On the model-based monitoring of industrial batch crystallizers

Crystallization is an important separation process to obtain high value-added chemicals in crystalline form from liquid solution in pharmaceutical, food and fine chemical industries. As most of the particulate processes, the quality of the solid product is determined by its particle size distribution (PSD). The achievement of the desired quality targets of the fine crystalline products relies on an efficient online process monitoring for separation supervision and control. However, hardware analyzers able to online measure the PSD and the solute concentration are rarely available, due to their costs \cite{Multi}. These unmeasured process variables can be estimated by state estimators that combine information from the process model and secondary measurements. The problem of designing state observers for online monitoring the PSD evolution has been mostly addressed under the assumption that some PSD measurements were available (see \cite{Mesb} and literature therein), which is not likely in practice. This work proposes a methodology to asses the feasibility of using common measurements (e.g. temperature and liquid fraction) for estimation purposes based on local observability \cite{Herm} and detectability \cite{AlFer} arguments. The results are supported using a data-derived technique, with data generated by a simulation model of the industrial crystallizer. Based on the results of the observability analysis, the structure of a state estimator is proposed.