Crystallization process simulation in the chemical industry

Reducing crystal size and increasing product quality variability for terephthalic acid

Fig. 1 – Terephthalic acid

PTA process flowsheet – click to enlarge

Fig. 2 – PTA flowsheet

Measurement variability – click to enlarge

Fig. 3 – Example of measurement variability

Supersaturation profiles – click to enlarge

Fig. 4 – PTA supersaturation profiles

Terephthalic Acid (PTA) crystal mass production – click to enlarge

Fig. 5 – PTA crystal mass production

PTA startup crystal size distribution – click to enlarge

Fig. 6 – PTA start-up crystal size distribution

PTA quantiles at startup – click to enlarge

Fig. 7 – PTA quantiles at start-up

The price that can be achieved for terephthalic acid tends to depend on the quality, as epitomised by the median crystal size. Products with a median crystal size of greater than 100µm command a price premium.

There are a number of ways to increase median size and improve the crystal size distribution (CSD), but because of the complex interaction between process and the complex phenomena of crystallisation it is not obvious how this can be achieved.

Increasingly, predictive modelling, combining rigorous physics and validation against experimental data, is providing the answers.

This case was undertaken for a major producer of terephthalic acid.

The customer required a convenient tool for model development, with the ability easily to switch between different process configurations.

They also required a powerful tool for validation, with parameter estimation and experiment design capabilities.

The process

PTA is produced using a continuously-operated multi-stage flash cooling process. A typical process is shown on the right.

Crystallisation is a key step in producing product of the desired characteristics with respect to crystal size distribution and median crystal size.

Requirements

The objectives of the project were to:

Troubleshoot existing operation
Design new crystalliser trains
Optimise crystallisers and downstream operations.

The key requirement for the model was that it should be able to:

explain observed process behaviour
predict behaviour at different operating conditions on same plant
predict behaviour on different plants

The model

The basic crystallisation model was taken from PSE's Advanced Model Library for Solution Crystallisation (AML:SC) and customised to meet the project requirements, given the available data.

Two different kinetic models were implemented in order to determine the dominant kinetic effects:

a model with power law expressions for primary nucleation, secondary nucleation and crystal growth. No size dependency of crystal growth or attrition (large crystals are more prone to attrition than smaller ones) was taken into account
a more mechanistic model which included a detailed attrition model that not only accounted for the crystal size dependency of the attrition rate but also for the size distribution of the attrition fragments formed.

The results showed that the dominant nucleation mechanisms are:

primary nucleation, which occurs mainly in the first stage
secondary nucleation due to attrition (all stages)

Validation

A key challenge was lack of independent measurements of crystallisation kinetic parameters.

Using the developed model in conjunction with the available data, parameter estimation was used to find the optimum values of unknown parameters.

The objective function included weighting related to variances in validation data as far as these were known (see example in Fig. 3).

In order to identify and eradicate the systematic errors that contribute to variability, model-based experiment design and parameter estimation were used in an iterative loop to determine new validation experiments in order to maximize model fidelity.

Results

The model was used for numerous calculations, ranging from dynamic (e.g. start-up) to steady-state studies.

In particular it was used to determine a more appropriate pressure profile along the multiple crystallisation stages. The aim was to reduce the volume fraction of crystals in the coarse tail of the PSD whilst not decreasing the product median size.

Selected results are shown on the right:

Fig. 4 shows the supersaturation curves for each of the stages over time
Fig. 5 shows similar profiles for crystal mass production
Fig. 6 is a 3-dimensional plot of size mass index vs volume mass density vs time
Fig. 7 shows various quantiles in a key stage over the start-up period

Conclusions

The result of the model validation and resolution of the speed and robustness issues was a versatile and descriptive model that executed robustly.

This demonstrated for the first time that models of the complexity and fidelity of the AML:SC models – even when used in multizonal configurations – could be applied successfully to a large-scale, multiple-stage crystallization process.