Integrated engineering framework for fatigue damage prediction of fighter aircraft using machine learning

Le résumé fourni par la source

• Integrated framework developed for aircraft fatigue damage prediction. • CFD and FE analyses identify fatigue-critical locations validated by GSS. • Real in-flight strain data from five missions used to train ML models. • GPR outperformed RF and XGB with R² > 0.99 and lowest RMSE values. • Framework supports Individual Aircraft Tracking and predictive maintenance. Accurate fatigue damage prediction is essential to maintain structural integrity and airworthiness of aging fighter aircraft. Traditional fatigue-assessment methods often overlook the combined effects of aerodynamic loads, flight variability and material degradation. This study presents a novel physics-informed, data-driven framework that integrates computational modeling, experimental validation and Machine Learning (ML) using in-flight strain data to predict fatigue damage in a real fighter aircraft structure. Computational Fluid Dynamics (CFD) and Finite Element (FE) analysis were employed to identify Fatigue-Critical Locations (FCLs) and strain gauges were installed at identified FCLs. Ground Strain Survey (GSS) was performed to calibrate and validate FE model, yielding a percentage error in the range of 5.3 to 5.6 %. Five flight tests were conducted to capture real time strain data. Fatigue-relevant statistical features were extracted from strain histories and used to train seven ML models. Model performance was evaluated using Leave-One-Flight-Out (LOFO) cross validation to ensure generalizability. Root Mean Square Error (RMSE), R 2 score, 95 % confidence intervals (CI), Pearson correlation coefficient (r) and corresponding p-values were used to evaluate model performance. Random Forest (RF), Extreme Gradient Boosting (XGB) and Gaussian Process Regression (GPR) models achieved high predictive accuracy. RF provided stable and consistent predictions, XGB effectively captured nonlinear interactions but showed sensitivity to certain flight conditions, and GPR achieved the best overall performance. For the main spar and rear spar FCL, GPR achieved RMSE (R²) of 0.095 (0.993) and 0.0155 (0.991) respectively. The framework demonstrates strong potential to support structural health monitoring and Individual Aircraft Tracking (IAT).

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.