Key to this improvement was use of new CFX computational fluid dynamics (CFD) modeling software, which provides a multi-size bubble model and manages the complexities involved with modeling a rotating agitator.
PTA is a key ingredient in polyester manufacturing. Reacted with a diol, typically ethylene glycol, it is used to produce polyester resins and fibers, powder coatings, specialty polymers and adhesives, as well as tire cord, magnetic tape, plastic film and other products.
DuPont's process relies on paraxylene oxidation. In this process, paraxylene is fed into a reactor, dissolved in acetic acid and water, and a soluble catalyst is added. The most common catalyst is a combination of manganese and cobalt salt with bromine added as a promoter.
Large tubes are used to sparge the vessel with air, and a mechanical agitator breaks the air up into small bubbles and distributes it throughout the reactor. In the process, which is based on propagating free-radical reactions, oxygen is absorbed very rapidly. Because any shortage of oxygen can lead to the formation of undesirable byproducts, it is critical to have an excess of oxygen in the reaction zone. The gas holdup in the reactor is typically in the 40% to 60% range. The amount of catalyst and solvent used, as well as the operating conditions in the reactor, are carefully tuned to optimize process yield and quality.
Reaction time is typically between 30 and 60 minutes, and yield is normally about 98%. The PTA is produced in the liquid phase and precipitates as a solid. Since the oxidation process is highly exothermic, excess heat is removed by condensing and refluxing the water and acetic acid vapor in order to maintain reactor conditions. The reactor products are continuously discharged as a hot slurry to a multistage crystallization process, where cooling takes place by flashing off acetic acid and water as pressure is reduced. The resulting slurry is separated to recover terephthalic acid solids. The solids are dried to recover the solvents before being sent to a purification process.
In the past, DuPont engineers used proprietary "network of zones" models to optimize the manufacturing process. These models use a fixed flow field, but incorporate mass transfer and chemical reactions. They typically have up to 120 cells, to allow reactor optimization without capturing the full detail of the flow fields within the reactor.
Recently, however, engineers have been using CFD to allow the reactor to be modeled with much higher levels of detail, depending on the amount of computing power and length of time available to solve the model. Its more fundamental modeling approach allows CFD to model varied geometries more effectively, said DuPont research associate John Jurgensen. "Because of these advantages," he says, "our future direction is to eliminate network of zones modeling and move exclusively to CFD reactor modeling platform."
CFD involves the solution of the governing equations for fluid flow, mass transfer, and chemistry at many thousands of discrete points on a computational grid representing the flow domain. A CFD analysis yields values for fluid velocity, fluid temperature, and fluid pressure throughout the solution domain. Based on the analysis, a designer or an engineer can optimize fluid flow patterns by adjusting either the geometry of the system or the boundary conditions such as inlet velocity or temperature. DuPont chose CFX software from ANSYS Inc. of Canonsburg, Pa. because the program can simulate multi-size bubbles, moving grids, two-phase flows, multiphase species mass transfer, and complicated chemical reactions.
Without numerical simulation, the only way to address a problem of this type is to build an experimental model. Generally, the model may be built of acrylic and the two fluids colored so that their mixing can be observed. The problem with this approach is that the model has to be scaled to keep its cost reasonable, and scaling can lead to errors that are very difficult to estimate. Any changes to the model, which would be required to investigate alternative configurations, take additional time and money.
Modeling the reactor
DuPont's early work with CFD for PTA manufacturing involved calculating flow rates and other parameters that were used as input to the network of zones model, Jurgensen said. After gaining experience with the CFD tool, though, engineers felt that they were ready to use it by itself to model a full reactor. Significant improvements had already been made to DuPont's proprietary reactor technology but engineers felt that further improvements could be achieved if they could more precisely model the conditions within the reactor. In particular, they wanted to explore methods to further enhance the distribution of air and oxygen phase transfer in the reactor, leading to a further reduction in byproducts and consumption of raw materials.
Modeling Manager Duncan Housley, Senior Process Engineer Martin Davis, and Jurgensen perform reactor modeling for clients and for DuPont's reactor development program. They began by modeling the reactor using the software's sliding mesh method, in which the domain was divided into two regions, a thin stationary outer part and an inner section containing the shaft and agitator. The inlet was represented by an inlet boundary condition. The outlet was specified as a pressure boundary condition. They used CFX's multiple-size-group multiphase modeling feature to simulate the behavior of the bubbles, including break-up and coalescence. The entire model has about 100,000 cells and can be solved on one processor of a Silicon Graphics Origin server.
Improving air distribution
"Our initial results indicated that there was indeed scope to improve the reactor oxygenation in several regions," Housley said. "Because this was the first time we had modeled a complete reactor with CFD, we weren't sure whether we could trust the predictions or not. A breakthrough came when a licensee reported a problem running the process in a different way than it was intended and modeled. We modeled the modified reactor. A comparison of the two models showed that they precisely predicted the difference in the performance of the two reactors, including the problems experienced by the licensee, confirming the predictive power of the original model and demonstrating the value of the model to improve performance."
"After several iterations," Davis said, "we concluded that we could substantially improve oxygenation in the reactor by changing the agitator's geometry. We fine-tuned our design until we optimized distribution of air throughout the reactor." This improvement, he says, is reflected either in process quality, through reduction in byproduct, or in operating cost savings, since the need for acetic acid is reduced, or both.
