Frontier Materials & Technologies (Sep 2023)
Concerning the selection of areas with a dominant type of dependence when analyzing production control data
Abstract
The formation of representative databases determines the interest in forecasting and managing the quality of metal based on data mining using special software products often based on regression analysis and not always taking into account the statistical nature of an object of study itself. This can lead to misinterpretation of the results or incomplete extracted information reducing the efficiency of statistical processing. Based on the analysis of the production database of the technology for producing 13G1S-U sheet steel, the authors evaluated the possibilities of multiple linear regression for predicting the quality of a steel sheet. The study shows that the type of distribution of the values of control parameters, the distribution nature of which was estimated based on the determination of the skewness and kurtosis coefficients, limits the regression forecast depth. Due to the great deviation of the predicted models from the experimental values in the right tail area of the distribution of the impact strength values, in this work, the authors developed the methods for separating data arrays and proposed criteria to compare the obtained results. To assess the accuracy of the results obtained, arrays with a deliberately asymmetric distribution were selected from the initial sample, against which the statistical characteristics were also compared. Based on the proposed techniques, the authors identified the dominant chemical elements that contribute to the difference in the distribution of the values of acceptance properties existing within the same standard technology. The study shows that the proposed separation method can be used as a variation of cognitive graphics techniques to identify areas with a dependence dominant type based on the correlation of skewness and kurtosis coefficients.
Keywords
