A positive linear decomposition for identifying patterns in dynamic process measurements
Oh Sang Woo, Alvis Salenieks, Michael Loizos Mavrovouniotis · 2005
The purpose of this article is to show the effectiveness of a pattern recognition method for high-dimensional dynamic measurements. The method consists of the following steps. SVD projects a matrix of dynamic process measurements on a low-dimensional subspace. A convex cone, defined by the non-negativity of measurements, is then created. For normalization purposes, a polygon is formed by intersecting the cone with a plane; its corners specify the feature vectors of the data. The polygon is reduced to a triangle with only the three most representative corners, enabling the automated selection of a feature vector for pattern recognition. Finally, a spanning tree created from the feature vectors classifies the patterns. In a case study, the feature vectors proved to be invariant to the width of the time window, and classification was possible even with feature vectors of differing time windows.