【作者】 鲁文波；

【导师】 蒋伟康；

【作者基本信息】 上海交通大学 ， 机械设计及理论， 2012， 博士

【摘要】 现代工业的发展使得机械设备的状态监测与故障诊断越来越受到重视，传统的故障诊断技术面临着新的挑战。目前故障诊断主要基于振动信号的测试与分析，但振动传感器在某些设备上或工况环境中安装不便，使得振动故障诊断方法的应用受到限制。结构振动与辐射噪声有着紧密联系，机械噪声蕴含着丰富的机械状态信息，噪声信号也能像振动信号一样进行机械设备的状态监测与故障诊断。声学诊断技术具有非接触测量、不影响设备运行、操作简单便捷等优点，可部分地代替振动信号作为故障诊断的补充手段。传统声学诊断技术主要基于单通道测试与分析，只能得到机械局部的声学特征随时间或频率的变化规律，也不易选择测点位置，而且声信号易受干扰和污染。声学故障诊断技术虽具有很大发展潜力，但当测点位置选择不当，局部的声学特征对故障不敏感时诊断效果就会受到影响，特别是在相干工况下，故障源的声信号如果被干扰源的声信号淹没，则传统声学诊断方法将不再适用，基于单通道声信号所提取的故障特征不能稳定地反映机械本身的运行状态，很大程度上影响了诊断效果。声成像技术通过传声器阵列获取测量面处的声信号，按照相应重建算法进行声源反演，可重建结构表面的声场分布并预测整个辐射声场，将重建声场的整体信息用于故障诊断，能克服传统声学诊断中局部诊断的缺陷，也可避免测点选择难的问题，这为新型声学故障诊断技术的发展创造了条件。基于整个重建面上的声学信息显然比基于单点测试的声信号所含有的故障信息要丰富，后者只具备一个或几个局部测点的声学信息，而前者还获得了整个声场的分布信息，从空间上体现了潜藏在声场分布中的故障模式，从所重建的声场信息中挖掘出的声场空间分布特征必然较常规声学诊断方法所提取的特征更全面也更稳定。在较强的干扰噪声环境下，即使局部单通道声信号并不稳定，但整体上声场的空间分布模式能保持较好的稳定性。机器运行时会产生辐射声场，不同的运行状态具有不同模式的声场分布，因此，若能有效挖掘机器在不同故障状态下辐射声场的分布模式规律，提取对故障模式敏感的特征量，将比传统基于单通道信号分析的声学诊断能更有效更可靠地诊断机械故障。鉴于这一思想，本文提出了基于声场空间分布特征的故障特征提取方法，应用图像处理技术从声像图中提取反映声场空间分布特性的纹理特征，揭示机械在不同运行状态下声场的分布规律；并在此基础上，提出了基于声场空间分布特征的机械故障诊断方法，分别发展了适于中高频分析的基于远场波束形成声成像的故障诊断方法和适于中低频分析的基于近场声全息声成像的故障诊断方法。数值仿真和实验研究证实了机器设备在不同运行状态下，声像图中的纹理信息反映了相应辐射声场的空间分布特性，能够揭示机器的不同故障模式状态而进行故障诊断。基于声场空间分布特征的机械故障诊断方法有效综合了阵列测量、声成像、图像处理、特征提取、模式识别等多学科研究成果，能对机械设备辐射声场可视化的同时，挖掘出潜藏于声场中的故障模式特征，有效进行故障诊断。滚动轴承和齿轮箱的故障诊断实验研究进一步验证了方法的有效性和实用性，干扰噪声的影响研究进一步体现了方法的优越性，这一新型的声学诊断方法不仅拓展了声成像技术的应用范围，也为声学故障诊断提供了新的思路和选择，促进了故障诊断技术的发展及其在实践中的应用。本文的具体研究内容如下：（1）综述了机械故障诊断技术的发展概况，特别是声学故障诊断与智能诊断；概述了声成像技术的发展，并对目前常用的波束形成和近场声全息技术进行了总结。（2）分析了机械设备辐射声场的产生，对结构声辐射问题进行了数学描述；推导了平面近场声全息的基本公式，总结了近场声全息在实际应用中的参数选择问题；介绍了波束形成的基本原理，总结了参数选择问题；研究了几种典型的纹理分析方法，用于提取声场空间分布特征，并对支持向量机模式识别作了研究介绍。（3）提出了基于声场空间分布信息的特征提取与机械故障诊断方法，并发展了适于中高频分析的基于远场波束形成声成像的故障诊断方法和适于中低频分析的基于近场声全息声成像的故障诊断方法，分别进行了具体研究。（4）在对声成像技术和特征提取方法的研究基础上，提出了基于NAH声成像和Hist+GLGCM特征提取的机械故障诊断方法，并应用于滚动轴承与齿轮箱的故障诊断，对其分别搭建故障诊断实验台，进行多类故障诊断实验研究，并与传统声学诊断方法进行对比，还考察了干扰噪声对声成像和诊断结果的影响。（5）研究了齿轮箱啮合频率及其边频处的声像图特性，考察了不同特征频率对诊断结果的影响，总结了基于声成像的齿轮箱故障诊断在特征频率选择上的问题，指出边频处的声场分布模式不稳定，不适于故障模式的辨识。（6）研究了不同灰度量化级和各纹理特征提取方法的诊断性能，给出了灰度量化级的选取原则，并总结了各特征提取方法在声场空间分布特征提取上的特点，给出了在故障诊断应用中的具体建议。（7）研究了声成像技术在机械故障诊断测试系统中的应用，在虚拟仪器平台下设计了基于声成像的故障诊断测试系统通用平台，开发了系统原理样机，齿轮箱故障诊断应用实例验证了其有效性。更多还原

【Abstract】 The development of modern industry makes condition monitoring and fault diagnosisof machinery more and more significant, and the traditional fault diagnosis technology isfacing new challenges. Currenly, fault diagnosis is mainly based on measurement andanalysis of vibration signal, whereas vibration sensors can not be installed conveniently onsome equipment or in some conditions, which makes the application of vibration-basedfault diagnosis get limitation. There is close contact between structural vibration and itsradiated noise. Mechanical noise contains abundant state information, and noise signal candiagnose equipment fault like vibration signal. Acoustic-based diagnosis (ABD) has theadvantage of non-contact measurement, not affecting equipment operation, operatingquickly and easily, which can partially substitute for vibration signal as a supplementarymeans for fault diagnosis. The traditional ABD based on single-channel testing andanalysis can only get local acoustic characteristics with the changes of time or frequency. Itis not easy to choose proper measuring points and the acoustic signal is easilycontaminated. ABD has great potential, but when inappropriate measuring points arechosen or local acoustic characteristics are not sensitive to fault pattern, diagnosis resultswill be affected. Especially in coherent condition, acoustic signal from fault sources issubmerged by that from interference sources, the ABD will no longer apply. The reason isthat fault features based on single-channel acoustic signal can not stablely reflect theoperational status of machinary, which can affect diagnosis results to a large extent.Acoustic imaging techniques can get sound signal at the measuring surface through themicrophone array. It can reconstruct sound field distribution of structure surfaces andpredict the radiated sound field by corresponding reconstruction algorithms. The scheme ofapplying overall information of reconstructed sound field can overcome the limitations intraditional ABD, which creates a condition for the development of new ABD technique.Clearly, Acoustic information from entire reconstruction sound field can contain more faultpatter information than acoustic information based on single point. The latter has only oneor a few local points of acoustic information, while the former get the distributioninformation of entire sound field, which can spatially reflect failure pattern hidden in soundfield. Spatial distribution featrues extracted from the reconstructed sound field are surelymore comprehensive and more stable than that extracted from the conventional ABD. Withstrong interference noise, even if a local single-channel acoustic signal is not stable, but, onthe whole, the spatial distribution pattern of sound field is able to maintain good stability.Running machines can generate radiated sound field, and different operating stateshave different patterns of sound field distribution. Therefore, if the radiation sound fielddistribution pattern can be effectively minied from different states, and the features sensitive to patterns can be extracted, the fault diagnosis will be more efficient and reliablecomparing to the traditional ABD based on single channel signal analysis. In view of thisidea, a feature extraction method based on sound field spatial distribution characteristics isproposed, which extractes textural features from acoustic images to reveal distribution lawof sound field for different machinery operating conditions. Based on this thought, a faultdiagnosis scheme based on space distribution characteristics of sound field is proposed.Then, specifically, a fault diagnosis scheme suitable for medium-high frequency analysisbased on far-field beamforming acoustic images is developed, and a fault diagnosis schemesuitable for medium-low frequency analysis based on near-field acoustic holographyacoustic images is developed. Numerical simulation and experimental research confirmthat in different equipment operating conditions, the textural information can reflect thecorresponding space distribution characteristics of sound field and reveal the differentfailure patterns for fault diagnosis. The fault diagnosis method based on sound field spacedistribution characteristics effectively integrates microphone array measurement, acousticimaging, image processing, feature extraction, pattern identification and othermultidisciplinary research. This method can visualize the radiated sound field of machinery,and mine failure patterns underlying sound field. The experimental research on faultdiagnosis for rolling element bearing and gearbox further validates the effectiveness andpracticality of the proposed method, and the research on effect of noise interference furthershows the superiority of the proposed method. This new ABD method expands theapplication scope of acoustic imaging techniques, provides new ideas and options for ABD,and promotes the development of fault diagnosis technology and its application in practice.The research content of this dissertation can be summarized as follows.(1) The progress of machinery fault diagnosis is reviewed firstly, especially for theABD and intelligent diagnosis. The development of acoustic imaging techniques isoutlined, and the commonly used beamforming and near-field acoustic holographytechnique are summarized.(2) Generation mechanism of machinery radiated sound field is analyzed, and thestructural acoustic radiation question is presented mathematically. The basic formulas ofthe plane near-field acoustic holography are deduced. Several problems of selectingparameters in practical application of near-field acoustic holography are summarized. Thebasic principle of beamforming is introduced, and several problems of selecting parametersin beamforming are concluded. Several typical textural analysis methods are investigatedfor extracting spatial distribution characteristics of sound field. Pattern recognition basedon support vector machine is also introduced.(3) Fault feature extraction and fault diagnosis based on spatial distributioninformation of sound field is put forward. Based on this idea, specifically, a fault diagnosis scheme suitable for medium-high frequency analysis based on acoustic imaging byfar-field beamforming is developed, and a fault diagnosis scheme suitable for medium-lowfrequency analysis based on acoustic imaging by near-field acoustic holography isdeveloped. Then, specific research is carried on, repectively.(4) Based on the investigation on acoustic imaging and feature extraction, amechanical fault diagnosis method based on near-field acoustic holography andHist+GLGCM feature extraction is put forward, and the method is applied to rollingelement bearing fault diagnosis and gearbox fault diagnosis. Rolling element bearing faultdiagnosis test-bed and gearbox fault diagnosis test-bed are established for research onmulti-class fault diagnosis, respectively. Diagnosis results are compared to traditional ABD.The effect of acoustic imaging and diagnosis results with noise interference is also studied.(5) Acoustic images at meshing frequency and its side frequencies are studied, andthe effect of choosing different characteristic frequencies is investigated. For applying thefault diagnosis method based on acoustic imaging tequniques, the noticeable problemsabout choosing characteristic frequencies are concluded. Sound field distributions at sidefrequencies are not stable, and not suitable for fault pattern identification.(6) Diagnosis performances of different gray levels and different textural extractionmethods are investigated. The principle of choosing gray level and the performances ofextracting spatial distribution characteristics of sound field for different textural extractionmethods are concluded, and concrete suggestions are provided for practical applicaiton.(7) The application of acoustic imaging techniques in mechanical diagnosis and testsystem is studied. The common platform of fault diagnosis and test system based onacoustic imaging is designed by virtual instrument, and the principle prototype isdeveloped. The application example of gearbodx fault diagnosis shows its effectiveness.更多还原