A conclusive model to predict PBF-LB of AlSi10Mg thin-walled structures: digitally reconstructed models guided by both artificial neural networks and computational fluid dynamics

Abstract Metal thin-walled structures (TWS) are critical to all engineering industries, however, their complex manufacture has prevented large-scale adoption. The additive manufacturing process of laser-based powder bed fusion (PBF-LB) can produce high-resolution metal TWS with micro-scale geometries and intricate features. However, the thermal nature of PBF-LB increases the likelihood of defect formation. By modeling the manufacturability of metal TWS using a digitally reconstructed model guided by Multi-Layer Perceptron (MLP) Artificial Neural Networks (ANN) and Computational Fluid Dynamics (CFD), this problem can be predicted and then resolved. This research presents the first experimentally validated digital reconstructions guided by ANN-CFD as a predictive tool for PBF-LB manufacturability of metal TWS. This work contributes to the body of knowledge by introducing a new AI-based model to predict TWS porosity, dimensional deviations and distortion. To generate the MLP-ANN model, three control factors were selected: inclination angle, laser power, and the number of laser scan passes. To train the ANN, a full factorial dataset of AlSi10Mg samples was produced. Results show the MLP-ANN model as a precise tool to predict the manufacturability of TWS produced by PBF-LB with accuracy exceeding 90%. The most effective factor for the thickness, dimensional deviations and distortion was found to be the number of laser passes. The results also showed inclination angle was the main driving factor for the porosity of the TWS. The outcomes from this study highlight the value of ANN networks in the prediction and eventual certification of AM processes for global engineering interests.