Application of acoustic spectroscopy and neural networks to on-line size measurement of pharmaceutical nanocrystals

Abstract

Acoustic spectroscopy and neural networks (NNs) are applied to on-line real-time measurement of particle size distribution (PSD) during wet milling of pharmaceutical nanocrystals. A method for modeling the relationship between acoustic attenuation spectra and PSD is proposed that is based on NNs and principal component analysis (PCA). PCA reduces the dimensions of both the spectra and the PSD; then, a neural network model of 2 × 2 × 2 (input, hidden, output layer nodes) with only eight connection weights is built. Compared with previous instrument models that could require as many as 14 physical properties, the current approach does not need any prior knowledge of the system's properties. In addition, the time taken to complete a PSD measurement is reduced from minutes to seconds and it always generates a single solution, rather than possible multiple PSD solutions as in early methods. Application to hydrotalcite nanomilling found good agreement between the on-line measurements and off-line analysis.