Design-of-experiments based Modeling & Optimization of LGA Cooling Crystallization via Continuous Oscillatory Baffled Crystallizer

A novel data-driven modeling and optimization method is proposed in this paper for cooling crystallization of L-glutamic acid (LGA) via a continuous oscillatory baffled crystallizer (COBC), based on the design of experiments (DoEs) for the main operating conditions of zone temperature setting and volume net flowrate. The crystal size distribution (CSD) can be effectively predicted by constructing a data-mapping model with double-layer basis functions, where the first layer is composed of wavelet basis functions for reshaping the steady-state CSD in each operating zone of COBC, and the second layer consists of polynomial basis functions for reflecting the nonlinear relationship between the above operating conditions and the corresponding CSD in each zone. Furthermore, a comprehensive cost function related to the desired crystal size, the distribution variance of product crystals and throughput is introduced to design an optimization method for the above operating conditions. A guaranteed convergence particle swarm optimization (GCPSO) algorithm is offered to solve the nonconvex optimization problem based on the established CSD prediction model. Experimental results on the continuous crystallization of LGA demonstrate that the above cost function and the desired crystal product yield can be improved over 23% and 9%, respectively, in comparison with all tests by DoEs.