Fig. 3: Comprehensive analysis and optimization of metastructure designs using neural networks and particle swarm optimization (PSO).
a Procedure for generating training data for the ML model: key geometric parameters (w1, w2, l1, l2, h1) of the metastructure are the inputs of the workflow. FEA simulations are performed across a range of these parameters, generating a comprehensive dataset of output displacements for each design that forms the basis for training the neural network. b The distribution of output displacements presents the diverse range of responses captured within the dataset. c The architecture of the MLP neural network model, designed to predict the deformation behavior of the metastructure based on input variables. d The high accuracy of the MLP neural network in predicting deformation, as evidenced by near-perfect R2 values achieved during the training and validation phases across various design configurations. e The algorithm is employed to optimize metastructure designs for specified output characteristics by integrating the MLP model with a defined objective function. f The alternative algorithm employed to optimize metastructure designs for output characteristics by integrating the PSO in conjunction with FEA simulations. g The alternative algorithm employed to optimize metastructure designs for output characteristics by integrating the PSO in conjunction with NN
