Plant Data Visualization Using Non-Negative Matrix Factorization

Authors:	Tapson Jonathan, University of Cape Town, South Africa Greene John, University of Cape Town, South Africa
Topic:	6.4 Safeprocess
Session:	Fault Diagnosis and Fault Tolerant Control: Theory
Keywords:	Failure detection, data reduction, spectrum analysis, computer aided diagnosis

Abstract

Non-negative matrix factorization (NMF) is a method for dimensionality reduction and simplification of large data sets. Unlike tools such as principal components analysis (PCA) and factor analysis, NMF produces basis vectors that correspond to perceptible features in the original data. This is particularly useful when working with data where visual interpretation of the simplified representation is required. Typical data of this type is condition monitoring (CM) data, where visual interpretation of vibration spectra is a standard diagnostic tool. The results suggest that NMF processing of CM data simplifies the visual interpretation process, and opens the way for automation of this task.