基于不确定信息的轧制过程故障预测与诊断关键技术研究

The desire and need for accurate diagnosis and real predictive prognosis is urgent for all industrial processes, in order to enhance process safety, reduce failure rate and improve product quality. Rolling process plays an important role in iron and steel industry. It is often featured by high-speed, continuous production, long flow and complex control systems, etc., which make it prone to various malfunctions, faults and failures. System-level fault needs more awareness because it is highly possible to evolve into a serious safety event and cause heavy loss to the enterprise. However, in practice, there exists much uncertain information in process models, measurements and knowledge in real industrial processes. The fault prognosis and diagnosis methods could have satisfying performances only when the uncertain information is taken into full consideration and handled in right way. Uncertainty problem is a bottleneck in fault diagnosis and prognosis. One major task in this project is to apply various advanced techniques to solve uncertain information in the rolling process. Considering that there are plenty of first-principle models in control loops of rolling processes, uncertain- model-based fault prognosis and diagnosis will be studied for loop-level faults. On the other hand, in view of the insurmountable difficulty in developing a system-level first-principle process model and the available vast amount of process data in rolling processes, uncertain-data-driven fault prognosis and diagnosis will be studied for system-level faults. The two technological branches will be merged in an intelligent maintenance system developed for a specific rolling process. The research task belongs to the popular cross areas, which covers the disciplines of information science, control theory and metallurgical engineering. The expected achievements can offer new perspectives, methodologies and technologies in the field of intelligent maintenance for complex industrial processes, in hopes that, the theoretical system of fault prognosis and diagnosis with uncertain information can be developed and applied into complex industrial processes.

以提高过程安全性、减少设备故障率、提升产品质量稳定性为目标的故障预测与诊断技术具有明确的科学意义和应用价值。轧制过程是冶金行业的重要生产过程，具有高速、连续、长流程、控制系统复杂等特点，故障率高且系统级故障危害性严重。轧制过程中模型、数据和知识都具有显著不确定性，只有充分考虑并正确处理信息的不确定性，才能取得良好的故障预测和诊断能力。本项目针对轧制过程的不确定信息，考虑了控制回路级和系统级故障的不同特点，研究过程建模、故障传播机理分析、海量数据特征提取、故障预测与诊断的关键理论与应用技术。项目研究内容是控制科学与工业信息化技术与冶金工业典型过程的深度交叉，属于复杂工程系统运行安全领域的新观点、新方法、新技术的基础研究与应用探索，不仅可以丰富不确定信息下的故障预测与诊断理论的研究体系，理论算法在轧制过程中的应用探索成果也可推广到其它流程工业过程，为先进理论的工业技术实践提供经验与指导.

轧制过程是冶金行业的重要生产过程，具有高速、连续、长流程、机-电-液多能域耦合、多产品、多工况等特点，故障发生率高，微小故障难以检测，复合故障难以隔离。另外，不同产品或工况下的过程机理、知识和数据特征均存在显著差异，这种强不确定性给现有的故障预测与诊断技术带来诸多挑战。.本项目主要考虑了轧制过程的四个重要特征：多变量、多阶段、多工况、多产品，研究了五类问题：系统建模、特征提取、故障诊断、故障预测和产品质量控制及可视化，提出了若干可行的理论方法和有效的技术手段。其中，基于模型移植的多产品/多工况系统快速建模方法、基于自然梯度无模型优化的产品质量控制方法、基于混合模型的双粒度故障诊断方法、基于改进全局可测故障残差（ToMFIR）算法的微小故障诊断与隔离方法、基于多信号模型和盲源分离的复合故障诊断方法、基于局部保持投影和混合概率回归模型的板形预测方法都具有良好的原始创新性。.项目组一方面采集了宝钢2030冷轧机组DSR板形控制系统的大量生产过程数据，另一方面搭建了热轧过程液压自动板厚控制（AGC）系统的仿真模型，比较充分地验证了理论算法的可行性和有效性。目前部分核心技术成果已成功推广应用至轨道车辆门系统中，说明本项目组具有良好的成果转化潜力和用科技服务社会的能力。.项目执行期间，项目组成员出版学术专著3部；在国内外核心学术期刊和重要学术会议发表和录用相关论文25篇，其中被SCI检索10篇，EI 检索19篇；申请发明专利13项；获“宝钢技术创新重大成果三等奖”（2015）和“第22届全国技术发明展览会金奖”（2017）；参加国内外会议10人次，邀请国内外相关领域专家讲学5场次。.项目研究成果可为不确定信息下的故障预测与诊断理论的研究体系提供一些新思想和新方法，理论算法在轧制过程中的应用探索成果已成果推广到其它工程系统，为先进理论成果的转化提供了成功经验。

内容获取失败，请点击重试