Transformer-Based Deep Learning for Optimized Vehicular Engine Health Monitoring

The growing reliance on industrial machinery and critical infrastructure has increased the need for real-time, accurate engine health monitoring to prevent failures and optimize performance. Traditional fault detection methods often rely on rule-based systems or shallow machine learning models, which may fail to capture complex patterns in engine behavior. In this study, we suggest a combination of Long Short-Term Memory (LSTM) and Temporal Fusion Transformer (TFT) model for engine health classification. The LSTM captures sequential dependencies in sensor data, while the TFT enhances interpretability by focusing on multi-horizon forecasting and attention-based feature selection. The system is trained on a dataset of engine parameters, preprocessed using normalization techniques, and optimized with a real-time adaptive retraining mechanism to improve model robustness over time. Experimental results demonstrate that our hybrid model achieves high accuracy, outperforming conventional machine learning approaches. Additionally, we integrate our model into a Flask-based web application to enable real-time monitoring and user-friendly fault detection. This research contributes to the advancement of intelligent predictive maintenance, reducing operational costs and enhancing industrial safety.