DATA-DRIVEN PREDICTIVE MAINTENANCE OF DIESEL ENGINES USING ADVANCED MACHINE LEARNING AND AI-BASED REGRESSION ALGORITHMS FOR ACCURATE FAULT DETECTION AND REAL-TIME CONDITION MONITORING

Abstract

Diesel engines remain integral to numerous industrial sectors, including transportation, power generation, and heavy-duty equipment. However, their complex mechanical configurations and exposure to variable environmental and load conditions often lead to unanticipated faults, resulting in costly downtimes, reduced performance, and increased maintenance overheads. These challenges are further intensified by the dynamic nature of engine operations, where traditional rule-based diagnostics frequently fail to detect subtle degradation patterns or early fault symptoms. Moreover, the increasing demand for operational efficiency, reliability, and environmental compliance underscores the need for intelligent, real-time fault prediction solutions. To address this challenge, this research presents a comprehensive, data-driven framework for predictive maintenance and fault diagnosis of diesel engines using advanced artificial intelligence (AI) regression algorithms. By analyzing multivariate sensor data and historical operational logs, we implement and evaluate a suite of machine learning models including Support Vector Regression (SVR), Random Forest Regression (RFR), and Artificial Neural Networks (ANN) to capture the intricate, nonlinear relationships between engine inputs and fault indicators. The study also explores model sensitivity and the influence of various hyperparameters on prediction performance, optimizing configurations for real-world deployment. A systematic training and validation process is applied using real-world engine datasets, ensuring the models are both accurate and generalizable across diverse operating scenarios. The proposed AI-based framework supports early fault detection, real-time condition monitoring, and prognostic decision-making to facilitate intelligent maintenance scheduling. Furthermore, feature importance analysis is employed to identify the most influential parameters contributing to fault occurrence, enhancing interpretability and model transparency. Comparative performance metrics including root mean square error (RMSE), mean absolute error (MAE), and R² score demonstrate that the AI models significantly outperform conventional threshold-based and rule-based diagnostic systems in both predictive precision and operational efficiency. Ultimately, this research contributes to the advancement of intelligent engine health management systems, reducing unplanned outages, minimizing lifecycle costs, and accelerating the digital transformation of diesel engine maintenance strategies.