Iterative Multidimensional Scaling for Industrial Process Monitoring

Authors

Abstract

Monitoring high dimensional industrial processes for performance analysis and improvement should be done in a low-dimensional space where human-computer interaction is effective and where the human operator can easily indentify the main actual features of the process. For this purpose space reduction must keep the relevant and informative geometric characteristics of the original space, using proper metrics. In this work the reduction of n-dimensional space to bidimensional one is developed through multidimensional scaling with a proposed iterative capability. In the 2-D process map, named POM- Projected Orientation Map, the operational regions of the process under specific conditions can be classified. This classification is made by recursive clustering Dignet algorithm, giving information to the monitoring system about the possible quality of the running operating conditions. This strategy is applied to the process of Hydro Desulfuration (HDS) from Refinery of Petrogal at Sines (Galp Energia

Keywords

process monitoring; data mining; refinery

Subject

supervision and monitoring data mining

Related Project

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

Conference

2006 IEEE International Conference on Systems Man and Cybernetics, October 2006