For industrial processes

there are not only correlations among variables

but also autocorrelation in temporal structure of these variables

therfore

a new sensor fault detection and diagnosis method based on min/max autocorrelation factors (MAF) was proposed in this work. Firstly

MAF analysis of historical normal data was made. Then

strong autocorrelation factors and weak autocorrelation factors were obtained. Based on these factors

the statistics for fault detection were constructed and corresponding contribution plots were derived. The proposed method was applied to the continuous stirred tank reactor (CSTR) and compared with the principal component analysis method. Simulation results demonstrated that the proposed method could detect sensor faults with slow variation more quickly with less false-alarm. The contribution plots based on MAF can explain complicated sensor fault more reasonably than principal component analysis (PCA)

which is a classical multivariate statistical method for process monitoring.