Prediction and Sensitivity Analysis of Decarburization Layer Based on Functional Kernel Fisher Discriminant Analysis and Extended Morris Method

Abstract

As a representative of high-speed wire, cord steel has the advantages of high strength, high toughness and so on, and is widely used in transportation and other industries. Decarburization layer is one of the key factors affecting the quality of cord steel products. Excessive thickness of decarburized layer leads to the decrease of fatigue strength and wear resistance of the material, which affects the service life and performance stability of the product. The decarburization process is affected by many factors such as heating temperature, holding time, and the interaction between the factors makes the thickness of the decarburization layer difficult to be accurately controlled. In this paper, an online prediction model of decarburization layer based on functional kernel Fisher discriminant analysis (FKFDA) method is constructed, and an extended Morris screening method is proposed. First, scalar data and multivariable time series data are combined, and then a nonlinear classification model is constructed using FKFDA to establish the corresponding relationship between heterogeneous production data and actual decarburization layer. Then, the proposed FKFDA method is applied to the actual cord steel production dataset. The accuracy of the proposed method is 75.6%, and the G-mean value is 0.722 on the cord steel production dataset, which has higher prediction accuracy than other methods. Finally, the extended Morris method based on the curve shape change is proposed to get the key factors affecting the decarburization layer, and the results are consistent with the actual situation.