Signal Separation Of Raman Spectra And Its Applications In Industrial Online Analysis

With the improvement of automation and intellectualization in factories,traditional off-line analytical methods in industrial production processes are gradually replaced by on-line analytical instruments.Online analytical instruments can monitor key parameters and attributes that influence quality of products during industrial production processes.The results can guide the adjustment and optimization of the production units,ensure the stable operation of the production units,enhance productivity and reduce energy consumption.Online Raman spectroscopy has been rapidly developed and widely used recently due to its fast and non-invasive measurement.Signal separation of Raman spectra is one of the difficulties in the application of Raman spectroscopy.The research of the thesis focuses on the signal separation technology of Raman spectra and studies the separation of background signal,instrumental broadening signal and independent peak signal in Raman spectra,respectively.The thesis not only proposes several novel sigjnal separation methods to solve the related problems but also realizes quantitative analysis of an acetone cyanohydrin product through the development of an online Raman analytical system.The system can provide accurate quality information of the product for process operations.This solves the problem of manual sampling in off-line analysis and thus improves automation level and production safety.The main works of this thesis are listed below:1.A novel algorithm for background signal separation of Raman spectra based on morphological operations is proposed The algorithm can separate Raman signal and background signal automatically and effectively.The algorithm has a robust performance under different parameter settings which reduces the requirement of prior knowledge.The algorithm can handle different types of background signal and achieves high accuracy and fast speed for background signal separation.2.A novel algorithm for peak signal separation of Raman spectra based on a convolution model is proposed.The algorithm can not only separate instrumental broadening signal from Raman signal after baseline removal,but also obtain each independent Raman peak signal.The algorithm can reduce peak overlap and enhance the quality of Raman spectral signal by separating the instrumental broadening signal.The algorithm can also extract the original Lorentzian peak signal.A mathematical description of Raman spectra is presented,so the parameters of each independent peak can be easily obtained The algorithm can help the subsequent qualitative or quantitative analysis.3.A novel method for the quantitative analysis of a key component in complex mixtures is proposed based on Raman signal separation.The method is applicable when not all of the spectra of pure components in the mixture are available while we need to quantitatively analyze a key component which is one of the unknown components.Experiments demonstrate that the method can achieve good prediction results even when the number of training samples is very small.4.An online Raman analytical system is developed for quantitative analysis of an acetone cyanohydrin product and has been applied in a production unit of a chemical plant.Online analysis of multiple components of the acetone cyanohydrin product is achieved based on signal separation of their Raman spectra.Long-time running of the system demonstrates that the system is stable and can offer fast and accurate analytical results.The system not only provides valuable quality information of the product which is critical to the adjustment of the production unit,but also greatly reduces the requirement of manual sampling and off-line analysis of toxic samples which improves the automation level and production safety of the plant.