太赫兹无损检测的多特征参数神经网络分析技术

发布时间：2018-03-15 22:20

  本文选题：光谱学　切入点：太赫兹时域光谱　出处：《光子学报》2017年04期 　论文类型：期刊论文

【摘要】：提出一种基于太赫兹无损检测的多特征参数神经网络分析技术,用于分析耐高温复合材料的粘贴质量无损检测.采用抽片式方法设计了一种耐高温复合材料的脱粘缺陷样品,抽片厚度为0.1mm.采用太赫兹时域光谱无损检测技术对耐高温复合材料的多层脱粘缺陷进行了检测试验研究,对比了上下脱粘缺陷所对应的太赫兹时域波形及频谱信息的异同,针对性地建立了耐高温复合材料粘贴质量的上层脱粘参数、下层脱粘参数、频域吸收质心参数等多特征参数,将特征参数进行优化作为反向传播神经网络的输入并对其进行上下脱粘分类识别.通过对反向传播神经网络的训练测试,实现了耐高温复合材料上层脱粘0.1mm、下层脱粘0.1mm的脱粘缺陷的识别.
[Abstract]:A multi-characteristic parameter neural network analysis technique based on terahertz nondestructive testing (THz) is proposed to analyze the adhesive quality of high temperature resistant composites. The thickness of the strip is 0.1 mm. The multilayer debonding defects of high temperature resistant composites have been tested by using THz time-domain spectroscopy nondestructive testing technique, and the similarities and differences of terahertz time-domain waveforms and spectrum information corresponding to the upper and lower debonding defects have been compared. Several characteristic parameters, such as the upper layer debonding parameter, the lower layer debonding parameter, the frequency-domain absorption centroid parameter and so on, are established. The feature parameters are optimized as input of backpropagation neural network and identified by up-and-down debonding classification. The training test of backpropagation neural network is carried out. The debonding defects of the upper layer and the lower layer of the high temperature resistant composite are realized.
【作者单位】： 长春理工大学光电工程学院光电测控技术研究所;长春理工大学机电工程学院;
【基金】：国家高技术研究发展计划(No.2015AA6036A) 国防技术基础科研(No.JSZL2015411C002)资助~~
【分类号】：O441.4;TB33;TP183
，

本文编号：1617091