Research On Quality Control Of Cement Raw Meal Based On Near Infrared Spectroscopy Detection Technology

As the first step in cement production,the quality of raw meal preparation has an important impact on the subsequent clinker firing.As the main index for evaluating raw material quality,fast and accurate detection of raw material composition is a prerequisite for realizing real-time quality control.At present,most domestic cement enterprises still use XRF fluorescence analyzer for raw material quality testing off-line,which cannot guide the production in a timely and effective manner in case of frequent changes in raw material composition.A few enterprises have introduced online elemental analyzers for real-time online testing and real-time quality control,but this type of testing uses a radioactive neutron source,which is not only a safety hazard,but also has high maintenance costs.To this end,near-infrared spectroscopy was introduced for online detection of cement raw material components,which in turn enables real-time quality control of cement raw materials.This detection technology has been widely used in quantitative testing as a fast,safe and pollution-free detection method.However,the following problems exist in near infrared spectroscopy detection technology for cement raw material composition testing:most of the cement raw material composition content is metal oxides,whose near infrared spectral absorption peaks are small and not easy to be quantitatively detected;changes in the density of cement raw material pile affect the intensity of diffuse reflected light,thus reducing the accuracy of near infrared spectroscopy detection;changes in raw material type or origin make the fitness of the detection model poor;there are changes in the composition of raw materials during the production process,and there is a certain lag time in the post-grinding detection of near infrared spectroscopy.In response to the above problems,this paper investigates the detection and quality control of cement raw materials composition based on near-infrared spectroscopic detection technology,with the following main contents and results.(1)In the quantitative detection of four components of cement raw materials,SiO₂,Al₂O₃,Fe₂O₃ and Ca O,the absorption peaks of the same component are not uniquely located due to the small and wide distribution of each component.The backward interval partial least squares(Bi PLS)and synergy interval partial least squares(Si PLS)methods were used for band selection of the raw and pre-processed spectra,and by comparing the algorithms of principal component regression(PCR),partial least squares(PLS),support vector machine(SVM),Multiple Linear Regression(MLR),Classical Least Squares(CLS),artificial neural networks(ANN)modeling effects,the optimal detection models for different components of cement raw materials were determined.For SiO₂ component,the original spectrum was used for Bi PLS band selection to determine the optimal PLS detection model with a predicted root mean square error of 0.136;for Al₂O₃ component,the Savitzky-Golay preprocessed spectrum was used for Bi PLS-Si PLS band selection to determine the optimal PLS detection model with a predicted root mean square error of 0.068;for Fe₂O₃ component the optimal PLS detection model was determined using Savitzky-Golay pretreatment spectra,selected by Bi PLS-Si PLS bands,with a predicted root mean square error of 0.031;for Ca O the optimal PCR detection model was determined using Savitzky-Golay pretreatment spectra,selected by Bi PLS bands,with a predicted root mean square error of 0.113.(2)For the influence of cement raw meal pile density on the detection of cement raw material near infrared spectra,the pile density was compensated by mixed sample correction.The correlation coefficients R² were improved by 19.10%,17.65%,20.37%,and 27.40%,respectively,and the root mean square error of prediction was reduced by 7.36%,4.14%,12.90%,and 20.67%,respectively,when compared with the detection models established for the samples without heap density information.and 20.67%,respectively.The comparative study of different cement companies shows that when the raw material of cement raw material changes or the origin is different,the cement raw material cannot be detected by the modified model alone,and the near infrared spectral model needs to be re-established,and the spectral band selection must be changed.(3)To address the hysteresis and nonlinearity problems in cement raw material quality control,based on the mechanism and characteristics of cement raw material batching process,combined with the results of near-infrared spectroscopy,the working conditions of cement raw material production process are classified and a working condition template is established.For the case that the quality of cement raw material is temporarily qualified,but its trend tends to be worse,trend control is adopted;for the case that the quality of cement raw material is not qualified,predictive control is adopted;a data-driven raw material composition estimation method is used to estimate the raw material composition of cement raw material.(4)Based on the above research content,cement raw material quality control software was developed and applied to the actual production in the field.The field application results showed that the average absolute deviation values of the detection values of SiO₂,Al₂O₃,Fe₂O₃ and Ca O composition content of cement raw material and XRF fluorescence detection values by near infrared spectroscopy detection technology were 0.11,0.033,0.045 and 0.157,respectively.Based on the near infrared spectral detection data and field data,the cement raw material quality control software was used for control,which led to a 17.25%increase in the qualified KH rate of cement raw material.