基于PSO-BP神经网络的红外无损检测缺陷定量识别
|
摘要
为解决红外无损检测缺陷定量识别困难的问题,提出了一种粒子群算法(PSO)优化反向传播(BP)神经网络的缺陷定量识别方法。以最佳检测时间与最大温差为模型的输入,孔洞缺陷的深度与直径大小为模型的输出,建立粒子群优化的BP神经网络缺陷定量识别模型。使用ANSYS软件对带有平底孔洞缺陷的金属平板进行脉冲热分析,提取金属平板检测表面的最大温差与最佳检测时间,作为神经网络模型训练与检验的数据样本,使用神经网络进行预测。计算结果表明:预测值的最大误差为5.5%,最小误差为1%,证明了粒子群优化BP神经网络方法进行红外无损检测定量识别的可行性。
In order to solve the problem of quantitative recognition of defects in infrared nondestructive testing, a particle swarm optimization(PSO) algorithm for quantitative defect recognition with back propagation(BP) neural network was proposed. With the input of the optimal detection time and maximum temperature difference, the depth and diameter of hole defects were the output of the model, and the BP neural network defect identification model of particle swarm optimization was established. ANSYS was used to analyze the metal plate with flat bottom hole defects by pulse thermal analysis. The maximum temperature difference and the best detection time were extracted from the surface of the metal plate. The neural network model was used to train, test and predict the data. Results showed that the maximum error of neural network prediction was 5.5%, and the minimum error was 1%. The feasibility of PSO-BP neural network for quantitative identification of infrared nondestructive testing was proved.
In order to solve the problem of quantitative recognition of defects in infrared nondestructive testing, a particle swarm optimization(PSO) algorithm for quantitative defect recognition with back propagation(BP) neural network was proposed. With the input of the optimal detection time and maximum temperature difference, the depth and diameter of hole defects were the output of the model, and the BP neural network defect identification model of particle swarm optimization was established. ANSYS was used to analyze the metal plate with flat bottom hole defects by pulse thermal analysis. The maximum temperature difference and the best detection time were extracted from the surface of the metal plate. The neural network model was used to train, test and predict the data. Results showed that the maximum error of neural network prediction was 5.5%, and the minimum error was 1%. The feasibility of PSO-BP neural network for quantitative identification of infrared nondestructive testing was proved.
引文
[1] 张金玉,杨正伟,田干,等.红外热波检测及其图像序列处理技术[M].北京;国防工业出版社,2015:8.
[2] 赵景媛,王黎明,刘宾,等.铸件内部缺陷红外无损检测的有限元模拟及分析[J].红外技术,2008,30(7):429.
[3] CARLSLAW H S.Conduction of heat in solids[M].London:Oxford University Press,1959.
[4] 刘颖韬,郭广平,杨党纲,等.脉冲相位法用于复合材料层析检测的研究[J].激光与红外,2009,39(4):390.
[5] 刘俊岩,戴景民,王扬,等.红外锁相法热波检测技术及缺陷深度测量[J].光学精密工程,2010,18(1):37.
[6] 刘波,张存林,冯立春,等.热波检测碳纤维蜂窝材料脱粘缺陷的边缘识别[J].红外与激光工程,2007,36(2):211.
[7] 韩力群.人工神经网络教程[M].北京:北京邮电大学出版社,2006.
[8] 范春利.几何形状导热反问题方法与应用[M].北京:科学出版社,2015.
[9] 陆金桂,奚天鹏,胡全斌.神经网络的遗传繁殖策略[J].南京化工大学学报(自然科学版),2001,23(5):1.
[10] 毕大强.粒子群优化算法及其在电力电子控制中的应用[M].北京:科学出版社,2016.
[11] 刘涛,李永峰,黄威,等.BP神经网络在红外热波无损检测定量识别中的应用[J].红外与激光工程,2012,41(9):2304.
[2] 赵景媛,王黎明,刘宾,等.铸件内部缺陷红外无损检测的有限元模拟及分析[J].红外技术,2008,30(7):429.
[3] CARLSLAW H S.Conduction of heat in solids[M].London:Oxford University Press,1959.
[4] 刘颖韬,郭广平,杨党纲,等.脉冲相位法用于复合材料层析检测的研究[J].激光与红外,2009,39(4):390.
[5] 刘俊岩,戴景民,王扬,等.红外锁相法热波检测技术及缺陷深度测量[J].光学精密工程,2010,18(1):37.
[6] 刘波,张存林,冯立春,等.热波检测碳纤维蜂窝材料脱粘缺陷的边缘识别[J].红外与激光工程,2007,36(2):211.
[7] 韩力群.人工神经网络教程[M].北京:北京邮电大学出版社,2006.
[8] 范春利.几何形状导热反问题方法与应用[M].北京:科学出版社,2015.
[9] 陆金桂,奚天鹏,胡全斌.神经网络的遗传繁殖策略[J].南京化工大学学报(自然科学版),2001,23(5):1.
[10] 毕大强.粒子群优化算法及其在电力电子控制中的应用[M].北京:科学出版社,2016.
[11] 刘涛,李永峰,黄威,等.BP神经网络在红外热波无损检测定量识别中的应用[J].红外与激光工程,2012,41(9):2304.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |