多模态过程质量有关故障的检测及诊断方法研究

Due to different production schemes and diversified market demands, the industry process always has multiple operation modes. Product quality is very important, however, not all faults would affect the quality. Since there is the feedback loop in the system, the effects of some faults would be compensated. This project studies multimode process fault detection and diagnosis, judgments whether the fault is relevant to the quality and focuses on quality relevant faults. For modes division, the method based on the neighborhood information and the method which applies the time information and the spatial information are developed for improving the accuracy of modes division. In order to extract the quality relevant information and determine whether the fault is relevant to quality, the quality relevant information selection strategy based on the mutual information is presented to improve the purpose of detection and diagnosis. Moreover, the double fault detection method is proposed, where the full variable information is monitored in the first level and the quality variable information is monitored in the second level. For establishing the fault detection model, the subspace division method according to data characteristics is studied and the feature extraction method which can maintain the global structure and local structure is developed. For fault diagnosis, the semi-supervised diagnosis model which fuses process mechanism knowledge is constructed.

不同的生产方案以及多样化的市场需求使流程工业过程往往具有多种操作模态。产品质量是十分重要的，然而，并非所有故障均会影响质量。由于系统中往往存在反馈回路，某些故障的影响会被补偿。本课题研究多模态过程故障检测与诊断，判断故障是否与质量有关，重点关注与质量有关的故障。针对模态划分问题，研究基于邻域信息的方法；研究同时利用时间信息及空间信息的方法，提高模态划分精度。针对质量有关信息提取以及故障是否与质量有关的判断问题，研究基于互信息等策略提取与质量有关信息的方法，提高检测目的性；研究双层故障检测方法，第一层采用全变量信息，第二层仅采用与质量有关信息。针对单一模态故障检测模型的构建问题，研究根据数据特性将变量划分子空间的方法；研究能同时保持全局结构以及局部结构的特征提取方法，全面有效提取信息。针对故障诊断问题，研究融合过程机理知识的半监督诊断模型，提高故障诊断精度。

本项目通过系统分析多模态过程的数据特征，研究了离线模态划分及在线结果确定方法，提出了模态划分过程中划分标准的确定策略，对多模态数据集进行准确的模态划分；研究了质量有关信息提取以及与质量有关故障判断方法，设计了过程变量与质量指标相关关系的计算方式，准确提取质量相关信息；研究了单一模态中故障检测模型的建立方法，提出了子空间划分过程中数据特性的选择及检验策略，进而建立准确的故障检测模型；研究了故障诊断模型的建立方法以在故障发生后准确识别诊断故障变量。通过深入研究分析，针对多模态过程质量有关故障的检测及诊断问题，取得了一些具有创新性的研究成果，为具有多模态特性的现代流程工业过程提供了有效的模态划分、特征提取、检测指标、诊断方法，总结整理后已分别在国内外专业期刊、学术会议发表。此外，本项目的研究成果已经应用于宝武钢铁集团的热轧生产过程。