数据驱动的复杂装备异常检测方法

发布时间：2018-02-27 10:25

本文关键词： 复杂装备状态监测数据数据驱动装备异常检测　出处：《清华大学》2015年博士论文　论文类型：学位论文

【摘要】：近年来,随着大数据技术以及物联网技术在制造业的广泛使用,以工业大数据为基础的现代状态监测系统,被广泛应用于复杂装备工作状态的监测。利用状态监测系统采集的复杂装备的海量工作状态数据,可以有效分析装备的异常工作运行状态,实现装备的维修任务调度优化、运行效率提升以及设计制造改进等。因此,基于装备工作状态数据的装备异常检测技术,越来越受到工业界尤其是复杂装备制造业的关注。本文以复杂装备的工作状态数据为研究对象,围绕装备异常检测这个核心问题展开。论文的主要内容和贡献如下:?针对海量的复杂装备工作状态数据,提出一个面向装备异常检测的装备状态监测数据空间的数据逻辑组织结构。基于该状态监测数据空间,采用Cassandra系统设计了一种基于自由表的工作状态数据物理存储结构,实现了“监测装备→状态参数→采集时间→采集数值”的层级存储结构。?针对单装备单状态监测数据的装备异常检测问题,提出一个装备状态监测指标空间。该指标空间包含每个状态参数对应的状态监测序列的一级指标和二级指标。通过调整每个状态参数的二级指标的异常阈值,可以有效检测出状态参数中的异常的状态监测序列。?针对单装备多状态监测数据的装备异常检测问题,提出一种基于装备工作周期中状态监测序列组潜在关系的异常检测方法。该方法针对每个装备工作周期中的状态监测序列组提取一个潜在关系向量,建立一个潜在关系概率模型。通过基于该潜在关系概率模型的最大似然分类器,可以有效检测装备工作周期中的异常状态监测序列组。?针对多装备多状态监测数据的装备异常检测问题,提出一种基于多装备群体协同的装备异常检测方法。该方法从时间维度和地区维度提取装备的工作状况矩阵,结合装备异常矩阵,在协同过滤的框架下采用矩阵分解的技术,可以有效检测每个地区的装备异常数量。
[Abstract]:In recent years, with the extensive use of big data technology and Internet of things technology in the manufacturing industry, a modern condition monitoring system based on industrial big data has been developed. It is widely used in monitoring the working state of complex equipment. By using the mass working state data of complex equipment collected by the state monitoring system, the abnormal working state of equipment can be effectively analyzed, and the maintenance task scheduling of equipment can be optimized. Therefore, equipment anomaly detection technology based on equipment working state data, This paper focuses on the working state data of complex equipment, focusing on the core problem of equipment anomaly detection. The main contents and contributions of this paper are as follows:? This paper presents a logical data organization structure of equipment state monitoring data space for equipment anomaly detection, based on the data space of equipment state monitoring, based on the large amount of complex equipment working state data. A physical storage structure of working state data based on free table is designed by using Cassandra system, and the monitoring equipment is realized. 鈫扴tate parameter. 鈫扐cquisition time. 鈫扖ollection value "hierarchical storage structure.? Aiming at the problem of equipment anomaly detection of single equipment single state monitoring data, A state monitoring index space for equipment is proposed. The index space includes the first class index and the second level index of the state monitoring sequence corresponding to each state parameter. By adjusting the abnormal threshold of the second level index of each state parameter, Can effectively detect the abnormal state in the parameters of the state monitoring sequence.? Aiming at the problem of equipment anomaly detection with single equipment multi-state monitoring data, An anomaly detection method based on the potential relationship between state monitoring sequence groups in the equipment working cycle is proposed, which extracts a potential relationship vector for the state monitoring sequence groups in each equipment working cycle. The maximum likelihood classifier based on the probabilistic model of latent relation can be used to detect outliers in the equipment working cycle effectively. Aiming at the problem of equipment anomaly detection based on multi-equipment multi-state monitoring data, a method of equipment anomaly detection based on multi-equipment group cooperation is proposed, which extracts the working status matrix of equipment from time dimension and area dimension. Combined with the equipment anomaly matrix, the matrix decomposition technique is adopted in the framework of collaborative filtering, which can effectively detect the number of outliers in each region.
【学位授予单位】：清华大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TU646

【相似文献】