IMPROVING THE PERFORMANCE OF PROCESS HEATERS THROUGH FIRESIDE MODELING

Transcription

1 Committed Individuals Solving Challenging Problems IMPROVING THE PERFORMANCE OF PROCESS HEATERS THROUGH FIRESIDE MODELING by M. Cremer, B. Adams, & M. Heap, REI P. Smith, Universtiy of Utah D. Brown, Stone & Webster Presented at the AFRC International Symposium, September, 1997, Chicago Illinois, USA. 77 West 200 South, Suite 210 Salt Lake City, Utah Telephone: FAX:

2 Improving the Performance of Process Heaters through Fireside Modeling 1.0 INTRODUCTION Page 1 Since 1990, Reaction Engineering International (REI) has been working with industrial clients to help solve difficult problems related to industrial systems encompassing utility boilers, pyrolysis furnaces, gas turbine combustors, rotary kilns, waste incinerators, smelting cyclones and others. The computational tools currently used by REI are based on software developed over the last seventeen years by Dr. Philip J. Smith, vice president of REI. These computer codes use a computational fluid dynamic (CFD)-based turbulent reacting flow solver which couples together the effects of turbulent fluid mechanics, homogeneous gas-phase and heterogeneous particle phase combustion chemistry, and conductive, convective, and radiative heat transfer. These tools simulate reacting and nonreacting flow of gases and particles, including gaseous diffusion flames, pulverized-coal flames, liquid sprays, coal slurries, isothermal and reacting two-phase flows, injected sorbents, and other oxidation/reduction systems. In this paper, discussion will be limited to the application of this software to current problems related to process heaters. The operation of process heaters requires exacting control of process fluid coil outlet temperatures and the heat flux distribution to the coils in order to prevent overheating which will accelerate the formation of coke on the inside walls of the process coils. Also, variations in coil outlet temperatures over time and from coil to coil are detrimental to process efficiency. The utility of these computational tools to the design of process heaters lies in their ability to provide full coupling of all relevant physical processes within the radiant firebox. Computer simulations can predict process outlet temperatures under different furnace operating conditions and improve the understanding of the dependence of a particular process on furnace conditions and, thus, lead to improved performance. The use of computer modeling has the potential to substantially reduce the risks associated with operating problems such as localized coil overheating by providing a means for establishing a reliable prediction of burner effects on furnace performance. This allows the designer to determine sensitivities to design changes without relying on simplified empirical models that are often extrapolated beyond the conditions for which their validity has been proven. Process heater case studies presented in this paper include investigation of: 1) the effect of flow pass balancing on variation of coil outlet temperature, 2) the effect of decoke nozzle location on overall firebox conditions and coke burnout during decoking operations, 3) predictions of tube temperatures and heat fluxes within the radiant and convective sections within a xylene splitter unit, and 4) the effect of refractory wall emissivity on conditions within the radiant firebox of a process heater. Case studies 1 and 2 were performed through an alliance between REI and Stone & Webster Engineering Corporation (SWEC), a worldwide leader in the design and manufacture of ethylene cracking furnaces.

3 Page 2 The remainder of the paper is organized as follows. In section 2, an overview of the modeling approach is given to highlight specific components of the process heater model. This is followed in section 3 by a discussion of the four case studies listed in the previous paragraph. The paper is summarized in section 4 where overall conclusions of this work are discussed. 2.0 BACKGROUND 2.1 Process Heater Model Overview The model used specifically for the simulation of process heaters provides full coupling between the "fireside" and "process side" computations. Heat transfer between the process coils and the hot radiant firebox depends on the process fluid temperature, the overall heat transfer coefficient from the process fluid to the hot coil surface, the fireside temperature distribution, and the local flow field. The heat transferred to the process fluid affects the local chemical composition and temperature which is then fed back to the fireside computations. Further details concerning how this is accomplished in the model are discussed in the following two sections. 2.2 Fireside Modeling Gas Phase Combustion The computational approach involves numerical discretization of the partial differential equation set which describes the physics of the system. Typically discrete computational nodes are needed to resolve the most relevant features of a three-dimensional combustion process. Around 60 variables (representing, e.g., gas velocity, temperature, concentration of various chemical species) are tracked at each node. Accurate simulation of the combustion processes requires accurate modeling of the dominant or controlling physical mechanisms in the process. Simulation of process temperatures in a process heater requires modeling of the flow patterns, reaction chemistry, gas and wall temperatures, heat transfer in the furnace (fireside), and heat transfer and temperature change in the process fluid (process side). In the computer model used here, coupled equations of chemical reaction, turbulent fluid flow, and convective and radiative heat transfer are solved to give a realistic and detailed model of the processes taking place within the fired zone. Turbulence is modeled using traditional methods of moment closure including Prandtl s mixing length model, the two-equation k-ε model (Launder and Spalding, 1972) and the nonlinear k-ε model (Speziale, 1987). In all simulations discussed in this report, the nonlinear k-ε model was used due to its more accurate prediction of normal Reynolds stress effects, allowing the prediction of secondary flows as occur in non-circular ducts. Within the model, the rate at which the combustion reactions occur is assumed to be limited by the rate of mixing between the fuel and the oxidizer. That is, the rate of chemical reactions is assumed to be fast compared to the rate of mixing (i.e. full chemical equilibrium is assumed), which is a reasonable assumption for the chemical reactions governing heat release. The

4 Page 3 thermochemical state at each spatial position is a function of the degree of mixing (parametrized by the mixture fraction, f), the mass fraction of particle off-gas (η), and the enthalpy (parametrized by the degree of heat loss, HL). The effect of turbulence on mean chemical composition is incorporated by assuming that the mixture fraction at each spatial position is described by a "clipped-gaussian" probability density function (PDF) having spatially varying mean and variance. Mean species concentrations are obtained by convolution over this assumed PDF. Since the rates governing the formation and destruction of NO x are of the same order of magnitude as those governing mixing, the assumption that chemical reaction is fast compared to mixing is inaccurate for these species. To account for the rate of formation of these species in the model, additional equations governing the formation of thermal, prompt, and fuel NO x are solved in a "post process" analysis. These equations are de-coupled from the equations governing the turbulent flow field since the formation of these trace species has a negligible effect on the velocity field. Thus, the converged velocity field is used in the equations governing NO x. Since radiation is typically the most significant mode of heat transfer to the process coils in a process furnace, it is critical that the radiation field be accurately represented. Accurately simulating radiative transfer to specific regions in a system requires a model which can account for both absorbing-emitting radiation processes and complex system geometries, including arbitrary structures such as process coils. Additionally, it is desirable that any radiative model selected be computationally efficient in terms of execution time and storage to allow coupling with other routines in a comprehensive combustion model. REI s process heater model utilizes the discrete-ordinates method which has been shown to be a good choice for modeling radiation in combustion systems, both in terms of computational efficiency and accuracy. This method retains the directional dependency of the radiation intensity in a way that other flux models are unable to achieve, yet provides for a finite-difference or finite-volume solution that is more computationally efficient than zone methods and more deterministic than Monte Carlo methods. The development of the discrete-ordinates method and its application to a number of complex geometries (e.g., Adams, 1993; Adams and Smith, 1993) have been presented in the literature and serve to validate the use of this method in accurately modeling radiative heat transfer in process heaters (Adams, 1993; Adams & Smith, 1993; Adams & Smith, 1995) Particle Phase Combustion Along with the capability of modeling gas phase combustion and turbulent transport, the model is also equipped to handle condensed phase turbulent transport, dispersion, and reaction. This capability makes possible the simulation of practical combustion processes which use solid particles, liquid droplets, or slurries as fuels in turbulent environments. In the context of modeling cracking furnaces, this capability makes possible the simulation of oil or coal fired furnaces or coke particle burnout within the radiant firebox during decoking operations. Predictions of simulations of this latter application are discussed in section 3.2.

5 Page 4 The turbulent transport of particles is solved for in a Lagrangian reference frame by modeling the time evolution of a probability density function (PDF) for the particle position. The value of this time evolving PDF, at any location, represents the probability of finding particles of the corresponding type and starting position, with that residence time, at that location in the flow field. This probability is used to obtain the expected number density of particles with the corresponding properties in each computational volume. The contribution to mass, momentum, and energy by these particles in each Eulerian computational cell is added dynamically while tracking the particle PDF to provide source terms which are coupled into the Eulerian gas phase transport equations. The mean particle position and its variance are expressed as ordinary differential equations, with the aerodynamic drag force and weight providing the main driving force for mean particle position change and the fluctuating component of velocity determining the variance about this value. Further details describing the methodology used for modeling turbulent particle transport are available elsewhere (Baxter, 1989; Jain, 1996, Smoot, 1985). Within the model, the three mechanisms for mass loss of each particle are vaporization, devolatilization, and heterogeneous reaction. The overall rate of reaction of each type of particle is found by solution of governing continuity equations for each of these processes. The kinetic parameters that are used are based on independent experimental observations and kinetic parameters deduced from these observations. In the context of coke particles in a process heater, the particle is considered to consist of four components: liquid, raw coke, char, and ash. Reaction rates for these components are computed based on local gas properties which affect heat transfer and mass transfer to the particle. On the other hand, mass evolved from each particle due to vaporization, devolatilization, or char oxidation is locally incorporated into the Eulerian gas phase computations. In particular, this approach leads to a comprehensive strategy for representing coke particle transport and burnout within the radiant firebox of a full-scale process heater. Further specific details concerning particle phase reactions are available elsewhere (Baxter, 1989; Smoot, 1985). 2.3 Process Side Modeling To accurately predict heat transfer to the process side fluid, the fireside conditions must be coupled with the process side. The energy absorbed into the process coils acts to heat up the fluid and provide energy for the endothermic cracking reactions. This process temperature affects the convective heat transfer on the inside tube wall as well as the outside tube metal temperature and finally, the net heat flux to the coil. Thus, the process and fireside conditions are tightly coupled and must be modeled as such. The approach used for providing this coupling in the simulations discussed in this report was as follows. Using a detailed simulation of the kinetics and heat transfer inside the tubes coupled with a relatively simple model of external heat transfer to the tubes, SWEC obtained tabulated data of process temperature and heat transfer coefficient versus cumulative absorbed duty for each firing condition. Since the overall heat transfer coefficient is dependent on tube diameter, this correlation was also tabulated as a function of tube diameter. This relationship was then used within the comprehensive model to provide the link between absorbed duty, heat transfer

6 Page 5 coefficient, and process temperature. In other words, given the absorbed duty as predicted by the comprehensive model, this relationship provided the appropriate process temperature and heat transfer coefficient at each grid location along the process coil. In principle, the process side chemistry model should be fully coupled with the fireside combustion and heat transfer model since variation in the external heat transfer to the process coils, as predicted by the comprehensive model, could potentially affect the relationship between absorbed duty and process temperature. However, calculations performed by SWEC indicate that this level of coupling is not necessary and that the simplified model of external heat transfer, based on a given heat flux profile, is sufficiently accurate to identify the relationship between absorbed duty and process temperature. This is not to say that the external heat flux profile is not important with respect to its effect on process temperature, but that the distribution of heat transfer to the tubes has a relatively small effect on the correlation between cumulative absorbed duty and process temperature. 2.4 Model Inputs In addition to the necessary geometrical information, simulation of conditions within the radiant section of a process heater requires the specification of several model inputs. Among these are the following: Tube model Tube thermal resistance Refractory wall thermal resistance Tube and refractory wall emissivity Fuel composition, temperature, and mass flow rate Oxidizer composition, temperature, and mass flow rate The manner in which conditions on the process side are coupled with those on the fireside is what is here termed the "tube model". The model used in the present simulations involves using a tabulated correlation between cumulative absorbed energy, process temperature, and the process side heat transfer coefficient. The internal heat transfer coefficient as well as the tube wall thickness and thermal conductivity are used to determine the overall tube thermal resistance, R t, between the fireside tube wall and the process side fluid. This wall resistance represents the total resistance to heat transfer by series conduction through the tube wall, the coke layer formed on the inside wall, and by convection from the inside tube wall to the process fluid. Generally, the fireside tube metal temperature (TMT) is not known, a priori, and must be determined from an energy balance at the surface. The balance may be written as q" rad + q" conv = q" cond = (T s - T b ) / R t (1)

7 Page 6 where q" rad, q" conv, and q" cond represent net radiative, convective, and conductive fluxes, respectively, at the surface. Assuming that the flameside convection coefficient (computed by the model), incident radiative flux, surface emissivity, backside temperature (T b, process temperature), and the conductive thermal resistance (R t ) are known, the surface temperature, T s, can be found. In a similar manner, thermal boundary conditions for the furnace refractory walls must be specified. As a percentage of the furnace fired duty, heat loss through the furnace walls typically ranges from 1-3%. Although small, this heat loss is not negligible and the spatial dependency of these losses within the furnace should be modeled. The thermal resistance, R t, between the fireside refractory wall and the outside ambient air is specified taking into account the conduction through the refractory material and convection from the outside wall to ambient conditions. As R t approaches infinity, heat loss through the furnace walls approaches zero. Using Eq. 1, along with assumed data for refractory wall emissivity, the furnace wall temperature, T s, can be computed. Finally, thermo-chemical data for all inlets must be specified. Usually this requires specification of temperature, composition, and mass flow rate of a fuel stream and an air stream. When twophase flows are modeled, as in the simulation of decoking operations inside the firebox of a process heater, the composition, temperature, and mass flow rate of the particulate phase must also be specified. 3.0 CASE STUDIES Computer simulations were recently carried out for three full scale process heaters. Two of these simulations, case studies #1 and 2, were for full scale ethylene cracking furnaces, and the other, case study #3, was for a full scale xylene splitter reboiler (XSR) unit. The primary objectives of the ethylene furnace simulations were to: Confirm required fuel firing rate to achieve the specified heat transfer to the coils Predict the coil to coil variation in process fluid outlet temperature Identify hot spots on process tubes Quantify the amount of flow pass balancing required to minimize coil outlet temperature (COT) deviations Predict the overall flow field, coke particle trajectories, and burnout during decoking back to the firebox The simulations of the ethylene furnaces were carried out as part of the design process prior to the manufacture of the units.

8 The simulations of the XSR were conducted to aid in solving problems which were being experienced during normal unit operation. Areas of importance were: Hot spots on process tubes High bridgewall/crossover temperatures Improved thermal efficiency and product yield Page 7 In addition to these three case studies, results from some preliminary work related to model advancement is presented as case study #4. This work involves the implementation of a twoband, weighted sum of gray gases radiation model to account for spectral dependency of radiation heat transfer. The current radiation model is based on a gray gas approach which does not explicitly account for variation of absorption and transmission through participating media as a function of wavelength. The new approach takes this spectral dependence into account so that the effects of changing furnace wall emissivity on radiant efficiency can be more accurately represented. 3.1 Case Study #1 - Normal Firebox Operation The first radiant firebox modeled is approximately 35 ft. high, 60 ft. long, and 10.0 ft. wide. There are twelve simple serpentine M coils aligned vertically in a plane passing through the center of the furnace as shown in Fig. 1. Each coil consists of three vertical inlet passes or tubes and three vertical outlet passes. The coil is swaged, so that the three final passes are larger in diameter than the three inlet passes. The flue-gas exit into the convective pass runs the length of the furnace along one sidewall and is adjacent to the radiant roof. The radiant firebox is symmetrical about a plane passing through the center of the furnace running perpendicular to the sidewalls. The advantage of this symmetry to the simulation is that only one-half of the furnace needed to be modeled. This was accomplished by placing a symmetry boundary condition through the center as shown in Fig. 1. The firebox is completely floor-fired using 24 burners distributed in two rows placed adjacent to the furnace walls. The burners are spaced so that the centerline of each burner corresponded with the centerline of the opposing process coil as shown in Fig. 1. The burners are nonpremixed staged gas burners with internal flue gas recirculation (John Zink designation PSFFR) as shown in Fig. 2. These burners are designed to fire up along the walls to produce relatively long flames so that the heat release profile is relatively smooth. In addition, the burners are designed to recirculate flue gas to reduce NO x formation Baseline Simulations For this furnace, baseline simulations were performed for three firing conditions: Normal firing conditions Maximum firing conditions Design firing conditions

9 Page 8 Model inputs which were varied for these three cases included the overall fuel firing rate, the feedstock, both quantity and composition, (this affected the tube model), and the fuel composition. The correlations between process side temperature and cumulative absorbed duty, based on SWEC analyses, that were used for these simulations are shown in Fig. 3. These correlations were used in the tube model as discussed in Sec. 2.4 to represent the relationship between cumulative absorbed duty and process temperature and thermal resistance. The simulations for all three cases indicated the presence of a significant cool end wall effect at a constant firing rate and constant flow to each coil. The predicted COTs indicating this behavior are shown in Fig. 4. Coils 1 and 2, adjacent to the end wall, run coolest and coils 3 and 4 run hottest. This variation can be seen to be related to the distribution of refractory wall temperature and the local gas temperature. For all three modeled conditions, Fig. 5 shows that as a function of furnace length, the refractory wall temperature mirrors the temperature distribution of the coils. The wall temperature increases from a minimum near the end wall and peaks in the vicinity of the third burner from the end wall. This behavior is also apparent in the distribution of gas temperature as seen in Fig. 6 for the normal firing condition. The cooler end wall temperature is explicitly shown in Fig. 7 in which its temperature is compared with that of the front and back furnace wall. Fig. 7 shows that the end wall temperature is significantly cooler than the adjacent walls. The overall temperature of the front wall is also predicted to be noticeably cooler than the back wall (where the back wall is defined to be the wall having the flue gas take off running along the top). It is believed that the temperature distribution within the furnace can be explained in part by the velocity field within the firebox. The three-dimensional flow field predicted for this furnace is relatively complex. The hot gas adjacent to the furnace walls flows upward from the floor burners. Some of this gas proceeds out the flue gas exit, but a portion of it turns toward the center of the furnace and flows downward adjacent to the tubes and is later re-entrained by the burners near the floor. The asymmetrical placement of the flue gas outlet results in the asymmetrical flow on either side of the row of coils. One effect of this asymmetry is a difference in temperature between the front wall of the furnace (the wall opposite the flue gas outlet) and the back wall, as can be seen in Fig. 7. This difference in temperature is most dramatic in the upper half of the furnace. This effect can be seen in Fig. 8 which shows front and back wall temperature versus furnace height for the normal firing condition. The suspected reason for the temperature difference is that hot gas from the floor burners travels up the entire length of the back wall prior to turning and exiting out the flue gas outlet along this wall. On the other hand, gas from the floor burners along the front wall travels only part of the distance up the wall prior to separating from the wall to turn and exit out the flue gas outlet along the opposite side. This separation causes a region of recirculating gas along the top of the front wall which is relatively cool, and thereby cools the upper half of the front wall. In a similar manner, the noticeably cooler temperature of the furnace end wall may be explained. Fig. 7 shows refractory temperature versus wall length for the end wall, front wall, and back wall. Clearly, the end wall of the furnace is significantly cooler than the other two walls. The predicted velocity field shows the existence of relatively cool recirculating gas toward the top of

10 Page 9 the end wall. This gas has a similar cooling effect on the end wall as was discussed before for the front wall. The effect of this relatively cool end wall is reduced radiative heat transfer to the end coils. The effects of this are witnessed as a reduction in process fluid temperature, coil outlet temperature, etc Flow Pass Balancing An important control objective is to minimize the deviation in COTs from coil to coil so that, as far as possible, each coil operates at the same conversion and cokes at the same rate. In a previous paper (Brown, Smith and Adams, 1994) the efficacy of zone firing was evaluated. For this furnace flow pass balancing is used with one convection pass for each symmetrical, adjacent pair of back to back coils. Simulations for all three of the firing conditions modeled in the baseline study predicted a significant effect of the end walls on heat transfer to the coils. The predictions indicated a dominant profile in the COTs in which coils 1 and 2 have the lowest COT and coils 3 and 4 the highest. The mass flow rates through the process coils are increased if the COT is too high or decreased if the COT is too low. The relative amount of modulation is governed by: q 2 = q 1 (m 2 / m 1 ) (2) where the process fluid temperature T corresponds to the cumulative absorbed duty q 1 (q 2 ) and process mass flow rate m1 (m 2 ). This expression reflects conservation of energy in that when the process mass flow rate is increased, (m 2 / m 1 > 1) a larger amount of energy, q 2, must be absorbed to bring the process temperature up to the same level, T. Simulations to test the feasibility of using flow rate modulation had to abide by the constraints of the expected control system for the overall fuel firing rate and the process mass flow rates. The overall fuel firing rate is governed by the average of the 12 coil outlet temperatures, and the mass flow rates to adjacent coil pairs is varied depending on the pair-wise averages of the COTs. The objective of this system is to obtain: for i=1,3,5,7,9, and 11. i / 2 COT( n) = 1/ 2 COT( n) i (3) To demonstrate that the statement of Eq. 3 is achievable using the control system described above, a simulation was performed using model inputs for the normal firing condition. The level of flow rate modulation predicted prior to the simulation was based on the energy transferred to each pair of coils in the baseline simulation of normal firing conditions. Based on this 1

11 Page 10 information and application of Eq. 2, a prediction of the percentage decrease or increase in the mass flow rate was made and input to the model. The total process flow rate was not changed. Flow rate modulation to adjacent coil pairs did not exceed 1.3% of the baseline mass flow rates. The mass flow rate to coils 1 and 2 was decreased, that to coils 3 and 4 was increased, and that to coils 5 and 6 was decreased. Fig. 9 shows the predicted pairwise average COTs from the simulation employing flow pass balancing compared with those obtained in the baseline simulation in which no flow pass balancing was used. This simulation clearly indicated that improved uniformity in COTs would be achieved using small levels of flow pass balancing. 3.2 Case Study #2 - Decoking Operations Thermal cracking of hydrocarbons is always accompanied by coke formation. During normal operation, coke slowly deposits on the inside walls of the process coils, leading to reduced heat transfer to the process fluid and increased pressure drop, and eventually plugged coils in the most extreme case. Depending on the feedstock, the severity of cracking, and the type of furnace, the run length can vary from days to months. To burn off the coke from the tube walls, the furnace has to be periodically shut down to perform decoking operations. Typically, a combination of high temperature air and steam is passed through the process coils to both blast the coke off the walls and partially oxidize it inside the coils. The unburned coke and gaseous products can then either be collected for disposal or rerouted back into the firebox for subsequent burnout. Simulations discussed in this section relate to modeling conditions inside the firebox during the latter type of decoke operation. Issues of concern in burning out the residual coke inside the firebox that were addressed in these simulations included: Effect of the decoke effluent on the overall flow and heat transfer inside the firebox Trajectories of the coke particles as they relate to impingement on tube and refractory walls The efficiency of coke particle burnout inside the radiant firebox Although process side conditions were coupled with fireside conditions as previously discussed, particular issues related to coke burnout inside the process tubes were not addressed in these simulations. The effluent exiting the decoke nozzles was assumed to be at a fixed mass flow rate, temperature, gas composition, particle mass fraction, and particle size distribution based on SWEC analysis and empirical data. Thus, the decoke effluent properties were not directly coupled with the predicted conditions inside the process coils. Process temperature inside the process coils as well as the tube metal temperature was computed as before using the approach outlined in section 2.3. Decoking conditions were simulated in the firebox shown in Fig. 10. This furnace is approximately 35 ft. high, 100 ft. long, and 10 ft. wide and has 96 U-coils grouped into 8 coil

12 Page 11 modules (of 12 coils each) aligned along the center of the furnace. The furnace is fired by 32 John Zink Co. PSFFR burners positioned uniformly along the opposing furnace walls as shown in Fig. 11. The eight decoke nozzles were placed in the floor and positioned between the first and second burners and the fifth and sixth burners from the end wall as also shown in Fig. 11. Due to the symmetry of this furnace only one-half was modeled and a symmetry boundary condition was used at the furnace centerline. Notable observations that were made in the simulation of decoking were the following: The high momentum flow from the decoke nozzles governs the overall flow pattern in the furnace Nozzle size significantly alters the overall fluid mechanics in the furnace The smallest coke particles burnout quite efficiently while being convected by the recirculating gases in the furnace. The largest coke particles fall back to the floor where they burnout without ever reaching the radiant roof. A portion of the intermediate particle sizes have the potential of exiting the radiant firebox prior to completely burning out. 3.3 Case Study #3 - Xylene Splitter Reboiler The primary objective of this study was to demonstrate the capability of computer simulation for operation, design optimization, and trouble shooting by comparing the simulation results with measurements provided by the client and suggesting performance improvements and solutions to current problems Furnace Description The function of the xylene splitter reboiler (XSR) is to vaporize liquid xylene. It is a cylindrical process heater with coils that have both a radiant and convective component. The heater is operated by controlling the coil outlet temperature (COT) and using the crossover (bridge wall) temperature as a reference point. The COT is normally controlled to be about 5 to 6 F above the coil inlet temperature (CIT) and the furnace yields approximately forty percent xylene vapor in the process. Fig. 12 shows a schematic of the XSR heater. The radiant furnace is a cylindrical furnace with six up-fired burners located on the furnace floor as shown in the top view in Fig. 12. The burners are natural-draft, duel-fuel burners firing 95% No. 6 fuel oil and the remainder fuel gas. The combustion gases exiting the radiant furnace flow through the rectangular convective section which is placed above the radiant furnace as seen in the side view in Fig. 12. Four process coils enter from the top of the convective section and pass horizontally back and forth through the convective section as they move down vertically (connected by jumpers between each elevation). There are 9 tube elevations (horizontal tube rows) in the convective section. Fig. 13 illustrates how the individual tubes are numbered in the convective section. The bottom three rows of tubes

13 Page 12 are bare tubes and the top six rows are finned. After exiting the convective section, the four coils leave the furnace, and travel to the top of the radiant section (two coils on each side, 180 degrees apart). The coils enter the top of the radiant section and move down and up near the outer wall of the radiant furnace (the burners are in the middle of the furnace). Each coil covers one-fourth (90 degrees) of the furnace circumference as it passes around the outer wall. All four coils exit the bottom of the radiant furnace. Liquid xylene enters the convective section and a mixture of liquid xylene and xylene vapor exit the radiant furnace Simulation Results Four simulations of the XSR were carried out: Baseline High excess air Low excess air All gaseous fuel Predictions were obtained on pressure drop through the furnace, furnace flow patterns, heat release profiles, gas temperature profiles, pollutant emissions, incident and net heat flux and tube skin temperature as a function of tube length, and percent vapor as a function of tube length. The predicted overall flow field in the radiant furnace can be characterized by upward flow in the center of the furnace generated by the six burners. The flames were predicted to be relatively long. The upward flow from the burners entrains furnace gases from the outside walls causing downward flow along the walls. Although the velocity profile smooths out with elevation in the radiant furnace, it remains significantly nonuniform at the inlet to the convective section. The nonuniform flow field at the entrance to the convective section has a significant effect on gas properties. The radial profiles shown in Fig. 14 compare mean gas temperature and oxygen. These radial profiles are taken through the center of one burner. It can be seen that the oxygen concentration dips almost to zero where the gas temperature peaks at the ten foot level. This corresponds to a region of high carbon monoxide concentration. Moving up the furnace the profiles become more smooth but neither the temperature nor the oxygen concentration is uniform at the entry to the convective section. This is clearly shown in the contour plot of temperature in Fig. 15. The gas temperatures are predicted to vary by greater than 80 F at the entry to the convective section. Predicted tube skin temperature and incident and net heat flux are shown in Fig. 16 for one of the process coils (D) in the radiant furnace. The incident heat flux is simply the incident radiant flux while the net heat flux is the net radiant heat flux plus the convective flux. Peak incident heat flux to the process coils in the radiant furnace varies with tube height and location. Since there are six burners, the furnace is not symmetric with respect to the individual tubes and the burners. Therefore, all tubes are not equal distances from the radiant source. The peak heat flux occurs at approximately one third of the distance from the floor to the roof, which corresponds to the

14 Page 13 maximum flame emission. The peak skin temperature occurs at the same location as the peak net heat flux. Maximum measured tube skin temperatures in the radiant furnace were approximately 760 F and the predicted maximum was 750 F. This is very good agreement considering possible errors in measurement and burner input conditions. Predicted tube skin temperature and incident and net heat flux for coil D in the convective section are shown in Fig. 17. To understand these predictions, it is important to understand the tube arrangement that is illustrated in Fig. 13. Four coils pass through the furnace: A, B, C, and D. Each coil has two passes in each of nine rows as shown. The bottom three rows of tubes are bare and the top 6 rows are finned. In Fig. 17 it can be seen that tubes 17 and 18 have the highest incident heat flux since they can "see" the radiant furnace. The incident flux drops off quickly moving upwards through the bundle because successive rows are shielded from the furnace by the lower tubes. The convective contribution to the net heat flux to the bare tubes is negligible, but for the finned tubes it is dominant. As shown in Fig. 17, the even numbered finned tubes, located adjacent to the wall, have higher tube skin temperatures than their adjacent odd numbered tubes. This is a consequence of higher convective heat transfer to these tubes due to the higher velocity gas traveling through the gap adjacent to the wall. It can be seen that the highest tube skin temperature occurs in the convective section and is located on tube pass 12 which is in the lowest row of finned tubes. 3.4 Case Study #4 - Effect of Wall Emissivity Theory and Background There has currently been much discussion within the ethylene producer's community about the effect of high emissivity refractory coatings on heat transfer to the process coils in a radiant heater. The intent of these coatings is to raise the emissivity of the refractory walls from a nominal value of 0.65 (at high temperature) to 0.9 or above. Based on actual furnace measurements, one ethylene manufacturer has reported that the application of these types of coatings results in greater overall furnace performance and efficiency by increased radiative heat transfer from the refractory walls to the process coils (Owen, et al., 1997). Figure 18 shows the expected theoretical effect of wall emissivity on the partitioning between the energy that is radiated versus the energy that is reflected by the walls. Assuming that the heat lost through the walls is the same in both cases, approximately 5%, then when the refractory wall emissivity is increased from 0.65 to 0.9, the amount of incident energy that is absorbed and reradiated increases from 60% to 85%. The amount of reflected energy decreases from 35% to 10%. The underlying reason why this may be significant regarding the amount of energy eventually absorbed by the process coils can be explained in terms of spectral radiation effects (Hottel and Sarofim, 1967) Results REI is in the process of adding capability to their simulation software to accurately model the effects of spectral radiation heat transfer to account for changes in refractory emissivity. The new

15 Page 14 model is a 2-band weighted sum of gray gases model which improves on the previous 1-band gray model. Preliminary spectral model comparisons with theory and with the old model demonstrate the expected improvement. Fig. 19 shows plotted predictions of net heat flux versus wall emissivity. Theoretical values were obtained for a parallel plate case in which one wall is assumed adiabatic with variable emissivity and the opposing wall (absorbing wall) has a fixed temperature and a fixed emissivity of 1.0. The model calculations were performed for a box with a 20:1 aspect ratio to approximate the parallel plate case. As expected, the gray model predicts no difference in net heat flux to the absorbing surface as a function of adiabatic wall emissivity even though theory predicts an increase in net flux from 64 to 89 kw/m 2 when the emissivity changes from 0 to 1. It can be seen in Fig. 19 that the 2-band spectral model predictions agree very well with the theoretical values at emissivities of 0 and 1 and that the trend for intermediate emissivities is as expected. REI is currently investigating the effect of wall emissivity on radiant efficiency in full scale process heaters using the improved spectral radiation model within its 3-D turbulent reacting flow software. These predictions will be reported at a later date. 4.0 CONCLUSIONS The three dimensional simulation of turbulent combustion and heat transfer within process heaters provides for a detailed examination of the effects of design and operating changes on process side conditions. Since the fireside conditions are coupled with the process side conditions, nonlinear effects associated with changing firing rates, coil and furnace geometry, firing distribution, feedstock, and process flow rates can be examined in a rigorous manner. In this paper, it was shown how these simulations can be used to improve performance and efficiency in process heaters. The simulations of case study #1 indicated that due to the nature of the fluid mechanics and heat transfer within the radiant firebox, COTs could vary significantly under uniform firing conditions and process flow rates. The simulations indicated that flow pass balancing should provide an acceptable solution to achieving uniform COTs in this system. Case study #2 and other simulations like it are providing useful information that is helping to determine the best design and placement of the decoking nozzles to minimize coke particle impingement on walls and enhance burnout in the radiant box. The simulations of case study #3 were useful in explaining variation in the bridgewall temperature as well as providing a solution to reducing excessive tube skin temperatures in the convective pass of the xylene splitter reboiler. Case study #4 illustrates how modeling improvements continue to be made to enhance predictive capability. In particular, this work is leading to providing meaningful predictions of the effect of refractory wall emissivity on radiant efficiency in process heaters.

16 Page 15 Fig. 1 Fig. 2

17 Page 16 Fig. 3

18 Page 17 Fig. 4

19 Page 18 Fig. 5

20 Page 19 Fig. 6

21 Page 20 Fig. 7

22 Page 21 Fig. 8

23 Page 22 Fig. 9 Fig. 10

24 Page 23 Fig. 11

25 Page 24 Fig. 12

26 Page 25 Fig. 13

27 Page 26 Fig. 14

28 Page 27 Fig. 15

29 Page 28 Fig. 16

30 Page 29 Fig. 17

31 Page 30 Fig. 18 Fig. 19

32 REFERENCES Page 31 Adams, B. R., "Computational evaluation of mechanisms affecting radiation in gas and coal fired industrial furnaces," Ph.D. Dissertation, Department of Mechanical Engineering, University of Utah, (1993). Adams, B. R., and Smith, P. J., "Three-Dimensional discrete-ordinates modeling of radiative transfer in a geometrically complex furnace," Combust. Sci. and Tech., 88, 293 (1993). Adams, B. R., and Smith, P. J., "Modeling effects of soot and turbulence-radiation coupling on radiative transfer in turbulent gaseous combustion," Combust. Sci. and Tech., 109, 121 (1995). Baxter, L. L., "Turbulent transport of particles," Ph.D. Dissertation, Department of Chemical Engineering, Brigham Young University, Provo, UT (1989). Brown, D. J., Smith, P. J. and Adams, B. R., "Cracking Furnace Fireside Modeling Advances," Paper presented at the AIChE Spring National Meeting, EPC Conference, Atlanta, GA (1994). Brown, D. J., Cremer, M. A., Smith, P. J., and Waibel, R. T., "Fireside Modeling in Cracking Furnaces," Paper presented at the AIChE Spring National Meeting, EPC Conference, Houston, TX (1997). Jain, S., "Three-Dimensional simulation of turbulent particle dispersion applications," Ph.D. Dissertation, Department of Chemical and Fuels Engineering, University of Utah (1996). Hottel, H. C., and Sarofim, A. F., Radiative Transfer, McGraw-Hill, NY, 1967, pp Owen, S., Wray, C., and Ennis, M., "High Emissivity Refractory Coating - The effect on efficiency improvement in a pyrolysis furnace," presented at AIChE Spring National Meeting - Eighth annual ethylene producerõs conference, March 11, 1997, Houston, TX. Smoot, L. D. and Smith, P. J., Coal Combustion and Gasification, Plenum Press, New York (1985).