A Systems Approach to Adding Natural Gas Firing to Two 600 MW Units

Presented at the Joint ISA/EPRI Conference

June 8, 1993

Wesley J. Schulze, Coen Company, Inc.


Two emerging practices in the design and implementation of combustion equipment systems are Computational Fluid Dynamic (CFD) Modeling and Distributed Control System (DCS) based Burner Management Systems (BMS). CFD Modeling greatly aids in firing equipment design, giving confidence in the proposed design solutions. CFD allows for optimization of these designs. Improved combustion efficiency and operational reliability, while meeting emissions requirements, are the end result.

Implementation of the BMS in DCS hardware leads to a system with extensive operational advantages. DCS hardware selection, architecture considerations, system design and integration of ancillary equipment were key elements of the engineering of this system.


This paper is divided into two major parts that are based upon a contract that included adding natural gas firing capability to two 600 Megawatt units for Central Hudson Gas & Electric Company, Poughkeepsie, New York. First, the computer modeling is explored and then the burner management system design.

Each unit at Roseton is a balanced draft, tilting tangentially fired furnace. The boilers are single furnace type with five levels of heavy oil mechanical atomizers. In January, 1991, Coen Company was awarded the contract to supply a variety of equipment and services as part of a project to add full load natural gas firing capability to these units. The contract was based upon an engineered systems approach. Included in the contract was computational fluid dynamic modeling, all fuel and air admission nozzles (buckets), refractory windbox sections, air flow diffusion panels, gas igniters, fiber optic extended flame scanners, safety shutoff valving and a distributed control system based burner management system. Installation was completed on the first unit in the fall of 1991 and the second in the spring of 1992.


Computational Fluid Dynamics (CFD) can be used in a number of ways to aid in the design of combustion equipment for the large furnaces encountered at Utility generating stations. This section will give several examples of how CFD was used in the design process for the Roseton station.


Equalizing air flow to each burner is always a key element in minimizing emissions and optimizing unit performance. Insufficient air flow to any burner results in the formation of excess Carbon Monoxide or Particulate Matter. Too much excess air to any burner increases NOx and requires higher overall excess air levels for a given level of CO or Particulate Matter. The units in question have had a history of air flow problems in the windbox. One problem was the maldistribution from the bottom to the top of each windbox. The second problem originated from insufficient windbox air velocity entering the furnace.

The engineering and construction work associated with the addition of natural gas firing gave an ideal opportunity to analyze these air flow problems, test the solutions and implement the improvements. During the field modification effort, the first part of the windbox modeling effort involved determining the air velocity profile entering the top of each of the two windboxes, each of which furnish combustion air to two corners of the furnace. This was done by finding the fan outlet velocity profile and modeling the effect of the connecting ductwork. The profile at the top of the windbox is the key to determining the difference in distribution to the two corners supplied by that windbox. Because the flow is unequally distributed prior to the last corner before entry to the windbox, the turning vanes in that corner only pass along that maldistribution. The extent of the difference in velocity through the corners was found to be 1.7% in our model. Although this could have been corrected by modifying the turning vanes, it was decided these modifications were not required at the time. Even though corner to corner air flow differences were not corrected, air flow from top to bottom of each windbox was modified as discussed below.

There are several ways to display the results of modeling in order to help us to interpret the output. We start with an overview of the ducting leaving the air heater and entering the top of the burner windboxes (Figure 3A). Using the example of air flow into the top of the windbox, we can use velocity vectors that are color coded. Unfortunately, much information is lost when black and white copies for this paper are mode from the color plots. In addition to color coding, the vectors show direction and their size is proportional to velocity. We can show this as a cross section plan view (Figure 3C) or a cross section side view (Figure 3D). Also, using color coding we can show velocity as a profile grid (Figure 4A). Velocity is also depicted by the relation deflection of each grid element. Flow vectors can also be used in a three dimensional isometric transparent view (Figure 4D). This air flow velocity can be shown in still another way by plotting a velocity grid on three axes (Figure 3F). A cross section is shown on two (X and Y) axes. Velocity is shown as a function of vertical position on the third axis. Once the model is converged, it is quite easy to view the display or displays that most clearly convey the information needed to understand the results.

After the velocity profile of the air flow entering the top of the windbox was determined it was possible to arrive at the velocity through each corner as a function of elevation (Figure 4F). The elevation velocity profile varied from the average velocity by 108.0 feet per second ± 13% at corner 4 and 109.9 feet per second ± 6% at corner 3 with the higher velocities at the lower burner levels.

Two candidate designs were evaluated as potential ways to improve this maldistribution. Both involved arrangement of perforated plate sections in the lower portions of the windbox just ahead of the secondary air dampers. The first arrangement simply consists of 48% open perf in the lower half of the windbox. A second arrangement included perfs with three different open area percentages. The two arrangements were modeled. In addition to the other ways we have of depicting graphically our modeling results, a tabular method was used for this computation (Table 2). The average velocity and maldistribution percentage was also calculated. The first arrangement resulted in a theoretical reduction on Corner #4 from 13% to 5%. Since the more complicated second arrangement resulted in only a marginally better improvement (to 4%), Arrangement No. 1 was selected.


Now to discuss the air flow velocity problem through the nozzles, the part history of retrofit work and the solution implemented as part of this contract. The basis of the problem is excess windbox cross sectional area at the air admission nozzles or furnace entrance. This oversizing produced insufficient combustion air velocity in the area surrounding the oil atomizers. Poor combustion with the wide range mechanical atomizers resulted. After twenty years of various efforts, including design study and resulting modifications, a compromise was reached. The auxiliary air sections between the fuel-air sections were blocked off with refractory sections. This retrofit work also resulted in different corner equipment in the two units. Although the refractory sections resulted in acceptable combustion performance, recirculation zones downstream of each refractory section created unacceptable degradation of the air nozzles immediately above and below these refractory sections. The damage was particularly severe on the top nozzle of each fuel air compartment where fly ash depositions insulated the nozzle from the combustion air flow allowing the furnace radiant hear to overheat the nozzles (Figures 3 & 4).

The first step in the modeling process was to create a model that reproduced the recirculation zones that caused the nozzle degradation. We used a velocity vector display and temperature contours to illustrate the base condition. The preferred solution centered on new refractory sections with 20% open area. This produces enough air flow to eliminate the recirculation pattern (Figures 11 & 12). At the same time, this solution does not materially affect the air flow velocity through the fuel compartments so that combustion quality is maintained.


The next phase of the modeling included the combustion reactions inside the furnace. Here, we modeled the flame front and fuel burnout patterns. Stability and the potential for furnace vibration or rumble is a very important issue in gas conversions. A key element in the design of a rumble free natural gas flame is the controlled mixing of fuel and combustion air. The contrasting situation is one in which the fuel and air are rapidly mixed, ready to be ignited. In the latter case, variations in radiant heating can result in multiple points where the flame front could occur. This is the basic cause of rumble. The desired pattern is a staged combustion with the flame front well anchored at the nozzle front. This stability is accomplished by the diffuser and gas drilling patterns to create recirculation at the burner front. The furnace model predicts the effect of combustion in the fireball upon ignition points within a given corner combustion zone (Figure 16). A scale model was constructed to verify the results of our computer model, confirm the staged, vibration free flame and fine tune the gas drilling.

As we complete the model further up in the furnace we predict burnout of the fuel and resulting concentrations of oxygen and methane (Figures 28 & 29). This is useful in predicting CO versus oxygen concentration. For our fuel oil combustion modeling key input data was obtained from a detailed particle size testing of one of the mechanical atomizing tips from these units. Modeling computations for fuel oil particle heat up, boiler and evaporation are used in addition to the process described below. With this modeling Particulate Matter formation is calculated. By considering oxygen concentration and temperature among other variables throughout the furnace, NOx is calculated. Velocity and temperature profiles are also available (Figures 30 & 31). As combustion equipment and/or combustion air flow distribution modifications are considered, the downstream effects in the furnace can be evaluated.


In a computation fluid dynamics model, the domain of interest (windbox, furnace, etc.) is formulated on a computer workstation. This volume is divided into discrete volumes, or cells, ranging in number from around 2,000 to over 50,000, depending on the complexity and the required level of accuracy of the model (Figure 1). Models for this project for various air flow and furnace sections were comprised of up to 32,000 cells. Physical properties of the fluid (temperatures, pressures, velocity components, compositions, etc.) are constant within each cell. The governing differential equations, which solve for the variations in fluid properties across the cell boundaries, are iterated across the entire model until a converged solution is obtained. These equations cover the complex relationship between fuel/air mixing and chemical kinetics as well as gas temperature and flow distribution patterns. Typically, thousands of iterations are required to reach convergence.

In the last few years, there have been tremendous advances in the analysis of combustion flow fields and reactions. This improved knowledge and success can be directly related to improved knowledge of turbulence, chemical reactions, instrumentation and the advancement of numerical methods through the use of the computer.

Today it has become common practice to use CFD as a tool to predict performance, trouble shoot and synthesize new products in the field of combustion. Fluent is one of several general purpose three dimensional CFD programs that are commercially available. We have used Fluent and added experimental and field data and combined the results with other proprietary programs to produce the output useful in our combustion design work. CFD using Fluent is a technique based on mathematical theory coupled with experimental data and numerical techniques to find solutions to reacting fluid flow problems that in the past could only be obtained by full or small scale testing.

With many iterations over thousands of cells, it is apparent significant computer power is required to produce these models. In fact, just a few years ago it would not have been practical to perform modeling of this extent for this project. We used a network of Sun Microsystems computers. Many of the computer runs were performed unmanned over nights and weekends. The total modeling effect was completed over a period of about two months.


Prediction of fluid flow processes can be obtained by two main methods: Small scale tests and theoretical calculation using CFD methods. Each of the methods has advantages and disadvantages.

The most reliable information about a physical process is often given by actual measurement. Such full scale tests are, in most cases, prohibitively expensive and often impossible. The alternative then is to perform experiments on small scale models. The resulting information, however, must be extrapolated to full scale and general rules for doing this are not always fully scaleable over the range of interest. Further, the small scale models do not always simulate all the features of the full scale equipment. This further reduces the usefulness of the test results. There are also serious difficulties of measurement in many situations. In fact, measuring of required data may be prohibitively expensive or impossible. Finally, the errors of some measuring techniques may be so excessive that the data obtained is of limited usefulness.

An important advantage of a computational prediction is low cost. In most applications, the cost of a computer run is many orders of magnitude lower than the corresponding experiment. Further, whereas the cost of equipment is increasing, computing costs are almost certain to be even lower in the future. In addition, many different configurations can be “tested” virtually overnight.

A computer solution of a problem results in detailed and complete information. It can provide all the relevant variables (such as velocity, pressure, temperature, concentration, turbulence, intensity) over the entire domain of interest. Unlike the situation in an experiment, there are a few inaccessible locations in a computation and there is no counterpart to the flow disturbance caused by the probes. Obviously, no experimental study can be expected to measure the distributions of all variables over the entire domain. For this reason, even when an experiment is performed, there is great value in obtaining a companion computer solution to supplement the experimental information.

In many cases, where detailed information is not required, small scale experiments or tests may be more cost effective than the CFD method. Validation of the computed results by comparison with real word data is always required. Conversely, for the design of scale models preliminary computation are often helpful and the amount of experimentation can usually be significantly reduced if the test firing is supplemented by modeling. An optimal prediction effort is therefore a judicious combination of computation and testing.


The CFD Modeling has proved a valuable design tool for this project in the same way it has proved an important resource in completing our other projects and design of our standard burner designs. The combustion equipment at Roseton has performed as expected, is reliable and firing is without rumble. Opacity, CO, NOx, Particulate Matter and Excess Oxygen have met targets. In large part, this performance is due to the combination of CFD and scale modeling performed in the design process.

Ongoing combustion modifications required to meet ever more stringent emission mandates will be able to take advantage of the investment made in the CFD Model. The model can be reactivated and reconverged for various NOx reduction strategies such as air flow biasing between upper and lower elevations to achieve off stoichiometric firing, addition of overfire air and addition of bulk mixed flue gas to the combustion air. NOx, CO, Particulate Matter and Furnace Exit Gas Temperature can be a calculated output for each strategy.


One of the very important decisions made early in the project was whether it was possible to reuse the existing Burner Management System for oil. Often, our first inclination is to try to salvage existing hardware. For a number of reasons, it was decided to supply a new BMS for both gas and oil on this project.

  1. There was not sufficient space in the existing electronics room to add new cabinets for the gas logic.
  2. The 17 year old BMS was approaching the end of its design life.
  3. Interface and Upgrade of the existing system would favor the original manufacturer, tending to eliminate competitive bidding.
  4. Performance guarantees would be difficult to enforce.
  5. As built drawings for the existing system are rarely completely accurate.
  6. A complete, integrated factory checkout would not be possible.

When considered together, it was clear there was not a real cost benefit to reusing the old oil BMS. In addition, the operational benefits made a new system an easy choice.


The design process can be divided into several discrete, but related segments.

From the customer operations standpoint, the most important document is the sequence of operations. This is a step by step narration of the information supplied to the operator and the actions he will take in the proper startup and shutdown of the boiler. This text serves to insure there is a meeting of the minds as to how the system should properly operate. It also serves as a basis for the remainder of the system engineering design, as well as a tool for training and continuing operations.

The input/output list is a key element in documenting the exact interface to field devices. For example, one area documented would be the actuation of a safety shutoff valve and the corresponding limit switch position feedback. This list is then used as a design basis for point assignment to the I/O cards, cable conductor allocation and production of the detailed schematic diagrams.

With the sequence of operation and I/O list complete, the remainder of BMS design can be divided into two remaining major components: Wiring schematics and logic design. An important characteristic of this project was the use of prefabricated cables between the field junction boxes and the DCS. Naturally, the initial manufactured cost was higher than if conventional wiring was to be run in the field. Advantages accrued by testing the system, including cables, during the factory test. This streamlined startup during the critical unit outages. Field junction boxes, along with the control insert panel, were also included in the factory test allowing for a thorough checkout. The cable manufacture, DCS hardware manufacture and integrated systems test were subcontracted to Westinghouse Electric Corporation, Pittsburgh, PA. The result was excellent hardware integrity.

Logic design was the other major aspect of the BMS. The first step was to depict graphically the logic for operation of the firing equipment. After approval of these logic drawings, the detailed logic that would be loaded into the DCS hardware was prepared. Relay ladder logic format was used since the format is relatively standardized and familiar to our project people, startup engineers and most plant technicians. Using a Power Flow Diagram it is easy to see the current status of each page of logic on line. This feature is invaluable to understanding the operation of the logic and trouble shooting problems. Further, inputs and outputs can be forced to a desired state to test the impact on a logic section or test the actuation of a field device.


As part of our field checkout, a special graphic page was constructed (Figure B2). This allowed a person at the engineer’s console to read the status of all inputs and control all outputs of a given burner. By checking with his counterpart at the field device, the engineer could verify the integrity of each input/output. This constituted the first phase of the commissioning of the electronics after the initial power up and system functional tests. I/O count for the BMS was approximately 1500 so that this loop testing was an important part of the startup. In checking each loop, it was possible to determine if each part of each loop was operating correctly. In this way, the field device, its connections, cables to the DCS and hardware and configuration all the way to the logic address was verified.

The next step was to simulate as much of the system operation as possible without firing the boiler. By “Forcing” some of the inputs it was possible to emulate much of the firing sequence for each igniter and burner. With this advanced testing, the majority of any problems were found prior to boiler startup. This testing was able to be done while other boiler/turbine work was underway. Besides shortening the critical station startup cycle, this orderly troubleshooting brings a systematic approach to finding any of the simple problems. When the actual startup is underway, and the more difficult problems surface, most of the problems have already been resolved and the parties involved have confidence in the system in general.


The contract was originally based upon one engineer’s console for each unit. The console was placed in the electronics room for trouble shooting and supervisor’s use. Also included was a set of operator’s graphics to be available on that console. Operations were to be exclusively from the conventional control insert panel. As the operating personnel reviewed the system more extensively, it was decided to include an additional console for each unit in the control room. The additional information available and convenience made the operator’s console an easy choice.

The customer is making good use of the features of the system. The BMS is operating as intended. The information and interactive capabilities of the operator and engineer’s consoles are being used. The startup was completed according to schedule in concert with the other plant outage activities. Our ability to coordinate combustion equipment design with the burner management system and ancillary hardware helped to achieve the overall project objectives.

  1. Denike, A.C., Keller, G.Y. & Thorn, G.H. Gas Conversion At Roseton Station, Challenges And Solutions, American Power Conference, Chicago, Illinois, April 1991.
  2. Denike, A.C., Keller, G.Y, Thorn, G.H. & Uhl, G.L., The Roseton Gas Conversion – How To Convert 1200 MW To Gas Firing, Power-Gen ’92 Conference, Orlando, Florida, November 1992.
  3. Facchiano, Anthony, Use Of Computational Fluid Dynamics Modeling To Study Fuel Burnout In A Low NOx Retrofit Application.
  4. Londerville, Steve, Sources And Solutions Of Burner Related Rumbling Problems In Boilers, The American Society Of Mechanical Engineers, 1990.