Image-based multi-resolution-ANN approach for on-line particle size characterization

In this work an image-based multi-resolution sensor for online prediction of crystal size distribution (CSD) is proposed. The mean and standard deviation of lognormal probability density function as the CSD can be predicted through the on-line sensor. In the proposed approach, texture analysis (fractal dimension (FD) and energy signatures) as characteristic parameters to follow the crystal growth is utilized. The methodology consists of a combination of thresholding and wavelet-texture algorithms. The thresholding method is used to identify crystal clusters and substrate empty backgrounds. Wavelet-fractal and energy signatures are performed afterwards to estimate texture on crystal clusters. Following the texture information extraction, a nonlinear mapping consisting of an artificial neural network (ANN) is incorporated using as inputs the texture information in conjunction with the available on-line process conditions (flowrate and temperature). A software framework developed in MATLAB enables the configuration of the image acquisition parameters as well as the processing of the on-line images. Validations against experimental data are presented for the NaCl-water-ethanol anti-solvent crystallization system