Development of a Model for Predicting the Thermophysical Properties of Carbon Materials and Proposal of Manufacturing Conditions Using the Model

Abstract

A steelmaking method using electric furnaces is attracting attention in the iron and steel industry, and a carbon material called needle coke is used as an aggregate for the electrode in electric steelmaking. The performance of needle coke as an aggregate for electrodes in steelmaking is greatly affected by the quality of the needle coke, which depends on the ingredients of the raw materials and the process conditions. Because the raw material ingredients are not always constant and depend on the place and time they are produced, the quality of the needle coke is not stable under the same process conditions. Therefore, it is necessary to optimise the process conditions. In this study, to optimise the process conditions using machine learning, a model was constructed to predict the thermophysical properties of needle coke from the raw material ingredients and process conditions based on previous data. Because the subject plant is operated in a dynamic process and there is a time delay in the previous data, the genetic-algorithm-based process variables and dynamics selection method, which selects the time delays and process variable regionally, was studied. Furthermore, inverse analysis was performed on a sample whose quality was considered to be outside the specifications based on the previous data, with the aim of controlling the quality within the product specifications by changing only the process conditions.