CODLAG propulsion systems.