Data-driven non-intrusive reduced order modelling of selective laser melting additive manufacturing process using proper orthogonal decomposition and convolutional autoencoder

Abstract

Two data-driven, non-intrusive, reduced-order models (ROMs): a convolutional autoencoder-multilayer perceptron (CAE-MLP) and a combined proper orthogonal decomposition-artificial neural network (POD-ANN) are proposed and compared for additive manufacturing (AM) processes. The CAE-MLP uses a 1D convolutional autoencoder for spatial dimension reduction of a high-fidelity snapshot matrix constructed from high-fidelity numerical simulations. The reduced latent space after compression is projected to the input variables using a multilayer perceptron (MLP) regression model. The POD-ANN uses proper orthogonal decomposition-based, reduced-order modeling with the artificial neural network to construct a surrogate model between the snapshot matrix and the input parameters. The accuracy and efficiency of both models are compared based on the thermo-mechanical analysis of an AM-built part. A comparison between the statistical moments from the high-fidelity simulations results and the ROMs predictions reveals a good correlation. Additionally, the predictions are compared with the experimental results at different locations. While both models show good comparison with the experimental results, the CAE-MLP predictions have proven to be better performing than those of the POD-ANN.