Understanding the Effects of Process Conditions on Thermal-Defect Relationship: A Transfer Machine Learning Approach

This study aims to develop an intelligent, rapid porosity prediction methodology for varying process conditions based on knowledge transfer from the existing process conditions. Conventional machine learning algorithms are extensively used in porosity prediction. However, these approaches assume that the training (source) and testing (target) data follow the same probability distribution, and the labeled data are available in both source and target domains. The source and target do not follow the same distribution in real-world manufacturing environments. The diversity of industrialization processes leads to heterogeneous data collection in different production conditions, and labeling is costly. Transfer learning is one of the robust techniques that enables transferring learned knowledge between source and target to establish a relationship while the target has less data. Therefore, this paper presents similarity-based multi-source transfer learning(SiMuS-TL) method to develop a relationship between a source and an unknown target. The similarities between sources and targets are learned by forming a new domain called the mixed domain, which organizes data into identity groups. Then, a group-based learning process is designated to transfer knowledge to make target predictions. The effectiveness of the SiMuS-TL is explored with the application of porosity prediction in additively manufactured parts in realistic situations, i.e., single-source and multi-sources transfer to unknown target porosity prediction. The porosity prediction accuracies are approximately 90% for both scenarios with the SiMuS-TL method, but conventional SVM and CNN classifiers barely perform well in predicting porosity while process condition varies.