Terahertz Time-domain Spectroscopy and Imaging Technology of Rubber Materials

Author:Xu Feng

Supervisor:duan mu qing zuo

Database:Doctor

Degree Year:2019

Download:48

Pages:159

Size:7499K

Keyword:

Rubber composites are widely used in industry as a kind of material with high elasticity,ductility and sealing.However,in the process of production and use,rubber composites inevitably have various defects,resulting in potential safety hazards.In 2003,the crash of the US space shuttle Columbia caused the defect of the thermal insulation foam in the fuel tank due to the failure of the rubber seal.In 2018,Chongqing’s 3U8633 flight to Lhasa caused rupture of the right side of the cockpit due to the aging of the rubber seal ring.As a result,more and more rubber defects and other issues have attracted attention in the aviation and industrial fields.Terahertz time domain spectroscopy(THz-TDS)is a new THz band spectral measurement technology based on femtosecond ultrafast laser technology.Compared with other non-destructive testing technologies,its low energy and strong penetration characteristics show obvious advantages,become one of the most promising scientific development technologies in the 21 st century.The research on terahertz time domain spectroscopy technology of rubber composites has been carried out both in China and other countries,mainly on the classification analysis of rubber materials,additives analysis and aging research,rarely on the internal defects of rubber and its multilayer composites.In this paper,terahertz time-domain spectroscopy is used to study the internal debonding,inclusion and cavity defects of rubber multilayer composite materials.This topic comes from the National 863 Program "Research on terahertz nondestructive testing technology of composite materials",which provides a large number of supporting data for the imaging and quality evaluation of rubber composite materials.In this paper,three different types of rubber materials were analyzed by terahertz time-domain spectroscopy,and the refractive index of sample rubber was obtained.Preset the debonding and voids of NBR-Al monolayer adhesive and NBR-Al multilayer structure(hereinafter referred to as rubber foam multilayer structure)were obtained.In this paper,terahertz time-domain spectral characteristics and imaging technology of internal defects in rubber composites are studied.On the basis of the classical BP neural network,a comparative analysis is made of the adjustment learning efficiency,the introduction of momentum factors and the introduction of steepness factors to identify the defects of the NBR-Al monolayer adhesive material.The feature weight support vector machine is used to quantitatively identify the debonding defects of the three layer adhesive material of the rubber multilayer composites.Finally,Terahertz image enhancement of rubber materials using convolution neural network is studied.The following results are obtained:(1)Terahertz time-domain spectroscopy was used to extract optical parameters of butadiene-acrylic rubber,ethylene-propylene-diene-diene rubber and chloroprene rubber.In the 0.2-1.6 THz band,the variation of refractive index and absorption coefficient spectra of three kinds of rubber were obtained: the refractive index of ethylene-propylene-diene rubber and chloroprene rubber decreased with the increase of frequency,and the refractive index of butadiene-acrylic rubber decreased with the increase of In terms of absorption coefficient,there is an absorption peak in the absorption spectrum of NBR,but there is no absorption peak in chloroprene rubber and EPDM.It shows that there are differences in physical and chemical properties of rubber materials,which can be used to distinguish the three kinds of common rubber.(2)By extracting the optical parameters of rubber with the same thickness and different thickness,the refractive index of 9 mm thick NBR,EPDM and chloroprene rubber are 1.3612,2.4637 and 3.0521,respectively.By extracting the optical parameters of NBR with different thickness,the refractive index of 3 mm,6 mm and 9 mm NBR is 1.31,respectively.23,1.3306 and 1.3612;the refractive index of 3mmA2 foam interlayer is 1.0412.(3)A NBR-Aluminum Rubber monolayer adhesive material was constructed.The debonding,void and inclusion defects were preset.The spectral characteristics of each defect in time and frequency domain were studied under the reflection mode,and the debonding,void and inclusion defects were broken.The optimal imaging frequencies are 0.215 THz,0.2508 THz and 0.262 THz,respectively.(4)Based on the classical BP neural network,the improved algorithm is analyzed and compared three algorithms,which are the adjust learning efficiency,introducing momentum factor and introducing steepness factor.The analysis shows that under the same number of iterations,the method of introducing steepness factor has faster convergence rate and higher learning rate than the other two methods.By introducing steepness factor to classify the sample defects,the system defect recognition rate can reach 90%.(5)The debonding,voids and inclusion defects of rubber foam multilayer structures are preset.The imaging methods of defects are studied through two modes of reflection and transmission.The feature spectrum information of time domain windows is proposed to study the underlying defects of the rubber multilayer structure.The results show that terahertz reflection mode can clearly image debonding and inclusions,but the effect of cavity imaging is not obvious;transmission mode can detect voids and debonding,but the effect of inclusion imaging is not ideal.(6)Comparative study of support vector machine and feature weighted support vector machine for 10 levels of debonding rubber foam multilayer structure recognition rate,support vector machine recognition rate of 78%.On this basis,Relief-F algorithm is selected to weigh the features,and the recognition rate is 84%.The recognition performance of the weighted support vector machine for debonding grade is improved obviously.(7)Established a deep learning convolution neural network model,enhanced the defect image of the rubber aluminum monolayer adhesive and rubber foam multilayer structure,increased the gray value dynamic interval of the pixel,enhanced the contrast gray tone change,made the image clearer,improved the image recognition rate,and passed the Canny Quantitative analysis of image local defects by operator provides a new solution for terahertz time domain spectral imaging.The innovation of this paper is as follows:(1)An extended time-domain window information delay method is proposed to detect the bottom defects of rubber three-layer composites,which overcomes the problem that the main reflection wave can not detect the bottom defects.(2)Based on the classical BP neural network,the improved algorithm is analyzed and compared three algorithms,which are the adjust learning efficiency,introducing momentum factor and introducing steepness factor.The analysis shows that under the same number of iterations,the method of introducing steepness factor has faster convergence rate and higher learning rate than the other two methods.By introducing steepness factor to classify the sample defects,the system defect recognition rate can reach 90%.(3)In view of the quantitative identification of ten debonding grades of rubber foam core three layer adhesive material,based on support vector machine,the Relief-F algorithm is introduced to weight the features,and the recognition rate is increased from 78% to 84%,which provides reliable for terahertz time domain system for the quantitative identification of the debonding of the three layer adhesive material of rubber foam core basis.(4)Established a deep learning convolution neural network model,enhanced the defect image of the rubber aluminum monolayer adhesive and rubber foam multilayer structure,increased the gray value dynamic interval of the pixel,enhanced the contrast gray tone change,made the image clearer,improved the image recognition rate,and passed the Canny Quantitative analysis of image local defects by operator provides a new solution for terahertz time domain spectral imaging.