【作者】 王向红；

【导师】 朱昌明；

【作者基本信息】 上海交通大学 ， 机械设计及理论， 2009， 博士

【摘要】 混流式水轮机是水电站普遍采用的机型。其转轮担负着把水能转化成机械能的重要任务,但极易产生裂纹。转轮出现裂纹将严重地危及电站的运行稳定性和运行安全,而频繁停机焊补裂纹又给电站造很大的经济损失。尽管从设计和制造方面进行了大量的研究,但仍不能完全解决混流式转轮的裂纹问题。若能监测裂纹的产生、掌握裂纹尤其贯穿裂纹的扩展趋势,将是预报转轮工作寿命的重要措施,它事关电站的高效、稳定和安全运行。材料在裂纹产生和扩展过程中,伴随着局部源快速释放能量而产生瞬态弹性波的现象,即声发射(Acoustic Emission,AE)。通过检测此波而发展起来的AE技术目前已有很多的研究和应用。本文对AE技术监测转轮叶片裂纹的可行性进行探讨,主要对AE信号在转轮上的衰减特性、强大背景噪声下弱小裂纹AE信号的提取、AE信号特征参数的提取与叶片上裂纹源的定位,以及疲劳过程中叶片材料的裂纹AE特性进行研究。主要取得了以下成果:1.研究了转轮上AE信号的衰减特性。根据AE信号在转轮上传输时的主要衰减因素,进行了两个方面的研究:传输距离对信号的衰减和结构界面对信号的衰减,并利用小波包技术对衰减结果进行了分析。采用信号能量和幅值这两个AE参数对信号的衰减性能进行描述,通过两组实验得出信号在转轮上传输6m后,仍然可检测到。距离为影响衰减的主要因素,界面也对信号衰减存在影响,但依赖于构成界面的构件相对尺寸的大小,因而实际应用中传感器的安装位置应选择在构件相对尺寸接近且结构较简单的部位。通过小波包分析得出特征小波包的信息与原信号的AE参数的最大偏差较小,特征小波包能有效反映转轮上AE信号的衰减特性。利用特征小波包代替原信号,可减少现场监测过程中数据保存和传输的压力。2.强背景噪声下AE信号的提取。针对水轮机组运行环境背景噪声强的特点,为从接收到的信号中滤除噪声而提取出有用信号,分两种情况进行去噪:1)当噪声成分和独立成分总数不大于提取的混合信号数目时,把噪声成分同时看作独立源,利用独立成分分析(ICA)的盲源分离,将有用成分提取出来,再经过(伪)逆变换还原出无噪信号。通过对混合不同强度白高斯噪声的断铅信号和混合水轮机组背景噪声的断铅信号的提取,实验结果表明,此方法基本不受输入信噪比以及信号所处频率范围的影响,均能较好的分离出信号。2)当源信号总数大于混合信号数时,利用基于ICA的稀疏编码收缩(SCS)方法进行去噪。由最大后验估计对独立成分进行提取,采用泛化高斯模型捕获独立成分的概率密度函数,并利用非线性收缩函数进行去噪,实验结果与小波阈值去噪方法进行比较。结果表明,基于ICA的SCS去噪方法能够从水轮机组运行的背景噪声中提取出叶片裂纹信号和断铅信号。虽然去噪效果没第一种情况好,但比第一种方法更适用,且去噪效果高于小波阈值去噪效果。3.提出了AE特征参数的提取方法以及转轮裂纹源的定位方法。针对转轮结构大而复杂且出现裂纹区域比较集中等特点,采用智能定位方法。为提高定位精度,首先必须对输入的大量相关AE参数进行特征提取。本文分别利用ICA、核主成分分析(KPCA)以及核独立成分分析(KICA)对AE参数进行特征提取。结果表明KICA提取的贡献率和超过90%的9维特征参数包含的信息量最大,且参数间的冗余最小。利用提取的特征参数分别通过小波神经网络和支持向量机这两种智能定位方法进行裂纹源位置的识别。仿真得出SVM对裂纹源的识别结果好于小波网络的结果。因而,在实际利用AE技术对水轮机转轮裂纹进行定位时,采用KICA特征提取降低了数据的维数,既减少在线监测过程中声发射参数存储和传输的压力,同时定位精度也得到了提高,KICA结合SVM是一种较好的定位方法。4.研究了转轮叶片材料的疲劳AE特性。水轮机叶片绝大多数裂纹为交变载荷作用下的疲劳裂纹,而叶片疲劳裂纹扩展速率是评估水轮机转轮损伤程度的基本数据。因此在实验室环境下,使叶片材料试件在标准三点弯曲疲劳实验下产生裂纹,研究此材料的疲劳裂纹扩展速率及相应的AE特性,并与水轮机组现场背景噪声的AE特征进行比较。研究结果表明,AE参数与疲劳裂纹状态存在一定的对应关系,可用AE参数表征裂纹的状态。得到了疲劳裂纹扩展过程中的AE计数率、能量变化率、裂纹扩展率与应力强度因子的关系,通过这些关系式可预测叶片的疲劳寿命。同时,得出了AE计数率、能量变化率与裂纹扩展率的关系,即可由检测到的AE参数的变化率推算出裂纹扩展率的大小,从而大致评估叶片的安全性,避免了实际结构中应力强度因子测量困难的问题。而且,发现疲劳裂纹信号与水轮机组现场背景噪声的AE参数值相差较大。更多还原

【Abstract】 Francis turbine is popularly used in power stations. The runner of the turbine bears the important task of transferring the water energy to mechanical energy, which makes cracks appeared easily. Initiation of blade cracks endangers the safety operation of power stations. And frequently shutting down to weld cracks brings a lot of economic loss for power stations. Although many studies have been done at the stage of design and manufacture, the problem of cracks can’t be resolved completely. It will be an important measure of predicting the working life of runners if we can monitor the crack initiation and master the developing trend of cracks, especially the throughout cracks, which is an enormous problem relevant to highly efficient, stationary and safety operation.Acoustic emission (AE) is a phenomenon of generating transient elastic wave caused by the fast energy release of local source during the initiation and growth of cracks. There are many researches and applications using AE technique to collect the signals of cracks. This paper is a preliminary study of applying AE technique to detecting the large-scale turbine runner. The major investigative contents include: attenuation characteristics of AE signals across the runner, extraction of weak signals buried in big background noise, feature extraction of AE parameters and source location of cracks in blades and the fatigue AE characteristics of the blade material. The following achievements had been obtained:1. The attenuation characteristics of AE signals propagated across the runner were researched. The attenuation characteristics due to the propagation distance and the existence of a structure interface were studied according to the major affecting factors on AE signals across the runner. At the same time, the attenuation results were analyzed by the wavelet packet technique. Two commonly used AE parameters, energy and maximum amplitude, were used to describe the attenuation performance. From the tests, it is concluded that AE signals are detectable after propagating at a distance of 6 m. The propagation distance is the major factor of attenuation and the interface composed by the same type of material has effects on attenuation, which depends on the relative size of structures. It is a better way to mount sensors on a simple structure that has a possibly equivalent size with the structure incurred AE sources when the sources propagate across the interface. Furthermore, through the wavelet packet analysis, it is concluded that the maximum deviations of the attenuation slopes of the energy and amplitude between the feature packet and raw signal are small, which shows that the attenuation characteristics of the packet and the corresponding raw signal are substantial agreement. The feature packet can reflect the attenuation characteristics of signals propagated across the runner. In light of this, the pressure on data transmission and storage can be decreased by extracting feature packet coefficients.2. Extraction of AE signals mixed with the strong background noise was studied. The operating noise of turbine units is strong. In order to extract the wanted signal from the received noisy signal, there are two ways to denoise according to the different conditions: 1) When the total number of noise components and independent components is not more than that of the received mixed signals (i.e. the noise component is also regarded as an independent source), the blind source separation of the independent component analysis (ICA) was used to extract the wanted components. Then the free-noise signal was obtained by the (pseudo) inverse transformation of the wanted components. Through extraction for the pressing lead signal mixed with white Gaussian noise of different strengths and the operating noise of the turbine unit, it indicates that this method is not affected by the inputted signal-noise-ratio (SNR) and the frequency range of signals. The wanted signals can be extracted preferably.2) When the number of sources is more than the mixed signals, the denoising method of sparse coding shrinkage (SCS) based on ICA was used. The independent components were estimated by the maximum a posteriori (MAP) estimation and the probability density functions (PDFs) of the independent components were fitted by generalized Gaussian model. Then the nonlinear shrinkage functions were used to denoise. The results were compared with those obtained by the wavelet threshold value denoising method. It shows that the SCS method can extract the crack signal of the blade and the pressing lead signal buried in the operating noise of the turbine unit. Although the results of the SCS method are not better than those of the first method, it is more applicable. Furthermore, its denoising results are better than the wavelet method.3. The feature extraction of the AE parameters and source location of the runner cracks were studied. Hydraulic turbine runner has a complex structure and the crack regions are centered, so the intelligent location methods were used. In order to improve the accuracy of location, it is necessary to extract the feature parameters of the AE signal. The ICA, kernel principle component analysis (KPCA) and kernel ICA (KICA) were used to extract features. The result shows that the information of the first nine feature parameters with above 90% contribution rate is the largest and that the redundancy between the feature parameters is the least. The wavelet neural network (WNN) and support vector machine (SVM) were used to recognize the crack regions according to the extracted feature parameters, respectively. The simulation shows that recognition rate of SVM is better than that of WNN. As a result, in real world applications, doing feature extraction by KICA can decrease the dimension of input signals, which reduces the pressure on AE parameters transmission and storage and improves the accuracy of location as well. In light of these, it is a good method for source location in complex big-size structures to combine KICA with SVM.4. The fatigue AE characteristics of the runner blade material were studied. Some researches indicated that most of the regular cracks were fatigue cracks when blades were subjected to vibratory alternate stress. The fatigue crack growth rate of blade is the basic data to evaluate the degree of damage of the turbine runner. The crack signals of the blade material from three-point bending fatigue tests were received, and the fatigue crack growth rate and the corresponding AE characteristics were studied in the laboratory. Furthermore, the characteristics of the cracks were compared with those of background noise received from the locale of a hydraulic turbine unit. The results show that the AE parameters and fatigue cracks have a coincidence relation and that the parameters can represent the state of crack. The correlations of the crack growth rates, AE count rates and AE energy changed rates versus the stress intensity factor (SIF) range were obtained. The fatigue life can be predicted by the correlations. At the same time, the correlations of the crack growth rates versus AE count rates and AE energy changed rates were obtained. As a result, the crack growth rates of can be deduced by the AE parameters changed rates, which avoids the problem of measuring the SIF in real applications. In addition, the ranges of the AE parameters of fatigue crack signals and the background noise onsite are different.更多还原