基于LLTSA算法的转子故障特征数据集降维方法研究
发布时间：2018-09-13 06:02
【摘要】：随着旋转机械向超大型化方向快速发展,利用运行中采集到的机械振动信号实施设备的状态监测与故障诊断,是确保其稳定、安全、高效运行的主要措施。故障诊断存在着信息量大、知识匮乏的问题,因此如何从故障特征数据中筛选出对故障诊断有用的特征数据子集,已成为故障诊断研究中亟需尽快解决的问题。传统的线性降维方法无法对非线性故障特征数据做出正确的降维处理,因此非线性故障特征数据的降维方法研究已成为当前研究的重点。此外,当机械系统需要实施在线状态监测时,正确判别新增数据的类别是保证监测结果正确性的必要前提条件。针对以上两问题,本研究对局部切空间排列算法(LTSA)进行改进,并将改进后的算法用于解决故障统计特征数据集的降维问题。还对通过借助增量LTSA学习算法对新增特征数据进行正确辨识的方法进行了探讨。开展的具体研究内容与得到的研究结论情况如下： 1)针对原始故障数据无法直接用于故障分析的问题,研究了常用于故障分析的时域特征,这些特征包括均值、标准差、峰峰值和裕度指标等。针对经典的线性降维方法不能满足非线性故障数据降维要求的问题,分析比较了几种常用的流形学习算法的特点。分析显示出,该类方法在处理非线性故障数据时,拥有能够较好地保留数据本质信息特征的优点。 2)针对传统的线性降维方法难以对非线性故障统计特征实施有效降维的问题,将局部切空间排列算法(LTSA)用于非线性故障特征数据的降维中。该方法能有效的对非线性数据进行降维处理。但由于在降维过程中,邻域k值的选取没有统一的标准,因此仅凭经验和试验法无法满足快速正确处理数据的要求。针对此不足,在本问题研究中引入了线性分块算法。应用情况表明：构造出的新算法能对非线性数据进行合理的局部线性分块,从而可解决LTSA算法邻域K值选取问题,使得降维结果得到了良好的改善。 3)针对在线监测与诊断需要及时正确的分析与处理新增数据的问题,采用增量局部切空间排列算法(LTSA)对新增数据进行处理。设计出的新方法在利用历史数据信息的同时还能够对新增数据进行判别分析。本项研究工作对设备当前状况的分析与未来的趋势发展判断具有重要参考价值。 4)将LabVIEW与MATLAB进行结合,充分利用两者的优势,成功的将LLTSA降维算法嵌入到了基于这两种软件的混合编程中。突出了良好的分析效果和可视化效果。 研究表明,机械系统非线性数据的特征生成、选择与降维和新增数据的有效辨识是机械故障诊断研究的新发展方向。该工作能够为数据驱动的智能诊断实现提供新思路。
[Abstract]:With the rapid development of rotating machinery in the direction of super-large-scale, it is the main measure to ensure its stability, safety and high efficiency to use the mechanical vibration signals collected in operation to implement the state monitoring and fault diagnosis of the equipment. There are many problems in fault diagnosis, such as large amount of information and lack of knowledge. Therefore, how to select a subset of feature data useful for fault diagnosis from fault feature data has become a problem that needs to be solved as soon as possible in fault diagnosis research. The traditional linear dimensionality reduction method can not deal with the nonlinear fault feature data correctly, so the research of nonlinear fault feature data dimensionality reduction method has become the focus of current research. In addition, when the mechanical system needs to carry out on-line state monitoring, it is a necessary prerequisite to correctly judge the new data categories to ensure the correctness of the monitoring results. In view of the above two problems, this paper improves the local tangent space arrangement algorithm (LTSA), and applies the improved algorithm to reduce the dimension of the fault statistical feature data set. The method of correct identification of new feature data by means of incremental LTSA learning algorithm is also discussed. The specific research contents and conclusions obtained are as follows: 1) aiming at the problem that the original fault data can not be directly used in fault analysis, the time-domain features commonly used in fault analysis are studied, which include mean value. Standard deviation, peak value and margin index, etc. Aiming at the problem that the classical linear dimensionality reduction method can not meet the dimensionality reduction requirements of nonlinear fault data, the characteristics of several commonly used manifold learning algorithms are analyzed and compared. The analysis shows that this kind of method is used to deal with nonlinear fault data. It has the advantage that the essential information features of data can be preserved well. 2) aiming at the problem that traditional linear dimensionality reduction method is difficult to effectively reduce the dimension of nonlinear fault statistical features, The local tangent space arrangement algorithm (LTSA) is used to reduce the dimension of nonlinear fault feature data. This method can effectively reduce the dimension of nonlinear data. However, in the process of dimensionality reduction, there is no uniform criterion for the selection of neighborhood k value, so the experience and experimental method alone can not meet the requirements of fast and correct data processing. To solve this problem, a linear block algorithm is introduced in this paper. The application results show that the new algorithm can reasonably divide the nonlinear data into local linear blocks, thus solving the problem of selecting the neighborhood K value of the LTSA algorithm. The dimensionality reduction results are improved well. 3) aiming at the problem that on-line monitoring and diagnosis need to analyze and process the new data correctly, the incremental local tangent space arrangement algorithm (LTSA) is used to process the new data. The new method not only uses the historical data but also discriminates the new data. This work has important reference value for the analysis of the current situation of the equipment and the judgement of the future trend development. 4) combine LabVIEW with MATLAB and make full use of the advantages of both. Successfully embed the LLTSA dimensionality reduction algorithm into the hybrid programming based on these two kinds of software. Good analysis effect and visualization effect are highlighted. It is shown that the feature generation, selection and reduction of the nonlinear data and the effective identification of the new data are the new development directions in the research of mechanical fault diagnosis. This work can provide a new idea for the realization of data-driven intelligent diagnosis.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TH165.3
本文编号：2240291
[Abstract]:With the rapid development of rotating machinery in the direction of super-large-scale, it is the main measure to ensure its stability, safety and high efficiency to use the mechanical vibration signals collected in operation to implement the state monitoring and fault diagnosis of the equipment. There are many problems in fault diagnosis, such as large amount of information and lack of knowledge. Therefore, how to select a subset of feature data useful for fault diagnosis from fault feature data has become a problem that needs to be solved as soon as possible in fault diagnosis research. The traditional linear dimensionality reduction method can not deal with the nonlinear fault feature data correctly, so the research of nonlinear fault feature data dimensionality reduction method has become the focus of current research. In addition, when the mechanical system needs to carry out on-line state monitoring, it is a necessary prerequisite to correctly judge the new data categories to ensure the correctness of the monitoring results. In view of the above two problems, this paper improves the local tangent space arrangement algorithm (LTSA), and applies the improved algorithm to reduce the dimension of the fault statistical feature data set. The method of correct identification of new feature data by means of incremental LTSA learning algorithm is also discussed. The specific research contents and conclusions obtained are as follows: 1) aiming at the problem that the original fault data can not be directly used in fault analysis, the time-domain features commonly used in fault analysis are studied, which include mean value. Standard deviation, peak value and margin index, etc. Aiming at the problem that the classical linear dimensionality reduction method can not meet the dimensionality reduction requirements of nonlinear fault data, the characteristics of several commonly used manifold learning algorithms are analyzed and compared. The analysis shows that this kind of method is used to deal with nonlinear fault data. It has the advantage that the essential information features of data can be preserved well. 2) aiming at the problem that traditional linear dimensionality reduction method is difficult to effectively reduce the dimension of nonlinear fault statistical features, The local tangent space arrangement algorithm (LTSA) is used to reduce the dimension of nonlinear fault feature data. This method can effectively reduce the dimension of nonlinear data. However, in the process of dimensionality reduction, there is no uniform criterion for the selection of neighborhood k value, so the experience and experimental method alone can not meet the requirements of fast and correct data processing. To solve this problem, a linear block algorithm is introduced in this paper. The application results show that the new algorithm can reasonably divide the nonlinear data into local linear blocks, thus solving the problem of selecting the neighborhood K value of the LTSA algorithm. The dimensionality reduction results are improved well. 3) aiming at the problem that on-line monitoring and diagnosis need to analyze and process the new data correctly, the incremental local tangent space arrangement algorithm (LTSA) is used to process the new data. The new method not only uses the historical data but also discriminates the new data. This work has important reference value for the analysis of the current situation of the equipment and the judgement of the future trend development. 4) combine LabVIEW with MATLAB and make full use of the advantages of both. Successfully embed the LLTSA dimensionality reduction algorithm into the hybrid programming based on these two kinds of software. Good analysis effect and visualization effect are highlighted. It is shown that the feature generation, selection and reduction of the nonlinear data and the effective identification of the new data are the new development directions in the research of mechanical fault diagnosis. This work can provide a new idea for the realization of data-driven intelligent diagnosis.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TH165.3
【参考文献】
相关期刊论文 前7条
1 阳建宏;徐金梧;杨德斌;黎敏;;基于主流形识别的非线性时间序列降噪方法及其在故障诊断中的应用[J];机械工程学报;2006年08期
2 杨庆;陈桂明;童兴民;何庆飞;;增量式局部切空间排列算法在滚动轴承故障诊断中的应用[J];机械工程学报;2012年05期
3 杨剑,李伏欣,王珏;一种改进的局部切空间排列算法[J];软件学报;2005年09期
4 张振跃,查宏远;Principal Manifolds and Nonlinear Dimensionality Reduction via Tangent Space Alignment[J];Journal of Shanghai University;2004年04期
5 梁霖;徐光华;栗茂林;张熠卓;梁小影;;冲击故障特征提取的非线性流形学习方法[J];西安交通大学学报;2009年11期
6 蒋全胜;贾民平;胡建中;许飞云;;基于拉普拉斯特征映射的故障模式识别方法[J];系统仿真学报;2008年20期
7 徐蓉;姜峰;姚鸿勋;;流形学习概述[J];智能系统学报;2006年01期
，本文编号：2240291
本文链接：https://www.wllwen.com/kejilunwen/jixiegongcheng/2240291.html
