【作者】 魏志刚；

【导师】 程建政；

【作者基本信息】 中国科学院研究生院（武汉物理与数学研究所） ， 声学， 2006， 硕士

【摘要】 轮箍是铁路机车行走的重要部件，在制造和使用过程中容易出现各种危害性缺陷，严重威胁到列车的行驶安全。在轮箍缺陷的超声无损检测中，缺陷信号的识别会受到轮箍标记、闸瓦、轮轨接触点等位置的反射信号及表面波等多种因素的干扰。本文在超声横波探伤方法的基础上，将模糊模式识别方法应用到机车轮箍的无损检测中。以铁路内燃机车轮箍为实验检测对象，并依据国内现行检测标准，通过多个标准人工伤模拟轮箍自然缺陷。在分析了轮箍的结构特征以及轮箍缺陷产生规律的基础上，选择了大角度横波检测方法进行检测。并根据声学原理，对探头声场分布、轮箍检测方法作了一定的理论分析及实验验证。运用大角度横波检测方法对轮箍进行检测，并通过谱分析，提取缺陷回波信号的频域特征。利用模糊模式识别相关技术，以模糊聚类方法建立典型缺陷模糊子集，再对未知缺陷应用基于择近原则模糊识别方法进行归类识别，实验结果证明了该方法的有效性。在对同一缺陷重复检测中，正确识别率高达92.5％，为该方法进一步应用于轮箍缺陷的实际检测和分类提供了依据。在利用信号频域信息基础上，进一步结合小波变换的多尺度分析优点，首先对信号进行连续小波变换，然后提取各尺度下的幅度谱作为信号特征，对未知缺陷使用模糊聚类方法进行识别。实验结果表明，基于连续小波变换的幅度谱特征能有效地区分回波信号的局部差异，提高了不同模式间的可分性。更多还原

【Abstract】 Wheel is the most important part of running locomotive. During manufacturingand running, all kinds of flaws may appear in it, which highly affect the traffic safetyof train. During the ultrasonic non-destructive testing of wheel flaws, reflection fromthe wheel brake, sign, touch point, and the surface wave all may influence the flawechoes recognition result. This paper presents an application of fuzzy patternrecognition method in ultrasonic transverse wave detection of wheel flaws.Using a gas engine wheel as the test experiment sample, in order to simulate thenatural flaws, several typical artificial flaws were made in it according to the nationalwheel testing standard. By analyzing the structure of wheels and the law of flawappearance, an ultrasonic transverse wave detection method is selected. Referring tocorrelative acoustics theory, the probe sound field and the wheel testing rule arestudied, and the experiments verify the computing result.Using the frequency domain values as the pattern features, the fuzzy clusteringmethod is presented to construct the fuzzy sets of typical flaws. The concept ofsimilarity degree and the choosing near principle are employed in the fuzzy patternrecognition. Experiments show that the method is efficiency. The correctlyrecognition ratio is up to 92.5％.Further more, considering the multi-scale analysis merit of CWT, the echosignals are transformed by continuous wavelet first, then the spectrum of which ischosen as the pattern’s feature values. The fuzzy cluster analysis method is applied inrecognition. Experiment results show that the method can enhance thediscriminability of different flaws, and achieve more promising performance.更多还原