Property prediction of continuous annealed steels

Wigley, Nicholas Roy 2012. Property prediction of continuous annealed steels. EngD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (6MB) | Preview
	PDF - Supplemental Material Restricted to Repository staff only Download (91kB)
Preview	PDF - Accepted Post-Print Version Download (6MB) | Preview

Abstract

To compete in the current economic climate steel companies are striving to reduce costs and tighten process windows. It was with this in mind that a property prediction model for continuous annealed steels produced at Tata Steel’s plants in South Wales was developed. As continuous annealing is one of the final processes that strip steel undergoes before being dispatched to the customer the final properties of the strip are dependent on many factors. These include the annealing conditions, previous thermo-­‐mechanical processing and the steel chemistry. Currently these properties, proof stress, ultimate tensile strength, elongation, strain ratio and strain hardening exponent, are found using a tensile test at the tail end of the coil. This thesis describes the development of a model to predict the final properties of continuous annealed steel. Actual process data along with mechanical properties derived using tensile testing were used to create the model. A generalised regression network was used as the main predictive mechanism. The non-­‐linear generalised regression approach was shown to exceed the predictive accuracy of multiple regression techniques. The use of a genetic algorithm to reduce the number of inputs was shown to increase the accuracy of the model when compared to those trained with all available inputs and those trained using correlation derived inputs. Further work is shown where the fully trained models were used to predict the relationships that exist between the processing conditions and mechanical properties. This was extended to predict the interaction between two process conditions varying at the same time. Using this approach produced predictions that mirrored known relationships within continuous annealed steels and gives predictions specific to the plant that could be used to optimise the process.

Item Type:	Thesis (EngD)
Status:	Unpublished
Schools:	Engineering
Subjects:	T Technology > TJ Mechanical engineering and machinery
Uncontrolled Keywords:	Steel; annealing; property prediction; generalised regression neural network; genetic algorithm; modelling.
Funders:	EPSRC/TATA Steel
Date of First Compliant Deposit:	30 March 2016
Last Modified:	16 Oct 2023 14:54
URI:	https://orca.cardiff.ac.uk/id/eprint/38602

Actions (repository staff only)

Edit Item