Combining multidimensional scaling and computational intelligence for industrial monitoring

Authors

Abstract

Large industrial complexes with hundreds of variables must be tightly monitored for safety, quality and resources optimization. Multidimensional scaling and computational intelligence are proposed in this work as effective tools for building classifiers of the operating state of the industrial process into normal / abnormal working regions. The VisRed, Visualization by Data Reduction computational framework, is extended with techniques from computational intelligence, such as neural networks (several architectures), support vector machines and neuro-fuzzy systems (in an evolving adaptive implementation) to build such classifiers. The Visbreaker plant of an oil refinery is taken as case study and some scenarios show the potentiality of the combined approach.

Keywords

computational intelligence, multidimensional scaling, neural networks, support vector machines, neuro-fuzzy systems, industrial monitoring

Subject

Industrial monitoring

Related Project

CLASSE - Classificação Sintética para Supervisão Industrial (Synthetic Classification for Industrial Supervision)

Conference

90th Industrial Conference on Data Mining ICDM 2009, January 2009