联合数据驱动的轧钢过程质量相关故障诊断方法及其应用

With modern rolling industrial process increasingly complex and large, prevention monitoring and fault diagnosis have become the key to ensure safe operation, improve product quality and achieve economic. Due to the complex operation mechanism, sheer size, complex operation conditions, chaotic environment and vague boundary conditions, it’s quite tough to realize effective process monitoring based on traditional data-driven approach. As a result, fault diagnosis technology based on joint data-driven is increasingly concerned. This project takes the hot strip mill process (HSMP), which is a multiple batches, multi-mode and complex dynamic batch process, as the research background and attempts to adopt joint data-driven to investigate theory and techniques of quality-related fault detection, fault propagation path identification and fault location based on data and knowledge. The problems addressed in the project include: 1) Extraction method of quality causality topology for rolling industrial processes; 2) Quality-related monitoring mode for multi-batch and multi-mode processes based on joint data-driven; 3) Quality-related fault detection, fault propagation path identification and fault location based on joint data-driven; 4) Evaluation index and application verification for quality-related fault diagnosis based on data-driven. The project aims at establishing a set of quality-related fault diagnosis theory and method, which would form some fundamental theory and application benchmark on data and knowledge based joint data-driven fault diagnosis for complex batch processes.

现代轧钢工业过程日趋规模化、复杂化，预防监测与故障诊断已成为保证运行安全、提高产品质量、实现经济效益的关键。由于轧钢过程机理复杂、规模庞大、工况多变、运行环境恶劣、边界条件模糊等原因，传统的数据驱动方法难以实现有效的过程监控，基于数据与知识的联合数据驱动的故障诊断技术日益受到关注。为此，本项目以带钢热连轧这一多批次多模态复杂动态间歇过程为背景，基于数据与知识采用联合数据驱动思路研究质量相关的故障检测、故障传播路径辨识与故障定位理论与技术，拟研究和解决以下几方面问题：1）轧钢工业过程质量因果拓扑图提取方法；2）联合数据驱动多批次多模态质量相关的监测模型研究；3）联合数据驱动的质量相关故障检测、故障传播路径辨识与故障定位方法；4）数据驱动的质量相关故障诊断评价指标及应用验证。旨在建立一套质量相关故障诊断理论与方法，为解决复杂间歇过程基于数据与知识的联合数据驱动故障诊断提供基础理论和应用范例。

现代流程工业正逐渐倚重于生产小批量、多品种、高附加值产品的间歇生产过程，通过过程监测来保障动态过程的安全可靠运行以及连续稳定的产品质量逐渐成为过程控制的首要任务。复杂工业过程具有机理复杂、规模庞大、工况多变、运行环境恶劣、边界条件模糊的特点。单一数据驱动的建模与监测方法难以解析质量相关故障的多源和多征兆特性，易造成质量相关故障传播路径的虚假连接。本项目以带钢热连轧这一多工况复杂动态间歇过程为研究背景，采用知识与数据联合驱动的思想研究了质量相关的故障检测、故障传播路径辨识与故障定位理论与技术。面向实际流程工业过程的非线性、动态性、多工况等实际复杂特性，主要解决了：1）如何设计变量相关性指标及其阈值构建规模适度有向无环的质量因果拓扑图，2）如何在知识指导下实现质量因果拓扑图的修剪和优化，3）如何将基于数据的定量分析思想与基于知识的定性分析思想相结合，建立知识与数据联合驱动的质量监测模型，4）如何利用定量技术搜索故障传播路径和故障定位以及多源故障诊断的关键技术问题，从而为复杂间歇过程质量相关的故障诊断提供了一套新技术和解决方案。上述研究成果在鞍山钢铁集团公司带钢热连轧生产线上进行了实际应用验证，取得了良好的过程监测效果。此外，项目组还对面向系统层级多工况下带钢热轧全流程质量监测与故障诊断方法进行了探索性研究，取得一定的研究成果。在应用前景方面，研究成果将为以带钢热轧为代表的流程工业过程的安全监控及自主保障提供理论支撑与技术保证。