FT-NIR online detection performance improvement and evaluation for 2,6-dimethylphenol purity

Abstract

In this paper, Fourier transform near-infrared (FT-NIR) spectroscopy was used to online detect the 2,6-dimethylphenol(2,6-DMP) purity of different rectifying columns in the distillation separation process. Considering the similarity of spectral data between different columns, transfer learning could be introduced to improve the online detection performance of 2,6-DMP purity by reusing the spectral data from other columns. Furthermore, in order to analyze the effect of the quantity and quality of the FT-NIR spectra on the detection accuracy, the influence function was used to construct the quantitative relationship between the detection accuracy and the spectral information to be transferred. The effectiveness of the approach was analyzed on the FT-NIR spectral datasets from the distillation purification process in the 2,6-DMP monomer separation section. By transferring more favorable spectral data to avoid negative transfer, the detection accuracy will be obviously increased.