(80b) Model Based Plant-Wide Optimization of an Industrial Scale Continuous Tank Crystallizer Network for Dextrose Monohydrate Crystallization | AIChE

(80b) Model Based Plant-Wide Optimization of an Industrial Scale Continuous Tank Crystallizer Network for Dextrose Monohydrate Crystallization


Szilagyi, B. - Presenter, Purdue University
Nagy, Z. K., Purdue University
Pal, K., Purdue University
Beheshti Tabar, I., Purdue University
This work is aimed to present a successful industrial-academic collaboration whose aim was to optimize a continuous large-scale dextrose crystallization network producing crystalline dextrose monohydrate. To put into the context, the throughput of the plant is in order of metric tons’/h solid crystals, whereas the combined nominal volume of the continuous tank crystallizers is in order of hundreds of m3. The main quality attribute of the product is the fine index, which represents the volume fraction of crystals under 75 µm. The maximally allowed fine index is 25 %, and the average fine index of the product significantly exceeded this value. Hence, the goal of the work was to carry out a model based fine index minimization for the current operation as well as the technically feasible combinations of the available crystallizers, also including technically realizable recirculation streams. A significant challenge was the time restriction (order of months) and limited plant scale measurement availability which, in addition to the objective, significantly impacted both the high and low level research decisions.

There are four key steps in the dextrose production technology: the fermentation, which is followed by the crystallization, centrifugation and drying. The fermentation of starch impacts significantly the crystallization process: the higher degree of polymerization (DP) oligomers of dextrose acts as growth rate inhibitor, which was confirmed by laboratory scale crystallization experiments. However, the concentration of higher DP components may vary as a function of the fermentation unit operation. Beyond the fine index criteria, the centrifugation is also impacted by the CSD, and, indirectly by the fermentation operation: the smaller crystals increase the cake resistance and also dissolve faster, hence, decreases the overall yield. Although we were able to make minor recommendation with respect to the feeding stream, such as feeding concentration in certain range, adjusting the higher DP concentration and the stream temperature was not possible.

Population balance models (PBMs) are widely applied to describe the crystal size distribution (CSD) variation with time during solution crystallization processes. The dextrose crystallization process in the mixed suspension mixed product removal (MSMPR) crystallizers was described by PBMs involving secondary nucleation and crystal growth, also accounting to growth rate inhibition. The PBMs are solved with the high resolution finite volume method (HR-FVM) in Matlab environment. The kinetic parameters from the literature were used as starting point, and the nucleation and growth rate constants were re-adjusted based on batch laboratory scale measurements [1]. The growth rate equation was extended with a simple empirical growth rate inhibition term, which was estimated using crystallization experiments with pure dextrose solution (no growth rate inhibition) and the plant syrup (known growth rate inhibitor concentration). The performance of the kinetic parameters was evaluated using an uncertain simulation of the current plant operation by a Monte-Carlo sampling of kinetic parameters from the uncertainty space defined by the hyper ellipsoid obtained from parameter estimation. The simulated fine index distribution, after an output error correction, matched the historical fine index data produced by the plant.

The calibrated process model was applied to minimize the fine index taking as decision variables the feeding concentration, temperatures in the crystallizers as well as the recirculation flowrates of each technically feasible configuration. All the optimizations were carried out using 125 randomly sampled kinetic parameters from the known uncertainty space, and the mean values of decision variables were considered as the final solution. The uncertain kinetics based optimization also allowed to compare the various operation scenarios in terms of expected fine index variability: the lower simulated variance of the fine index distribution obtained with the same kinetic parameter combination suggests that the given operation is more robust with respect to the uncertainties and disturbances.

Due to the lack of exact plant-scale model validation, the accuracy of optimizations for the exact fine index prediction might be fair to moderate, however, the results are directly comparable to each other, which allowed to provide a semi-quantitative expected performance of each operation mode but exact comparison of the feasible configurations. The optimization revealed that the existence of recirculation stream has significantly greater positive impact on the fine index minimization than the overall system volume. This was also validated with laboratory scale experiments by a cascade of three MSMPR crystallizers, which clearly showed that the recirculation increased significantly the mean size of crystals in the product stream. This observation enables to decommission the least efficient crystallization units (~40 % of the total volume), leading to significant operation and maintenance cost savings. The optimum feeding concentration, recirculation ratio and temperatures of (operationally) preferred configuration aligned well with the operation of other dextrose crystallization plants of the company which, however, operated at different scales involving different type of crystallizers. Ultimately, the research results gave sufficient support to the company to initiate a 10 year long internal development plan for operating point switch.


[1] A. Markande, A. Nezzal, J.J. Fitzpatrick, L. Aerts, Investigation of the crystallization kinetics of dextrose monohydrate using in situ particle size and supersaturation monitoring, Part. Sci. Technol. 27 (2009) 373–388. doi:10.1080/02726350902994050.