Development and validation of a digital twin framework for SMT manufacturing

Electronics manufacturing is a global industry and key to innovation in the virtual world because it is the physical backbone. The cost-effective development and manufacture of such electronic modules are critical to maintaining the competitiveness of high-wage manufacturing countries. Therefore, two approaches suggest themselves: Design for Manufacturing (DfM) and manufacturing optimization. To apply both approaches in industry, it is necessary to introduce models that describe the relationship between process input and process output. Applied to electronics manufacturing processes, this means that design, process parameters, and material properties must be mapped to process quality criteria in end-of-line testing. Recently, machine learning (ML) algorithms have been emerging and dominating other modeling methods such as numerical simulation. To develop such complex ML models, a unified data structure for each input and output must be defined. This paper proposes an extensible ML-enabled framework that provides direct and structured access to the printed circuit board (PCB) design and process parameters. This framework is used to perform structured data acquisition using a custom data mining board. During the manufacturing of these PCBs on a full surface mount technology (SMT) process line, all available process machine-level data is collected, archived, and parsed into a uniform, standardized, and flat data structure. This enables fast analysis of the correlation between inputs and quality criteria, direct access by ML algorithms, and training of models. Through these measures, the quality of the framework is validated and correlations are revealed and compared to previous literature. This shows the great importance of a data framework for the integration of data analysis technologies into industrial processes.