Optimizing additive manufacturing parameters for martensitic stainless steel via machine learning

Abstract

In the Laser Powder Bed Fusion (LPBF), the relative density of the printed parts is influenced by melt pool defects, which is critical to the performance of the printed parts. Due to the complexity of the LPBF process, it is challenging to achieve highly dense printed parts through experimental trials of process parameters. Machine Learning (ML) techniques are capable to correlate material performance with process parameters and the properties of the printed parts. In this work, we introduce a ML approach to predict the relative density of printed parts. Data was collected from over 60 papers on powder bed fusion, including information on material composition, process parameters, melt pool dimensions, and relative density. Five machine learning (ML) models were built to predict the relative density of LPBF martensitic stainless steels so as to obtain the optimized process parameters and corresponding powder conditions for full density. Among them, the Gradient Boosting Decision Tree (GBDT) model achieved a prediction accuracy of 98.30 % for the relative density, which was subsequently experimentally validated. This demonstrates that machine learning can optimize manufacturing processes, reduce production costs, improve production efficiency, and may also be applicable to other materials with AM process.