Determination of fault propagation by time delay estimation using k nearest neighbour imputation

Abstract

The identification of faults in control loops is sometimes a very difficult procedure, but even if it has been achieved, one very important topic will often be unsolved in the presence of plant-wide faults. This topic is the analysis of fault propagation. Via fault propagation analysis it is possible to determine which control loop is either the source of fault or at least nearest to the unmeasured source of fault. In the present paper, some well-known methods for time delay estimation will be shortly summarized and a new and more effective one based on k nearest neighbour imputation will be shown. Knowledge about plant-wide time delays allows an analysis of fault propagation and hence identification of fault source. Advantage of the new method is that it can be used for time-invariant nonlinear and MISO (Multi-Input, Single-Output) systems as well. Moreover, this method can be completely automated as it does not use any parameters that have to be manually adjusted.