MODELING BATCH COOLING CRYSTALLIZATION PROCESSES BY MEANS OF NEURAL NETWORKS

摘要

The use of neural networks as a modeling and analysis tool in chemical engineering has increased significantly in recent years, especially in applications involving complex processes, in which phenomenological models are difficult to apply. In the case of industrial crystallization, not only the evolution of the supersaturation, but also a number of processes must be considered, such as primary and secondary nucleation, crystal growth or agglomeration, which are difficult to describe by phenomenological models. In this work, experimental laboratory-scale data with aqueous solutions of copper sulfate (CuSO_4·5H2_O) and zync sulfate (ZnSO_4·7H_2O) were used in modeling batch-cooling crystallization by means of a feedforward neural network. Tests with the fitted neural network model showed good agreement with experimental results. The relative deviation between experimental and model-predicted values was smaller than 5% for the total mass of crystals and in average about 10% for the crystal size distribution. Simulations were then carried out with the model in order to verify the process sensitivity in terms of the variables considered.