Shallow and Deep Evolutionary Neural Networks Applications in Solid Mechanics

Data-driven models are now successfully used to solve complex multi-objective optimisation problems in materials mechanics. This chapter deals with the creation of surrogate models and its subsequent application for optimisation by evolution-based algorithms. Models representing curved beams or frames fabricated from composite tubes with circular cross sections are discussed here. Some learning strategies based on Evolutionary Neural Network (EvoNN), Bi-objective Genetic Programming (BioGP), and Evolutionary Deep Neural Net (EvoDN2) algorithms were applied. The surrogate models obtained from these algorithms were subsequently subjected to multi-optimisation for computing the Pareto fronts as the resulting output. Simple geometries could be very efficiently trained and optimised by each of the above-mentioned approaches, but more complex configurations (e.g., curved beams and frames made up of more tubes or tubes with additional layers) were effectively solved only by the EvoDN2 algorithm, as demonstrated in examples.