Predicting Process Performance

The Problem Titanium dioxide (titania) is a white pigment used widely in a variety of applications. Manufacture involves a reaction in which gaseous titanium tetrachloride and oxygen are mixed in a flame reactor to produce titanium dioxide powder and chlorine gas. Pigment performance depends critically on mean particle size and on particle size distribution. The control of the production process is complex and computer modelling has been used routinely to determine both reactor configuration and the relationship between operating conditions and product properties. Because of the complexity of the chemical environment and the reactor dynamics simulations can take up to a month to converge and this work was undertaken to investigate the degree to which an improved model linked with parallel computing could deliver an improved predictive capability for powder properties from significantly shorter run times (ie a factor of ~200 - hours not months).

The Challenge The oxidation process involves several complicated stages including nucleation, growth and coagulation of the titania particles and the dissociation and recombination of chlorine. The reactors have multiple feed ports and operate at various temperatures, velocities and mixture ratios. The challenge was to achieve the shorter run times whilst including new modelling features to address the phase behaviour, the physical and chemical theory and operation in three dimensions with real reactor geometries

The Solution
The particle growth process was modelled by discretising the size distribution in terms of a fixed number (up to 100) of size intervals characterised by a mean diameter. Each size class was then treated as a distinct transported variable. The source terms in the titania size class linear equations were applied for each process stage: nucleation increased the population of the smallest size class; growth transported particles from one size class to the next; coagulation conserved mass within size classes and reduced the standard deviation of the final size distribution. In addition, the strong exotherm influenced the temperature dependence of the flow parameters viscosity, density and specific heat. Integration over the area at the end of the reactor pipe then gave the emerging particle size distribution with typical run times of 10000 processor hours which can be achieved quickly on parallel computer architects.

The Benefits
• The customer was able to improve the predictive capability of a complex process simulation using parallel computing to achieve both higher run speeds and enhanced model characteristics giving savings in both time and money.
• Several disparate phenomena were modelled sufficiently accurately to predict end product properties successfully allowing a way forward for further improvements in reactor design and process control.
• The customer has realised a step change in its modelling capability which can be applied to benefit other product and process development projects.