Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.719165
Title: Optimisation of the heat treatment of steel using neural networks
Author: Tenner, Jonathan
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2000
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Heat treatments are used to develop the required mechanical properties in a range of alloy steels. The typical process involves a hardening stage (including a quench) and a tempering stage. The variation in mechanical properties achieved is influenced by a large number of parameters including tempering temperature, alloying elements added to the cast, quench media and product geometry, along with measurement and process errors. The project aim was to predict the mechanical properties, such as Ultimate Tensile Strength, Proof Stress, Impact Energy, Reduction of Area and Elongation, that would be obtained from the treatment for a wide range of steel types. The project initially investigated a number of data modelling techniques, however, the neural network technique was found to provide the best modelling accuracy, particularly when the data set of heat treatment examples was expanded to include an increased variety of examples. The total data collected through the project comprised over 6000 heat treatment examples, drawn from 6 sites. Having defined a target modelling accuracy, a variety of modelling and data decomposition techniques were employed to try and cope with an uneven data distribution between variables, which encompassed nonlinearity and complex interactions. Having not reached the target accuracy required the quality of the data set was brought into question and a structured procedure for improving data quality was developed using a combination of existing and novel techniques. IV The stability of model predictions was then further improved through the use of an ensemble approach, where multiple networks contribute to each predicted data point. This technique also had the advantage of enabling the reliability of a given prediction to be indicated. Methods of extracting information from the model were then investigated, and a graphical user interface was developed to enable industrial evaluation of the modelling technique. This led to further improvements enabling a user to be provided with an indication of prediction reliability, which is particularly important in an industrial situation. Application areas of the models developed were then demonstrated together with a genetic algorithm optimisation technique, which demonstrates that automatic alloy design under optimal constraints can now be performed.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.719165  DOI: Not available
Share: