PREDICTING THE ELASTICITY OF METAL ADDITIVE MANUFACTURING PARTS USING MACHINE LEARNING

Metal additive manufacturing applications are growing quickly in various industries such as automotive, aerospace, biomedical, etc. To support this growing trend, it is necessary to estimate the material properties of metal AM parts to ensure that they meet necessary industry standards. Many research studies focus on determining the material properties of metal AM parts using inspection methods. However, inspection can be costly and time-consuming, especially if the inspection method is destructive and alters the part’s physical form. Accordingly, data-driven machine learning methods are being investigated to predict the material properties of AM parts. While these studies are promising, these methods fail to consider process conditions such as process interruptions and delays. For AM, the material properties are highly influenced by the continuity of the printing process due to the layer-by-layer printing mechanism. For this, the melting and solidification cycles must be controlled to avoid pore generation and defects, that will deteriorate the material properties of metal AM parts. In this paper, data is taken from an experiment in which the printing process was interrupted. The test samples are classified according to three regions: pre-interruption, amidst-interruption, and post-interruption. Moreover, heat treatment using an annealing process is performed. Afterward, data is collected on the region, heat treatment, hardness, and elasticity. To understand and model the complicated interactions between the parameters (i.e., printing region, heat treatment, hardness) and the material's elasticity, a prediction model is developed using the AdaBoost machine learning algorithm. The developed model is able to yield predictions with a high degree of accuracy. Such a model can help reduce inspection costs and avoid unnecessary destructive testing methods needed to determine the elasticity of the part.