ABSTRACT
The reliability of chemical processes can be greatly improved by implementing inline monitoring systems. Combining multivariate analysis with non-destructive sensors can enhance the process without interfering with the operation. We present here hierarchical models using both principal component analysis and partial least square analysis developed for different chemical components representative of solvent extraction process streams. A training set of 380 samples and an external validation set of 95 samples were prepared and Near infrared and Raman spectral data as well as conductivity under variable temperature conditions were collected. The results from the models indicate that careful selection of the spectral range is important. By compressing the data through Principal Component Analysis (PCA), we lower the rank of the data set to its most dominant features while maintaining the key principal components to be used in the regression analysis. Within the studied data set, concentration of five chemical components were modeled; total nitrate (NO3-), total acid (H+), neodymium (Nd3+), sodium (Na+), and ionic strength (I.S.). The best overall model prediction for each of the species studied used a combined data set comprised of complementary techniques including NIR, Raman, and conductivity. Our study shows that chemometric models are powerful but requires significant amount of carefully analyzed data to capture variations in the chemistry.