Prediction of Longitudinal Facial Crack and Optimization of Process Parameters during Continuous Casting of Slabs Using Artificial Neural Network Models

Steel undergoing the peritectic transformation are prone to cracking during continuous casting. Although many researches and developments offered solutions to prevent the occurrence of these cracks, the subject is still beset with uncertainties. Several technologies and methods are available for prediction of cracks after slabs casting though, a significant number of the semi finished product are still affected by defects. It leads to rejection of slabs and results in subsequent losses. The paper presents a neural network model to optimize a set of process variables that can avoid the formation of longitudinal facial crack. The neural network models entail training of neural nets by employing the plant data collected from a continuous casting process viz casting speed, super heat, heat fluxes from two broad faces and two narrow faces of mould as inputs and the probable location of crack as output. The model is then validated by testing of network for its accuracy against 1/3rd of data, which were not used in training the neural network. These were fed to the trained network and invoked to predict the possible location of crack as an output. The model verification strategy involved analyzing the output for its abnormality in crack formation and subsequently specifying their corresponding fluctuation in the input parameters with respect to the optimized set point values of input process variables. The process variables were optimized by conducting their sensitivity analysis and using connection weight approach. The ANN model was validated with different casting conditions. The model depicts the predicted crack location obtained by the back-propagation neural network methods. The paper demonstrates that it can satisfactorily predict the conditions leading to longitudinal facial crack generation