基于工业大数据的设备健康与故障分析方法研究与应用

发布时间：2018-06-13 06:33

本文选题：故障诊断 + 健康管理　；参考：《中国科学院大学(中国科学院沈阳计算技术研究所)》2017年硕士论文

【摘要】：设备健康与故障分析,一般依托于PHM(Prognostics and Health Management故障诊断与健康管理)系统。在一般的PHM系统中,常常利用大数据技术与统计学习算法,对生产过程中产生的各类数据进行分析,通过分析结论去量化与评估工业设备的健康状态,同时预测故障的发生过程与失效时间。PHM技术将设备的健康管理从传统的故障管理转变为衰退管理,通过预测性维护实现设备的零宕机和持续可靠的运行。目前大部分PHM系统是底层传感器与数据分析工具相关联的架构设计,往往对于每一类设备都有专用的分析套件,具有很强的专用度,但是缺乏重构性与算法通用性。另一方面,对于产生数据规模不大的设备均采用单机模式,由于设计模式限制,分析设备往往与监测设备之间直接相连。对于大量设备产生的大规模数据分析任务,在低响应时间需求下,单台分析设备的处理能力往往是不够的。基于以上需求,本文提出了可重构,通用化的PHM系统流程,同时实现了分布式集群下的工业大数据分析平台,提高了系统的适用性与运行效率。本文以工业设备的健康管理与故障诊断分析方法为主要研究内容,首先介绍了课题背景与PHM系统的发展现状,分析了其设计架构与平台特性。随后对论文中涉及到数据挖掘算法模型与Hadoop,Spark,Django等平台技术做了详细阐述。之后本文对典型的工业设备包括轴承与工业电容的预后算法进行研究,了解了一般PHM系统的基本分析流程。接着对通用的PHM平台进行了顶层架构设计,其中包括数据获取模式,算法平台的六层架构与评价测试方法等基本模块。最后提出了基于分布式平台的PHM系统实现方案,包括集群的搭建,运维,调优,算法的分布式移植等。
[Abstract]:Equipment health and fault analysis generally rely on the PHP Prognostics and Health Management Fault diagnosis and Health Management system. In a general big data system, big data technology and statistical learning algorithm are often used to analyze all kinds of data produced in the production process, and to quantify and evaluate the health status of industrial equipment through the conclusion of the analysis. At the same time, the technology of predicting the process of failure and failure time. PHM changes the health management of equipment from traditional fault management to decline management, and realizes the zero downtime and continuous reliable operation of the equipment through predictive maintenance. At present, most PHM systems are related to the underlying sensors and data analysis tools. They often have a special analysis suite for each type of equipment, and have a strong degree of specificity, but lack of reconfiguration and generality of the algorithm. On the other hand, the single machine mode is used for the equipment with small scale of generating data. Because of the limitation of design mode, the analysis equipment is often directly connected with the monitoring equipment. For large scale data analysis tasks generated by a large number of devices, the processing capacity of a single analysis device is often insufficient under low response time requirements. Based on the above requirements, this paper proposes a reconfigurable and generalizable big data system flow, and implements the industrial big data analysis platform under distributed cluster, which improves the applicability and running efficiency of the system. In this paper, the health management and fault diagnosis methods of industrial equipment are taken as the main research contents. Firstly, the background of the subject and the development status of PHM system are introduced, and its design framework and platform characteristics are analyzed. Then the data mining algorithm model and the technology of Hadoop Sparkor Django platform are described in detail. Then the prognostic algorithm of typical industrial equipment including bearing and industrial capacitance is studied and the basic analysis flow of general PHM system is understood. Then the general PHM platform is designed at the top level, which includes the data acquisition mode, the six-layer architecture of the algorithm platform and the evaluation and testing method. Finally, the realization scheme of PHM system based on distributed platform is put forward, including cluster construction, operation and maintenance, optimization, distributed migration of algorithm, etc.
【学位授予单位】：中国科学院大学(中国科学院沈阳计算技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【参考文献】