Spectral Graph Analysis for Process Monitoring

Abstract

Process monitoring is a fundamental task to support operator decisions under ab- normal situations. Most process monitoring approaches, such as Principal Components Analysis and Locality Preserving Projections, are based on dimensionality reduction. In this paper Spectral Graph Analysis Monitoring (SGAM) is introduced. SGAM is a new process monitoring technique that does not require dimensionality reduction techniques. The approach it is based on the spectral graph analysis theory. Firstly, a weighted graph representation of process measurements is developed. Secondly, the process behavior is parameterized by means of graph spectral features, in particular the graph algebraic connectivity and the graph spectral energy. The developed methodology has been illustrated in autocorrelated and non-linear synthetic cases, and applied to the well known Tennessee Eastman process benchmark with promising results.