Research On Content Measurement Of Textile Mixture By Fourier Transform Near Infrared Spectroscopy

Near infrared spectroscopy is widely applied in many fields because of the characteristics of fast analysis, good effect, low price and no pollution. The measurements on contents of textile mixture by Fourier transform near infrared spectroscopy and chemometrics were thoroughly investigated and lots of research work has been done in aspects of cotton/terylene and cotton/wool samples preparation, collection and preprocessing of near infrared spectrum data, models building and upgrading. Finally, satisfied results were achieved.(1) Basic theory, detection methods, analysis procedures of near infrared spectroscopy were introduced. Data preprocessing and quantitative analysis method were investigated.(2) The experiment scheme of measurements of contents of textile mixture by Fourier transform near infrared spectroscopy was designed. The whole process of sample preparation, data collection and preprocessing, models building and upgrading was initially determined.(3) Based on experiment scheme, fifty-one samples of cotton/terylene and fifty-one cotton/wool were obtained. Two kinds of samples were separated into three sets for calibration, validation and prediction respectively. Near infrared spectrum data of samples were gathered experimentally which provided the original data for preprocessing and models building.(4) On the basis of data preprocessing technology, the original data were arranged according to their characteristics and analysed spectrum range was chosen. Wavelet transform was adopted for spectral data de-noising and compression processing. Principal component analysis was used for spectral data processing and dimensionality reduction.The data quantity was reduced and data quality was optimized by data preprocessing,.(5) Linear and nonlinear models which included multivariable linear regression model(MLR), Principal component regression model(PCR), BP neural network models (PCA-BP, WT-ca3-BP, WT-ca4-BP and WT-ca5-BP) were developed. Former two models are linear, the rest are non-linear. Three evaluation indicators Absolute Error, Mean Absolute Error and Root Mean Square Error (AE, MAE and RMSE) were made which provided theoretical support for analyzing and appraising the performance of calibration models.(6)Compared the linear models with the nonlinear ones, it could be derived that the linear models'principles were simple and helpful for understanding near infrared spectra analysis technology, while the prediction accuracy and flexibility of nonlinear models are better than linear models. Especially, the prediction accuracy of WT-ca3-BP (41-17-2) network model is the highest in all models, MAE is within 1.8% and RMSE is within 2.4% for cotton /terylene samples, and MAE is within 2% and RMSE is within 2.5% for cotton/wool samples, which is most suitable for the contents prediction of unknown samples. On the basis of not changing the structure of the WT-ca3-BP network model, calibration and validation samples were combined fully to be re-set to the new calibration samples, the model was upgraded. The prediction accuracy and predicted effect of upgraded model were best and satisfied. At last, it was applied to predict the contents of five unknown samples.(7) The source of error was analysed and near infrared detection system of components for textile mixture material was established.