Research On Optimization Of Quantitative Calibration Model For Fast Determination Method Of Cement Raw Meal Composition Based On Near Infrared Spectroscopy

Cement is an important building material,which plays an irreplaceable role in the national economic construction.In recent years,the annual yield of cement in China has been increasing.A series of documents have been issued in China,which put forward the requirements for the development of cement industry:improving the quality of cement,saving energy and improving the efficiency of cement production.Raw meal batching process is a critical aspect in cement production,and the quality of cement is largely determined by the quality of raw meal.Calcium oxide,silicon dioxide,aluminum oxide and iron oxide are the effective components of cement raw meal,in the cement industry,the proportion of raw materials are adjusted by determinating the content of these oxides to ensure the stability of raw meal quality.How to quantitatively analyze the effective ingredients of cement raw meal quickly,accurately and stably is an urgent problem in cement industry.In this paper,a method based on near infrared spectroscopy is proposed to determine the content of calcium oxide,silicon dioxide,aluminum oxide and iron oxide in cement raw meal,with the advantages of rapid analysis,simultaneous determination of multicomponent,no sample pretreatment,no sample consumption,low cost,nondestructive and environmental protection.The accuracy of the proposed method mainly depends on the performance of quantitative calibration model,therefore,in order to obtain more accurate results to meet the requirements of raw meal proportioning control process in cement industry,the modeling process is improved and the quantitative calibration model is optimized from four aspects:sample set partitioning,outlier elimination,spectral preprocessing and characteristic wavenumber variable selection,and the corresponding tests are carried out.The main research contents are as follows:（1）The quantitative calibration model is optimized based on sample set partitioning.A method,named sample set partition based on joint X-Y distance（SPXY）,is used to divide cement raw meal sample set,which provides a reliable method for selecting representative calibration set samples.In order to evaluate the effect of this method,random sampling,Kennard-Stone partition method and Duplex method are used for comparison.Based on the calibration set selected by the above methods,partial least squares models for the four components are established respectively.The model parameters based on SPXY are much better than other methods,demonstrating that this method can improve the determining accuracy of oxide content in cement raw meal.（2）The quantitative calibration model is optimized based on outlier elimination.An algorithm,called cross-validation-absolute-deviation-F-test（CVADF）,is proposed to eliminate the outliers existing in the cement raw meal calibration set,and the partial least squares models of the four oxides are established based on the remaining samples.In order to evaluate whether the proposed method can accurately identify the outliers,Classical techniques,leverage diagnostic,Euclidean distance diagnostic,Mahalanobis distance diagnostic and principal component score diagnostic are used for comparison.The model parameters obtained by the proposed method are optimal,showing the superiority of this method,and the prediction ability of cement raw meal quantitative calibration models can be improved by using this method.（3）The quantitative calibration model is optimized based on spectral preprocessing and characteristic wavenumber variable selection.Different spectral pretreatment methods,multiple scattering correction,standard normal variate transformation,first derivative,Savitzky-Golay smoothing,are utilized to process the near infrared spectra of cement raw meal samples,then the partial least squares models of the four oxides are developed,and the optimal preprocessing method for each component is determined.After pretreatment,the prediction ability of the model is enhanced and the average prediction error is reduced.An algorithm,named backward interval partial least squares-genetic algorithm,is proposed to select the characteristic wavenumber variables of cement raw meal near infrared spectroscopy,which provides a reliable method for eliminating irrelevant and redundant variables in the spectra.The wavenumber intervals which are closely related to the oxides content are obtained by backward interval partial least squares algorithm,then the variables in the intervals are further selected by genetic algorithm,finally,the characteristic wavenumber variables are acquired.The partial least squares models of the four oxides are established respectively,the dimension of spectral matrix is reduced,the model are simplified,the performance of the models is enhanced,and the accuracy of the results is improved.（4）According to the improved modeling process,the sample set is divided by sample set partitioning based on joint X-Y distances method,the outliers in the calibration set are eliminated by cross validation absolute deviation F-test method,the spectra are preprocessed,and the feature variables closely related to the component content are selected according to the backward interval partial least square-genetic algorithm,and the partial least square models are established,202 raw meal samples collected from the production line of Shandong Qufu Zhonglian cement factory are modeled and predicted.The improved application effect was verified.Then 119 raw meal samples with different raw materials and origins produced by Shandong Linyi Zhonglian cement factory are analyzed,and the universality of the modeling process was tested.In this paper,the modeling process of near infrared spectroscopy analysis of cement raw meal is improved,and the quantitative calibration model is optimized from four aspects:sample set partitioning,outlier elimination,spectral preprocessing and characteristic wavenumber variable selection.The prediction ability of the model is significantly enhanced,and the accuracy of oxide content determination is improved.The content of calcium oxide,silicon dioxide,aluminum oxide and iron oxide in cement raw meal is determined quickly and accurately.If the methods in this paper is applied in actual production,the raw meal batching process will be guided in time,and the requirements of raw meal proportioning control on precise accuracy will be satisfied.There are important research significance and application value to ensure high quality,low cost and green operation of cement production line.