Neural Network Breakout Prediction Model for Continuous Casting
Lameck Mugwagwa1, Lungile Nyanga2, Samson Mhlanga3
1Lameck Mugwagwa, Industrial and Manufacturing Engineering, National University of Science and Technology Bulawayo, Zimbabwe.
2Lungile Nyanga, Industrial and Manufacturing Engineering, National University of Science and Technology Bulawayo, Zimbabwe.
3Samson Mhlanga, Industrial and Manufacturing Engineering, National University of Science and Technology, Bulawayo, Zimbabwe.
Manuscript received on November 22, 2012. | Revised Manuscript received on December 03, 2012. | Manuscript published on December 30, 2012. | PP: 380-383 | Volume-2, Issue-2, December 2012. | Retrieval Number: B0958112212 /2012©BEIESP
Open Access | Ethics and Policies | Cite
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Continuous casting is a process in which liquid steel is cooled in a bottomless mould into semi-finished steel products called billets, blooms or slabs depending on their cross section. In the process of continuous casting, two of the major problems encountered are cracks and breakouts. Breakouts usually result in temporary shutdown of the caster and huge amounts of downtime. Primary cracks which form before the solidifying strand exits the mould, are invariably linked to breakouts. Controlling primary cracks results in reduced chances of breakouts. This work aims at designing a breakout prediction neural network model. In this paper, a two-layer feed forward backpropagation neural network model is developed for predicting the existence of primary cracks that might lead to a breakout. The network obtains its inputs in form of temperature values from rows of thermocouples attached to the mould tube. Based on solidification characteristics of steel, the neural network is supplied with various inputs (of temperature values) and targets and is trained to predict the crack status in the mould. Training is performed using the Levernberg-Marquardt (trainlm) training algorithm, and the log sigmoid transfer function was used for both the hidden and output layer. The output from this neural network was a logical 1 (if a primary crack is present) and a logical 0 (if no primary crack is present). The neural network model is validated by simulating in MatLab/Simulink.
Keywords: Continuous casting, breakout prediction, neural network.