Prediction of nonlinear mechanical behaviour of the corrugated flexible composite skin for morphing wing via PCA-enhanced deep learning method

Abstract

Tensile curves of corrugated flexible composite skin (FCS) significantly represent morphing wings’ mechanical properties, from which critical mechanical descriptors such as tensile stiffness, strength, and toughness are defined. This study aims to develop an efficient surrogate model that predicts the tensile load-displacement (T-D) curves of corrugated FCS with varied geometry and stacking sequences. A database of T-D curves was generated via finite element analysis (FEA) for different structural parameters; these parameter sets serve as DNN inputs, and their corresponding curves as outputs. The accuracy of the FEA results based on the Hashin-Puck progressive failure criteria was verified by mechanical tests and the proposed analytical model. T-D data were projected into a lower-dimensional space to reduce dimensionality via principal component analysis (PCA). Key DNN hyperparameters were then optimized using a Bayesian Optimization and HyperBand algorithm. As a result, the proposed PCA-DNN data-driven approach predicts T-D curves for various FCS designs within a fraction of a second, achieving high accuracy. Mean absolute errors for key descriptors remain below 5 % of the range of values in the dataset. Finally, we extended the model via transfer learning to accurately predict compressive behavior using minimal additional data, demonstrating strong generalization across different loading modes. Owing to the universality of its construction principles, the method has broad applicability in morphing wings with different corrugated structures.