Defect Detection in Industrial Soldering Processes Using Machine Learning: A Critical Literature Review

As electrical devices take on more life-critical roles, such as in autonomous driving, ensuring the quality of solder joints during production becomes increasingly important. Recently, there has been a growing interest in using machine learning techniques for this purpose. However, current research lacks a comprehensive overview that categorizes and analyzes relevant studies based on their specific intervention points within the production process. This literature review aims to examine and evaluate research coverage along three dimensions: intervention points in the process, non-destructive testing methods, and machine learning techniques employed. For this review, 112 conference papers and journal articles published since 2010 were selected from three databases using the PRISMA methodology. These publications were classified into the three dimensions previously mentioned, summarized, and analyzed. Furthermore, the literature core is critically evaluated to identify research gaps and limitations. The analysis shows that most studies focus on solder joint control, with few addressing intervention points in solder paste and component placement. Visual imaging and neural networks are the dominant techniques for non-destructive testing and machine learning, respectively. Despite a variety of literature that uses high-performance neural networks, meeting industrial detection standards often requires tolerating high false alarm rates. The findings contribute to structuring existing research and identifying research needs, particularly in validating these systems and integrating data from various testing methods and intervention points.