Crystallization process simulation in the food industry

Batch evaporative crystallization of Lactitol

Fig. 1 – Lactitol crystals

Lactitol crystallization process flowsheet – click to enlarge

Fig. 2 – The process flowsheet

$Brix refractometer$

$Trend from Brix refractometer – click to enlarge$

Fig. 3a – Brix refractometer (above) and trend of concentration measurement

$Trend from Brix refractometer – click to enlarge$

Fig. 3b – OPUS on-line particle analyser (above) and trend of particle size measurements

Supersaturation setpoints for original and optimized batch recipes – click to enlarge

Fig. 4 – Supersaturation setpoints for original and optimised batch recipes

Temperature setpoints for original and optimized batch recipes – click to enlarge

Fig. 5 – Temperature setpoints for original and optimised batch recipes

This case involved PSE working in conjunction with IPCOS and Delft University of Technology to optimise the batch recipe and implement online control on a batch evaporative crystallisation of artificial sweetener Lactitol.

The client was Purac Biochem B.V., a subsidiary of CSM, and a leader in food ingredient manufacture. Lactitol monohydrate is used as a bulk sweetener in calorie-controlled foods.

The process

The main process is a is a forced circulation crystalliser, consisting of a stirred crystallisation vessel, pumps and heat exchangers as shown in Fig. 2.

The batch recipe involves four steps: concentration, seeding, cooling and evaporation.

Pressure, temperature are measured throughout. In addition:

A Brix refractometer supplied by K-PATENTS Process Instruments is used for in-line measurement of concentration (strictly speaking, total dried solids).
A Sympatec GmbH OPUS on-line particle-size analyser with UltraSonic extinction is used for continuous measurement of the PSD.

The stirred-tank batch cooling process was seeded using crystals of 60 µm median size produced by milling and sieving. The crystals formed were needle-shaped, with length of up to 350µm (Fig. 1).

Requirements

The project objectives were to:

improve product filterability by increasing the median crystal size and reducing the amount of fines
determine the optimal batch recipe to maximize the median crystal size within the same or shorter batch time of the existing recipe
reduce batch-to-batch variability by implementing online control.

The model

The basic crystallisation model was taken from PSE's Advanced Model Library for Solution Crystallisation (AML:SC).

This modelled the crystal population balance, taking into account the following phenomena:

crystal growth (in the main vessel) and dissolution (in the external circulation loop through the heat exchanger)
secondary nucleation, occurring as a result of attrition caused by crystal–impeller collisions.

The model was validated against both laboratory and plant data to ensure that it accurately represented these phenomena.

Once the initial model had been validated, a base case simulation was performed. This gave sufficiently good agreement with experimental results to proceed with the optimisation.

Recipe optimisation

Following initial simulation runs for the batch, the problem was formulated as a dynamic optimisation with the following characteristics:

Objective function	Maximise crystal median size
Constraint	Batch duration less than or equal to original batch duration
Optimisation variable 1	Supersaturation setpoint trajectory
Optimisation variable 2	Temperature setpoint trajectory

Optimisation results

The linear-piecewise calculated trajectories for the supersaturation and temperature setpoint are shown in Figs. 4 and 5.

Fig. 4 shows the supersaturation setpoint ramping up to higher than the current setpoint initially, then dropping below the current setpoint trajectory about halfway through the batch
Fig. 5 shows the temperature setpoint remaining lower than the current setpoint trajectory initially, though peaking slightly above the current setpoint halfway through the batch.

As a result of the optimisation, the median crystal size increased from 350 to 415 µm, a significant enhancement which greatly improved filterability.

Moreover, capacity was increased by 20% while reducing batch time, mostly by reducing the number of failed batches.

Because of the narrower CSD, quality – in particular filterability and flowability of the final product – was improved.

Conclusions

The project demonstrated that the use of a first-principles model of a crystallisation process, validated against experimental data, is capable of improving operation significantly with relatively small alterations to batch procedures.

Also, having created a well-validated model that gives good agreement with experimental results, it is straightforward and a small effort to configure the problem description as a dynamic optimisation and execute an optimisation.

Dynamic optimisation is the most practical way to determine optimal operating policy, because of the large number of variables that need to be considered simultaneously.

Implementation of online control

These successful off-line applications were followed by the implementation of a Model-based Predictive Control (MPC) system. This:

improved process observability and controllability
increased the degree of automation of the batch process
reduced operator interaction from 20 to 4 times per batch
strongly reduced batch-to-batch variability

In addition it enabled operators to gain a better theoretical knowledge of the crystallisation process.