LAUSR

Abstract For in-situ measurement of the granule moisture content (MC) during a fluidized bed drying (FBD) process by using the near-infrared (NIR) spectroscopy, a novel spectral calibration model building method is proposed in this paper for improving the real-time measurement accuracy. To tackle the serious high-dimensional parameter estimation problem arising from a large number of NIR spectral variables in the wavelength range for measurement, a small number of wavelet functions are constructed to closely approximate the measured NIR spectral curve for each sample, such that a functional regression model is established based on these wavelet functions, which facilitates reducing the model parameters for output prediction while the spectral nonlinearity can be conveniently addressed. An active learning strategy is given to choose these wavelet basis functions with respect to a user specified fitting accuracy. Owing to the orthogonal property of wavelet basis functions, the established calibration model can be efficiently used for real-time measurement. Numerical studies and experimental results on in-situ measurement of the silica gel granule moisture under different FBD operating conditions well demonstrate the effectiveness of the proposed method.