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0.1 mm.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0.1 mm、下层脱粘0.1 mm的脱粘缺陷的识别.

A multi-feature parameter neural network analysis technique based on terahertz nondestructive testing for the analysis of the adhesion quality of the heat-resistant composites was proposed. A film (thickness is 0.1mm) extraction method was put forward to simulate the bonding defect of the heat-resistant composites. The terahertz time-domain spectroscopy nondestructive testing technology was used to detect the multilayer high temperature resistant composite bonding defects. Compared the similarities and differences of terahertz time-domain and the frequency domain information between the upper debond defect and the lower debond defect, the multi characteristic parameters were proposed for the adhesive quality, such as upper debond parameter, lower debond parameter and centroid absorption parameter for frequency domain. The characteristic parameters were optimized as the input of back propagation neural network to recognize the upper and the lower debond defects. Based on the back propagation neural network training test, the identification was realized for the 0.1mm thickness upper debond defect and the 0.1mm thickness lower debond defect.