节点文献

声发射信号处理系统与源识别方法的研究

Acoustic Emission Signal Processing System and Source Recognition Methods

【作者】 赵静荣

【导师】 王珂;

【作者基本信息】 吉林大学 , 通信与信息系统, 2010, 博士

【摘要】 声发射信号处理系统是声发射检测的关键环节,也是声发射技术的重点研究内容。本文对声发射信号特性进行分析,采用小波分析和支持向量机等信号处理技术,针对声发射处理系统软硬件设计、声发射信号信噪比提高、声发射源定位时延估计和声发射源分类识别等关键技术展开研究,主要工作如下:设计了基于波形分析的高速多通道声发射信号采集处理系统,构建了具有通用性、可扩展性的嵌入式硬件结构,以满足声发射信号处理快速性和大计算量的需求,提高了声发射信号采集的精度,并为后续研究工作的开展提供了完备的软硬件平台。研究了声发射信号信噪分离方法,针对传统小波变换算法复杂度高、运算速度慢不利于实时声发射信号去噪预处理的问题,提出了一种运算速度快、结构简单的基于提升小波的声发射信号去噪算法,有效地提高信噪比并易于硬件实现。系统分析了时延估计对声发射源定位精度的影响,利用小波分析良好的滤波特性,将小波分析与相关分析相结合,提出一种基于小波变换的相关时延估计方法,构建了基于Coif5小波的时延相关估计算法,解决传统时延相关估计方法易受噪声影响,导致时差定位精度低的问题。通过对复合材料声发射信号传播特性的分析,结合基于小波的相关时延估计算法,提出了基于声速修正的改进时差定位算法,解决了在复合材料声发射频度过高、传播衰减过大或检测通道数有限时无法采用声发射时差定位的问题,提高了复合材料声发射源定位精度。以提高小样本数据集下声发射源识别准确率为目的,构建了声发射信号小波包特征参量提取算法,将小波包特征能量分布系数作为分类器的输入特征向量,提出了一种基于二叉决策树分类策略的支持向量机多分类方法,提高了声发射源的识别准确率。

【Abstract】 Acoustic emission (AE) is an important non-destructive testing technology and its major objective is to locate and identify AE source in order to detect the damage degree and the service life of testing objects. AE signal analyzing and processing is the only way to achieve this goal. Research about AE signal processing system with higher performance and more effective AE sources identification method which helps to improve the identification, assessment and positioning accuracy of AE source has important theoretical significance and practical value.At present, with the AE detection range expanding and test object more diversification, the traditional AE signal processing techniques and testing instruments based on the parameters analysis hardly improve the ability of resolution, filtering and classification. Thus waveform analysis technology became a new research direction and research emphases of the AE signal processing. However the signal processing techniques based on waveform analysis are the difficulty and bottleneck of AE testing because the AE signal is a kind of weak signal hidden in the strong background noise which has the characteristic such as nondeterminacy, unpredictability, transient and multiformity. Combined with the characteristics of AE signal, new information processing technologies such as the wavelet analysis and support vector machine (SVM) model identification are introduced into the AE signal processing field to solve the problem above. The research focus on such key technical problems as processing system hardware and software design, filtering processing, time delay estimation, source location and classification.1. High-speed multi-channel acoustic emission signal processing system design based on the waveform analysisAiming at the shortcoming of existing AE signal processing system in improving AE testing performance, a high-speed multi-channel acoustic emission signal processing system and its software and hardware function module were designed with the general and expansibility.Based on PCI bus data communication mode between DSP and computer, the acoustic emission data acquisition and pretreatment system were designed by using DSP and FPGA embedded hardware structure. Various functions such as acoustic emission data analysis, display and output were realized by using powerful display function and rich software programming resources of the computer system. The system satisfies the requires of high speed and large amount of calculation for acoustic emission signal processing. It improved the acoustic emission signal acquisition and processing precision and provided a complete software and hardware platform for subsequent research work.2. The study of acoustic emission signal de-noising method based on wavelet analysisAfter studying denoising methods and steps based on wavelet analysis, three denoising methods including wavelet modulus maxima algorithm, wavelet scales related law algorithm and wavelet threshold method were compared for their quality and applicable conditions. Then the wavelet threshold denoising method was identified as the key research content for its simple structure and small amount of calculation.After analyzing the key technical issues about the threshold function selection and wavelet threshold optimization, comparative characteristics between the soft and hard thresholding function was studied and a half soft thresholding function was adopted for denoising. The signal-to-noise ratio and minimum mean square error were used as denoising effect evaluation index. The denoising effects of Wavelet functions including db6、sym8 and coif5 were analyzed through computer simulation experiment.Aiming at the shortcoming of traditional wavelet de-noising methods like poor flexibility, slow speed and hardware implementation difficulties, and a denoising method based lift wavelet was proposed. The denoising method was improved by using adaptive lifting schem and combined with the half soft thresholding function for AE signal denoising. And the results of the improved method are compared with that of the traditional wavelet transform and the lift wavelet transform. The simulated results showed that the adaptive lifting scheme had the best denoising results and has the advantages of simple structure, fast speed and easy hardware implementation.3. The study of cross correlated time delay estimation method based on wavelet analysisAfter studying the principle and characteristics of time difference location method for acoustic emission source, the influence of time delay estimation for positioning accuracy was analyzed. And a cross correlated time delay estimation method based on wavelet transform is proposed. The wavelet analysis is combined with the correlation analysis in the proposed method.From the wavelet transforming of acoustic emission signal, theory formula of the proposed method was deduced. Considering the nonstationary random characteristic of AE signal, the cross correlated time delay estimation algorithm based on the Coif5 wavelet is proposed after analyzing of the orthogonality, symmetry and smoothness and regularity of the wavelet function. The proposed algorithm was compared with the original correlated time delay estimation algorithm by computer simulation analysis under the condition of different noise. The results showed that. the proposed algorithm reduced the impact caused by the acoustic emission wave frequency dispersion and the noise, and improved the positioning accuracy without restriction of the signal noise correlation.4. Research and Improvement of AE time difference location methodThe propagation properties analysis of AE signal in composite materials showed that the sound velocity was very different at different direction of propagation because of the non-uniformity and anisotropic. But the sound velocity is set for a constant in the existing time difference location methods, which resulted in low positioning accuracy when the method was used for processing AE signal in composite materials. Even the method can not work under the condition with high frequency of AE event, serious transmission attenuation or limited inspection channels.Combining the time difference estimation method based on wavelet transform, an improved time difference location method was proposed to solve the problem. The influence caused by the sound velocity difference was eliminated by using a sound velocity- angle relationship equation. Comparative experiments of the time difference location method and its improved method were completed on the Carbon fiber composite material plate. The results showed that the proposed method improved the acoustic emission source location accuracy effectively, and the positioning error within 3%.5. Study of AE source classification method based on SVMAfter studying the existing AE signal classification method by analyzing the advantages and disadvantages, a classification method for AE source based on SVM is proposed. The limitations of the existing classification methods for AE signal include that researchers need to have rich background knowledge and data analysis experience for the method such as amplitude identification, frequency identification and statistical pattern recognition, and fuzzy diagnosis and artificial neural network method relies on statistical characteristics under the large sample data set. But in practical application, acoustic emission signal has irreversibility that result in the acoustic emission signal of pattern recognition usually does not have large amounts of sample data. The contradiction causes that the methods above have the problem such as getting in local minimization and poor generalization ability which restrict the acoustic emission sources identification accuracy improvement.After studying the application of wavelet packet analysis in signal processing, this paper proposed a wavelet packet-based characteristic parameter extraction method for AE signal. Combing the method with the multi-classification methods of SVM, the AE signals caused by the damage modes including fiber fracture, matrix cracking and interface separation were classified. The experiment result proved the feasibility of the AE source classification method based on SVM, and the identification accuracy effectively was effectively improved.

  • 【网络出版投稿人】 吉林大学
  • 【网络出版年期】2010年 08期
  • 【分类号】TN912.34
  • 【被引频次】27
  • 【下载频次】1899
  • 攻读期成果
节点文献中: 

本文链接的文献网络图示:

本文的引文网络