Discovery of fault-introducing tool groups with a numerical association rule mining method in a printed circuit board production line
提出一种基于遗传算法的数值关联规则挖掘方法，用于发现PCB生产线中因工具交互作用导致故障的工具组，在模拟和实际产线中优于机器学习算法，最多多识别36.5%的故障芯片。
Large-scale manufacturers aim to reduce the number of faulty products by finding tools or process factors that cause product faults through tool monitoring. The causes of faulty electronic components such as semiconductor chips and printed circuit boards (PCBs) include abnormalities in single tools and abnormalities caused by interactions between the tools of a specific process and a related process. Here, the tools exhibiting an interaction effect are called the fault-introducing tool group. This study presents a numerical association rule mining method for discovering the fault-introducing tool groups based on a genetic algorithm. A novel fitness function and rule pruning process are developed to identify the fault-introducing tool groups. The effectiveness of the method is verified using simulations and a case study of actual PCB production lines. The proposed method can discover fault-introducing tool groups better than machine learning algorithms. Additionally, the method can accurately identify fault-introducing tool groups in various manufacturing environments, such as those with highly skewed yield distributions or variations in yield distributions over time. In an actual PCB production line, the groups identified by the proposed method produced up to 36.5% more faulty chips than those identified by the comparison models.