The Development of a BPN Data Mining Calibration System in a SMT PCB Assembly Line

Printed Circuit Boards (PCBs) are extensively used in diverse electronic products today. The placement process tends to be a bottleneck in a PCB assembly line, and many SMT PCB assembly researches focused on improving the process perspective of the machine operations. This research proposed a BPN data mining model to analyze the effect of a head calibration process on the defects of placement operation in a PCB assembly line. The analysis was able to generate a set of values for head calibration which reduced component placement defects caused by inappropriate calibration processes. Based on the data collected from assembly lines, this research first uses a regression analysis to predict the total defects caused by the placement heads of the machine—the process is even able to predict the defects of single Placement head. Next, the Back-Propagation Network (BPN) is used as a framework for parameter training. The development of BPN is able to predict the placement defects caused by each placement head of the machine. Furthermore, the placement head with highest defect number is selected and a further sensitivity analysis performed. Based on the results of the analysis, the individual parameter calibration of the head is performed again to reduce the expected defects and increase production efficiency. The developed model can assist engineers to perform the calibration process of placement heads in a PCB assembly line and is able to reduce defects in placement operations so as to improve productivity.