Optimised decision-making model for vehicle health monitoring system leveraging deep learning algorithm

• Published in: Advanced engineering informatics Vol. 69; p. 103896
• Main Authors: Hossain, Md Naeem, Rahman, Md Mustafizur, Tusher, Ekramul Haque, Ramasamy, Devarajan, Rahim, Md. Abdur
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2026
• Subjects: Decision model; Deep learning; Machine learning; Vehicle health monitoring systems; Vulnerable components; Deep learning; Vehicle health monitoring systems; Vulnerable components; Decision model; Machine learning
• ISSN: 1474-0346
• DOI: 10.1016/j.aei.2025.103896

•A deep learning-based decision model is proposed for vehicle health monitoring.•Vulnerable components are identified using a systematic assessment framework.•Severity values are calculated combining sensor data and vulnerability analysis.•Neural network with random weights outperforms conventional ML algorithms.•Achieved 95.5% accuracy in vehicle health classification using NNRW algorithm. Traditional vehicle health monitoring systems suffer from high false alarm rates, inability to predict specific failure types, and inefficient handling of large-scale data, compromising vehicle safety and operational reliability. Existing approaches overlook vulnerability assessment of critical components across entire vehicle systems. This research develops an optimised deep learning (DL)-based decision-making model for real-time vehicle health monitoring and diagnosis by analysing vulnerable components (VCs). A structured vulnerability identification framework was developed through literature synthesis to identify critical VCs from various vehicle subsystems. A decision model was constructed integrating vulnerability assessment with sensor data to classify vehicle conditions into four categories: good, minor, moderate, and critical. A deep learning approach utilising a neural network with a random weight algorithm (NNRW) was implemented as the core predictive engine, featuring four hidden layers with nine nodes each, TanH activation functions, and a Softmax output layer. The NNRW deep learning model was trained and validated against conventional machine learning (ML) algorithms. Nine VCs were identified across engine, suspension, tire, gearbox, and brake subsystems. The NNRW algorithm significantly outperformed conventional ML techniques, achieving 95.5% decision accuracy (vs. 85.5% ML average), 94.4% precision (vs. 86.99% ML average), 93.9% recall (vs. 86.91% ML average), and 94.1% F1-score (vs. 82.8% ML average) using 80% training and 20% test data configuration. This model addresses ML limitations in handling vast data volumes, providing superior accuracy for real-time vehicle health diagnosis, potentially revolutionising predictive maintenance strategies. The results demonstrate that deep learning and vulnerability-based evaluation have potential applications in vehicle health management.