Towards the next generation of machine learning models in additive manufacturing: A review of process dependent material evolution

Abstract

Additive manufacturing facilitates producing of complex parts due to its design freedom in a wide range of applications. Despite considerable advancements in additive manufacturing techniques, 3D-printed parts are still suffering from durability, repeatability, and reliability. One main reason, which increases the complexity of the problem and causes defects during manufacturing, is the high number of processing parameters. Machine learning approaches seem to be a promising solution to tackle the challenges in the additive manufacturing field. This paper employs a systematic literature review by employing natural language processing and text mining techniques to analyze the recent advancement in the application of machine learning in porosity detection and prediction in 3D-printed parts. Two methods of text analytics are used to evaluate different avenues of research in additive manufacturing. Most frequent machine learning methods employed to evaluate the porosity formed in the 3D-printed parts are introduced and classified based on their applications. Recent advancements in developing hybrid machine learning models reveal the importance of physical domain knowledge (e.g., thermomechanical laws and constraints) in these models and their accuracy. Eventually, challenges and opportunities that exist for the next generation of machine learning techniques in the AM field are identified and summarized.