基于数据挖掘的冷连轧过程板形缺陷预测与诊断方法研究

发布时间：2018-09-03 09:07

【摘要】：冷连轧过程具有多工况、多变量、非线性、大数据等特点,板形缺陷成因复杂,基于机理模型的方法难以胜任板形缺陷的在线预测和诊断。本文以宝钢2030mm冷轧过程DSR板形控制系统为应用对象,采取多元统计分析、数据挖掘等数据驱动技术,研究多工况下冷轧过程的板形预测及缺陷识别问题。针对DSR动态板形控制系统的多工况以及某些工况下的产品缺陷数据具有小样本的特点,提出了一种基于支持向量机(SVM)的板形在线监测方法。首先,将过程变量进行主成分分析(PCA),所得到的主成分作为支持向量机的输入,将板形变量正交分解后作为支持向量机的输出,通过迭代计算、训练,最终得到多输入单输出的板形回归预测模型;然后,利用贝叶斯准则优化支持向量机的系数,更新支持向量机回归模型,克服不确定性信息对板形预测精度的影响。实验结果表明,该方法可以有效解决多工况下的快速板形建模问题,板形预测精度高。针对DSR动态板形控制系统的多工况、海量数据等特点,改进了传统的频繁模式挖掘算法——Apriori算法,实现多工况下板形缺陷数据的频繁模式挖掘。首先,利用主成分分析(PCA)进行过程变量降维;然后,利用PCA方法中的SPE统计量界限设置阈值,挑选故障数据;再用改进的Apriori算法对板形缺陷的关联性特征进行挖掘;最终得到每种缺陷的频繁项,形成相应的诊断知识用于故障原因的识别。实验结果表明,此方法诊断结果和现场专家的诊断结果吻合,方法可信度高,具有良好的工程应用前景。
[Abstract]:The cold continuous rolling process has the characteristics of multi-working condition, multi-variable, nonlinear, big data and so on. The formation of shape defects is complex, and the method based on mechanism model is difficult to predict and diagnose the shape defects on line. Taking the DSR shape control system of Baosteel 2030mm cold rolling process as the application object, this paper studies the shape prediction and defect identification of cold rolling process under multiple working conditions by using data driven techniques such as multivariate statistical analysis and data mining. In view of the fact that the product defect data of DSR dynamic flatness control system has a small sample under many working conditions and some working conditions, an on-line flatness monitoring method based on support vector machine (SVM) is proposed. Firstly, the principal components obtained by (PCA), are used as the input of support vector machine, and the shape variables are decomposed as the output of support vector machine. Finally, the shape regression prediction model with multiple input and single output is obtained. Then, the coefficients of support vector machine are optimized by Bayesian criterion, and the regression model of support vector machine is updated to overcome the influence of uncertainty information on shape prediction accuracy. The experimental results show that this method can effectively solve the problem of rapid shape modeling under multiple working conditions, and the shape prediction accuracy is high. In view of the characteristics of DSR dynamic flatness control system such as multi-condition and massive data, this paper improves the traditional algorithm of frequent pattern mining, and realizes the frequent pattern mining of shape defect data under multiple working conditions. First, the process variables are reduced by principal component analysis (PCA); then, the threshold of SPE statistics in PCA method is set to select fault data; and then the correlation features of shape defects are mined by improved Apriori algorithm. Finally, the frequent items of each defect are obtained, and the corresponding diagnosis knowledge is used to identify the fault cause. The experimental results show that the diagnostic results of this method are in good agreement with those of experts in the field, and the method has high reliability and has a good prospect of engineering application.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG335.9;TP311.13

【参考文献】