Deep learning based porosity prediction for additively manufactured laser powder-bed fusion parts

Abstract

Machine learning techniques are extensively used to understand and predict complex non-linear phenomena across various applications. Moreover, these techniques minimize the time and costs associated with experimental and numerical analysis. In this work, a deep learning technique, specifically artificial neural networks (ANN), was employed to predict the density/porosity of laser powder-bed fusion (LPBF) additively manufactured (AM) parts by training the ANN model with X-ray computed tomography (CT) images. In addition to the experimental data, synthetic CT data was generated and used to improve the performance of the ANN model. The ANN model was then optimized for the number of hidden layers and neurons. Different errors like mean absolute error (MAE), root mean square error (RMSE), and square of co-relation coefficient (R2) were used as performance metrics to determine the accuracy and effectiveness of the network. The proposed ANN model was validated and showed excellent predictions (R2 = 0.9981, MAE = 1.6944 x 10−5). The framework proposed in this work can be used to speed-up the quality assurance of AM parts by reducing the time required for the analysis of CT data.