基于主元分析与偏最小二乘的故障诊断方法研究

发布时间：2018-01-19 10:20

本文关键词： 故障诊断多元统计主成分分析偏最小二乘法　出处：《河北师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着工业水平朝着复杂化的发展与计算机技术的大规模应用,故障诊断方法成为了工业生产中的重点问题。通过对生产过程的的状态进行观察与预测,能够在故障发生之前及时发现工况异常,诊断出故障发生位置,合理消除噪声干扰,从而保证工业生产过程正常、安全运行。伴随着计算机的发展,工业生产过程中大量的信息都可以完整保存下来。因此,如何合理运用这些信息,通过分析大数据来判断工况是否正常,成为了研究领域的热点问题。在这样的时代背景下,基于数据驱动的故障诊断方法收到了广泛关注,并成功的应用于冶金、制药的生产过程中,在监控与检测领域体现了优越性。多元统计分析以主成分分析(PCA)与偏最小二乘法(PLS)为主,该方法不需要建立具体的数学模型,适用于产生大量数据的工况。通过对高维数据进行降维处理,可以提高计算效率,因此具有重要的研究价值与实用性。本文通过对主成分分析与偏最小二乘法的研究,并结合风力发电机的的工况进行仿真研究。论文的主要内容如下:说明了故障诊断技术在自动控制领域的重要性及其意义。介绍了故障诊断与数据驱动的方法在最近半个世纪来的发展历史。建立了主元模型的具体表达形式,描述了主元分析的基本理论,以及基于主元分析的建模方法,与利用统计图和贡献图进行基本的故障定位。针对传统主元分析方法的不足,研究了重构主元方法,并且针对重构方法无法诊断多故障同时发生的情况,对重构法做了改变。介绍了偏最小二乘法的基本理论,并且仿真对比了主元分析与偏最小二乘法的诊断效果,分析了两种方法在今后的研究趋势。总结与展望。对本文内容进行了总结,并对数据驱动的下一步研究计划作出了规划。
[Abstract]:With the development of industrial level towards complexity and the large-scale application of computer technology, fault diagnosis method has become a key problem in industrial production. The state of production process is observed and predicted. It can find the abnormal working condition in time before the fault occurs, diagnose the location of the fault, reasonably eliminate the noise interference, so as to ensure the normal and safe operation of the industrial production process, along with the development of the computer. A large amount of information in the industrial production process can be completely preserved. Therefore, how to use the information reasonably, through the analysis of big data to determine whether the working conditions are normal. It has become a hot issue in the field of research. In such an era background, data-driven fault diagnosis method has received extensive attention, and has been successfully applied in metallurgical and pharmaceutical production process. In the field of monitoring and detection, the main advantages of multivariate statistical analysis are principal component analysis (PCA) and partial least square method (PLS). This method does not need to establish a specific mathematical model. It is suitable for the condition of producing a large amount of data. By reducing the dimension of the high-dimensional data, the computational efficiency can be improved. Therefore, it has important research value and practicability. In this paper, principal component analysis and partial least square method are studied. The main contents of this paper are as follows:. The importance and significance of fault diagnosis technology in the field of automatic control are explained. The development history of fault diagnosis and data-driven method in the last half century is introduced, and the concrete expression form of principal component model is established. The basic theory of principal component analysis (PCA), the modeling method based on PCA, and the basic fault location based on statistical diagram and contribution diagram are described. The reconstruction principal component method is studied, and the reconstruction method is changed in view of the fact that the reconstruction method can not diagnose multiple faults at the same time. The basic theory of partial least square method is introduced. And the simulation results of principal component analysis and partial least square method are compared, and the research trend of the two methods in the future is analyzed. Summary and prospect. The content of this paper is summarized. And the next step of data-driven research plan to make a plan.
【学位授予单位】：河北师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212;TP277

【相似文献】