An Integrated Multioutput Classification-Based Defect Diagnosis Model for Pick-and-Place Machines

Abstract

Surface mount technology (SMT) is a method to mount components directly onto printed circuit boards (PCBs) and is widely used in low-cost and high-density electronic assemblies. Pick-and-place (P&P) is a core procedure for component placing after the solder paste printing (SPP) process in SMT. Generally, the industry uses an automated optical inspection (AOI) machine to detect defects after the components are mounted. However, the AOI machine cannot discern the failures’ root causes and offer reliable P&P machine maintenance references. With the advent of Industry 4.0, machine learning (ML) methods can be applied to improve production line maintenance. Therefore, the traditional check-up process can be changed into a data-driven, predictive, and condition-based maintenance process. Production efficiency can be significantly increased. In this article, a multioutput classification-based defect diagnosis (MCDD) model has been developed to trace the root causes of defects by using the patterns discovered from the experiment data. The experiments with initial machine errors are conducted and investigation information is collected. Compared with the traditional root cause identification model, the developed model is easier to adjust and can achieve an overall classification accuracy of 84.5%. Furthermore, the transfer learning method has been used to apply the trained model for one component to other components and can achieve an accuracy of 81.74%.