Aircraft transonic buffet load prediction using artificial neural networks
This paper presents the results of a large-scale parametric investigation of deterministic and non-deterministic “black–box” modeling methods for the prediction of dynamic loading spectra, induced by transonic buffet, on the main wing of a fighter aircraft configuration. The aim is to consolidate practised regression and artificial neural network based modeling approaches, and their relevant functional parameter space, drawing comparisons between over–all performance, robustness and computational burden are communicated, intended to provide guidance and recommendations for future research efforts towards MISO modeling of complex chaotic mechanical systems. It is shown that Artificial Neural Networks (ANN) outperform the Auto-RegressiveMoving-Average model with eXogenous inputs model (ARMAX) method with respect to robustness and overall accuracy in the prediction of bending and torsion loads at wing root, mid-span and tip. In comparison to a baseline regression-based prediction, the highest performing ANN results provide a decrease in estimated load spectrum error by more than 99.5% for all load cases, with estimated difference between measured and predicted load spectra of less than 0.01%.
