Abstract
This chapter reports five experiences in successfully applying different data mining techniques in a hotdip galvanizing line. Engineers working in steelmaking have traditionally built mathematical models either for their processes or products using classical techniques. Their need to continuously cut costs down while increasing productivity and product quality is now pushing the industry into using data mining techniques so as to gain deeper insights into their manufacturing processes. The authors’ work was aimed at extracting hidden knowledge from massive data bases in order to improve the existing control systems. The results obtained, though small at first glance, lead to huge savings at such high volume production environment. The effective solutions provided by the use of data mining techniques along these projects encourages the authors to continue applying this data driven approach to frequent hard-to-solve problems in the steel industry.
TopBackground
Galvanized steel is a product experiencing a growing demand in sectors like the automotive industry, domestic appliances manufacturing and construction industry, due to its anticorrosive properties. As the requirements of the clients are increasingly harder to meet, companies pursuing to lead the market need to follow a continual improvement strategy covering every stage of the galvanizing process (Tian, Hou & Gao, 2000; Fabrellas, Ruiz, Martínez, Sanz, & Larrazábal, 2002; Díaz, Cuadrado, Diez, Rodríguez, Obeso & Rodríguez, 2003; Rendueles, González, Díaz, Diez, Seijo & Cuadrado, 2006). Two are the key aspects that determine the quality of the product (Schiefer, Jörgl, Rubenzucker & Aberl, 1999; Tenner, Linkens, Morris & Bailey, 2001; Domínguez, Miranda-Ariz & Salvador, 2002):
- •
The anticorrosive properties conferred by the zinc coating. The protection against corrosion requires that the zinc layer has uniform thickness. A strict control of the base metal surface preparation, the temperature, composition and homogeneity of the coating, the speed of the band and the air flow throw the blades is mandatory.
- •
The mechanical properties. They basically depend on the chemical composition of the steel, the rolling process and the thermal treatments before immersion in hot liquid zinc.
Key Terms in this Chapter
Genetic Algorithm: A systematic method used to solve search and optimization problems and apply to such problems the principles of biological evolution, namely, selection of the ‘fittest’, sexual reproduction (crossover) and mutation.
Closed Loop Control System: A system that utilizes feedback as a mean to act over the driving signal according to the differences found amongst the observed and expected behavior.
Decision Tree: A classifier in the form of a tree structure, where each node is either a leaf node or a decision node. A leaf node indicates the value of the target attribute (class) of examples.
Steady-State Regime: Status of a stabilized system.
Learning Algorithm: In the artificial neural network area, the procedure for adjusting the network parameters in order to mimic the expected behavior.
Artificial Neuron: A mathematical model that generally applies a non-linear function to a linear combination of the input variables in order to provide an output value.
Overfitting: Fitting a model to best match the available data while loosing the capability of describing the general behaviour.
Artificial Neural Network: A set of connected artificial neurons.
Multilayer Perceptron: A specific kind of artificial neural network whose neurons are organized in sequential layers and where the connections amongst neurons are established only amongst the neurons of two subsequent layers.
Robust: Not affected by the presence of outliers.
Open Loop Control System: A system that does not rely on feedback to establish the control strategies.
Outlier: An observation that does not follow the model or pattern of the majority of the data.
Transient Regime: Status of a system that is changing from one steady-state regime to another.