Application of Machine Learning on the Additive Manufacturing of AlSi10Mg

Aluminium alloys have a wide range of applications, especially in the fields of aviation and aerospace. However, the complexity of the manufactured parts is sometimes limited when produced with conventional manufacturing processes. With the use of additive manufacturing techniques, near net-shape parts can be produced with a great structural complexity. However, understanding the printing process itself and the correlations between the different process parameters and the mechanical properties is quite challenging. Therefore, design of experiment (DoE) is often applied to obtain an insight into the whole process. The focus of this study lies on the laser powder bed fusion (L-PBF) process understanding and optimization of AlSi10Mg with the help of machine learning (ML). The understanding of the influences of different printing parameters - i.e. laser power, laser spot size, hatching distance, layer height and scanning speed - on the mechanical properties of this particular material was achieved through ML algorithms from a process parameter window generated through a DoE (Box-Behnken). Different ML models – e.g. Linear Regression, Extra Trees Regression, etc. – were generated for the purpose of predicting relative density, ultimate tensile strength and hardness of the manufactured parts. Moreover, the relative density was chosen for a direct comparison of different ML models. Finally, the models that performed best are discussed with regard to the printing process and the applied testing procedures for measuring the mechanical properties.