Research On Key Technologies Of On-line Raman System And Industrial Applications

Online analytical instruments are widely used in the industrial process. It provides timely and accurate analysis for optimizing operation, playing an irreplaceable role in saving energy. The capacity of online Raman spectroscopy to measure samples of a variety of forms with flexible sampling and without sample pretreatment makes it an important component of the Process Analytical Technology (PAT). In this thesis, several key technologies of online Raman technology are present; the development of online Raman system for the PX plants is described, which have been put into industrial applications. This thesis specifically including:1、Online Raman system using a back-illuminated CCD array spectrometer with interference signal, this thesis proposes a signal correction method based on the interference of non-specific fluorescent substance. The method was applied to monitoring an oil refinery catalytic reforming unit. In continuous operation, part of the characteristics of the spectrometer changed. The standard relative intensity correction method can not get the information and failed to achieve the correction of the interference signal, leading to the problem of spectrum distortion. The proposed interference signal correction method based on non-specific fluorescent substance can accurately track the changes, achieving an effective solution for on-line Raman system using back-illuminated CCD.2、A major problem in Raman spectroscopy is that the spectrum is often suffered from intrinsic fluorescence which is orders of magnitude greater than the Raman signal. Background subtraction is essential for further analysis, particularly for quantitative analysis using multivariate calibration. In this thesis, we propose a background removal algorithm which approximates the background by a polynomial and estimates the polynomial coefficients by iterative weighted least squares. The performance of the algorithm accompanied with two comparative methods are evaluated both on simulated and real Raman spectra. The results show that the proposed algorithm provides the best result using R2between the actual and extracted Raman peaks. It also improves the performance of background removal in quantitative Raman spectroscopy. Further more, the algorithm is least dependent on the choice of polynomial order.3、In the multivariate calibration, the model performance not only depends on the structure and parameters, but also influenced by the calibration sample distribution. In practical application, the calibration samples often distribute unevenly in space. Modeling based on the whole calibration set degrades the performance. To address the limitation, a new hybrid model based on support vector classification and regression is proposed in this thesis. The classification decision tree with the form of the binary tree is firstly built by least-squares support vector classifier; then the least-squares support vector regression is used to construct the regression model for each class. This model is applied to Raman spectral analysis of gasoline octane number. The result shows that the proposed model has greatly improved the accuracy of quantitative analysis.4、Based on Indirect hard modeling method, an extended peak resolution algorithm is proposed. A fast algorithm of the Voigt function is applied in the method, which improves the stability of the optimization results. By using numerical vector form for component spectrum without non-linear changes, it reduced the optimization complexity. Moreover, by alternating iterative optimization, the baseline estimation and peak resolution is separated, which improves the stability of parametric optimization with high order polynomial. The extended peak resolution algorithm is applied to Raman spectroscopy for quantitative analysis of benzene, toluene content in gasoline. The experimental results show that the proposed algorithm can effectively extract the spectral characteristics. Compared with the multivariate calibration methods, the proposed algorithm is of higher accuracy in quantitative analysis. 5、Development of on-line Raman system for the PX plants and putting into industrial applications. For the online analysis of the PX plants in petrochemical factory, the thesis described details of the on-line Raman analyzer including the principle, system compostion, model-based analysis and field application. The quantative analysis of the mixture of C8aromatics components is based on the extended peak resolution algorithm. The on-line Raman analyzer has been successfully applied to the composition analysis of the C8aromatics at the feed of adsorption tower in PX plants. According to the result of field operation, the repeatability error does not exceed±0.1%and the accuracy is±0.5%, which fully meets the requirements of the adsorption and separation processes. Moreover, the factory is gradually replacing the conventional gas chromatographic analysis. The Raman analyzer provided key information for the real-time process optimization.