Improving the Performance of Automated Optical Inspection (AOI) Using Machine Learning Classifiers

Automated Optical Inspection (AOI) machines inspect the Printed Circuit Board (PCB) manufacturing visually using a camera autonomously scans the device under test for both catastrophic failure (e.g. missing component) and quality defects (e.g. fillet size, shape or component skew). High false call rate is a fundamental concern of AOI machines that occurs when a component is considered as a ‘fail’ incorrectly that then have to be verified manually. In order to alleviate this problem, we train and compare different machine learning models (Decision Tree, Random Forest, K-Nearest Neighbors and Artificial Neural Network) and thresholds using logged fail data and extracting the efficient categorical and numerical features. The results show that the trained classifiers are able to identify the false calls well and increase the accuracy without increasing the error slip much. The K-Nearest Neighbor model, with a low threshold achieves the best result