In the past few decades, additive manufacturing has evolved for the one-step fabrication of various complex, customized metallic components that cannot be easily and economically produced by other means. However, widespread applications and market penetration of such components are often hindered by the formation of common defects that affect part quality, reliability, and serviceability, and increase the cost. Here, for the first time, we show that a combination of physics-informed machine learning, mechanistic modeling, and experimental data can reduce the occurrence of common defects in additive manufacturing. By analyzing experimental data on the defect formation for commonly used alloys available in the disjointed, peer-reviewed literature, we identify several important variables that reveal the physics behind the defect formation. Values of those variables computed using a mechanistic model, when used in a physics-informed machine learning, provide the hierarchical importance of the variables on defect formation. In addition, based on the results of the physics-informed machine learning, we provide easy-to-use, verifiable, quantitative formalism that can be used in real-time to predict defects before experiments. The proposed methodology can help in reducing common defects such as balling, cracking, lack of fusion, porosity, and surface roughness, and solve other complex engineering problems beyond additive manufacturing.
All Science Journal Classification (ASJC) codes
- General Materials Science