Influence of Variation/Response Space Complexity and Variable Completeness on BP-ANN Model Establishment: Case Study of Steel Ladle Lining

Abstract

Artificial neural network (ANN) is widely applied as a predictive tool to solve complex problems. The performance of an ANN model is significantly affected by the applied architectural parameters such as the node number in a hidden layer, which is largely determined by the complexity of cases, the quality of the dataset, and the sufficiency of variables. In the present study, the impact of variation/response space complexity and variable completeness on backpropagation (BP) ANN model establishment was investigated, with a steel ladle lining from secondary steel metallurgy as the case study. The variation dataset for analysis comprised 160 lining configurations of ten variables. Thermal and thermomechanical responses were obtained via finite element (FE) modeling with elastic material behavior. Guidelines were proposed to define node numbers in the hidden layer for each response as a function of the node number in the input layer weighted with the percent value of the significant variables contributing above 90% to the response, as well as the node number in the output layer. The minimum numbers of input variables required to achieve acceptable prediction performance were three, five, and six for the maximum compressive stress, the end temperature, and the maximum tensile stress. empirical equations into other fields are far less satisfying. Equations (1) and (2) show more robust applications, yet they yield a rather larger number for trials.

Heat Transfer in Steel Ladles: Models and Applications

“…In particular, this type of modeling can be divided into two classes: hybrids, also known as gray-box modeling, which start from the premise that information inherent to the process derived from physical modeling can provide important gains to the model and, therefore, should be used to compose the solution to the problem (Tian et al, 2008;Ahmad et al, 2014;He et al, 2014;Botnikov et al, 2019;Song et al, 2019); and non-hybrids, called blackbox modeling, in which little or no prior knowledge of the system is needed. In general, this type of modeling is composed of algorithms based purely on machine learning, capable of identifying patterns between information and then predicting and executing tasks (Laha et al, 2015;Klanke et al, 2017;Wang et al, 2018;Hou et al, 2019;Jo et al, 2019).…”

“…The results of the application revealed that the models presented considerable precision in the prediction and are satisfactory for the practical production process. Hou et al (2019) applied a BPNN to predict the thermal and thermomechanical responses of a steel ladle, consider-ing variables related to the properties of refractory linings and ladle geometry. To this end, five orthogonal matrices were used for finite element simulations and training of the neural model, with the aim of organizing the combination of ten characteristics in the variables space.…”

“…In order to define the main factors that most impact heat losses from ladles, Fredman (2000) reviewed part of the literature, starting in 1956, which deals with the control of the thermal status of liquid steel, not only in ladles, but also in all stages between the tapping of the converter and the continuous casting. From the beginning of the 21st century, the works found in the literature began to apply different types of models, especially physical, mathematical, empirical and statistical ones (Gupta and Chandra, 2004;Jormalainen and Louhenkilpi, 2006;Tian et al, 2008;Wu et al, 2012;Ahmad et al, 2014;He et al, 2014;Laha et al, 2015;Wang et al, 2018;Hou et al, 2019).…”

Method of raw materials selection for production of the MgO‐C bricks of comparable properties using PCA and K‐medoids

“…With the use of selected supervised techniques it is possible to predict the wear rate of refractories installed in heat devices and its parameters. [28][29][30][31] Unsupervised techniques are also used when it is necessary to find comparable groups of data in large data sets. 32 The aim of this work is to apply unsupervised machine learning algorithms to identify MgO-C materials of optimal quality considering basic properties.…”

“…Selection of such materials may be done with the use of statistical tools and machine learning techniques which are used widely in a field of refractory materials applications. With the use of selected supervised techniques it is possible to predict the wear rate of refractories installed in heat devices and its parameters 28–31 . Unsupervised techniques are also used when it is necessary to find comparable groups of data in large data sets 32 …”