Abstract
The quality of the surface mount technology (SMT) process directly impacts product efficiency and reliability. Solder paste printing and reflow soldering processes are vital for assembling high-quality electronic components. Effectively optimizing these process parameters to ensure product consistency and reliability has become a critical issue in the electronics manufacturing services industry. Motivated by realistic needs to enhance the quality of the SMT process. This study proposes a surrogate-based optimization framework to improve the quality and productivity of the SMT production line. It encompasses five stages: domain knowledge, design of experiment, data collection and analysis, modeling, and optimization. Statistical correlation analysis and experimental design are used to reduce experiment counts. Then neural networks and optimization algorithms are utilized to identify the optimal process parameters in solder paste printing process. Moreover, this study proposes transfer learning methods for cross-product and line parameter optimization, which not only reduces production changeover time but also offers valuable insights for developing the solder paste printing process. A heat transfer model derived from a single experiment is used to identify parameters for reflow soldering. The target function is then optimized to find the optimal reflow recipe. Additionally, a solder joint defect detection system is established using deep learning and image processing techniques, capable of real-time detection and classification of solder joint defects. To evaluate the validity of the proposed framework, the surrogated-based optimization framework was deployed in a leading networking solutions company in Taiwan. Indeed, the developed solution has been implemented in this case company.