플랜트 공정 FD FAN 설비의 이상 상태 예측을 위한 센서 데이터 분석

초록

This paper addresses the problem of anomaly detection and the system identification in an FD (Forced Draft) Fan monitored by performance sensors. The measured time series data were collected under the normal operating conditions, whereas the measured data included one failure duration. A technique of time-series methods for fault detection, identification and estimation in FD fan is presented. System identification is based upon Nonlinear ARX (NARX) models in which a multivariate regression method is employed for anomaly detection and fault magnitude estimation. NARX learned the system identification or regression model using the measured data in the normal state. Then, NARX predicts the feature signal using the input signals, and determines the abnormality by measuring the difference between the predicted feature signal and the measured feature signal. Dynamic time warping (DTW) is applied to estimate the abnormal score of the measured signal. Finally, it is shown that the proposed method can detect the initiation of the abnormal status.

키워드