Improved spectral model building for using ATR-FTIR spectroscopy to measure solution concentration during cooling crystallization

Abstract

To ensure in-situ measurement accuracy on the solution concentration during cooling crystallization process, an improved spectral model building method is proposed in this paper for application of the attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy. The traditional partial least-squares (PLS) model is modified into a functional regression form for expressing the relationship between the input variables of in-situ measured spectra and the output of solution concentration, so as to address the issue of insufficient samples for model calibration with respect to the high-dimension of spectral variables for measurement. The widely used DB4 wavelet functions are taken as basis functions to approximate the smooth functions in the proposed functional regression model, by virtue of their multi-scale and orthogonal properties to procure good accuracy. Accordingly, a parameter estimation algorithm named wavelet partial-least-squares is proposed for the spectral model calibration. The application to measure the solution concentration of L-glutamic acid (LGA) cooling crystallization process well demonstrates the effectiveness and merit of the proposed method.