Abstract: The quality of thick plate shape is essential to steel companies,and thick plates are processed through several phases in the manufacturing process,including many factors.The problems of quality fluctuation and unknown causes of abnormal shape exist in processed thick plates.For this reason,a model for forecasting the quality of thick plate shapes with root cause analysis was proposed to overcome this issue by employing big industrial data of the thick plate manufacturing process.Owing to the requirement of the small batch production and the significant nonlinearity of production process,a shape quality prediction model of thick plate based on Kernel Partial Least Squares (KPLS) was established.Aiming at the unknown reasons for the abnormal shape quality of thick plates,which was difficult to identify the variables that truly affect the shape quality of thick plates among many process variables,a Mixed Integer Genetic Algorithm (MIGA) was proposed to analyze the production processes and variables that affect the shape of thick plates.The production process and variables that were most effective in predicting the quality of thick plates were selected for the final thick plate shape prediction and root cause analysis by embedding MIGA into KPLS modeling.The actual industrial process data from a steel plant and production process experiment were verified,and the results showed that the proposed algorithm could accurately forecast the quality of thick plate shape,analyze the root cause,and thus identify the essential variables impacting thick plate quality,which was helpful for actual production operations.

Key words: thick plate, Kernal partial least squares, quality prediction, optimal feature subset, root cause analysis

摘要： 厚板板形质量对于钢铁企业十分重要,由于厚板在生产过程中经过多个工序加工,涉及众多变量,加工后的厚板板形存在质量波动和异常原因未知的问题。针对这一难题,本文利用厚板生产过程中的工业大数据,提出一种具有根因分析能力的厚板板形质量预测模型。首先针对厚板板形质量预测问题,由于厚板小批量生产、生产过程变量非线性强,本文采用基于核偏最小二乘(KPLS)的方法建立厚板板形质量预测模型。然后针对厚板板形质量异常原因未知的问题,利用混合整数遗传算法(MIGA)分析影响厚板板形的生产过程工序和变量。通过将MIGA集成到KPLS建模中,选择厚板板形质量预测效果最好时的生产过程工序及变量,用于最终厚板板形质量预测和根因分析。最后,采用来自某钢铁厂的实际工业过程数据进行实验验证,通过工业实验表明,所提算法(KPLS-MIGA)能够实现对厚板板形质量的准确预测,并能够对厚板板形质量异常进行根因分析,从而寻找到影响厚板板形质量的关键变量,对实际的生产操作具有较好的指导意义。

关键词: 厚板板形, 核偏最小二乘, 质量预测, 优选特征子集, 根因分析