AI-BASED AUTOMOTIVE ENGINE DIAGNOSTICS: EXPLORING THE ENGINEFAULTDB DATASET FOR FAULT DETECTION

Abstract

Automotive engine diagnostics are crucial for ensuring vehicle performance and safety, yet traditional tools often fail to detect complex faults, leading to costly repairs. This paper presents an AI-driven approach forengine fault detection using the EngineFaultDB dataset, which containssensor data from engines under various operating conditions. We propose an ensemble learning method combining Random Forest (RF) and Multi-Layer Perceptron (MLP) neural networks for multi-class classification of four fault types (0, 1, 2, and 3), employing the One-vs-Rest (OvR) strategy to handle themulti-class nature of the problem. While MLP achieves the highest accuracy(74.96%), it lags behind Random Forest (74.82%) and Ensemble (74.89%) interms of F1 score (0.694), suggesting that its precision does not alwaystranslate into effective fault detection. Random Forest provides a betterbalance of accuracy, precision, recall, and F1 score, emerging as the most robust model. The Ensemble model did not significantly outperform individual models, indicating that ensemble methods require furtheroptimization, such as advanced techniques like stacking or boosting, hyperparameter tuning, and feature selection. These results underscore thepotential of AI-based systems for predictive maintenance in the automotive industry. Future research should focus on refining ensemble models,expanding datasets, and integrating deep learning techniques to improve diagnostic accuracy and the reliability of automotive systems.