Spectral Graph Analysis for Process Monitoring

Abstract

Process monitoring is a fundamental task to support operator decisions under abnormal situations. In this article, spectral graph analysis monitoring (SGAM) is introduced. The approach is based on the spectral graph analysis theory. First, a weighted graph representation of process measurements is developed. Second, the process behavior is parametrized by means of graph spectral features, in particular, the algebraic connectivity and the spectral energy. The developed methodology is illustrated in autocorrelated and nonlinear synthetic cases and applied to the well-known Tennessee Eastman process benchmark with promising results.