Experiment on Turbofan Engine Performance Degradation Evaluation Based on LightGBM

Abstract

As a core component in the aerospace field, the performance degradation of turbofan engines directly impacts flight safety and airline operating costs. Traditional evaluation methods rely on empirical models and simple statistical analysis, which struggle to uncover the potential patterns in massive operational data and lack accuracy under complex operating conditions. This study aims to construct a turbofan engine performance degradation evaluation model based on LightGBM, overcoming the limitations of traditional methods and providing scientific support for predictive engine maintenance. Using 300 hours of sustained test data from a certain type of turbofan engine and four different operating condition subsets as samples, the study employs the LightGBM algorithm to construct the model after preprocessing such as outlier detection and data normalization. A sequential quadratic optimization algorithm is then used to optimize component performance degradation calculations. K-fold cross-validation is employed, and the model is evaluated using indicators such as the coefficient of determination (R²) and mean absolute error (MAE). The model is also compared with decision tree, random forest, and XGBoost models. The results show that the LightGBM model achieves a prediction accuracy of 98.8%, a judgment accuracy exceeding 95%, an AUC of 0.812, and an F1 score of 0.960. It improves accuracy by more than 10% compared to decision trees and random forests, and outperforms XGBoost. It combines high computational efficiency with low memory usage, effectively capturing complex nonlinear relationships between performance parameters. This model provides a reliable technical solution for intelligent diagnosis of aero-engines and has significant engineering application value for optimizing maintenance strategies, extending engine life, and reducing operating costs.