Variational Autoencoder를 활용한 제조 실행 시스템(MES) 오동작 탐지

Abstract

In order to cope with a rapidly changing manufacturing environment, we constantly improve the functions of themanufacturing execution system (MES) and change the master data of the system from time to time during plantoperation. However, this is a way to increase manufacturing efficiency and can also cause unexpected systemmalfunctions. Therefore, an anomaly detection system that can detect a malfunction of MES is needed. Thispaper analyzes MES log file to detect a system malfunction. A log file consisting of categorical data is convertedinto numerical data using Word2Vec and learns from the variable Autoencoder to determine if a system ismalfunctioning. Experimental results using a log file generated by a semiconductor MES simulator show that theproposed method has excellent system malfunction detection performance.