基于近红外光谱检测技术的水泥生料成分含量检测研究

Research on Detection of Cement Raw Material Content Based on Near-Infrared Spectroscopy

近红外光谱检测技术已经成功应用于水泥生料成分的快速检测,但我国水泥企业在生产水泥生料时所用原材料品种不一,使用不同的原材料进行生产时对近红外光谱建模带来一定影响。为了研究不同原料生产的水泥生料近红外光谱建模差异,对不同地区水泥生产线所生产的水泥生料进行建模研究。选取两个不同地区水泥生产线的水泥生料样本各95份和82份,各自选取80份和67份作为校正集,15份作为验证集。首先将两条水泥生产线的样本每份重复装样测3次光谱,取平均光谱做为样本的近红外光谱。然后通过采用S-G平滑法对两个不同地区所生产的水泥生料近红外光谱进行预处理。对比发现两个地区水泥生料近红外光谱存在一定差异,采用偏最小二乘回归算法建立检测模型,所建立的模型精度差异较大。采用CARS波段挑选法,分别对两种水泥生料近红外光谱进行挑选,生产线一的水泥生料样本SiO_(2),Al_(2)O_(3),Fe_(2)O_(3)和CaO近红外光谱波段由3113个变量分别保留了85,89,55和67个变量,生产线二的水泥生料近红外光谱则分别保留了51,55,55和55个变量,且保留的波段明显存在一定区别。最后分别建立了两个地区的水泥生料SiO_(2),Al_(2)O_(3),Fe_(2)O_(3)和CaO近红外光谱检测模型。通过对比发现原材料不同时所挑选的波段不同,且检测模型预测效果良好。生产线一的SiO_(2),Al_(2)O_(3),Fe_(2)O_(3)和CaO检测模型的RMSEP(预测均方根误差)分别为0.109,0.053,0.034和0.185,生产线二的SiO_(2),Al_(2)O_(3),Fe_(2)O_(3)和CaO检测模型的RMSEP分别为0.084,0.024,0.023和0.184。结果表明当水泥生料的原材料发生变化或者产地不一时,不能仅靠修正模型对水泥生料进行检测,而是需要重新进行近红外光谱建模,且光谱波段选择也会发生变化。采用波段挑选法对水泥生料近红外光谱进行波段挑选能够提高检测模型的模型精度。

Near-infrared spectroscopy has been successfully applied to the rapid detection of cement raw meal composition,but Chinese cement companies use different raw materials when producing cement raw meal.The use of different raw materials for production has a certain impact on near-infrared spectroscopy modeling.In order to study the difference of near-infrared spectral modeling of cement raw meal produced in different regions,this paper studies the modeling of cement raw meal produced by cement production lines in different regions.95 and 82 samples of cement raw materials from cement production lines in two different regions were selected respectively,80 and 67 samples were selected as calibration sets,and 15 samples were selected as verification sets.Firstly,the samples from the two cement production lines are repeatedly loaded and tested three times,and the average spectrum is taken as the near-infrared spectrum of the samples.Then the near-infrared spectra of cement raw materials produced in two different regions are then pretreated by the S-G smoothing method.The partial least squares regression algorithm is used to establish the detection model,and the comparison shows that there are certain differences in the near-infrared spectra of cement raw meal in the two regions,and the accuracy of the model established by the same method is quite different.Using the CARS band selection method,the near-infrared spectra of two kinds of cement raw materials were selected.The near-infrared spectra bands of SiO_(2),Al_(2)O_(3),Fe_(2)O_(3) and CaO samples from production line 1 retained 85,89,55 and 67 variables respectively from 3113 variables.The near infrared spectral bands of SiO_(2),Al_(2)O_(3),Fe_(2)O_(3) and CaO in the cement raw meal samples of production line 2 are 51,55,55 and 55 variables respectively retained from 3113 variables,and the retained bands are different.Finally,the near-infrared spectrum detection models of SiO_(2),Al_(2)O_(3),Fe_(2)O_(3) and CaO in cement raw materials in the two regions were established respectively.Through comparison,it is found that the selected wave bands are different when the raw materials are different,and the prediction effect of the detection model is good.The RMSEP(predicted root mean square error)of production line I SiO_(2),Al_(2)O_(3),Fe_(2)O_(3) and CaO detection models are 0.109,0.053,0.034 and 0.185 respectively,while the RMSEP(predicted root mean square error)of production lineⅡSiO_(2),Al_(2)O_(3),Fe_(2)O_(3)and CaO detection models are 0.084,0.024,0.023 and 0.184 respectively.The results show that when the raw materials of cement raw materials change or the place of production is different,the cement raw materials cannot be detected only by the modified model,but the near-infrared spectral modeling needs to be carried out again,and the spectral band selection will also change.Using the band selection method to select the band of near-infrared spectrum of cement raw meal can improve the model accuracy of the detection model.