An adaptive artificial neural network-based generative design method for layout designs

• An ANN-based generative design method is proposed for layout/material design. The integrates GAN, CNN and genetic algorithm to form a general approach. adaptive learning optimization strategy proposed, allowing optimal solutions be found efficiently. Excellent performance of the demonstrated on two heat source layout problems. Comparison with existing methods conducted. Layout designs are encountered in variety fields. For problems many degrees freedom, efficiency becomes major concern. In recent years, machine such as artificial neural networks have been used increasingly speed up process. A main issue approaches need large corpus training data that generated using high-dimensional simulations. high computational cost associated generation largely diminishes gained by methods. this work, an network-based approach developed. This uses adversarial network generate candidates thus number variables greatly reduced. To evaluation objective function, convolutional constructed surrogate model function evaluation. inverse carried out conjunction networks. novel which allows space effectively explored search solutions. As needed both problems, obtained. Compared several approaches, has best terms accuracy efficiency. Graphical Abstract .