【作者】 詹湘琳；

【导师】 靳世久；

【作者基本信息】 天津大学 ， 精密仪器及机械， 2007， 博士

【摘要】 超声检测是油气管道环焊缝缺陷检测的一项重要技术。超声相控阵由于具有独特的电子扫查、动态聚焦和扇形扫描特性,能实现对非平面表面及复杂结构物体的缺陷检测,成为近几年超声探伤领域中的一个研究热点。本文重点研究了超声相控阵在油气管道环焊缝缺陷检测中的应用及缺陷识别技术。基于平面矩形活塞阵元,采用数值分析方法研究了一维超声相控线阵的声场特性,对辐射声场进行仿真。重点探讨了阵元数量N、阵元间距d、阵元长度l、阵元宽度w、焦距F和声束偏转角度θs等线阵参数对发射声束聚焦偏转的影响,给出了最优检测性能的换能器尺寸,为相控阵换能器的选择提供了理论依据。针对目前超声相控阵探伤中对缺陷的定性分析仍然依靠探伤人员检测经验的现状,将模式识别方法引入到环焊缝缺陷检测。提出了采用提升小波变换提取缺陷信号特征的方法。针对超声相控阵系统采集的油气管道环焊缝试块中的缺陷信号,采用“频带-能量”特征提取形式,结合基于距离的类别可分性测度作为特征提取的评价标准,比较了小波包变换与提升小波变换提取缺陷信号能量特征的性能。实验结果表明,提升小波特征提取速度比小波包的特征提取速度快一倍。采用基于提升小波变换的分形技术提取缺陷信号的分形特征。实验证明,加入分形特征后特征的可分性测度比仅采用能量特征提高了6.28%,开辟了管道环焊缝缺陷信号特征提取的一条新途径。采用基本遗传算法对提取的油气管道环焊缝缺陷特征进行优选,确定了有利于环焊缝缺陷识别的最优缺陷特征子集,提高了后续缺陷识别的效率。采用人工神经网络技术实现管道环焊缝缺陷识别。对目前在人工神经网络中应用最广泛的多层前馈误差反向传播(BP)算法进行了性能分析和算法改进,较好地实现了管道环焊缝缺陷信号的分类识别。研究了基于db4提升小波分析的径向基函数(RBF)神经网络进行环焊缝缺陷识别的性能。实验结果表明,RBF网络极大地提高了缺陷识别的速度,但识别的准确率太低,不适合于油气管道环焊缝缺陷的识别。提出采用数据挖掘技术中的支持向量机(SVM)模型对油气管道环焊缝缺陷进行自动识别。建立了以RBF核为核函数,SMO算法为进化算法的SVM模型。比较了改进的BP网络、RBF网络和SVM模型的缺陷识别结果,表明SVM在识别效率和准确率方面具有明显优势。研制了一套针对油气管道环焊缝缺陷检测的超声相控阵系统。实验证明系统具有良好的缺陷检测性能。更多还原

【Abstract】 Ultrasonic testing is an important flaw detection technology of oil and gas pipeline girth welds. With its unique electronic scanning, dynamic deflection focusing (DDF) and sectorial scanning characters, ultrasonic phased array (UPA) technology can be used for defect detection of curved face objects or objects with complex structure. Thus, UPA technology has been becoming a focus of ultrasonic testing research. Flaw detection and flaw identification of oil and gas pipeline girth welds, using UPA system are deeply researched in this dissertation.Based on rectangular plane piston array element, numerical analysis method is applied to research on the ultrasonic field characteristics of linear UPA. The deflecting and focusing effects of some array parameters, such as array spacing d , on the transmission beam are thoroughly discussed, which supplys theory of UPA trancducer choosing. An optimum UPA transducer is accordingly achieved.To improve automatic level of flaw qualitative analysis, pattern recognition method is introduced. Lifting Wavelet Transform (LWT) is proposed to extract features of flaw echoes in pipeline girth weld block. Combined with feature evaluation standard based on divisibility measure of distance, frequency-energy feature extraction form is applied. Wavelet Packet Transform (WPT) and LWT are used to extract energy feature and their results are compared. The testing result indicates that extraction time of LWT is nearly half of WPT.Fractal theory combined with LWT is put forward to extract the flaw signals’fractal dimension. Experimental results validate that divisibility measure of features with fractal is 6.28% more than that of features without fractal. This establishes a new path to extract flaw echoes’feature of pipeline welding.Simple Genetic Algorithm (SGA) is used to choose better features in extracted flaw echoes features. A best flaw features subset is determined, improving succeeding flaw identification efficiency.Artificial neural network (ANN) is adopted to implement automatic flaw identification of pipeline girth weld. Performance analysis and algorithm improvement are carried out on BP neural network, which has been used widely in practical application. The testing result is preferable. Moreover, RBF neural network with db4 LWT is built to identify the flaw waves in pipeline girth welds. The result shows that identification speed is greatly increased, but the accuracy is dramatically decreased. So RBF network is unfit for flaw identification of pipeline girth weld.Support vector machine (SVM) is a new technology in data mining field, and is brought forward for automatic flaw identification in ultrasonic NDT field. A SVM model with RBF kernel and SMO evolution algorithm is built. The improved BP network, RBF network and SVM are all tested on the same acquisition data of flaw echos. The experiment result suggests that SVM is superior to others on identifying speed and accuracy.The current UPA system generally is heavy and bulky. To overcome this shortcoming, a UPA flaw detection system for pipeline girth weld is developed. Experimental results validate its excellent flaw detection capability.更多还原