You are on page 1of 6

Numerical Modeling for Design Optimization of SCR Applications

Kevin Rogers Mel Albrecht Michael Varner Babcock & Wilcox Barberton, Ohio, U.S.A. Presented to: ICAC NOx Forum March 23-24, 2000 Washington D.C., U.S.A.
BR-1699

Abstract
Babcock & Wilcox (B&W) has utilized numerical modeling to simulate fluid and combustion phenomena on commercial contracts for more than a decade. Verification of model predictions with actual field test results has provided valuable feedback for model development and refinement. Achieving a high level of agreement between field results and the numerical predictions provides confidence when applying these models as design tools. This confidence is demonstrated by years of modeling and design practice experience put to use for new and retrofit low NO x furnace combustion strategies. Similar challenges of development and verification exist when extending numerical modeling to include downstream SCR processes. Modeling approaches are undergoing continual improvement predicting component pressure drop, temperature and chemical specie distributions, mixing efficiencies, and reactor performance. Building upon prior boiler experience and success, numerical modeling is becoming an increasingly valuable tool for SCR system design optimization efforts.

gree of success and confidence grew through application experience and results validation. This type of experience, alongside the continuum of computer and software development, has helped to propel the acceptance of numerical modeling by the power generation industry. Selective catalytic reduction (SCR) system design has echoed a need for similar experience and results validation. Modeling also becomes an instrument capable of mapping the performance characteristics of discrete components or arrangements. Through optimization of the components performance, an improvement in the overall system design can be obtained. Retrofit projects in particular can exhibit a strong need for accurate modeling of flexible design approaches. They are classically burdened with process conditions and arrangements that are less than ideal.

Numerical vs Physical
Physical modeling has historically been the dominant technique that has used reduced scale models employed by the system suppliers, catalyst manufacturers, and end customers. Catalyst sizing and performance criteria have often been set or influenced by the expected flow and component profiles determined through physical flow modeling. When specified parameters achieve an acceptable value in the physical model, it is assumed that the field results will be similar. If the acceptance criteria are based on physical modeling experience, it then becomes important to understand how the results could differ when utilizing a numerical approach. In the end, both modeling approaches should lead to comparable conclusions that represent full-scale field operation.

Introduction
Numerical modeling is an economically effective design and analysis tool for the simulation of flow, heat transfer and combustion phenomena surrounding boiler components and auxiliary equipment. Furnaces, burners, windboxes, steam drums, pulverizers, electrostatic precipitators, coal nozzles and piping are only a few examples of the components and equipment that have been successfully modeled numerically by B&W. As the collection of project designs and case studies expanded, and as field test results became more available, the de-

Babcock & Wilcox

Numerical modeling offers some advantages over its physical model counterparts. In many cases, numerical modeling can more easily and economically predict flue gas temperature distribution and chemical component distributions. (i.e., flue gas NOx distribution, NH3/NO x ratio distribution and mixing phenomena). Due to its flexibility, numerical modeling permits the study of an increased number of geometric arrangements or modifications in a more timely manner than is possible with physical flow modeling. Numerical modeling can be done at full scale. Numerical models can be stored for future use and referenced far more efficiently than physical counterparts. Warehousing a physical model is not required. The input boundary conditions of temperature, flow and component concentration profiles can be specified to a higher level of detail. Capital investment of model instrumentation and its maintenance/calibration costs are not an issue. Input conditions can be varied easily to assess system sensitivity. The data is inherently in an electronic format. Independent of the modeling approach, the actual inlet conditions and the range of flow/temperature/component mal-distribution expected should be measured and described as accurately as possible. With downstream predictions influenced by the upstream inputs, it is sometimes beneficial to move the upstream model boundaries to locations where the inlet conditions are better known or documented.

With the longer history of use, physical models have achieved a certain level of acceptability in providing predictions. For this reason, some numerical validation studies are performed by comparing the results of the numerical approach to those of the physical. Ideally, each approach should be compared to actual field results so that each are compared to the real system, not to each other. Field data taken should be directly compared to model results. While field data provides the highest level of validation, it can also become quite costly and requires forethought on where to place test ports and measurement instrumentation on the field units. The task of actually collecting the data is time-consuming and sometimes difficult. Aside from validation issues, there is typically some given set of acceptance criteria. These criteria define the target values which, if achieved, will provide a high level of performance confidence.

Velocity Profile
The velocity distribution profile is often reported as a coefficient of variation (Cv) by the following equation: Cv = 100 /x where, Cv = Coefficient of variation expressed as a percent of the standard deviation about the mean. = standard deviation x = mean Depending on performance requirements, the velocity distribution requirements at the catalyst inlet can typically range from 10% to 20% standard deviation from the mean. The velocity profile through the plane of ammonia injection can be equally or more important. For a constant ammonia injection flux, variations in flue gas velocity through each injection zone will induce downstream NH3/NOx molar ratio maldistributions.

Validation and Acceptance Criteria


As modeling efforts on specific projects or case studies accumulate, a wealth of information begins to build. The result is a history and case study library that improves verification, as well as comparative and predictive capabilities. The variety of SCR system designs and arrangements modeled to date by B&W has aided in the building of such a reference library. To the extent possible and practical, these model studies are being augmented with field site data and comparative physical modeling results. Full-scope modeling experience appears increasingly important. With the entire process from the boiler through the SCR modeled, one can develop a more keen understanding of the direct and peripheral consequences of design changes from the NO x generation point to the stack. The ability to manipulate the design and uncover ways to improve performance with modifications to the burners, economizers, economizer bypass arrangements, flue arrangements, flue internals, reactor design, and even downstream flue gas cleaning equipment, is significantly improved. The computer models lend themselves to these library-building efforts, as they are completely and inherently storable in an electronic data format. Without significant hardship, base models can be reconfigured to analyze process results from field testing. It is important that field site measurements be planned to provide feedback for model refinement and validation work. However, a high level of comparability to field measurements is not always paramount. When performing gross level design optimizations, the use of a model to gauge the relative effectiveness of changes and optional arrangements is often a primary goal that can be satisfied at reduced accuracy.

Temperature
The criteria for temperature distribution are typically expressed as the minimum and maximum deviation about the mean, rather than as a standard deviation about the mean. The typical allowable values will range from 20F to 50F. The average gas temperature can influence the allowable minimum or maximum. For example, the minimum recommended catalyst operating or ammonia injection temperature will usually set the allowable minimum. If the average is high, the allowable deviation on the minus side is greater.

Ammonia to NOx Molar Ratio (NH3/NOx)


The numerical approach can easily utilize an inputted inlet boundary of a field-measured NOx profile and determine the NH3/NO x ratio based on predicted dispersion and/or mixing of injected NH3 downstream. The criterion for molar ratio at the SCR catalyst face is typically expressed as a coefficient of variation similar to that for velocity profiles. The target values can typically range from 5% to 15% standard deviation from the mean.

Babcock & Wilcox

Pressure Drop
Pressure drop was once the primary criterion for many system designs. Obscured somewhat by the velocity, temperature and component profile requirements, it is nonetheless extremely important. On retrofit projects, the allowable pressure drop is usually based on fan or furnace pressure limitations, and the system operating power-consumption goals. The pressure drop associated with the catalyst, and at times that associated with a rather contorted flue arrangement, will rob from pressure drop allowable towards the goals of ammonia injection and static mixing.

Comparisons to Physical Modeling


Comparisons between physical and numerical models have typically shown an acceptable level of agreement with pressure drop criteria. An example is a comparison of a physical study where scaling relationships were used to arrive at full-scale results from those measured at the model size. Two numerical models were developed for the comparisons. One was based on the physical models 1/10th scale, while the other was done in full scale. Figure 1 provides a view of the arrangement. Pressure comparison cases P1 through P3 represent comparable measurement points between the physical model and that of the numerical. The comparisons are presented in Tables 1 and 2 below.

The agreement between the predicted pressure drop in the numerical model and the measured pressure drop in the physical flow model suggests an adjustment in acceptance criteria between the two methods is not needed. Each method can be reasonably assured of providing similar results. Many numerically-based velocity and temperature profile results appear to give values that are more on the conservative side compared to those based on the physical model. Table 3 and Table 4 provide examples. Table 3 represents a particular arrangement where mixing devices and vane arrangements are varied from cases V-1 through V-4. Velocity similitude was not maintained from the physical to numercal models; however, for each case, the coefficients of variation for the physical and numerical models were very close in magnitude. The effect of velocity and Reynolds number appear to be more important when considering dispersion and mixing phenomena as compared to that of simple turbulent flow pressure drop determination. They type and order of the rurbulence model used in numerical models also influences the degree of potential conservatism in the results. In Table 4 below, cases T-1 through T-3 represent predictions at the catalyst for a constant overall arrangement where changes in flue internals were made to examining alternate methods of economizer bypass gas introduction. In this case the conservatism of the numerical model was significant when compared to predictions obtained from a physical approach.

Table 1 Physical Model Prediction and Full-Scale Numerical Model Prediction Pressure Drop Comparison Cases Numerical Full-Scale Model Prediction, inches H 2O Physical Model Full-Scale Prediction, inches H 2O P-1 0.48 0.50 P-2 0.18 0.19 P-3 1.8 1.9

Table 2 Physical Model Measured and 1/10th Scale Numerical Model Prediction Pressure Drop Comparison Cases Numerical 1/10th Scale Model Prediction, inches H 2O Physical Model Measurement, inches H 2O P-1 0.45 0.46 P-2 0.16 0.17 P-3 0.54 0.55

Table 3 Physical to Numerical Velocity Profile Comparisons Velocity Profile Comparison Cases Relative Numerical Model to Field Gas Velocity Numerical Model Velocity Coefficient of Variation, % rms Relative Physical Model to Field Gas Velocity Physical Model Velocity Coefficient of Variation, % rms V-1 1.0 17 0.6 15 V-2 1.0 20 0.6 17 V-3 1.0 19 0.6 18 V-4 1.0 22 0.2 17

Table 4 Physical to Numerical Temperature Extreme Comparisons Temperature Extreme Comparison Cases Relative Numerical Model to Field Gas Velocity Numerical Model Temperature, max & min +/- of the average Relative Physical Model to Field Gas Velocity Physical Model Temperature, max & min +/- of the average T-1 1.0 +59/-49 0.6 +10/-28 T-2 1.0 +148/-80 0.6 +60/-36 T-3 1.0 +56/-62 0.6 +32/-34

Babcock & Wilcox

System Design Optimization


With the high performance duty of many contemporary SCR designs, optimization techniques are of ever-increasing importance. Numerical modeling has been demonstrating itself as a very useful tool, not only for the design of a specific project, but also for product design development and generation of information to improve optimization methods. In the past, much less importance was placed on temperature and chemical component distributions. Flue systems were designed with splitters and/or vanes to achieve the simple goal of minimizing pressure drop. Today, the optimization exercise can become rather convoluted as attempts are made to improve temperature and component distributions in ways that minimize degradation of velocity profiles and system pressure loss characteristics. Achieving uniform distribution profiles at the SCR catalyst face with regard to NOx concentration, NH 3/NOx molar ratio, velocity and temperature is often complicated by physical space limitations, arrangement limitations and limits to boiler modifications. In difficult retrofit applications, a unit length of flue is a precious commodity for the system designer who is focused on performance goals. The higher the degree of uniformity required, the more impractical it is to leave a length of flue designed for the lowest pressure drop. Synergy between these mixing and flow distribution tasks is critical in restricted space system design, especially when complicated by the need for bypass gas introduction to satisfy minimum allowable temperature requirements. For large side-to-side or top-to-bottom imbalance, the task of minimizing flue length becomes quite challenging. While this challenge is being tackled with both physical and numerical modeling, it is the numerical approaches that appear equipped with the needed flexibility to address the issues in the time required.

However, a high blend number does not necessarily indicate a high blend or mixing efficiency. It may have been achieved by very long lengths or by high energy consumption. There are three primary degrees of freedom. One is the process result, such as the degree of homogeneity required. The others are length and energy. As the length variable drops, the energy requirement increases and vice-versa. If you find you are not achieving the process result within the allowable length and pressure drop, then a higher efficiency system must be sought or the required process result must be lowered. Both the spatial and energy efficiencies of devices and arrangements can be assessed to provide insight into the optimization exercise. Spatial efficiencies provide an indication of the degree of blend achievable per unit length or volume. Energy efficiency values provide an indication of the degree of blend per unit energy dissipation rate. A bend, an expansion or contraction, and even a straight length of flue can perform a static mixing function. Alone, their efficiency is often low in terms of achieving a blending process result. However, when augmented with internals designed to capitalize on the arrangement and the flow profiles throughout, their efficiency with regard to blending can be enhanced. Test programs utilizing numerical modeling as an analytical tool can be performed to map these characteristics.

Optimization Exercises
Model construction and run time are major considerations. Higher order turbulence models require both greater capital investments in software and computing power, as well as the time requirements to construct the model and run cases to convergence. With an increasing demand for fast project execution, the project schedule can often dictate the allowable model detail and number of iterations. While reductions in program detail can save time, they also tend to generate conservatism in the results. With simpler numerical models demonstrating themselves as compelling tools for assessment of the performance changes versus configuration changes, one has to be very astute to not over- or under-detail the model structure. The output can be configured in ways that facilitate postprocessing by either system performance programs or optimization exercises. For example, the output of the model can be configured for direct input into reactor performance profiling programs. In this way, the reactor performance based on the catalyst activity, volume and inlet profile variations can be predicted to assess allowable operating and control system response tolerances. Often a target value for NH3 /NOx molar ratio at the catalyst face is first attempted by adjustment to the ammonia injection grid design and/or flow profiles through the plane of injection. Figure 2 provides a hypothetical case of resultant molar ratio distribution at the catalyst face. In this example the NH3/NOx coefficient of variation at catalyst face is approximately 17% when no static mixing or ammonia injection zone flux adjustments are made. For visualization purposes, Figure 2 shows the concentration to each zone as a simple percent deviation about the mean for each of forty-two zones across a reactor inlet cross-section. In this case, the target molar ratio coefficient of variation at the catalyst face was 7%. Fortunately, the system design could tolerate some pressure drop expenditure for velocity profile adjustment upstream of the ammonia injection grid (AIG) and on a static mixing function downstream of the AIG. After several options were modeled, the best distribution achievable

Control Models
Specially designed models can be developed to substitute components from the various modeled systems. In this way, performance estimates of discrete components or sections on a side-by-side controlled basis can be made. These control models can be used to map the performance characteristics of the devices and arrangements, providing tools to manipulate designs and focus more quickly on an optimized design. These control models also allow shortcomings of the particular program software or model design approach to become more apparent and identifiable. An example of using numerical modeling to characterize components in ways that facilitate system design optimization is in the area of blend functions. In this case, a dimensionless blend number can be used to describe the degree of blending achieved. This blend number Bn can be described as follows: Bn = 1 - 1/0 where 0 is the standard deviation of the component at the inlet of the dispersion or mix zone and 1 is the standard deviation of the component at the outlet of the dispersion or mix zone. As the standard deviation at the outlet of the zone (1) drops with respect to the inlet standard deviation (0), the blend number will asymptotically approach a value of 1. Therefore, the higher the blend number, the greater the degree of blend.

Babcock & Wilcox

Deviation from Mean, %

Static Pressure (in. H 2O) 0.0 -0.5 -1.0 -1.5 -2.0

Figure 1 Contours of static pressure on a plane through the SCR system.

Figure 2 biasing.

NH3/NOx ratio without static mixing or injection zone

within the allowable pressure drop was a 9% coefficient of variation. Figure 3 depicts this distribution as the percent deviation about the mean. In both cases, no ammonia injection grid tuning is used; that is, the ammonia flow to each injection grid zone is identical and constant. The ammonia concentration profile produced at the reactor inlet cross-section from individual injection zones was predicted to facilitate AIG flow tuning. Figure 4 provides an indication of the coverage a single injection zone has in this particular arrangement. Modeling of a case where the flow of ammonia to each injection zone is biased to help account for NOx, and primarily velocity maldistribution at the ammonia injection plane, gave confidence that the target value could be achieved with a predicted 5.3% standard deviation from the mean. Figure 5 depicts this distribution as the percent deviation about the mean. Models are also valuable in optimizing ammonia injection zone quantity and geometry. The projected dispersion pattern at the catalyst face based on variations to ammonia injection zone design can be analyzed to provide the maximum controllability with the minimum zone quantity. Not only does this knowledge assist in design optimization efforts, but it can give insight to field start-up crews on the projected responsiveness of each ammonia zone flow control valve.

Deviation from Mean, %

Figure 3 biasing.

NH3/NOx ratio with static mixing and no injection zone

Summary
Numerical modeling will be continually influenced by advancements in programming and computing power. Equally important, however, is the degree of application-specific experience, which often proves to be the key to a successful optimization in a timely fashion. Verification of model predictions with actual field test results has and will continue to provide valuable feedback for model development and refinement. Confidence in the numerical modeling approach will be increased as further field results and numerical predictions are obtained and compared. As model libraries are expanded and comparisons are documented, the application of numerical models for design optimization will be similarly improved.

NH 3, ppmdv

Figure 4

AIG control profile.

Babcock & Wilcox

In addition to specific project modeling and optimization, numerical modeling is becoming an increasingly valuable tool for mapping and characterizing the performance of discrete components within the system. As the knowledge of the performance of discrete components and arrangements grows, so does the ability to select and choose the most appropriate design path in the most expeditious manner.
Deviation from Mean, %

References
1. K. Rogers, M. Milobowski, and B. Wooldridge, Perspectives on Ammonia Injection and Gaseous Static Mixing in SCR Retrofit Applications. Presented at EPRI-DOE-EPA Combined Utility Air Pollutant Control Symposium, Atlanta, Georgia, August 16-20, 1999. 2. A. Sayre and M. Milobowski, Validation of Numerical Models of Flow Through SCR Units. Presented at EPRI-DOEEPA Combined Utility Air Pollutant Control Symposium, Atlanta, Georgia, August 16-20, 1999.

Figure 5

Mixing plus biasing.

Copyright 2000 by The Babcock & Wilcox Company, All rights reserved. No part of this work may be published, translated or reproduced in any form or by any means, or incorporated into any information retrieval system, without the written permission of the copyright holder. Permission requests should be addressed to: Market Communications, The Babcock & Wilcox Company, P.O. Box 351, Barberton, Ohio, U.S.A. 44203-0351. Disclaimer Although the information presented in this work is believed to be reliable, this work is published with the understanding that The Babcock & Wilcox Company and the authors are supplying general information and are not attempting to render or provide engineering or professional services. Neither The Babcock & Wilcox Company nor any of its employees make any warranty, guarantee, or representation, whether expressed or implied, with respect to the accuracy, completeness or usefulness of any information, product, process or apparatus discussed in this work; and neither The Babcock & Wilcox Company nor any of its employees shall be liable for any losses or damages with respect to or resulting from the use of, or the inability to use, any information, product, process or apparatus discussed in this work.

Babcock & Wilcox

You might also like