Abstract Scope |
Since the phenomenological understanding of additive manufacturing is still evolving but a large volume of data is available, data-driven machine learning tools are gaining attention to solve many important issues without the need for mechanistic understanding. The availability of well-tested algorithms, open-source codes, and publicly accessible cloud computing clusters are the keys to this progress. Machine learning is used in different stages of the printed product life cycle including the process and product design, process control, defect reduction, improving microstructure and properties, and part qualification. This presentation aims to provide several recent examples of applications of machine learning in additive manufacturing. The examples include minimizations of common defects such as cracking, balling, and lack of fusion, reduction in residual stresses, and control of grain size and part geometry. We discuss how machine learning can provide simple, verifiable, easy-to-use predictive tools that can be used to forecast the formation of defects, grain structure, deposit geometry for new sets of processing conditions before performing experiments. These predictive tools are immensely useful because they minimize the need for time-consuming and expensive experimental trials to print high-quality parts. In addition, we explain how machine learning can evaluate the hierarchical importance of the variables so that engineers can know which variables to adjust to reduce defects, improve quality, and control the process and product attributes. Furthermore, we provide examples where the use of variables computed using well-tested mechanistic models of additive manufacturing has made the machine learning algorithms more accurate and computationally efficient. Finally, we show how well-trained machine learning algorithms can provide easy-to-use process maps for shop floor usage that can be used to solve important problems in additive manufacturing in a time-efficient and cost-effective way. |