다중 식별자를 이용한 Adversarial Autoencoder 기반 제조 공정 이상 탐지

Abstract

When unexpected problems ocur in manufacturing proces, it is necesary to configure an anomaly detectionsystem to monitor and control them. Abnormal data are critcal because they cause a decrease in yield and porquality. If abnormal data is not detected, the proces continues and the los becomes greater. Abnormal data havefewer numbers than normal data, resulting in clas imbalance problems. Therefore, we solve the data imbalanceproblem by learning distribution of normal data only. Unlike conventional methods, adversarial autoencoder(AAE) is able to create distributionsimilar to the original data through competive learning using discriminator. This paper proposes adversarial autoencoder with multiple discriminators, a method to learn the distribution ofnormal data more acurately by ading two discriminators to AAE. We use Long Short-Term Memory (LSTM)layer to fithe time series characteristics. Experiments confirm thathe method proposed in this paper show greatanomaly detection performance.