| As a green alternative for fossil diesel, biodiesel is mainly used by blending with fossil diesel. However, due to the uneven quality of blend fuel, biodiesel has not been widely used in China. To promote the use of biodiesel, we urgently need to improve its quality detection technology. The mixing ratio of blend fuel has a great influence on its performance. Rapid measurement of the mixing ratio of blend fuel not only helps to adjust the fuel-injection of engine, optimize the use performance of the fuel, but also provides method for manufacturers to control the quality of blend fuel, optimize the closed-loop operations.The measurement technology of biodiesel and blend fuel can be divided into chromatographic method and spectral method. Compare to the chromatographic method, the spectral method is rapider and cheaper, which is especially suitable for online measurement. Although some spectral methods have been reported in the measurement of mixing ratio, they have not study in depth the effect of different spectra and different chemometric methods. This thesis combines near infrared (NIR) and Raman spectra with various chemometric methods to set up the corresponding model for the measurement of the mixing ratio of biodiesel.Detailed research contents include:1. Construct a dual spectral acquisition system based on NIR spectroscopy and Raman spectroscopy, obtain the NIR and Raman spectra of 62 blend fuel samples. Then delete the spectral outlier samples and determine the samples used for modeling.2. Propose a statistical data-driven (SDD) based wavelength selection method. This method does not need to judge whether the wavelength variable is important and only requires setting the statistical number and the maximum screening time. Results show that this method can obtain the characteristic variables objectively and effectively. Meanwhile, a new comprehensive error evaluation index is proposed to guide the screening of variables. Then, this method is verified by two NIR data sets. Both of the results show the effective of this method. Finally, the setting of two parameters is discussed.3. Analyze the NIR spectra of the blend fuel samples. Various chemometric methods (including pretreatment, wavelength selection and regression methods) have been tested, and the most suitable chemometric methods are found out finally. The NIR spectra are firstly pretreated by first derivative and detrending. Then, the SDD based wavelength selection method is used for the screening of wavelength variables. Only 30 characteristic wavelength variables (5.859% of the previous variables) are obtained. Finally, partial least squares (PLS) regression method is used for building the calibration model. The model has a root mean square error of calibration (RMSEC) =0.0195 and a root mean square error of prediction (RMSEP)=0.0170. The correlation coefficient (R)=0.9974. The relative standard error of prediction is 7.15%.4. Analyze the Raman spectra of the blend fuel samples. The Raman spectra are firstly pretreated by baseline correction. Then, the SDD based wavelength selection method is used for the screening of wavelength variables. Only 35 characteristic wavelength variables (2.186% of the previous variables) are obtained. Finally, partial least squares (PLS) regression method is used for building the calibration model. The model has a root mean square error of calibration (RMSEC)=0.0080 and a root mean square error of prediction (RMSEP)= 0.0083. The correlation coefficient (R)=0.9996. The relative standard error of prediction is 2.34%.In conclusion, the SDD based wavelength selection method is suitable for the variable selection in both NIR and Raman spectra. This method not only simplify model effectively, but also improve the precision and stability of model. For the measurement of mixing ratio, both NIR spectra and Raman spectra can achieve the goal. The Raman spectroscopy method has higher precision, which is especially suitable for future online detection in industry. |
PDF Full Text Request