ARCHIVED - Application of Data Mining Techniques for Industrial Process Optimization

Download Full Document (PDF document, 435,5 Ko)

Radu Platon and Mouloud Amazouz, CanmetENERGY

CETC Number 2007-141 / 2007-07-30

Abstract

Data mining can be defined as the science of extracting useful information from large data sets or databases. Data mining is used for building empirical models, which are based not on the underlying theory about the process or mechanism that generated the data. Data mining, as the name suggests it, is data-driven, and it provides a description of the observed data. Its fundamental objective is to provide insight and understanding about the structure of the data and its important features, and to discover and extract patterns contained in the data set. This discipline also referred to as knowledge discovery in databases (KDD), is a process of extracting implicit, previously unknown, and potentially useful information from data (Shapiro et al, 1992). Data mining brings together a multitude of disciplines, such as database systems, statistics, artificial intelligence, data visualization, and others.

The discovered knowledge can be applied to information management, query processing, decision-making, process control, and many other applications. Mining information and knowledge from commercial or industrial data, and applying this information to new and innovative uses has been recognized by companies as an important area for generating revenues and increasing business opportunities.

The availability of high volume data in the industrial sector, have given rise to a new level of interest in the applications of knowledge discovery and data mining for industrial applications.

Data mining products are commercially available and numerous industrial applications are being developed.

The main objective of the report is to provide an overview of data mining methods suitable for industrial process applications, by examining the publications related to data mining applications that are being developed or that are already implemented in the industry. A survey of this published work provides not only current trends, but also a better understanding of the different applications required by the industry, and the data mining methods used for these applications.

In the context of the industrial applications, the scope of this paper covers industrial processes such as pulp & paper, and petrochemical operations, with applications geared mainly towards process monitoring and control, process parameter value inference (soft sensors), detection of abnormal situations (faults) and their diagnostic and a general improvement of the process understanding through discovery of correlations between process parameters.

A general overview of the main tasks performed by data mining and methods used to achieve these tasks is provided.

A summary of the data pre-processing steps performed for a data set containing historical process data of Smurfit Stone's pulp & paper plant at La Tuque (Québec) is presented. Commercial software packages used for developing and implementing these applications were also examined. This proved to be useful, since most companies provide information about industrial applications of their respective software packages.

To learn more about CanmetENERGY's activities related to industrial systems, visit the Industrial Systems Optimization section of the website.