基于QGA-LS-SVM的超声导波缺陷轮廓重构  

Defect profile reconstruction for ultrasonic-guided waves based on QGA-LS-SVM

作　　者：刘兵[1] 唐力伟[2] 王建斌[1] 王长龙[1] 张轩硕[1] 

机构地区：[1]军械工程学院无人机工程系,河北石家庄050003 [2]军械工程学院火炮工程系,河北石家庄050003

出　　处：《中国工程机械学报》2013年第3期205-210,215,共7页Chinese Journal of Construction Machinery

基　　金：军械工程学院科学技术研究基金资助项目(2010SY4309002)

摘　　要：超声导波缺陷轮廓重构是指由检测到的缺陷回波信号重构缺陷轮廓及参数,是实现超声导波信号反演的关键.探讨了应用最小二乘支持向量机(Least Squares Support Vector Machine,LS-SVM)对缺陷轮廓进行重构的方法,并利用量子遗传算法(Quantum Genetic Algorithm,QGA)优化LS-SVM及核函数的参数.LS-SVM的输入是缺陷所产生的回波信号,输出是缺陷轮廓数据,建立起由缺陷的回波信号到缺陷二维轮廓的映射关系.训练样本和测试样本由实验数据与仿真数据组成.该方法实现了缺陷的二维轮廓重构,并与BP(Back Propagation)神经网络、GRNN(Generalized Regression Neural Network)神经网络和常规遗传算法LS-SVM三种方法的重构效果进行了比较.结果表明,该方法速度快、精度高,并有很好的泛化能力,是一种行之有效的缺陷反演方法.The defect profile reconstruction for ultrasonic-guided waves is aimed to reconstruct the defect profiles and parameters via echo-wave signals, i. e., inversion of ultrasonic-guided wave signals. By using the LS-SVM as a defect profile reconstruction method, the quantum genetic algorithm （QGA） is adopted to optimize the parameters of LS-SVM and kernel function. Due that the LS-SVM inputs are defect-resulted echo signals, the outputs are defect profile data. Mterwards, the mapping between defect echo signals and 2D profiles is established. Subsequently, the training and testing samples are composed of experimental and simulation data. Consequently, the 2D defect profile reconstruction is implemented by comparing the reconstructed effects upon BP network, GRNN and LS-SVM. From experimental results, it is indicated that the proposed method, which possesses such advantages as high speed, high accuracy and generalization ability, is proven effective and feasible for defect inversion.

关 键 词：超声导波 轮廓重构 量子遗传算法 最小二乘支持向量机 

分 类 号：TG115.28[金属学及工艺—物理冶金]