Machine Learning-Based Prediction of Diesel Engine Health Using Operational Parameters: Comparison of SVM and BPNN Models

Abstract

Condition-based maintenance (CBM) is crucial for enhancing the reliability of diesel engines. This study evaluates the effectiveness of support vector machine (SVM) and backpropagation neural network (BPNN) in predicting engine faults using operational parameters, such as engine RPM, lubricating oil pressure, fuel pressure, coolant pressure, oil temperature, and coolant temperature. Unlike previous research, this study validates engine condition labels based on standardized operational parameter thresholds, ensuring a more reliable and realistic representation of data. Statistical analyses using Spearman correlation and ANOVA deviance reveal that engine RPM and coolant temperature are significant predictors of engine health (p < 0.05). The findings show a notable difference in performance between the two classification models assessed. The Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel achieved an accuracy of 85.46%. However, the BPNN, configured with a [2-3-3-2] layer architecture and utilizing the tansig activation function, significantly outperformed the SVM, achieving an accuracy of 97.16%. These results suggest that the BPNN is more adept at capturing nonlinear patterns and providing more accurate predictions. Overall, this study underscores the importance of integrating domain-based data validation with machine learning techniques to develop reliable predictive maintenance systems.