机械故障信号欠定源估计与盲提取方法研究

发布时间：2018-09-19 06:49

【摘要】：旋转机械设备运行时产生的振动信号和声信号都蕴含着大量用于状态监测和故障诊断的重要信息。当前运行设备的状态变化会时刻影响这些信号的相关特征参数,而常用的状态监测与故障诊断的主要方法就是通过处理传感器拾取的故障信号,进而根据相关特征参数的分布情况间接掌握设备的当前运行状态。因此,状态监测与故障诊断成功与否的前提和关键在于如何从强干扰的机械状态信号中提炼出绝对有用、能够客观的评价和判断诊断对象的状态特征。然而,实际工业现场存在大量背景干扰噪声、多种未知的机械结构源信号相互耦合、致使传感器数目小于源信号数目情况屡屡存在、加之传感器拾取的观测信号的传输过程未知等因素,致使传感器获取的观测信号每每都是所有可以影响的因素和故障信号经过多次混杂后的结果,待估计的故障源目标信号与其他干扰信号掺杂共处,通常几乎不能直接从观测故障信号中得到有价值的信息。因此,为了能够准确、高效地提取故障源目标信号,首先必须尽可能地将背景噪声和其他干扰信号进行抑制或排除。盲源分离已经成为信号处理学科领域的研究热点,机械信号处理也不例外。尤其是在几乎没有先验知识情况下,盲信号处理技术可以实现从混合信号中恢复或估计出源信号,体现它的优越性,是解决机械故障复合信号盲分离的一个有力的手段。但是,传统的盲信号处理方法对实际工况的机械故障信号的识别和提取还存在很多不足。因此,本学位论文在国家自然科学基金项目和云南省科技计划资助项目的资助下,以盲信号处理为研究基础,针对实际复杂环境下的复合机械振动信号和机械噪声信号提取和分离问题,使用理论研究和实验验证相结合的研究方式,初步建立了机械故障信号的欠定盲源估计和盲提取模型及方法,为机械复合故障信号的欠定源估计和分离提供一种研究思路,全文主要研究内容如下：(1)从工程实际出发,介绍本文的选题背景和研究意义。就盲信号处理技术和机械故障诊断盲处理应用的国内外研究现状进行了较为全面的综述,总结了目前盲信号处理技术在故障诊断领域应用现存问题。(2)针对传感器拾取工业现场机械设备故障信号的特点,提出了基于形态滤波和核独立分量分析结合算法和基于形态滤波和遗传模拟退火的模糊C均值聚类改进的稀疏分量分析算法。仿真和实验研究结果显示,结合形态滤波技术的两个算法在完备情况下可以解决复合故障盲分离,从而提高了算法的实用性和分离结果的可靠性。但在在欠定情况下,前者失效,而后者需要预先给定聚类数目。(3)针对工业现场源数目未知、欠定的问题和SCA算法需要事先给定源数目,建立了机械冲击信号的源数目估计方法研究框架。在此框架基础上,提出了基于总体经验模态和自适应阈值设置的奇异值分解算法。通过计算机仿真和复合故障轴承振动信号源数目估计来验证该理论框架的可行性,研究表明该算法有较好适应性。(4)针对工业现场强背景噪声和欠定问题,建立压缩感知和欠定盲解卷积等价的理论框架,在此框架基础上,提出改进形态滤波和频域压缩感知重构的欠定盲提取算法。利用前面提出的源估计算法估计源信号的数目；使用形态滤波滤除背景噪声；遗传模拟退火优化的FCM算法用于对混合矩阵的估计,进而根据混合矩阵构建传感矩阵；最后使用压缩感知重构算法的正交匹配算法在频域恢复源信号。计算机仿真和实验研究验证了上述算法的正确性,表明算法可以很好地分离复合故障信号。(5)针对现有盲解卷积算法对单一故障声信号有效,但可以很好提取复合故障冲击信号,而前面提出的压缩感知重构算法对复合故障声信号失效。提出盲解卷积和频域压缩重构结合算法,并对复合故障轴承声信号进行分离,得到很好的分离结果。
[Abstract]:Vibration signals and acoustic signals produced by rotating machinery equipments in operation contain a large number of important information for state monitoring and fault diagnosis. The state changes of current operating equipment will affect the relevant characteristic parameters of these signals at all times. The main method of state monitoring and fault diagnosis is picked up by processing sensors. Therefore, the prerequisite and key to the success of condition monitoring and fault diagnosis lies in how to extract absolutely useful mechanical state signals with strong interference, and can objectively evaluate and judge the state characteristics of the diagnostic object. There are a lot of background interference noises in the actual industrial field, and many unknown mechanical structure sources are coupled with each other, so that the number of sensors is less than the number of source signals. In addition, the transmission process of the observed signals picked up by the sensors is unknown, and other factors, so that the observed signals obtained by the sensors are always all factors that can be affected. As a result of mixing elements and fault signals for many times, the estimated fault source target signal doped with other interference signals can hardly get valuable information directly from the observed fault signals. Therefore, in order to extract fault source target signal accurately and efficiently, the background noise and its background noise must be combined as much as possible. Blind Source Separation (BSS) has become a research hotspot in the field of signal processing, and mechanical signal processing is no exception. Especially in the case of little prior knowledge, BSS can recover or estimate the source signal from the mixed signal, which shows its superiority and solves the mechanical problem. However, the traditional blind signal processing methods still have many deficiencies in identifying and extracting mechanical fault signals under actual working conditions. Therefore, this dissertation is based on blind signal processing with the support of the National Natural Science Foundation of China and the Yunnan Science and Technology Project. Aiming at the problem of extracting and separating vibration signals and mechanical noise signals in complex environment, the model and method of underdetermined blind source estimation and blind source extraction for mechanical fault signals are preliminarily established by combining theoretical research with experimental verification, which can provide underdetermined source estimation and separation for mechanical composite fault signals. The main contents of this paper are as follows: (1) Based on the engineering practice, the background and significance of this paper are introduced. The research status of blind signal processing technology and blind processing of mechanical fault diagnosis at home and abroad is summarized, and the application of blind signal processing technology in the field of fault diagnosis is summarized. Existing problems. (2) According to the characteristics of sensor picking up faulty signals of industrial machinery and equipment, a sparse component analysis algorithm based on morphological filtering and kernel independent component analysis and improved fuzzy C-means clustering algorithm based on morphological filtering and genetic simulated annealing are proposed. The two algorithms can solve the blind separation of complex faults under complete conditions, which improves the practicability of the algorithm and the reliability of the separation results. However, under undetermined conditions, the former is invalid, while the latter requires a predetermined number of clusters. Aim To establish a research framework for estimating the number of sources of mechanical shock signals. Based on this framework, a singular value decomposition algorithm based on total empirical mode and adaptive threshold setting is proposed. The feasibility of the theoretical framework is verified by computer simulation and the estimation of the number of sources of bearing vibration signals with compound faults. (4) A theoretical framework of compressed sensing and underdetermined blind deconvolution is proposed to solve the problem of strong background noise and underdetermined background noise. Based on this framework, an underdetermined blind extraction algorithm is proposed to improve morphological filtering and frequency domain compressed sensing reconstruction. Morphological filtering filters the background noise; genetic simulated annealing optimized FCM algorithm is used to estimate the mixed matrix, and then the sensor matrix is constructed according to the mixed matrix; finally, the orthogonal matching algorithm of compressed sensing reconstruction algorithm is used to restore the source signal in the frequency domain. (5) Blind deconvolution algorithm is effective for single fault acoustic signal, but it can extract complex fault impulse signal very well. The compressed sensing reconstruction algorithm proposed earlier is invalid for complex fault acoustic signal. A combination algorithm of blind deconvolution and frequency domain compression reconstruction is proposed, and the combined algorithm is applied to complex fault acoustic signal. The sound signals of bearings are separated and good separation results are obtained.
【学位授予单位】：昆明理工大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH165.3

【参考文献】