A design methodology of composite scarf repairs using artificial intelligence

Abstract

Composite Scarf Bonded (CSB) based techniques are highly effective in structural connections and structural repairs. In this article, a preliminary design methodology based on Machine Learning (ML) algorithms trained on databases obtained via a semi-analytical approach is proposed and used to generate the design space for CSB structures under tensile loads. This ML framework introduces the one-hot encoding technology to deal with discrete inputs, such as multiple stacking sequences. Four ML algorithms, Adaptive Boosting, Gradient Boosting Regression, Extreme Gradient Boosting, and Artificial Neural Networks are studied. The best-performing model is then used to generate the damage tolerance-based design space for CSB structures made from fabric and unidirectional prepregs, accounting for material and geometrical uncertainties. Very good representations of the design space and accuracy in structural strength and failure mode are obtained. An optimal scarf angle zone, where laminate and adhesive fail simultaneously, was identified using the proposed framework. This design framework opens new avenues for the selection of material and layup configuration in structural design and enables the fast estimation of the optimal scarf angle range for the preliminary design of CSB structures.