| dc.description.abstract | With the outbreak of Covid-19 in 2019, the cost of manufacturing, assembly, transportation, warehousing and sales in the supply chain has been rising, which is undoubtedly a great challenge for the low-margin PCB industry. How to monitor product quality, equipment stability, and Quickly analyzing product defects and then introducing corresponding improvement countermeasures is an important issue to improve the profits of the company.
The optical AOI inspection error is the largest in all the PCB inspection station. The reason is the limit of its equipment capabilities, and it is difficult to effectively detect small defects. Therefore, once the amount of defeats increase, more small defects will be missed under the same probability. The missed defeats will become short and open defeats in electrical test, which will lead to scraps.
It will take at least 20-30 days from AOI inspection to electrical defect board analysis, and the improvement of key processes is obviously lack of immediacy. More defects are continually generated before corrective action being taken, resulting in waste of manufacturing costs. Therefore, if we can effectively use the first-hand information of AOI inspection to monitor the quality and establish a model to predict the electrical test yield for improvement and validation, we can shorten the time of all improvements, and reduce the number of defective products and the manufacturing cost of WIP (Work In Process).
The purpose of this research is to establish an accurate and immediate model that can be widely used in various products. According to the AOI inspection results, corresponding to the final electrical test short-circuit and open-circuit data, find out the factors that affect the electrical test results.
Based on the CRIPS-DM process, this paper will use the defect data of the AOI inspection in the PCB manufacturing process and the results of the electrical test to establish a model, corroborate with the actual results, and confirm the correlation between the two inspections.
In this research, SAS Enterprise Guide will be used in discriminant analysis for data preprocessing, and SAS Enterprise miner will be used to compare the accuracy of each model of Logistic regression analysis, decision tree analysis, and neural network-like analysis results, and to select a suitable model for prediction. | en_US |