Multivariate Detection of Transient Disturbances for Uni- and Multirate Systems

Abstract

This paper presents a method to detect transient disturbances in a multivariate context, and an extension of that method to handle multirate systems. Both methods are based on a time series analysis technique known as nearest neighbors, and on multivariate statistics implemented as a singular value decomposition. The motivation for these developments is that there is an increasing industrial requirement for the analysis of data sets comprising measurements from industrial processes together with their associated electrical and mechanical equipment. These systems are increasingly affected by transient disturbances, and their measurements are commonly sampled at different rates. This paper demonstrates superior results with the multivariate method in comparison with the univariate approach, and with the multirate method in comparison to a unirate method, for which the fast-sampled measurements had to be downsampled. The method is demonstrated on experimental and industrial case studies.