Statistical monitoring of dynamic multivariate processes. Part 2. Identifying fault magnitude and signature

This paper presents the second part of the two-part analysis of statistical monitoring of complex multivariate processes. Part I introduced an effective method to remove both autocorrelation and cross-correlation from the monitored variables. This part shows that this method, i.e., the removal of correlation, can considerably influence the magnitude and signature of fault conditions. To overcome this problem, this part introduces a compensation scheme that retains the correct magnitude and signature of step-type faults. The paper extends the compensation scheme to provide accurate estimates of the magnitude and signature of general deterministic fault conditions. The utility of this scheme is shown using simple univariate and multivariate examples, as well as an application to a benchmark simulator, and recorded data from an industrial distillation unit.