Prediction of Charpy V-notch impact energy of low carbon steel by using shallow neural network and deep learning
17 July 2020
Revised: 6 August 2020
Accepted: 10 August 2020
Available online: 14 August 2020
Abstract: In the present work, the impact energy prediction model of low carbon steel was investigated based on the industrial data. A three layer neural network, extreme learning machine and deep neural network were compared with different activation functions, structure parameters and training functions. To determine the optimal hyper-parameters of deep neural network, Bayesian optimization was applied. The model with best performance was applied to investigate importance degree of process parameter variables on impact energy of low carbon steel. The results show that deep neural network obtains better prediction results than that of shallow neural network due to the multiple hidden layers improving the learning ability of the model. Among all the models, the Bayesian optimization deep neural network achieves the highest correlation coefficient of 0.9536, lowest mean absolute relative error of 0.0843 and lowest root mean square error of 17.34 J for predicting the impact energy of low carbon steel. Among all the variables, the main factors affecting the impact energy of low carbon steel with final thickness of 7.5 mm are the thickness of the original slab, the thickness of intermediate slab and rough rolling exit temperature on the specific hot rolling production line.