基于自适应采样方法的高维流数据在线过程监测的研究

Online process monitoring is quite important to process safety and product quality assurance. With the rapid development of sensing technology, sensors that generate real-time high-dimensional streaming data have been widely used in a variety of manufacturing systems. However, effective monitoring of such data is a challenging task for quality improvement due to large volume, complex fault patterns, and dynamics of changes. Currently, there is a lack of efficient statistical process control (SPC) methodologies to address these challenges. The target of this research is to develop scalable and adaptive methodologies for online monitoring of high-dimensional streaming data: Efficient scalable schemes via adaptive order shrinkage and its corresponding asymptotic properties will be investigated; and considering the resource constraints existing in workshops, adaptive sampling strategies will be studied then; meanwhile, we will incorporate engineering knowledge to enhance monitoring performance to finally address those scientific problems. It is hopeful that this research could provide a solid theoretical basis and key technologies for real-time high-dimensional streaming data processing and online process monitoring in manufacturing systems.

生产过程在线故障监测对生产的稳定运行和产品质量保障发挥着重要作用。当前复杂制造系统通过各类传感设备已能采集高维流数据用以对生产过程进行全面监控。然而，由于高维流数据存在复杂度高、数据量大、变化快速等特征，现有的过程监测系统尚缺乏能有效针对这类数据的在线监测方法。本课题以复杂系统产生的高维流数据为研究对象，针对“高维流数据难以实时处理”这一关键问题，提出了通过“自适应”监测局部过程变量以达到对全局监测目的的研究思想。课题充分考虑实际生产中存在的各种资源约束，拟利用自适应缩序理论建立一套快速分析高维流数据并能准确识别故障模式的在线监测方法。课题拟通过研究探索自适应采样方法的性质，并融合工程背景知识，解决基于高维流数据分析建模、参数估计、实时监测等科学问题。该研究将最终为解决复杂制造系统实时处理海量数据的问题奠定科学基础，同时也为生产过程在线监控提供理论依据与关键技术。

本项目以复杂生产和运营系统为研究对象，充分考虑当前生产和运营环境下高维流数据需要“实时处理与分析”的需求，提出了基于多源流数据自适应融合与建模的系统实时监测与诊断方法体系。研究重点解决了当前实际情况中数据存在的读取、传输、清洗资源约束下的数据分析与建模问题，提升了在实际工况下的复杂系统过程监测与诊断水平。在基于高维传感数据的建模方面，研究团队攻克了基于数据时空耦合条件下的建模理论，研发了满足流数据随机缺失条件下的数据融合方法，并依此开发了相应的全局过程监测与分析工具。研究方法在先进制造产品一致性、国家储备粮品质监测、产品运维服务和医疗服务领域均得到较好的验证。研究成果在工业工程领域、质量与可靠性工程领域等国际知名期刊发表了多篇高水平论文。