104 Application To Reactor Engineering

The possibility of using CFD models to screen configurations of industrial reactors will allow reactor engineers to spend more time evolving creative and innovative reactor designs. For example, the configuration proposed for a liquid phase oxidation reactor (LOR) can be studied using a computational flow model to help understand the relation between reactor configuration and claimed performance enhancement. Ranade (1998) discussed the use of CFD models to aid understanding of the formation of roll cells and the region of high turbulence just below the draft tube in the proposed LOR. Detailed analysis (using post-processing and visualization tools) of results predicted by such CFD models will be useful in tailoring LOR configurations to suit different process requirements. At this point, it is essential to re-emphasize the importance of understanding the limitations and assumptions involved in setting up the computational model when interpreting the predicted results. As discussed earlier, predictions of CFD models can go wrong for two main reasons (other than human error and machine malfunction): they may be based upon a physically incorrect mathematical representation or upon numerically deficient representation. Inadequacies of the second kind can be and should be eliminated or minimized by using careful numerical experimentation with grid distributions, discretization schemes etc. (refer to Chapters 6 and 7 for more discussion on these issues). Inadequacies of the first kind will almost always be present for the tractable simulation of complex industrial reactors. Reactor engineers therefore have to identify the most important aspects of the flow which are relevant to reactor performance, and to accurately represent those aspects in the mathematical model. In most cases, one has to break up the problem into several sub-problems and employ a hierarchy of modeling tools with appropriate degrees of sophistication.

Considering again the example of LOR, one of the most important functions of LOR hardware is to set up roll cells below the draft tube, which can capture gas bubbles to enhance the efficiency of oxygen use. As shown by Ranade (1998), it is possible to use CFD models to obtain qualitative as well as quantitative understanding of such roll cells. However, from the reactor engineering point of view, a major objective is to examine whether the roll cells formed provide adequate residence time for gas bubbles to achieve the desired oxygen consumption. It is therefore necessary to understand the phenomenon of bubble capture within these roll cells and its influence on mass transfer (with chemical reaction) from a single gas bubble. Before setting up the computational model to simulate turbulent gas-liquid flow in a complicated geometry, it may be worthwhile to develop a bubble-tracking model employing a Eulerian (for liquid phase)-Lagrangian (for bubbles) framework. Such a model will allow one to undertake a preliminary screening of alternative configurations without developing the complete model. It is possible to represent detailed mass transfer with reaction (based on validated reaction kinetics) by such a computational model. Apart from preliminary screening, results of such a model will also provide the basis for developing a realistic but tractable model in a Eulerian-Eulerian framework.

Development of a detailed model based on a Eulerian-Eulerian framework may be necessary to make realistic estimations of bubble size distribution, bubble flow, its influence on impeller power dissipation and flow field. Ranade and van den Akker (1994) have shown that gas-liquid flow in the bulk region of stirred reactors can be reasonably predicted using the computational snapshot approach. However, unless details of flow near the impeller blades are simulated adequately, it is not possible to determine the desired characteristics of gas-liquid flow (and their variation with configuration, scale and operating parameters). Recent work (Ranade et al., 2001c) shows that it is possible to simulate trailing vortices and gas accumulated in these vortices using a two-fluid model. However, these models are not able to predict the cavity formation behind blades without becoming computationally too demanding. In order to address the reactor engineering issues of interest, it may be necessary to break the overall problem into sub-problems and again use an appropriate modeling approach for each sub-problem. For example, the complex problem of trailing vortices behind impeller blades and their interaction with bubbles may be simplified by studying vortices behind a single blade. Ranade and Deshpande (1999) modeled and simulated gas-liquid flow over a single impeller blade. They were able to predict the details of trailing vortices and the capture of gas bubbles within these vortices.

Rigby et al. (1997) also applied a CFD-based model to understand bubble break-up from ventilated cavities in gas-liquid reactors. Ranade et al. (2001d) used a volume of fluid (VOF) approach to understand cavity formation behind blades. Observations and insight gained through such studies may be used to develop appropriate sub-models, which can then be incorporated in a detailed reactor-engineering model.

With appropriate validation (direct and indirect), detailed reactor engineering models can be used to screen and optimize reactor hardware. For most practical reactor engineering applications, validation of a computational model can be carried out only via indirect means. For single-phase, homogeneous flow applications it is often possible to use CFD models to optimize reactor hardware. For such applications, it is necessary to include species conservation equations in the mathematical model. The influence of species concentrations on fluid properties can easily be accounted for in such models. When reactions are slow compared to mixing, extension to include species conservation equations is straightforward. In such cases, CFD models can be used directly to simulate reactor performance and to evaluate the influence of design and operating parameters on reactor performance (Middleton et al., 1986; Brucato et al., 2000). When reactions are fast compared to mixing, special models to account for microscale segregation need to be developed (see Chapter 4 for more discussion on the modeling of mixing with fast reactions). Ranade (1993) described an example of the application of a CFD model (with a multi-environment micromixing model) to evaluate the influence of design and operating parameters of a stirred reactor on its performance. Some of his results are discussed here.

Ranade (1993) considered the case of a semi-batch stirred reactor to carry out diazotization reactions. The underlying chemistry can be represented by classical series-parallel reactions:

The first reaction is extremely fast compared to the second reaction and compared to the expected mixing rate in stirred reactors. Reactions are carried out in a semi-batch mode with reactant B fed over a time tfeed to a reactor containing pure component A. Thus, if the added reactant B mixes instantaneously with A, a second reaction cannot take place. All the added B will be consumed in the first reaction (since B is a limiting reactant). If the mixing is not fast enough, all the added B will not be in contact with A and will have the opportunity to undergo a second reaction to produce component S. The yield of component S can therefore be considered as a measure of mixing: the more the yield of S, the poorer will be the mixing. The objectives of Ranade's (1993) study were to establish relationships between reactor configuration (feed pipe location, scale) and operating parameters (impeller speed, feed flow rate) and reactor performance, that is, the yield of desired product, R. Since the physical properties of liquids were not strong functions of species concentrations and operation was practically isothermal, it was possible to decouple the flow and reactive mixing models. In the first phase, a computational model was developed to predict detailed mean and turbulence characteristics of the stirred reactor equipped with a standard Rushton turbine. Since the feed pipes were located in such a way that most of the reaction zone lies outside impeller swept volume, a black box approach was conveniently used to generate the desired results quickly. The predicted results were verified by comparing with the published data on flow generated by Rushton turbines. These results were then used to simulate reactive mixing in stirred reactors. The multi-environment model of Ranade and Bourne (1991), which is discussed in Chapter 4, was used. Comparison of the predicted influence of feed location, impeller speed and reactant concentrations with experimental observations is shown in Fig. 10.31. It can be seen that the computational model correctly captures the influence of all of these parameters. The validated model was then used to select an appropriate feed location and other operating conditions. It was further used to evaluate the possibility of using multiple feed inlets to enhance reactor capacity without reducing the yield of desired product R.

Even if the feed location is near the impeller stream, higher feed flow rate leads to reduction in the yield of the desired product as shown in Fig. 10.31 (feed rate and therefore reactor capacity will be inversely proportional to the feed time, tfeed shown in this figure). A possible alternative to increase reactor capacity is to use multiple feed pipes at the same radial and axial locations to ensure the same levels of turbulence at all feed pipes. If multiple feed pipes can be used, reactor capacity can be enhanced without changing the feed rate through each feed pipe. However, it must be noted that feed introduction via multiple inlets may lead to deterioration of selectivity of the desired product if the reaction plumes emanating from different feed inlets interact with each other. The computational model was used to evaluate the idea of introducing feed through multiple inlets. The predicted results are shown in Fig. 10.32. It can be seen that, for feed location A, selectivity remains unaffected by an increase in the number of inlets, up to eight feed inlets. This means it is possible to reduce feed time or increase capacity by a factor of eight without affecting selectivity towards the desired product! Thus, the computational model can be applied to optimize reactor configurations if the necessary cost data is available.

Computational models can be used for reactor engineering applications in a variety of other ways. Even for multiphase reactors, where direct verification of models is difficult, CFD-based models can be used to evaluate alternative reactor configurations and to characterize existing reactor hardware. Such characterizations or fluid dynamic audits of existing reactor hardware will be useful to identify the scope for potential improvement and ways of realizing this potential by evolving retrofit designs. Bakker and coworkers (Bakker et al., 1994a; 1994b; Fasano et al., 1994) cited several examples of using CFD models to enhance the performance of

0 0

Post a comment