주조 공정에서 제어 로직과 온도 데이터를 활용한 CNN Autoencoder 기반 공정 이상 진단

Abstract

Checking the normal operation of facilities through process analysis and quickly identifying and handling problems when abnormalities occur are key challenges in the manufacturing industry. <br>In this study, we propose a CNN autoencoder model to determine the status of the process based on equipment operation data and temperature data patterns of the casting process. By training the model with data that combines equipment operation Gantt charts and temperature change graphs, it recognizes change patterns within the process and determines abnormalities without relying on data parameters. The CNN autoencoder model, which self-learns the pat- terns of the training dataset, effectively classifies abnormal cycles by learning only the patterns of normal data, even in environments where process results are not collected. Additionally, it provides new solutions for process improvement by identifying problems that were difficult to detect with existing data analysis-based anomaly detection models. Our experimental model interprets the training dataset as images and presents a pattern analysis method that can be implemented without complex data processing, not only in the manufacturing industry but also in any field where data patterns exist.