连铸大方坯内部质量多元模糊模式预测研究

Prediction of internal quality of the continuous casting blooms based on multivariate fuzzy pattern recognition

连铸坯的质量控制对提高产品质量和降低生产成本具有重要作用,对生产过程中连铸坯的质量状况进行在线判定和预报已成为很多冶金学者和工程人员关心的热点问题。实际生产过程中,连铸坯质量缺陷包括铸坯表面、内部和形状等各种缺陷类型,且各类缺陷的成因复杂,难以全部采用机理模型进行描述。多元模糊模式识别是基于模糊集理论对具有不确定性和非线性关系的系统能够有效辨识的一种方法。本文将多元模糊模式识别应用于连铸坯质量缺陷的判定系统中,给出了缺陷类型判定的详细步骤和方法。以莱钢特殊钢厂20CrMnTiH齿轮钢连铸大方坯为研究对象,结合冶金理论分析和主成分分析法确定影响连铸坯内部质量的工艺参数,将铸坯无缺陷(合格)、角部裂纹、中间裂纹、中心裂纹以及中心偏析缺陷作为标准模式,并通过统计分析得到不同缺陷模式下各因素的隶属函数,采用最大隶属原则对20CrMnTi连铸坯的质量缺陷类型进行判定,预测准确率为81.82%。结果表明,该方法能够准确地对浇铸过程中连铸坯发生各类缺陷的类型进行预测,在实际生产中具有一定的实用性。

The quality control of continuously cast bloom plays an important role in improving the quality of products and reducing the production cost. Therefore, numerous metallurgist and scholars have paid great attention to the judgment and prediction of quality of continuously cast bloom. Generally speaking, the quality defects of continuously cast bloom include surface defects, internal defects, shape defects and so on. Due to the complex cause of the defects, it is hard to describe them all by mechanism model. The multivariate fuzzy pattern recognition, which is based on fuzzy set theory, is an effective method for the recognition of uncertain and nonlinear system. Based on above consideration, the multivariate fuzzy pattern recognition is applied to the judgment of quality defects of continuously cast bloom, and detailed procedures are clarified in this paper. Taking the production of 20CrMnTiH continuously cast bloom in Lai Wu special steel plants for example, the influential technological parameters of internal defects are fixed through combining the methods of theoretical analysis and principal component analysis. Meanwhile, the continuously cast bloom with zero defects, comer crack, central crack, eenterline crack and center segregation are set as standard patterns, and the membership functions of each factor under different defects patterns are acquired with statistical analysis. The judgments of defect types of 20CrMnTi with maximtun membership principle indicate that the hit rate is 81.82%. In other words, this method could predict the defect types of continuously cast bloom with high precision, and could provide a relatively good reference for real production.