مهدي صادقي سليمي

With the increasing complexity of modern vehicles an‎d the growing emphasis on reliability, safety, an‎d efficiency, predictive maintenance has become particularly vital for the engine subsystem, a critical component whose unexpected failures can lead to significant costs an‎d operational disruptions. In this study, a precise engine health prediction model is developed by leveraging both sensor data an‎d operational parameters. Among several machine learning algorithms eva‎luated, CatBoost achieved the highest performance across all metrics, with an F1 score of 0.7583, outperforming the baseline k-nearest neighbors model by approximately 1%. Furthermore, extensive statistical analyses, including the fitting of data distributions to well-known theoretical models, correlation analysis, an‎d classical hypothesis testing, were conducted to provide a more robust foundation for modeling. The core innovation of this research lies in its dedicated focus on the challenge of class separability in engine fault detection. Improvements were achieved through deliberate algorithm selec‎tion, dimensionality techniques, restructuring of classification strategies, an‎d, most notably, a redefinition of class labels to optimize data distribution. This reclassification resulted in an approximate 9% increase in eva‎luation metrics an‎d significantly enhanced the distinction between fault categories. Additionally, the integration of these techniques resulted in a further 2% performance gain, enhancing the model’s effectiveness an‎d generalizability under real-world conditions.

تعمير و نگهداري پيشگويانه , يادگيري ماشين , پيش‌بيني خرابي , زيرسيستم موتور , تحليل داده‌هاي سنسور

Predictive Maintenance , Machine Learning , Fault Prediction , Engine Subsystem , Sensor Data Analysis

Mehdi Sadeghi Salimi

Morteza Mollajafari