통계적 기계학습법을 적용한 플랜트 공정 장비의 시계열 데이터 분석및 고장 상태 진단

Plant equipment are designed to operate in condition of minimized failures during its service life. Because of the uncertainties involved in operating process, the monitoring and maintenance are very important to avoid the catastrophic failure of the equipment during the service life. This research gives a diagnosis system for a main equipment of oil-plant industry to designing pattern recognition and classification scheme based on machine learning approach. The sensor data of the equipment was used for analysis after eliminating unwanted noise which may cause errors in pattern recognition through preprocessing. In addition, the sensor data that can be an indicator of failures is selected based on the maintenance history, and descriptive statistical values that can be a characteristic of each sensor is extracted at certain time intervals. And extracted feature values are projected on 2-dimensional space through PCA (Principal Component Analysis) that is one of dimensionality reduction algorithms. The failure history data was classified using the Naive Bayesian classification method. Finally, the suggested diagnosis system based on machine learning approach was tested by using some of the actual data and verified the availability.