A Proposal of Condition Monitoring with Missing Data and Small-Magnitude Faults in Industrial Plants

Abstract

The patterns of incipient and small-magnitude faults are easily masked by the effect of interferences, missing data, and noisy measurements which are common in the industrial environments. Therefore, a smart data analysis of these patterns is needed for effectively minimizing the false and missing alarm rates resulting from noise, uncertainty, and unknown disturbances with the goal to achieve high detection performances, even in presence of missing data in the observations. This paper provides a novel methodology for the on-line imputation of missing data by using three techniques: Fuzzy C-means (FCM), Singular Value Decomposition (SVD), and Partial Least Squares regression (PLSr). Afterward, a data preprocessing stage using the KPCA and Exponentially Weighted Moving Average (EWMA-ED) is developed. The effectiveness of the proposal to obtain satisfactory results in the detection of small-magnitude faults was validated by using the Tennessee Eastman (TE) process benchmark.