Prediction of Helicopter Rotor Loads and Fatigue Damage Evaluation with Neural Networks

Abstract In recent years, machine learning algorithms have experienced rapid advancement, driven by the exponential growth of data availability and computational capabilities. Among these algorithms, artificial neural networks stand out as one of the most renowned and effective classes, possessing the ability to discern relationships within data. In this study, we harness neural networks to deduce the relationship between flight mechanics parameters and resulting loads in an articulated rotor configuration. The accuracy of these algorithms hinges closely on the quality of the dataset used for training. Given that rotor loads manifest as time-periodic signals with precise harmonic content, we train dedicated neural networks to predict each harmonic individually. Subsequently, the load time history is reconstructed post hoc by amalgamating predictions from each individual network. Various network architectures are explored, and a sensitivity analysis is conducted on hyper-parameters to determine the optimal configuration for this specific application. Moreover, these predictions serve as input for a fatigue damage calculation algorithm.