| Moisture is a common impurity in organic solvents,and water content is an important monitoring parameter for achieving quality control of organic solvents in the chemical industry.Its index can directly affect the production efficiency of enterprises.In view of the operational and time-consuming shortcomings of the existing standard test method for the water content of organic solvents(the Karl Fischer method),there is an urgent need to study a new method for the determination of the water content of organic solvents.Terahertz time-domain spectroscopy is a fast,non-destructive,safe,and highly accurate detection technique.The hydrogen bond relaxation vibrations in water molecules make terahertz waves strongly absorbed by them,providing an excellent basis for the application of terahertz technology in the field of moisture content detection.The application of terahertz time-domain spectroscopy to the detection of water content in organic solvents is of great importance.In this paper,a study on the quantification of the water content of typical organic solvents(ethylene glycol,ethanol)is carried out based on terahertz time-domain spectroscopy,and the main elements of the study are as follows:(1)A study of terahertz time-domain spectroscopy and its detection methods was carried out to provide an experimental basis for quantitative analysis.The principles of different terahertz time-domain spectroscopy systems were analyzed in relation to the needs of water content detection in organic solvents,and a transmission terahertz time-domain spectroscopy system was selected.A liquid sample cell was designed according to the terahertz spectroscopic properties of organic solvents,and the effects of spectral system repeatability errors,sample position relationships,sampling operations,and reflection peaks on the spectral test results were investigated.(2)The terahertz spectra of two samples of aqueous organic solvents were obtained,and the variation in water content of organic solvents was analyzed.Based on Lambert’s theorem and effective medium theory,the intrinsic relationship between optical parameters and water content of aqueous organic solvents was investigated,and two models of the water content response of organic solvents based on a single optical parameter(absorption coefficient or refractive index)were constructed.In addition,based on the above two models,a water content response model combining absorption coefficient and refractive index is derived by transformation.The results show that both the absorption coefficient-based water content response model for aqueous glycol samples and the refractive index-based water content response model for aqueous ethanol samples have high fitting and prediction accuracy,with RMSEPs of 0.0086 and 0.0084,respectively;this demonstrates the feasibility of using terahertz technology for water content detection of organic solvents.The feasibility of the terahertz technique for the detection of water content in organic solvents was demonstrated.However,the stability of the spectral data at a single frequency point is easily affected by factors such as sample accuracy and reflection peaks,leading to a decrease in the prediction accuracy of water content at low frequency points.(3)In order to reduce the influence of low spectral signal-to-noise ratio on the stability and accuracy of water content prediction,the water content of organic solvent was predicted based on PCR,PLSR,and SVR regression analysis methods combined with optical parameters of the frequency band(0.5–1.5 THz),and the influence of spectral data preprocessing methods on the prediction performance of the regression model was studied.The results showed that the three regression prediction models could effectively predict the water content of the two organic solvents,and the best SVR model was obtained by comparing the fitting degree and prediction accuracy.The SVR model of glycol absorption coefficient after SNV treatment had the best prediction performance,and its R~2 and RMSEP were 0.994 and 0.0045,respectively.The prediction performance of the SVR model of ethanol refractive index was the best under the combination of S-G smoothing and z-score standardization,and its R~2 and RMSEP were0.992 and 0.0060,respectively.Compared with the theoretical response model of water content,the SVR regression prediction model not only improves the stability of water content prediction results but also improves the prediction accuracy to a certain extent.Combining the results of theoretical response models and regression models for the quantification of water content in organic solvents,the application of terahertz spectroscopy to the detection of water content in organic solvents has been demonstrated to be feasible,and this study can provide a theoretical basis and new detection ideas for the detection of water content in organic solvents and polar solutions. |
