You are on page 1of 10

Wear 252 (2002) 179188

On numerical modeling of particlewall impaction in relation to erosion prediction: Eulerian versus Lagrangian method
B.E. Lee, J.Y. Tu, C.A.J. Fletcher
Centre for Advanced Numerical Computation in Engineering and Science (CANCES), The University of New South Wales, Sydney 2052, Australia Received 31 January 2001; received in revised form 18 June 2001; accepted 31 July 2001

Abstract The modeling of particlewall impaction in a conned gas-particle ow using both Lagrangian and Eulerian approaches is reported. The Lagrangian method is based on a general computational uid dynamics (CFD) code, FLUENT (FLUENT-4.3, 1996). In the Eulerian method, based on our previously developed code [J. Eng. Gas Turb. Power 119 (1997) 709], a computational procedure by decomposing one Eulerian solution of particulate phase into two equivalent Lagrangian solutions for incident and reected particles has been developed. These two approaches are evaluated versus experimental data for particlewall impaction using spray droplets. Two test cases, a 45 ramp and an isolated single tube, have been studied using the above two approaches to determine the particle behavior and physical properties of impacting and reected particles near wall surface. Results show that both approaches are successful in predicting the main features of particulate ow near wall, however, the Eulerian approach is much less expensive than the Lagrangian approach in obtaining the ow solution of impacting particles. The particulate ow predictions using both approaches have been applied for predicting tube erosions that are compared with reported data. Good agreement between predictions using the two approaches and between the predicted and measured erosion results are observed. 2002 Published by Elsevier Science B.V.
Keywords: Impaction; Slip boundary; Impact velocity

1. Introduction The deposition and erosion of target metals by the impaction of particles are found in many industrial multiphase ow systems, such as in uidized beds and coal-red power utility boilers. It has been pointed out in literature [1] that for the same conditions of target tube material and properties of impacting particles, reduction in the momentum of the impinging particles and particle number density at impact can greatly reduce the erosive wear of tubes. Therefore, it is important to model particlewall impaction in conned gas-particle ow for predicting particle erosion and deposition on the wall surface. There are two basic approaches commonly used to predict gas-particle ows: Lagrangian and Eulerian formulations. In the Lagrangian formulation, the motion of a single particle is considered and relevant variables are calculated along the particle trajectory. One disadvantage of this approach is that it is computationally expensive for engineering applications because of the great number of particles which are required to determine the average behavior of the rela Corresponding author. Tel.: +61-2-93180004; fax: +61-2-93192328. E-mail address: clive.etcher@unsw.edu.au (C.A.J. Fletcher).

tively high particle loading. The Lagrangian formulation, however, is a more fundamental procedure to describe the particlewall collision process and can yield a detailed physical description of individual particle motion. The Eulerian formulation treats both gas and particulate ows as continua and the phases are regarded as two mutually interacting uids. The main advantage of using the Eulerian method is to make computation fairly economical for ows with relatively high concentrations of particles and for the purpose of engineering designs, such as power boiler design [2]. In addition, effects of interactions, particularly turbulence, between two phases (two-way coupling) are more easily considered by using the Eulerian approach [3]. However, some difculties in using the Eulerian method exist in the prediction of particle erosion and deposition because the Eulerian approach gives mean values of the particulate phase over a small control volume where both incident and reected particles contribute to this mean value near wall surface. Pourahmadi and Humphrey [4] predicted the erosion using the Eulerian approach with the slip boundary condition. This type of slip boundary condition may lead to a substantial error in the prediction of erosion. For instance, in the case of normal impaction, mean Eulerian velocity would be 0, indicating that no erosion occurs.

0043-1648/02/$ see front matter 2002 Published by Elsevier Science B.V. PII: S 0 0 4 3 - 1 6 4 8 ( 0 1 ) 0 0 8 3 8 - 9

180

B.E. Lee et al. / Wear 252 (2002) 179188

Nomenclature Cp D Dp E K n St Th Ui U0 particulate concentration diameter of cylinder particle size erosion rate erosion constant velocity exponent 2 Stokes number ((p Dp /18) (U0 /D)) homologous temperature ratio impact velocity inlet velocity

Greek symbols impact angle viscosity of gas p density of the particle chemical composition of y-ash particles Moreover, the slip boundary condition cannot directly give physical properties of impacting particles including impact velocity and angle as well as particle number at impact. In this paper, both Lagrangian and Eulerian approaches are used to model particlewall impaction in conned gas-particle ow. The Lagrangian method is based on a general computational uid dynamics (CFD) code, FLUENT [5], while the Eulerian method is based on our previously developed code [2]. A new computational procedure is proposed for the Eulerian approach to estimate the equivalent Lagrangian solutions for incident and reected particles near wall surface. Thus, the particle properties, such as impact velocity and angle, and concentration of impacting particles, which are important parameters in the prediction of erosion and deposition, can be obtained using the present Eulerian method. The predictions using both methods are compared with experimental data of Rudoff [6] for particle wall impaction using spray droplets. Results from these two approaches for gas-particle ow over a 45 ramp and an isolated single tube are compared and show that good agreement is achieved. Finally, the predictions of tube erosion based on the above particle solutions of both approaches are compared with the reported data.

algorithm resulting in a set of algebraic equations, which are solved using a line-by-line tridiagonal matrix algorithm, achieves pressure/velocity coupling. An RNG-based k turbulence model is chosen to model gas turbulent uctuations. A Lagrangian-formulated particle equation of motion is solved via an advanced RungeKutta method to predict particle velocity and trajectories once the gas ow eld is obtained. FLUENT uses a stochastic method that incorporates the instantaneous values of gas velocity including uctuation components appearing in the equation of particle motion. Upon striking a wall surface, a particle is forced to rebound according to the prescribed restitution coefcients. Particle fragmentation and/or particle rotation are not considered. By tracking a statistically signicant number of particles, overall pictures of the mean particulate ow eld can be obtained based on the concept of conservation. The mean velocity and concentration of incident and reected particles near wall surface can also be estimated from particle trajectories. More detailed descriptions of the Lagrangian method are provided in the FLUENT users guide manual [5]. 2.2. Eulerian method In the Eulerian approach, the two phases are considered to be separate interpenetrating continua and separate (but coupled) equations of motion with separate boundary conditions are solved for each phase. A previously developed computer code with a two-uid model [2,7] is employed to solve the complex conned gas-particle ow. The gas turbulence is modeled by an RNG-based k model [8] with a modication for conned gas-particle ow [3]. A two-layer wall function is used to avoid the need for very ne mesh distributions close to solid surfaces. The code uses a nite volume discretization in conjunction with a non-staggered generalized-coordinate grid. The grid generation process is handled independently with the corner point locations read in from a separate database, which can be generated by users or from a commercially available mesh generation package. The convective terms are discretized at the faces of control volumes with a generalized QUICK convective differencing method [9]. Second-order derivatives are evaluated using three-point symmetric formulae. A velocity potential correction [10] is introduced to ensure that the continuity equation of the gas phase is satised and to upgrade the gas pressure using the SIMPLEC algorithm [11]. The stored values at the centroids of control volumes are interpolated and modied to calculate the ux at faces of the control volumes using the moment interpolation method [12]. The governing equations for both the gas and particulate phases are solved sequentially at each iteration to obtain all the dependent variables. At every global step each equation is iterated, using a strongly implicit procedure [13]. This iteration process is continued until the equation residuals are sufciently smalltypically, the mass should be conserved to be approximately 0.001%.

2. Computational methods 2.1. Lagrangian method The Lagrangian method is based on a continuum model for gas phase and a discrete method for particulate phase. A general CFD code, FLUENT, is used for the Lagrangian simulation of conned gas-particle ow. FLUENT solves the governing equations of gas ow by a nite-volume formulation on a non-orthogonal, curvilinear coordinate grid system using a collocated variable arrangement. The SIMPLEC

B.E. Lee et al. / Wear 252 (2002) 179188

181

Fig. 1. Lagrangian simulations of particle trajectories over a cylindrical body at different Stokes numbers: (a) St = 0.17; (b) St = 1.58; (c) St = 4.40.

Fig. 2. Lagrangian particle trajectories of a 45 ramp ow (U 0 = 10 m/s, D p = 35 m).

As noted earlier, the Eulerian description of the particulate phase near the wall gives a mean value that contains the information contributed from both incident and reected particles. In order to decompose this Eulerian solution of the

particulate phase near the wall, the following observation is made from Lagrangian simulations as shown in Figs. 1 and 2. It is found from these simulations that the information extracted from the reected particles from the wall surface

Fig. 3. Illustration of computational domain and boundary conditions for a 45 ramp ow.

Fig. 4. Illustration of the experiment using spray droplets [6].

182

B.E. Lee et al. / Wear 252 (2002) 179188

Fig. 5. Comparison between predicted and measured axial velocities of droplets along the stagnation streamline (U 0 = 15 m/s).

Fig. 6. Predicted particulate ow eld using the Eulerian approach for incident particles: (a) velocity vectors; (b) particulate concentration.

B.E. Lee et al. / Wear 252 (2002) 179188

183

could be derived from the impaction solution of incident particles. Under this treatment, the reected particles can be independently simulated using this derived information as the inlet boundary condition for the reected particles on the wall surface. Based on this concept, a new computational procedure for the Eulerian method is proposed as follows (also see Fig. 3): Step 1: Gas ow eld is rst computed with a wall boundary condition on the wall surfaces and gas solution is then xed for all the following calculations of the particulate phase, when one-way coupling between two phases is assumed to be same as in the Lagrangian simulations.

Step 2: Particulate ow eld is computed by considering the impacted wall surface as an outlet boundary (note: the gas ow eld does not change) and then the equivalent Lagrangian solution of incident particles is obtained. Step 3: The inlet boundary condition of reected particles from the wall surface is determined by extrapolating the solution of incident particles (Step 2) at control volumes adjacent to the wall, and by considering the momentum loss of incident particles during impaction and the direction of reected particles due to restitution coefcient. Step 4: By applying the inlet boundary condition only for the reected particles on the wall surface, the Eulerian solution of reected particles is computed.

Fig. 7. Predicted particulate ow eld using the Eulerian approach for reected particles: (a) velocity vectors; (b) particulate concentration.

184

B.E. Lee et al. / Wear 252 (2002) 179188

Step 5: The mean ow eld of the particulate phase that contains both the incident and reected particles is produced by superposing the above two Eulerian solutions.

3. Results and discussion Gas-particle two-phase ows are characterized by strong coupling between the phases. However, it is well known that if concentration of particles is so small, the gas ow eld is not signicantly inuenced by the presence of particles. Computations with one-way coupling can yield quite accurate results for dilute gas-particle ows, which can be found in many industrial applications, e.g. boiler tube erosion by y-ash particles. The present study has been limited to this

type of ow because both Lagrangian and Eulerian solutions of particulate phase are obtained from a xed gas ow eld. Rudoff [6] experimentally studied the particlewall impaction on a cylinder with an afterbody using spray droplets. The experimental settlement and dimension is illustrated in Fig. 4. They measured the mean axial velocity of different droplet sizes along the stagnation streamline of the cylinder for a free stream velocity of 15 m/s. Numerical prediction is performed using both approaches for comparison with the experimental data. An assumption has been made in this numerical simulation that spray droplets will adhere to the wall surface after impaction. Fig. 5 shows the predicted and measured axial droplet velocity distribution along the stagnation streamline for the two particle sizes. It can be seen from this gure that good

Fig. 8. Comparison of mean particulate concentration distributions: (a) Eulerian prediction; (b) Lagrangian prediction.

B.E. Lee et al. / Wear 252 (2002) 179188

185

agreement between the experimental and numerical results by both approaches is obtained. The computational domain of the test case for a 45 ramp and its Lagrangian simulation have previously been illustrated in Figs. 2 and 3, respectively. Figs. 6 and 7 show the Eulerian solutions of the ramp ow in terms of velocity and concentration of incident and reected particles, respectively. It can be seen from these gures that using the computational procedure proposed in this paper, two equivalent Lagrangian solutions of incident and reected particles can be estimated by the present Eulerian method. The mean particulate ow eld can be obtained by superposing the above two Eulerian solutions. Comparison of mean particulate concentration using the Eulerian approach with that using the Lagrangian approach is shown in Fig. 8 and good agreement between two results is found. Fig. 9 shows the computational gird for the test case of a single tube. Both approaches produce almost the same pattern of mean particulate velocity and concentration distribution around the tube as shown in Fig. 10. Due to the effect of reected particles on the mean particulate ow led, it is found that the particle velocity suddenly changes in magnitude and direction, and a shock pattern of high particulate concentration is formed in the front of the tube. Since we are more interested in the prediction of tube erosion by impacting particles, a comparison of the predicted physical properties of impacting particles around the tube using two approaches is shown in Fig. 11. Three parameters in relation to erosion prediction, impacting particle velocity,

angle and particulate concentration, are chosen for comparison between the two approaches because those parameters were impossible to be derived from using the traditional Eulerian approach. It can be seen that numerical results using the two approaches are in good agreement. An erosion model was developed [14] and expressed as follows:
2

E=
j =1

Kj Cpj f (j )Ui j g(Dp )h(Th )j

(1)

where E is the erosion rate ((m/h) 108 ), j = 1, 2, the angular and spherical particle inuence, respectively, K the erosion constant, Cp the particulate concentration (number of impacting particles in the Lagrangian approach), the impact angle of the particle, Ui the impact velocity (m/s), n the velocity exponent, Dp the particle size ( m), Th the homologous temperature ratio and the chemical composition of y-ash particles. The data used for this study are well described elsewhere [14] and summarized as follows. The values of velocity exponent, n1 = 2.5 and n2 = 2.4, where subscripts 1 and 2 refer to angular and spherical particles, respectively. The impact angle (the constants are listed in Table 1) is calculated from f () = a + b + c 2 + d 3 + e 4 + f 5 The particle size is calibrated as g(x) = g Dp 200 = 1 e17.387(x0.025) (3) (2)

Fig. 9. Grid system for a single tube ow (161 m 26 m).

186

B.E. Lee et al. / Wear 252 (2002) 179188

Fig. 10. Mean particulate ow eld produced by using the Lagrangian approach: (a) velocity vectors; (b) particulate concentration.

The temperature of target metal is formulated by h(Th ) = h T Tm = 1 0.5633Th + 7.8167Th2 (4)

where Th and Tm are homologous temperature ratio and melting temperature, respectively. The chemical composition of y-ash, 1 is a fraction of angular quartz particles
Table 1 Values of constants for impact angle term a Angular Spherical 0.0 9.056 0.0 b 13.448 0.529 7.153 c 63.655 0.355 16.081 d

by weight and 2 = 1 1 the fraction of hematite by weight. By applying the above erosion model and using the numerical solutions of particulate phase described in Section 2, the erosion rates of the tube surface are calculated and compared (Fig. 12) with the measured data of Bauver et al. [15]. Very good agreement between both the numerical

e 121.287 0.0 2.525

f 42.451 0.0 0.0

Interval 0.0 < < 0.45 0.45 < <1.0 0.0 < <1.0

129.628 0.413 11.703

B.E. Lee et al. / Wear 252 (2002) 179188

187

Fig. 11. Comparison of the physical properties of particles at impact: (a) normalized impact velocity; (b) impact angle; (c) particulate concentration.

predictions and fairly good agreement between the predictions and data is observed. However, it should be noted that the current prediction of tube erosion using the Eulerian approach needs far less computational time than using the

Lagrangian approach, which requires a great number of particles to be tracked for statistically meaningful values of physical properties of particles at impact.

4. Conclusion Both the Eulerian and Lagrangian methods are used for the modeling of particlewall impaction, which is very important for the prediction of particle erosion and deposition on the wall surface. The traditional Eulerian formulation has difculty in determining the physical properties of the impacting particles, including impact velocity, impact angle and particle number at impact on the wall surface. A new computational procedure is proposed for the Eulerian approach to estimate the equivalent Lagrangian solutions for incident and reected particles near the wall surface. Numerical results of the physical properties of impacting particles using the present Eulerian method show good agreement with those predicted using the Lagrangian method. Comparisons of numerical predictions with reported data show that both approaches are successful in predicting the main feature of the particulate ow near the wall and the erosion rate

Fig. 12. Comparison of the measured and predicted erosion rates using both approaches.

188

B.E. Lee et al. / Wear 252 (2002) 179188 [8] A. Orszag, V. Yakhot, et al., Renormalization group modeling and turbulence simulations in near-wall turbulent ows, in: R.M.C. So, C.G. Speziale, B.E. Launder (Eds.), Elsevier, Amsterdam, 1993, pp. 10311046. [9] N.-H. Cho, C.A.J. Fletcher, Computation of turbulent conical diffuser ows using a non-orthogonal grid system, Comput. Fluids 19 (1991) 347361. [10] C.A.J. Fletcher, J.G. Bain, An approximate factorisation explicit method for CFD, Comput. Fluids 19 (1991) 6174. [11] J.P. Van Doormaal, G.D. Raithby, Enhancements for the SIMPLE method for predicting incompressible uid ow, Num. Heat Trans. 7 (1984) 147163. [12] C.M. Rhie, W.L. Chow, Numerical study of the turbulent ow past an airfoil with trailing edge separation, AIAA J. 21 (1983) 1525. [13] C.A.J. Fletcher, Computational Techniques for Fluid Dynamics, Specic Techniques for Different Flow Categories, Vol. 2, 2nd Edition, Springer, Heidelberg, 1991. [14] B.E. Lee, C.A.J. Fletcher, M. Behnia, Computational prediction of tube erosion in coal red power utility boilers, ASME J. Eng. Gas Turb. Power 121 (1999) 746750. [15] W.P. Bauver, J.D. Bianca, J.D. Fishburn, J.G. McGowan, Characterization of Erosion of Heat Transfer Tubes in Coal Fired Power Plant, ASME Paper, 84-JPGC-FU-3, 1984.

on the surface, however, the Eulerian approach needs far less computational time than using the Lagrangian approach. References
[1] J. Fan, D. Zhou, J. Jin, K. Chen, Numerical simulation of tube erosion by particle impaction, Wear 142 (1991) 171184. [2] J.Y. Tu, C.A.J. Fletcher, M. Behnia, J.A. Reizes, D. Owens, P. Jones, Prediction of ow and erosion in power utility boilers and comparison with measurement, ASME J. Eng. Gas Turb. Power 119 (1997) 709716. [3] J.Y. Tu, C.A.J. Fletcher, An improved model for particulate turbulence modulation in conned two-phase ows, Int. Commun. Heat Mass Transfer 21 (6) (1994) 775783. [4] F. Pourahmadi, J.A.C. Humphrey, Modelling soliduid turbulent ows with application to predicting erosive wear, PhysioChem. Hydrodyn. 4 (3) (1983) 191219. [5] FLUENT Inc., FLUENT Users Guide, Version 4.3, 1996, pp. 866869. [6] R.C. Rudoff, Effects of Turbulence and Number Density on the Drag Coefcient of Droplets, AIAA Paper 910074, 1991. [7] J.Y. Tu, C.A.J. Fletcher, Numerical computation of turbulent gas solid particle ow in a 90 bend, AIChE J. 41 (10) (1995) 21872197.

You might also like