Answers are grounded in this paper and cite their sources.
Defects in manufacturing not only reduce the quality of a produced part but also cause a waste of resources. Conventionally, quality control relies on manual inspection or the use of sensor technology to gather data to train machine learning models to identify defects. Although quite common, defects and anomalies in manufacturing do not occur often enough to gather a dataset with balanced representation of defective and normal samples. Using supervised classification models in such scenarios can be challenging. In this work, we propose a generative-adversarial-network-based approach to anomaly detection, where the models are trained only on normal samples. During testing, samples are classified as normal or anomalous based on the ability of the model to create samples as similar as possible to the test sample. We compare and evaluate the performance of the proposed model against the pretrained classification model AlexNet as well as an autoencoder model. Furthermore, we show that the proposed model performs on par with the pretrained classification model and allows localization of the anomalies in the data.
No takes yet. Share an insight, caveat, or question.
Ghansiyal et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: