应用激光, 2017, 37 (6): 888, 网络出版: 2018-01-10   

激光超声缺陷统计特征神经网络识别技术研究

Neural Network Identification of Defect Statististics Characteristics in Laser Ultrasonics
作者单位
1 中北大学计算机与控制工程学院, 山西 太原 030051
2 电子测量技术国家重点实验室, 山西 太原 030051
3 中北大学机电工程学院, 山西 太原 030051
摘要
针对激光超声检测过程中, 缺陷的分类、定位及特征难以识别的问题, 根据激光超声遇到不同裂纹呈规律性变化的特点, 采用神经网络对在激光超声检测中出现的缺陷特征进行统计识别。对采集到的激光超声信号归一化和规范化后, 首先计算每个信号的均值、均方根值、峰值、峭度等十个统计特征, 并将这些特征组合成一个等长的特征向量, 然后采用径向基(RBF)神经网络识别。经过计算发现总的识别正确率为95%, 部分类型的缺陷识别可以达到100%, 较低的识别正确率也在80%以上。实验结果表明, 该方法能精确、高效地识别裂纹缺陷且对环境的适应能力比较好, 有助于实现对裂纹的定量检测。
Abstract
In the process of laser ultrasonic inspection, the problems of defect classification, location and characteristics of defects are difficult to be identified. According to the characteristics of laser ultrasound to meet different crack show regular change, the neural network is used to perform statistical identification of the defect characteristics in laser ultrasonic testing. The laser ultrasonic signals were collected after normalization and standardization, it calculated the mean value, root mean square value, peak and kurtosis characteristics of ten, and combined these statistic features into an even-length feature vector, Then by using radial basis (RBF) neural network identification. After calculation, the total recognition accuracy is 95%, some types of defects can reach 100%, and the lower recognition accuracy is more than 80%. The experimental results show that the method can accurately and efficiently identify crack defects and better adaptability to the environment, it is helpful to realize quantitative detection of cracks.

郭华玲, 秦峰, 郑宾, 王余敬. 激光超声缺陷统计特征神经网络识别技术研究[J]. 应用激光, 2017, 37(6): 888. Guo Hualing, Qin Feng, Zheng Bin, Wang Yujing. Neural Network Identification of Defect Statististics Characteristics in Laser Ultrasonics[J]. APPLIED LASER, 2017, 37(6): 888.

本文已被 3 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!