2020 · 21 citations · 7 references
ComputingComputer ScienceManufacturing EngineeringAutomationMachine LearningStatistical LearningClassificationIndustrial InspectionPattern RecognitionSupervised LearningAutomated InspectionAutomated Inspection SystemsAutomated Optical InspectionK-nearest NeighborsMachine Learning For Production SystemsPrinted Circuit Board InspectionPrinted Circuit BoardsRandom Forest Classification
False calls are an undesired occurrence in manufacturing of printed circuit boards that have to be verified manually after the automated optical inspection} (AOI). In this paper, the feasibility of identifying false calls of the AOI using subsequent machine learning models is investigated. Different algorithms, data set sizes and thresholds integrated into the model are considered. Thus, depending on the use case, the accuracy of the model should be maximized without increasing the error slip. Therefore, the numerical measurement data generated by the AOI is usedfor model training, testing and validation. The investigated models use a k-nearest neighbors (knn) classifier as well as an artificial neural network (ANN). Since metrics of ML models can be strongly dependent on data quantity and quality the models are tested and compared on different data sets. While the knn-classifier achieves the best results for small data sets, its use becomes problematic for large data sets due to increasing computing time. For large data sets ANNs as well as Random Forest Classifiers surpass the performance of the knn-classifier.
7
Transfer learning for automated optical inspection
Seung‐Hyeon Kim, Wooyoung Kim, Yung‐Kyun Noh et al. · 2017 · 96 citations