Machine Learning-Based Grading of Engine Health for High-Performance Vehicles

This paper presents a machine learning-based approach to grade engine health and generate a respective score ranging from 0 to 100 for tuned high-performance vehicles. It integrates the technical intricacies of automotive engineering with machine learning practices in a clear and sequential process. Data is collected from sensors monitoring RPM, boost, rail pressure, timing, and temperature. The data is processed for supervised learning and analyzed using scatter, pair, heatmap, PCA, parallel coordinates, and t-SNE plots. Models are trained, innovatively tuned through hyperparameter optimization, and tested for their ability to grade new data logs. Results highlight K-Neighbors, Extra Trees, and Extreme Gradient Boosting as exceptional regressors for this task, with Gradient Boosting, Support Vector Machine, and Random Forest trailing behind. This automated grading of engine health and performance enhances objectivity and efficiency in the tuning process, and potentially serves as a basis for a digital twin. The developed methodology is discussed in the context of health evaluation for any sensor-based system, with practical applications extending across various domains and industries.