Improving dynamic energy absorption performance and stability loss prediction of an all-PET sandwich structure through artificial neural networks

Abstract One key characteristic that is frequently studied in composite sandwich structure mechanics is low-velocity impact performance. Single polymer composites present higher ductility when compared to traditional glass fibre and carbon fibre reinforced plastic composites (GFRP, CFRP), which makes them strong candidates for impact energy absorption. These lightweight structures are also fully recyclable due to their all-PET constituents, an important aspect when considering possible applications. In this paper an existing second order composite SrPET and PET foam sandwich structure (self-reinforced poly-ethylene terephthalate matrix and fibre) is numerically investigated for optimal structural design configuration under out-ofplane, low velocity, dynamic loading. The study explores the parametric geometry interdependencies of the structure’s configuration with respect to its energy absorption potential. The results are then compared to the quasi-static loading performance behaviour studied by M.N. Velea et all [1]. This work provides a deeper understanding of how geometry configurations can influence energy absorption capabilities highlighting structural strengths and weaknesses while exploring non-equilateral triangle configurations. The Design Space Exploration (DSE) is carried out by using an optimization algorithm for a dynamic out of plane 3D impact simulation based on the finite element (FE) method software. The synergy between DSE and FE generates a guided simulation loop that can successfully be used to train a neural network to predict the dynamic behaviour of different geometric configurations from the exploration space. Based on the data samples obtained, an artificial neural network (ANN) is built to predict the energy absorption capacity for a given set of structural design parameters, useful in experimental validation test cases.