Neural Networks to Predict The Properties of a New type of Batch Hot Dip Galvanized Steel
Date of Award
Doctor of Philosophy (PhD)
Graduate School of Computer and Information Sciences
Marlyn Kemper Littman
William Van Ooij
Zinc is used widely as a corrosion resistant coating on steel. However, in Europe, zinc is considered an environmental pollutant. Zinc-aluminum alloy coatings may be able to minimize the leaching of zinc into the environment. There are a number of thermodynamic and chemical variables that make the zinc-aluminum coating on steel more durable, in terms of thickness, hardness, corrosion rate, and roughness, thus minimizing zinc pollution and enhancing the durability of the galvanized product. Among these variables are the galvanizing bath temperature and chemical composition, and item immersion time and withdrawal rate. One way to investigate the interaction of these variables is via the use of neural networks. Neural networks are especially useful in mapping independent variables (such as temperature, percentage of alloying metals) to dependent variables (such as zinc layer hardness and thickness) that may be related to each other in a nonlinear fashion. Neural networks learn to recognize patterns, and subsequently store their knowledge after being exposed to a set of sample patterns. This stored knowledge can be used for prediction given a different set of input patterns. The ability of neural networks to predict output after the learning process has been completed makes them very useful in the study of galvanizing. Instead of having to use a great number of experiments to determine the characteristics of the galvanized zinc layer, a neural network can be used to model the galvanizing process, and can predict the properties of the galvanized layer, again, without having to resort to endless experiments - "a Virtual Galvanizing Laboratory." Actual laboratory experiments using zinc-aluminum alloy galvanizing baths were conducted in Rhesca Laboratories, Helsinki, Finland, using a Hot Dip Galvanizing Simulator. A total of thirty six samples were used in the study, thirty for training, and six for testing. A feedforward neural network was successfully trained using Neurodimension's Neurosolutions software. The results of the training and testing of the network exhibited correlation between the bath composition, dipping time, temperature and the sample thickness, roughness, corrosion resistance and hardness. These results indicate that it is possible to construct a "Virtual Galvanizing Laboratory" in which "virtual experiments" can be performed to predict the thickness, roughness, corrosion resistance and hardness properties of zinc-aluminum galvanized steel.
Ernest E. Klerks. 2002. Neural Networks to Predict The Properties of a New type of Batch Hot Dip Galvanized Steel. Doctoral dissertation. Nova Southeastern University. Retrieved from NSUWorks, Graduate School of Computer and Information Sciences. (640)