| Summary: | Anti-solvent crystallization is frequently applied in pharmaceutical processes for the separation and purification of intermediate compounds and active ingredients. The selection of optimal solvent types is important to improve the economic performance and sustainability of the process, but is challenged by the discrete nature and large number of possible solvent combinations and the inherent relations between solvent selection and optimal process design. A computational framework is presented for the simultaneous solvent selection and optimization for a continuous process involving crystallization and distillation for recycling of the anti-solvent. The method is based on the perturbed-chain statistical associated fluid theory (PC-SAFT) equation of state to predict relevant thermodynamic properties of mixtures within the process. Alternative process configurations were represented by a superstructure. Due to the high nonlinearity of the thermodynamic models and rigorous models for distillation, the resulting mixed-integer nonlinear programming (MINLP) problem is difficult to solve by state-of-the-art solvers. Therefore, a continuous mapping method was adopted to relax the integer variables related to solvent selection, which makes the scale of the problem formulation independent of the number of solvents under consideration. Furthermore, a genetic algorithm was used to optimize the integer variables related to the superstructure. The hybrid stochastic and deterministic optimization framework converts the original MINLP problem into a nonlinear programming (NLP) problem, which is computationally more tractable. The successful application of the proposed method was demonstrated by a case study on the continuous anti-solvent crystallization of paracetamol.
|
|---|
