【作者】 王衍学；

【导师】 何正嘉；

【作者基本信息】 西安交通大学 ， 机械工程， 2009， 博士

【摘要】 机械设备结构的复杂性,故障的发生以及设备工况的改变都将导致其振动信号具有非平稳特性。从这些非平稳信号中提取有用的故障信息是能否准确检测故障的关键。时频分析是非平稳信号分析的有力工具,但是传统时频分析方法存在各自局限性,难以完全适用于某些有用信息的捕捉,因而有关的研究和应用受到了很大的限制。本文提出了广义时频域平均、局部均值分解和双树复小波变换等新的时频特征提取方法,旨在为机械早期和复合故障诊断提供更加有效方法。超声波检测是各种超声波工程应用中的关键,本文试图研究一种具有广泛应用前景的超声波通用检测方法。本文提出基于光滑伪Wigner-Ville分布、连续小波变换和Hilbert-Huang变换(HHT)的广义时频域平均技术。该技术将时域平均技术扩展到时频域内,充分发挥时频分析良好的时频聚集性以及多次平均操作的抗干扰信号的能力。广义时频域平均技术非常适合于超声波检测应用,检测方法不受到各种噪声干扰和传输媒介的失真影响,数值仿真实验验证了其有效性。以超声波流量计应用为例,将广义时频域平均技术应用于实际超声波检测中,结果证实该技术具有较强超声波信号检测能力。另外,本文指出现行多种时频分析方法的组合技术,其本质也是一种广义时频域平均。局部均值分解(Local Mean Decomposition, LMD)是最近提出的一种新的信号自适应分解方法,本文提出一种基于局部均值分解的机械故障信号解调技术。针对局部均值分解的端点效应、滑动平均步长选择等关键问题进行详细研究,提出相应的具体改进方法。另外,基于LMD分解得到的瞬时幅值和瞬时频率,本文提出了能够同时全面反映信号时频域信息的瞬时时频谱构造方法。通对调幅、调频和混合调制三种仿真信号的分析,证实了所提的LMD改进理论的可行性和有效性,且相比HHT和小波映射Hilbert谱方法,局部均值分解的瞬时时频谱可更为全面、准确解调出信号中的调制和载波信息。利用局部均值分解后构造的瞬时时频谱对实验室转子和某重催机组碰摩故障进行分析,进一步证实了该技术可有效提取出设备运转过程中碰摩故障所导致的工频与亚频的调制特征。另外,通过对某轧机齿轮箱故障诊断得出LMD瞬时时频谱可准确解调齿轮调制以及故障的严重程度等信息。自适应信号分解方法近年来由于其在工程应用较少的认为参与受到研究人员的广泛关注。经验模式分解(Empirical Mode Decomposition, EMD)与LMD均为信号的自适应分解方法,为了在工程实践中很好地应用此两种方法,本文在局部均值求解、瞬时频率计算、分解分量和滤波性能等方面做了深入对比研究,结果表明:LMD在计算局部均值过程中不会出现EMD技术的过冲、欠冲等问题;LMD可以不借助Hilbert变换准确求得瞬时频率,且信号分解过程与瞬时频率求解一步完成;LMD分解所得到的乘积函数相比EMD的内禀模态函数物理意义更明显;LMD与EMD具有相似的小波类型的滤波特性,但是LMD不易出现EMD模态破裂问题。采用LMD与EMD对两个实际旋转机组分别存在碰摩和蒸汽激振故障进行诊断,结果表明LMD在早期故障检测方面要胜过EMD方法。为了弥补基于第一、二代小波变换、谱峭度和EMD等技术在机械信号噪声消除与冲击特征提取中的不足,本文提出了基于双树复小波变换的机械信号降噪和复合故障诊断技术。仿真信号分析结果表明双树复小波变换具有良好的抗频带混叠能力和平移不变性,这些特性分别保证了双树复小波变换在谐波信号提取和周期性复合冲击特征检测方面要优于经典离散小波变换、第二代小波变换和经验模式分解。双树小波变换的优良特性实质上来源于两个小波基之间的特殊关系,而并不是单纯地利用了经典小波变换的小波基与待分析信号之间的匹配性。机械信号存在各种噪声是不可避免,本文针对信噪比较小时机械信号特征提取问题,联合邻域相关系数降噪技术,提出了一种双树复小波领域相关系数降噪方法。对齿轮裂纹故障信号的降噪结果证实了所提降噪方法与基于第一、二代小波变换的领域系数降噪方法相比,可有效去除噪声,并且尽可能多地保留有用信息。在多重特征同时基提取方面,滚动轴承复合故障诊断和某实际工业机组的多重故障特征提取结果证实双树复小波变换方法要比第二代小波变换和快速谱峭度方法可更为有效。另外,双树复小波变换所具有的有限冗余特性和鲁棒性,使其具有较高运算效率,可完全用于机械故障的在线监测与诊断。开发了潜艇模型状态监测与故障诊断系统。通过系统的下位机监测与上位机的分析诊断功能,可实时、准确地监测模型内设备的运行状态,预测和诊断早期、潜在的故障,同时有效监测模型噪声辐射,识别主要噪声源,并据此提高了其隐身性能。利用双树复小波变换良好的滤波和平移不变性能,增强对潜艇模型故障的特征提取能力。联合应用双树复小波变换和局部均值分解技术,将双树复小波变换作为局部均值分解的前处理,提高了局部均值分解自适应滤波能力,可识别更多的潜在的故障信息。更多还原

【Abstract】 Vibration signals are often non-stationay due to the complexity of the mechanical systems, especially when the faults occur. Extracting features from those non-tationary signals is the key to successfully conducting machine condition monitoring and fault diagnosis. Time-frequency techniques have shown their advangtages for the analysis of non-stationary signals, however sometimes they can not provide comprehensive and accurate information because of their limations. The objective of this dissertation is to develop novel and effective time-frequency methods for fault detection and diagnosis.Ultrasonic detection is essential for the ultrasonic-based applications. Rather than focusing on a particular application area, we attempt to provide a general methodology for ultrasonic detection. Based on the smoothed pseudo-Wigner-Ville distribution, continuous wavelet transform and Hilbert-Huang transform, three extended time-frequency domain average (ETFDA) techniques are proposed in this dissertation. These techniques extend the time-domain average (TDA) and combine the localizing characteristics of time-frequency analysis with the abilities of the TDA to suppress noise interference. They are suitable to detect the ultrasonic even when the received signals are smeared by the noise or distorted in the medium. Numerical investigation on the performance of the ETFDA is carried out. A number of tests conducted on simulated and actual signals have demonstrated that ETFDA possesses satisfactory performances.Demodulation is an available method for fault diagnosis. Local mean decomposition (LMD) is a new adaptive signal decomposition technique, and a demodulation technique based on LMD is developed. A method of boundary processing and a strategy for determining the step size of moving average are presented to improve the algorithm of LMD. Instantaneous amplitude (IA) and instantaneous frequency (IF) of the signal can be computed independently of Hilbert transform using LMD method. A well-constructed description of the derived IA and IF is given in the form of instantaneous time-frequency spectrum (ITFS), which preserves the time and frequency information simultaneously. Results of three synthetic signals indicate that the proposed method is a better demodulation approach to extract the all the carrier and modulated components as well as the accurate IF, compared with Hilbert-Huang transform and wavelet project Hilbert spectrum. The validity of the technique is then demonstrated on the laboratory experiments and a real rotor system of a gas turbine with rub-impact fault. Due to the opposite friction during operation, the transient fluctuation of IF of the fundamental harmonic component is successfully identified in the ITFS. In addition, we find that the proposed technique is more effective and sensitive than other methods in detecting sub-harmonics and FM components contained in the rub-impact signals.In recent years, the adaptive decomposition methods have attracted many researchers’attention, because they are less influenced by human operator in practical applications. LMD and EMD are both adaptive decomposition methods. This dissertation compares LMD and EMD from four aspects through numerical simulation: local mean, decomposed components, instantaneous frequency and the wavelet-like filtering characteristic, and the results obtained are as follows: firstly, overshoot and undershoot of local mean can be avoid using LMD; secondly, more accurate IF of the signals can be acquired by LMD; thirdly, product functions contain more meaningful interpretation than IMFs and fourthly, mode separation may not occure using LMD. Then LMD and EMD are both applied in the health diagnosis of two industrial rotating machines with rub-impact and steam-excited vibration faults, respectively. The results reveal that LMD seems to be more suitable and have better performance than EMD for the incipient fault detection.In order to solve the defects of the first and second generation wavelet transform, kurtogram and EMD etc. in denoising and extracting impulse, a novel signal denoising technique and a multi-fault diagnosis technique based on the dual-tree complex wavelet transform (DTCWT) are proposed. Through numerical simulations, it is demonstrated that DTCWT enjoys better shift invariance and less spectral aliasing than first generation wavelet transform, second generation wavelet transform (SGWT) and empirical mode decomposition (EMD). These advantages arise from the relationship between two dual-tree wavelet basis functions, instead of the matching of the wavelet basis function to the signal being analyzed. Since noises inevitably exist in the measured signals, an enhanced vibration signal denoising algorithm incorporating DTCWT with NeighCoeff shrinkage is also developed. Denoising results of the vibration signals resulting from a cracked gear indicate the proposed denoising method can effectively remove noise and retain the valuable information as much as possible compared with DWT- and SGWT-based NeighCoeff shrinkage denoising methods. As is well known, the acquisition of comprehensive signatures embedded in the vibration signals is of practical importance to clearly identify the cause of the fault, especially the combined faults. In the case of multi-features detection, diagnosis results of rolling element bearings and an actual industrial equipment confirm that the proposed DTCWT-based method is a powerful and versatile tool and can consistently outperform SGWT and fast kurtogram widely used recently. Moreover, it must be noted, the proposed method is completely suitable for the on-line surveillance and diagnosis due to its good robustness and high efficiency.Monitoring and diagnosis system of vibration and acoustics for the submarine model is developed. By slave computer monitoring and the host computer analysis, running state of the equipment in the submarine model can be real-time and accurately monitored and prognosis on the incipent and potential faults can be conducted. Meanwhile, sound radiation of the model is monitored and main noise source is diagnosed, which may raise the concealment performance of the submarine model. DTCWT and LMD are used in the monitoring and diagnosis system. DTCWT is helpful to extract harmonic features of equipment in the submarine due to its good anti-aliasing filtering. Furthermore, DTCWT can be considered as a preprocessing method for LMD, which may enhance the performance of the filtering and detect more potential fault signatures.更多还原