냉동고 작동오류 진단방법 개발 : 기계학습 알고리즘 비교를 중심으로

This study aims to diagnose operation faults of freezer such as door left open by mistakes and refrigerant leaks by using machine learning approach. Machine learning algorithms can take training raw data and then output trained model that contains prediction rules. Active power of freezer, laboratory ambient temperature, and freezer inside surface temperature are selected as monitoring variables. Heat capacity, refrigerant mass, and door opening also varied upon actual operation scenarios. About 190,000 raw data were collected. We selected five machine learning algorithms: SVM, DT, KNN, ANN, and Naive Bayesian Classification. Kernel-based classification algorithms such as KNN and SVM were found to have better performance in diagnosing operation faults of freezer than other machine learning algorithms.