基于不精确概率的隔振平台故障诊断方法研究

发布时间：2018-06-17 08:31

本文选题：不精确概率 + 证据理论　；参考：《中国科学技术大学》2017年硕士论文

【摘要】：现代化工业的蓬勃发展,使得生产生活与各类机械设备密不可分。随着机械设备的大型化、自动化程度越来越高,一旦其发生故障就容易造成严重的危害,因此,对机械设备故障诊断的研究具有重要意义。本文调研了现有的故障诊断方法,其中,多源传感器信息融合的方法由于可以利用多源信息更为全面地反映系统的工作状态而受到广泛应用。故障发生时系统采集的数据并非是固定不变的,而是包含有大量的不确定性,为了对这些包含着不确定因素的多源信息处理而得到可靠的诊断结果,常用DS证据理论作为理论基础进行不确定性推理。本文通过对通用证据理论故障诊断框架的分析,发现该诊断框架在证据生成以及决策诊断方面尚存在不足之处。同样作为不确定性推理方法,不精确概率理论相对而言是一个更为一般化的模型,对不确定信息的表征更符合实际需求,在处理数据的过程中更符合人的思维习惯,其多种决策准则也适用于不同精度要求的系统。基于此,本文提出了基于不精确概率的故障诊断方法,并以隔振平台为实验对象验证了方法的可行性。首先,构建了不精确概率理论下的故障诊断框架。本文第三章通过对传统证据生成方法的改进得到了下概率的生成方法,设计了可以表示专家决策倾向的诊断价值函数,将故障诊断问题转化为在现有条件下对价值函数期望值的比较与决策问题。随后比较了特征级融合和决策级融合在本方法中的相同与不同之处,采取特征级融合对数据进行融合。最后,通过实例分析了不同决策准则的特点。本文第四章以隔振平台为研究对象,构造了多类故障,使用推进式窗口对数据采样,以三路传感器采样数据为原始数据,对不同特征参数进行了大量的分析和对比以选定特征参数,实现了对隔振平台故障类型的判断。第五章首先以第四章为基础,针对振动发散类故障对诊断时间要求比较高的问题,采用本文方法实现了对故障时间的诊断。针对特征参数质量较差时,使用不精确概率诊断时会出现误判的情况,设计了结合使用DS证据理论与不精确概率理论的方法,在保证诊断时间的情况下有效减少了误判。实验结果表明,本文提出的方法用于故障诊断是可行有效的。
[Abstract]:With the vigorous development of modern industry, production and life are closely related to all kinds of machinery and equipment. With the large scale of machinery and equipment, the degree of automation is becoming higher and higher. Once it breaks down, it is easy to cause serious harm. Therefore, it is of great significance to study the fault diagnosis of machinery and equipment. In this paper, the existing fault diagnosis methods are investigated. Among them, the multi-source sensor information fusion method is widely used because it can reflect the working state of the system more comprehensively by using the multi-source information. When the fault occurs, the data collected by the system is not fixed, but contains a lot of uncertainties. In order to process the multi-source information with uncertain factors, the reliable diagnosis results can be obtained. DS evidence theory is often used as the theoretical basis for uncertainty reasoning. Based on the analysis of the general evidence theory fault diagnosis framework, it is found that there are still shortcomings in the evidence generation and decision diagnosis. As an uncertain reasoning method, the theory of inexact probability is a more general model, and the representation of uncertain information is more in line with the actual needs, and is more in line with the thinking habits of people in the process of processing data. Many of its decision criteria are also applicable to systems with different precision requirements. Based on this, a fault diagnosis method based on imprecise probability is proposed, and the feasibility of the method is verified by taking the vibration isolation platform as the experimental object. Firstly, a fault diagnosis framework based on imprecise probability theory is constructed. In the third chapter, the method of generating the lower probability is obtained by improving the traditional method of evidence generation, and the diagnostic value function which can express the tendency of expert decision is designed. The problem of fault diagnosis is transformed into the problem of comparison and decision of the expected value of value function under existing conditions. Then, the similarity and difference of feature level fusion and decision level fusion in this method are compared, and the feature level fusion is adopted to fuse the data. Finally, the characteristics of different decision criteria are analyzed by examples. In the fourth chapter, the vibration isolation platform is taken as the research object, and many kinds of faults are constructed. The data are sampled by propulsive window, and the original data are sampled by three sensors. The different characteristic parameters are analyzed and compared in order to select the characteristic parameters, and the fault types of vibration isolation platform can be judged. In the fifth chapter, based on the fourth chapter, aiming at the problem that the diagnosis time of vibration divergence type faults is high, the fault time diagnosis is realized by the method of this paper. In order to solve the problem that inaccurate probability diagnosis will occur when the quality of feature parameters is poor, a method combining DS evidence theory and imprecise probability theory is designed, which can effectively reduce the misjudgment under the condition of ensuring the diagnosis time. The experimental results show that the proposed method is feasible and effective in fault diagnosis.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH17

【参考文献】