An incipient fault detection approach via detrending and denoising

Abstract

An incipient fault tends to be buried by either the process trend or the measurement noise. Fault–trend ratio (FTR) and fault–noise ratio (FNR) are two main factors that impact the detection performance. An incipient fault detection approach is proposed in this paper based on the detrending and denoising techniques. There are three main phases in this approach. First, to increase FTR, a detrending algorithm is implemented. The fault detection rate can be significantly enhanced, when the normal trend is eliminated from the testing residual. Second, to increase FNR, a denoising algorithm is realized. The residual obtained from this algorithm can avoid the incipient fault being buried by the widely oscillating noise. Therefore the fault detection performance can be further improved. Third, the new detection statistic is composed based on the two algorithms. The approach is applied to a simulated process, a satellite attitude control system process, and the Tennessee Eastman process. The comparison results show that the proposed method outperforms the traditional Hotelling method in detecting incipient faults.