Machine learning based response prediction and performance optimization of a trapezoidal corrugated plate under air blast

The current paper aims to enable rapid prediction of the air blast performance of a trapezoidal corrugated sandwich panel (TCSP) and facilitate subsequent optimization design. First, the Conventional Weapons Effects Program (CONWEP) is leveraged to compute the mechanical property of the TCSP under air blast loading, and the computation results are compared to tested data to verify the accuracy of the numerical approach. Subsequently, Abaqus/Explicit is secondary developed via Python scripting to perform batch computations of air blast responses under varying geometric and explosive parameters. On the basis the computational results, the effect of individual parameters on the structural response are elucidated, and then a comprehensive dataset characterizing the response of the TCSP is constructed. On this foundation, support vector machine (SVM) is utilized to classify whether structural fracture failure occurs under specific air blast loading conditions, following which a deep neural network (DNN) is deployed to facilitate a rapid prediction of the structural plastic deflection magnitude, with interpretability analysis conducted on the predictive model. Addressing the limited generalization capability inherent in conventional nodal-data-based contour generation methods, an image-to-image deformation contour prediction approach is proposed based on variational autoencoder (VAE). Finally, the intelligent prediction model for structural plastic deformation is incorporated as the fitness function, and the Particle Swarm Optimization (PSO) algorithm is leveraged to search the optimal geometric parameters of the TCSP. The results demonstrate that the deformation of the optimized TCSP under air blast load is reduced by 48% compared with the original configuration.