基于偏最小二乘法的三种光谱定量测定LDPE/PP共混物组分含量的比较研究

分别利用中红外(mid-infrared, MIR)、近红外(near-infrared, NIR)和拉曼光谱(Raman)采集了31种不同比例的低密度聚乙烯/聚丙烯(LDPE/PP)共混物样本的光谱，利用偏最小二乘法(partial least-square, PLS)建立了光谱数据与LDPE含量的模型，研究了不同光谱范围和预处理方法对模型准确性的影响，并比较了三种光谱定量测量的准确性。结果表明，对于光谱差异小、存在噪音或基线干扰的谱图，预处理方法和光谱范围对模型的准确性均有较大的影响;经过三种预处理以及选择合适的光谱范围建立的模型决定系数(R2)分别从未处理前的0.887 6，0.849 3和0.875 7提升到0.990 6，0.997 3和0.997 2，校正均方根误差(root mean square error of calibration, RMSEC)则分别从10.15，11.75和10.67降低到2.941，1.561和1.598;三种光谱在经过预处理之后均能够较好地定量测量LDPE的含量，NIR和Raman模型准确性更高，由于两者的测量速度快，因此尤其适合于进行快速、准确的定量测量。

Comparison of Three Spectroscopies for the Determination of Composition of LDPE/PP Blend with Partial Least-Squares

This paper presented an application of mid-infrared (MIR), near-infrared (NIR) and Raman spectroscopies for collecting the spectra of 31 kinds of low density polyethylene/polyprolene (LDPE/PP) samples with different proportions. The different pre-processing methods (multiplicative scatter correction, mean centering and Savitzky-Golay first derivative) and spectral region were explored to develop partial least-squares (PLS) model for LDPE, their influence on the accuracy of PLS model also being discussed. Three spectroscopies were compared about the accuracy of quantitative measurement. Consequently, the pre-processing methods and spectral region have a great impact on the accuracy of PLS model, especially the spectra with subtle difference, random noise and baseline variation. After being pre-processed and spectral region selected, the calibration model of MIR, NIR and Raman exhibited R2/RMSEC values of 0.990 6/2.941, 0.997 3/1.561 and 0.997 2/1.598 respectively, which corrsponding to 0.887 6/10.15, 0.849 3/11.75 and 0.875 7/10.67 before any treatment. The results also suggested MIR, NIR and Raman are three strong tools to predict the content of LDPE in LDPE/PP blend. However, NIR and Raman showed higher accuracy after being pre-processed and more suitability to fast quantitative characterization due to their high measuring speed.