In-Process Particle Characterization by Spectral Extinction

Abstract

Particle formation processes are important in many industries, for example, the manufacture of pigments, pharmaceuticals, cosmetics and foods. Most would benefit from improved control achievable with on-line measurement in real time of particle size, shape and composition. In spectral extinction, an electromagnetic beam is attenuated as it passes through a particulate suspension. The degree of extinction depends on the wavelength of the radiation and on the particles’ size, shape and refractive index. At low concentrations, extinction is directly proportional to concentration and generally it continues to increase monotonically with concentration at concentrations up to ∼25% by volume. Spectral extinction has been shown to be a feasible method of obtaining the particle size distribution for spheres of known refractive index over a range of particle sizes. However, the calculations required to invert the extinction spectrum into a particle size distribution are time-consuming and are not always successful. A neural network has been trained to identify mean and standard deviation of log normal particle size distributions from extinction measurements over a range of wavelengths. Tests have shown the method to be effective over particle mean sizes from 90 nm to 1000 nm using wavelengths from 350 nm to 900 nm. Narrow particle size distributions can be measured over a wider range of sizes. It has been established that surprisingly few extinction measurement inputs are sufficient to obtain reasonably accurate determinations of log normal distribution parameters. Improved accuracy was achieved by including the wavelength for maximum extinction as an input to the neural network.