关键词: 根因分析, 非线性相异度分析, 对比格兰杰因果关系分析, 热轧带钢, 头部拉窄

Abstract: A root cause diagnosis model of head narrowing in hot rolled strip was proposed based on the dissimilarity index and comparative Granger causality analysis. Firstly, kernel entropy component analysis (KECA) was used to extract the feature of the original data, and  the dissimilarity evaluation was realized based on the dimensionality reduction under the nonlinear relationship. Secondly, comparative Granger causality analysis model was established. Therefore, the allowable fluctuation range of each causal relationship was calculated by batch data under a large number of normal conditions, which was regarded as a criterion corresponding to judge whether the causal relationship was abnormal. The final root cause diagnosis of the abnormal was determined by the causal relationships. Finally, the diagnosis model of the head narrowing in hot rolling was established to verify the effectiveness of the method using real industrial data. Results show that the method has better performance, which may accurately detect abnormal batches, and locate root cause of abnormal batches.

Key words: root cause analysis, nonlinear dissimilarity analysis, comparative Granger causality analysis, hot rolled strip, head narrowing