Remote supervision of industrial processes based on PMML-defined SOM models

Abstract

Visualization techniques are the subject of a growing interest for processing and interpretation of large volumes of multidimensional data, giving rise to the so called visual data mining paradigm. A prominent approach for multidimensional data visualization is based on the Self-Organizing Map (SOM), which allows to project the input data on a 2D or 3D space that can be visualized. This approach has been used in this work for supervision and modeling of industrial processes. In this field, a large amount of variables have to be managed and joined. Most of them are located in distant places, suggesting the use of Information Technologies (IT) based on the Internet for remote supervision. In this work, a growable and modular client-server architecture is proposed for the remote supervision. In this architecture, the SOM models of the industrial processes are stored in the standard format PMML (Predictive Model Markup Language), so that any remote client that supports this standard can interpret them. The aim is to create an efficient structure of distributed supervision.