Adaptive MIMO neuro-fuzzy logic control of a seeded and an unseeded anti-solvent semi-batch crystallizer

Abstract

This study explores the implementation of a two input/two output adaptive neuro-fuzzy logic controller on an anti-solvent semi-batch crystallization process. The solution concentration and the solubility curve of paracetamol (PA) in a mixture of water and isopropanol in the range of temperatures between 10 and 40 ∘ C were determined using attenuated total reflection (ATR)-Fourier transform infrared (FTIR) spectroscopy. The in situ chord length distribution of crystals was obtained from laser backscattering by focus beam reflectance measurement (FBRM) probe. The controlled variables were the supersaturation and the difference in the chord length counts between two sampling times, and the manipulated variables were the cooling rate and anti-solvent flow rate. The ‘direct’ objectives of this study were to keep the controlled variables inside their predetermined ranges. The ‘indirect’ objectives were to improve the end-of-batch properties that included batch time, yield, and particle size distribution. Performance of the adaptive neuro-fuzzy logic controller for the closed-loop system was evaluated based on meeting the ‘direct’ and ‘indirect’ objectives. The best results in terms of batch time and product yield for unseeded experiments were 280 min and 95%, respectively. However, the most significant improvement was noted in the seeded set of experiments that resulted in 225 min batch time, an increase of the volume weighted mean size by 275.1 μ m , and 99% product yield.