Study On Detection Of Fiber Content For Blended Fabrics Based On Near Infrared Spectroscopy

Textile products are one of the main light industrial products for export in our country,and they are necessities for people living.The quality of textile products is directly related to manufacturers,trade related parties,government regulators,and every customer.Fiber content is one of the main quality indexes for evaluating textile products.As the important basis for consumer choice of textile products,it has important guiding role for washing and maintenance of textile products.Almost all the countries(regions)have issued a decree or compulsory standards,which textile products must be labeled fabric fiber types and content.The precision of traditional chemical dissolved is accurate.But it has disadvantages of time-consuming,low efficiency,pollution,and harm to the operator's physical and mental health.It is urgent to develop fast and efficient analysis technology with increasing of textile products year after year.Near infrared(NIR)spectroscopy,because of the advantages of rapid,nondestructive,high efficiency,green,no pollution,has been applied to textile products industry.The NIR spectra of textile products have the characteristics of weak absorption,wide spectral peak and peak overload.These problems lead to the difficulties of weak chemical component extraction and quantitative analysis.In this research,detection of fiber content was developed for blended fabrics by NIR spectroscopy,and the main research contents and conclusions were as follows:(1)Design and verification of scheme for enhancement of NIR spectral absorbance of fabric blends.A scheme was designed for enhancing NIR spectral absorbance of fabric blends.The characteristic of transflectance spectra was analyzed,and relationship between the fibers and NIR spectra was discussed.The influence to NIR average absorbance spectrum was compared between diffuse reflectance and transflectance detection mode.The absorbance of NIR spectra for textile was improved to 14.3% in transflectance detection mode,and the model precision of transflectance detection mode was slightly better than diffuse detection mode with correlation coefficient of 0.991 and root mean square error of cross validation(RMSECV)of 2.65%.The experimental results suggested that the NIR spectral absorbance was increased for the blended fabrics of cotton and polyester.(2)Validation and analysis for influence of thickness and porosity of blended fabrics to NIR spectroscopy.The influence of thickness and porosity was explored to NIR spectra.The effects were investigated for smoothing,multiple scattering corrections(MSC),conventional derivative processing,and the wavelet approximation derivative preprocessing.Derivative processing and MSC methods effectively eliminated the baseline drift of NIR spectra,and the optimal model was obtained by the combination of approximate derivative wavelet and MSC method.The RMSECV of the optimal model reduced from 6.43% to 5.28%.(3)Study on qualitative discriminant model of NIR spectra for identifying types of fabric fiber.The method of the combination of extreme learning machine(ELM)and NIR technology was investigated for identifying the types of fabric fiber.The influence of different parameters on the model discriminant accuracy was analyzed.And soft independent modeling of class analogy(SIMCA),least squares support vector machine(LSSVM)were analyzed too.The experimental results suggested that the optimal ELM model was obtained with identifying six kinds of sample precisely,and the running speed of ELM was superior to LSSVM and SIMCA models.(4)Study on quantitative analysis model of NIR spectroscopy for textile products.Thirteen NIR spectral variables were selected by Monte Carlo uninformation variable elimination(MC-UVE)and continuous projection algorithm(SPA).Then partial least squares(PLS),LSSVM and ELM models were developed.The correlation relationship was investigated between NIR spectra and fabric fiber content of textile products,and NIR quantitative analysis model was established.By comparison,LSSVM model was best with correlation coefficient of 0.99 and root mean square error of prediction(RMSEP)of 2.0% respectively.At the same time,there was no significant difference between NIR spectral quantitative analysis and the national standard method according to hypothesis testing of 95% confidence level.(5)Design of platform for quantitative analysis fabric fiber content by NIR spectroscopy.The influence of detection accuracy was explored for quantitative analysis of fiber content including the intensity of light source,spectrometer,light source and detector layout structure.A verification platform was designed including spectrum collection and quantitative analysis,calculation and control circuit,software.The bend of cotton and viscose was applied to test the performance of the verification platform.The correlation coefficient was 0.987,and RMSECV was 3.21%.From what had been discussed above,NIR spectra absorbance enhancement scheme was designed and verified,the fabric thickness and porosity to the influence of NIR spectrum was analyzed and corrected,qualitative discrimination and quantitative analysis models were established,the NIR quantitative analysis platform for fabric fiber content was designed,the cotton and viscose fiber blended fabric was used for test the performance of NIR spectral quantitative analysis platform.A reference was provided for the quantitative analysis of blended fabric by NIR spectroscopy.