J. of Thermal Science Vol.7, No.1 J l ~ ¢~ C o m p u t a t i o n a l S t u d i e s of L o b e d F o r c e d M i x e r F l o w s H. H u S.S. W u Jet Propulsion Laboratory, Beijing University of Aeronautics, Beijing, China S.C.M. Yu Thermal and Fluids Engineering Division School of Mechanical and Production Engineering, Nanyang Technological University, Singapore Full Navier Stokes Analyses have been conducted for the flows behind the trailing edge of a lobed forced mixer. The governing equations are derived from the time-dependent compressible Navier-Stokes equations and discretized in the finite-difference form. A simple two-layer eddy viscosity model has also been used to account for the turbulence. Computed results are compared with some of the velocity measurements using a laser-Doppler anemometer (Yu and Yip (1997)). In general, good agreement can be obtained in the streamwise mean velocity distribution but the decay of the streamwise circulation is underpredicted. Some suggestions to the discrepancy are proposed. Keywords: lobed forced mixer flows, eddy viscosity model, velocity distribution. INTRODUCTION Splitter plates with a convoluted trailing edge (commonly referred to as lobed forced mixers) are passive, fluid mechanical devices which generate in the coflowing streams a three-dimensional shear layer with strong secondary flow. As shown in Fig.l, the geometry of a lobed forced mixer is characterized by a periodically alternating, lobed trailing edge surface. The lobe causes large scale streamwise vortices to be shed at the trailing edge so that the downstream flow field is embedded with an array of streamwise vortices of alternating sign. The enhanced mixing of the lobed mixers is believed to be directly attributable to the large mixing scales generated by the streamwise vortices. Application of lobed mixers in the turbofan engine exhausts has been studied by Presz, Blinn and Morin [1J and Presz, Gousy and Morin [2], where the rapid mixing of the core and the bypass flows can achieve noise reduction and thrust enhancement. Received August, 1997. Cold 11111flou Inlel ixert{o~laust Lobed mixer Fig.1 Typical lobed (forced) mixer exhaust geometry Paterson [3] measured velocity and turbulence characteristics downstream of a lobed mixer using laserDoppler a n e m o m e t e r and concluded t h a t b o t h the lobe shape and lobe penetration were the important p a r a m e t e r s to determine the effectiveness of mixer performances. Paterson also found that the generation of streamwise vortices was due mainly to the secondary flow shed by the lobe and, therefore, concluded t h a t the streamwise vortices ,were inviscidly generated. The subsequent investigation by Barber, Paterson and Skebe [4] confirmed the inviscid nature of the secondary flow generated by a lobe and the strength of the secondary flow was also found to be higher when the H. Hu et al. Computational Studies of Lobed Forced Mixer Flows lobe penetration consisted of straight parallel sidewalls. They also found that higher strength would result in a faster mixing rate and hence would facilitate the condition of achieving spatial uniformity downstream of the mixer. Based on flow visualization tests in a water tunnel, Werle, Paterson and Presz [5] suggested that the flow structure of the wake region behind the lobed mixer follows a three-step process by which the streamwise vortex cells form, intensify, and then break down. Most intense mixing seemed to occur in the third region. By varying the velocity ratio across the lobe, the location of these three regions could be shifted. Velocity measurements by McCormick and Bennett [G] and Yu and Yip [7] concluded that intense small-scale turbulence and mixing occurred at about two to four wavelengths downstream of the mixer trailing edge is mainly due to the deformation of the normal vortex into a pinched-off structure by the streamwise vorticity. The complicated three-dimensional features of the flows behind the mixer trailing edge provide challenges to use CFD methods to predict them. Some early computation works were focused mainly on the region downstream of the lobe exit whereby the accuracy of the lobe exit velocity information was crucial to the success of the predictions. The methods have been improved later on by modelling the lobe itself (Refs.9). K o u t m o s and McGuirk Is], Tsui and Wu [9] solved fully elliptic equations using the finite volume method coupled with a k - e turbulence model to predict the flow behind the lobed forced mixer trailing edge. In general, good agreement can also be obtained against measured results. Although these calculations were promising, confidence in the predictions requires further rigorous testing against suitable measurements. The work presented here has the objective in achieving this goal. T h e following section describes briefly the mathematical model. It will be followed by presentation and discussion of the results. The paper ends with a summ a r y of important findings. 23 transformed to a b o d y conforming curvilinear coordinate system (~,77,(). Then, in the absence of b o d y forces, the equations for three-dimensional turbulent flow in non-dimensional form can be written as: -OQ* --+ Ot Jr O{F* - F ; } + O~ O{H* - H*} o( 0{C*-C;} 0~ --0 (1) where: Q, = j-1Q F* = J - l ( ~ F + ~yG + ~zH) G* = J-I(~?~F + ~?yG + ~?~H) (la) H* = J - Z ( ( ~ F + ( y G + (~H) G* = J-l(77xF. + ,TyG. + UzH,,) S:, = J-I((~F~ + (~C. + (~U~) j _ a(~,77,() _ 1/ o(x,y,z) Q = x~ xv x( Y~ Yv Y( z¢ zv z( (~b) p pu pu pu 2 + P puv F = pv pwl paw (E + P)u 0 Txx F~=~ 1 Txy rxz M A T H E M A T I C A L MODEL Urxx T vrxy T w1"xz Governing Flow Equations T h e governing equations are derived from the time-dependent, compressible Navier-Stokes equations (strong conservation law form) cast in terms of mass-averaged variables and Cartesian coordinates. In order to facilitate the implementation of a finitedifference solution procedure and the treatment of b o u n d a r y conditions, the Navier-Stokes equations are pv puv G -- pv u + P pvw (E + Ply , "r# O(P/p) ( ~ - 1)Pr Ox 24 Journal of Thermal Science, Vol. 7, No.l, 1998 is evaluated with Sutherland's law, and the turbulent viscosity is computed with a turbulence model, which will be discussed later. The molecular and turbulent Prandtl numbers are assumed to be constant at 0.72 and 0.9 respectively. 0 ~y 1 G Tyy = -ffe ry, 7# a(P/p) (~/- 1)Pr Oy Numerical Algorithm In integral form, the governing Eqs.(1) can be written as p12~ puw H = pvw + P , 0 pw 2 (E + P)w +ff ds=0 (6) for a fixed region V with boundary S. Here Q represents the conserved quantity and E is the corresponding flux terms, where "rxz 1 = ( F - F,,)~--) + ( a - G , ) 7 I4,, = -~ee + (H - H,)-~ (7) Tzz ur~z + Vry, + w%~ - 7~ O(P/p) Dz (lc) Flow stagnation conditions are used as reference quantities, the density p, the velocity components u, v, w, and the internal energy E are expressed by p' =--, PP Po n--~-----n= z + n y j (8) +nz ( 7 - 1)Pr p' p' , # =--, poa~ #o u', if, w I E' u,v,w=--, E=-T= ao poa~' T' a2~! where 'indicates a dimensional quantity and a0 is the stagnation speed of sound. Thus, the Reynolds number Re is based on stagnation condition. The temperature is given by E = p [ c , T + ~(u2 + v2 + w2)] The governing Eqs.(6) are discretised by first dividing the. computational domain into hexahedral cells (Fig.2) and then approximating Eq.(6) for mass, momentum, and energy conservation in each cell. We assume that the dependent variables Q are known at the point (i, j, k) where each such point is the center of one of the cells. If we apply a semi-discretisation (i.e. a discrete approximation to the spatial terms only), we obtain the following discrete analog of Eq.(6). (9) d(Q~a,k) + E~,~-k = 0 where (Io) (2) and the pressure can be determined with the ideal gas equation of state: P = pRT (3) The viscosity, p and the ratio p / P r , where P r is the Prandtl number, are defined as ,j,k+ ~ = ~.,. + ~,u~b i,j,k- (4) --4 ~-~lan ~£urb - + - (5) Pr Brian Prturb where the subscripts "lam" and "turb" refer to laminar and turbulent, respectively. The laminar viscosity T h e vector S i + ½,j,k denotes the cell face between the points (i, j, k) and (i + 1, j, k). The value of E at the cell face (i + 1 / 2 , j , k ) is taken as the average of E at the points (i, j, k) and (i + 1, j, k). H. Hu et al. Computational Studies of Lobed Forced Mixer Flows / .//~ i+1/2,/, 25 where Qn and Qn+l are the values of Q after the nth and ( n + l ) t h time steps respectively and k 1 a1=~,~2=~, 3".,i.k ~ 1 1 ~3=~,~4=1 In order to increase the Courant number, the residual averaging is used implicitly in the product form: ~ i (1 - e¢6~)(1 - ev6nv)(1 - e~g~i)Ri,j,k = Ri,j,k Fig.2 Hexahedral cell around point (i, j, k) In order to eliminate the numerical odd-even decoupling and capture shocks without any preshock oscillation, artificial viscosity Di,j,k is employed in the equations (Ref.10), so the governing Eqs.(9) can be written as d (Q~j,k)) + Eij,k - Di,j,k = 0 (12) where (15) where Ri,j,k is the residual at each cell, 5~,m5~¢ are the three second order differential operator and the ~ ~, e¢ are three coefficients correspondingly. Turbulence Modeling Because algebraic models are less CPU intensive than other models, previous works have shown that the turbulence model variations had little impact on the flow-field solution of the mixer flow, so the twolayer eddy viscosity model of Baldwin-Lomax (Ref. 11) is employed for wall boundary layers and mixing region. The turbulent viscosity is calculated as follows: D,j,k = (D~ + D . + D¢)Qij,k D~Qi,£k = di+t/2,j,k -- di-I/2,j,k hi + l /2,j,k di+l/2"j'k = At (4) X [e(2+)l/2j,kAxQi,j,t ~ -- £ i+l/2,j,k h 3 ~ x(~i-l,J, k]] (13) AQi,j,k = Qi+l,j,k - Qi,j,k i+l/2,j,k = k (2) m a x ( v i + l , j , k , E (2) (16) The internal formulation of the turbulent viscosity is based on a mixing length l, function of the local aerodynamic quantities of the flow and of the normal distance from the wall. vi,j,k) (#t)i,*n~r = Pl21al e(i4.l)l/2,j,k ~- m a x ( 0 , k (4) - e (2) i+l/2,j,k) In the outer region of the boundary layer, is expressed as below: Pi+l,j,k -- 2Pi,/,k + Pi-xj,k vij,,~ = I ~ ; ~ + P~-lj,k I K (2) = 0.5, { (ut)i ........... y < y~ (m)o., ........ y --- y~ u, = (Pt)o,~t~r : 0.618ccvpFwFk (18) F k ( y ) = [ 1 - 5 5 ( cky ~°1-1 (19) K (4) = 0.01 The governing Eqs.(12) can be solved in a class of four stage R u n g e - K u t t a schemes, so it can be written as where " \Ymax/ J U2 [ 1 Cwk -F~ma -DtFx )'~ F., = y m a x F m a x min ~, Q(O) = Q . ccp = 1.6; Q ( I ) __ Q;:j,k - °tl R(Q(°))i,J,k Q(2) = Q .i,j,k _ a2R(Q(1))ij,k (14) Q n + l ____Q(4) (20) c,~k = 0.25; ck = 0.3 The quantities Ymax and Fm~x are determined from the maximum value of the function Q(3) = Qni,j,k _ a3R(Q(2))ij,t~ Q(4) __ Q~,j,k - ot4R(Q(a))i,J, k (17) (•t)outer { yll~l[1-exp(-y+/26)] F(y) = yll21for boundary flow for wake (21) 26 Journal of Thermal Science, Vol. 7, No.l, 1998 Geometry Modeling The schematic of the studied lobed mixer is shown in Fig.3. Because of the symmetric character of the 2-D mixer, only one half of lobe of the geometry is modeled. According to the different boundary conditions, three computed domains are separated for the flow field of lobed mixer, namely the upper stream, the lower stream and the wake. Synmactric f- _ ..~"_ " ,~,- - - - ~ ;.1..~...4. : --" d, .e * ..... - , r ...... , . .- = 1 2 :" m / s " ' Domain z* ,'egio0. v .I.._ _ 2"~I... _-_--. phme .."2-~_~ . . . . D,,m,,,n 2 i.=66mm !,~"=~2 " " F~Co,nptLted Syntmetri,~ plane ~ ).-=-6~_~ ram" 3 .................... "~" Sy nuuetric (~ "'" plane d d,,mai,12 Fig.3 Schematic of the lobed mixer and studied three domain Grid G e n e r a t i o n The grid is generated in the three separate regions defined in Figs.4-6. The 3-D finite difference grid is constructed as a sequence of 2-D grid using T T M method. Mesh points in each cross-sectional plane in the physical space are related to those in a corresponding uniform computational space (A~ = At/ = A~ = 1) by the transformation derivatives and transformation Jacobian in Eq.(lb). The results presented in the following, obtained using a 28 x 20 (on a crosssectional plane)x 30 (in the streamwise direction). .v/2-0.25 (b) ,v/2-1,0 (c) x/A-2.0 (d) x / 2 - 3.0 (e) .v/2- 5,0 (f) F i g . 6 Grid in the studied domain 3 Boundary Conditions Z X a , 3 - d ~ t d , , f , , m d i e d d,~mam ] Ibl .'¢ - - S2 rren ,el x=-41 q,lJ mm x = ii F i g . 4 Grid in the studied domain 1 The boundary condition required for Navier-Stokes calculation are the total pressures (PT,p and PT,~) at the primary stream and secondary stream in flow b o u n d a r y of the computational domain, and static pressure (Pa) at the downstream b o u n d a r y of the computed domain 3. The wall boundary conditions are used at the top of the computational domain 1 and the b o t t o m of the computational domain 2. Symmetrical b o u n d a r y conditions are employed at the other sides of the computational domain. RESULTS AND DISCUSSION ( a '~ 3--d ~ t d o f s t u d i e d d o m a i n 2 x - - 82 m m x = - 41 m m x = cl S t r e a m w i s e M e a n Velocity C o n t o u r s F i g . 5 Grid in the studied domain 2 T h e axial velocity contours at selected downstream H. Hu et al. locations are presented in Figs.Ta-e. In the wake region the streamwise vortex convected the high speed stream into the lower speed region, and by the same token, the low speed flow also p e n e t r a t e d into the high speed region. Subsequently, turbulence was being generated and diffusion enable the s m o o t h i n g out of the velocity gradients. Effective mixing can be achieved in the wake between the two streams. T h e velocity distribution becomes more uniform at the further downstream station. In general, the c o m p u t e d results capture the m e a n flow field reasonably well when comparing with measurements. .--.. I f, •',C ii • (. . . . 'g" ., v/2 = o.25 (a) .v/2 = 1.0 (b) x/2 = 20 (e) v/2=5 o (e) .v/2 =3,o (d) Fig.7 Contour of the normalized streamwise mean velocity (U/U,.) at domain 3 (U~=IO.O m/s) Secondary Flow Velocity Vectors T h e secondary velocity vector plots are shown in Figs.8a-e. As expected, a streamwise vortex was formed immediately at the trailing edge with centre of the vortex core located at a r o u n d the inflexion point of the lobe (Fig.5a). T h e s t r e n g t h of the secondary m e a n velocity maintains nearly the same s t r e n g t h from the trailing edge to a b o u t x / A = 2 . 0 (see Figs.5b-c). At region after x/A -- 3.0 (Fig.5d) and to the end of the m e a s u r e m e n t range, the secondary flow decreases rapidly. C o m p a r i n g the corresponding vector s t r e n g t h with those measured, it was found t h a t the decay for the secondary m e a n velocity was actually slower t h a n t h a t found in the experiment. ..,.. . ...... ' .,,•. I (a) F = ~ Y . dS x/;~= I.o (b) x/,;, = 2.0 (¢) x/2=3.0 (d) :5:-- I x/),=5.0 (e) P i g . 8 Velocity vector field in domain 3 (22) where the circuit C is the outer p a t h of the c o m p u t a tional domain, as d e n o t e d by the s h a p e d area given in Fig.3. T h e result of which are shown in Fig.9. A slower decay is found for the calculation b u t the t r e n d is very similar. Based on the findings of Yu and Yip [7], the streamwise vortex development u n d e r w e n t a t h r e e - s t e p process by which it was formed, intensified and quickly dissipated towards the end of the near field. T h e spatial spacing between two rows of streamwise vortices of alternate signs within a lobe 1.l 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 O.1 [~ i i ~ , ~ i i :k I e a M L r e n ! e l l D , ] ] ' i ' I ' i , t 0.0 10 2.0 3.0 -ko 5.0 6.0 7.0 8.(1 9.0 100 .v/), Fig.9 Variation of normalized streamwise circulation with downstream distance Nonnal vortex line __ at~et~ a,:,,:e 1o t!l~ :2"" x/;,=o.~5 Decay of the Streamwise Circulation T h e streamwise vortex plays an i m p o r t a n t role in the mixing process. In the results presented next, the s t r e n g t h of this vortex is characterized by circulation defined by Initial pinched-off "7-:1 7 2~ Computational Studies of Lobed Forced Mixer Flows ]Peak Trough streanm lse v o r ~ e t ~ .kdditional pinched - off effects 0 D due to tile ,~ "1'~'~4.,(~", nonnal vorticity. ~ . . ~ k . ~ ( b ) End view ,/~ \component rides ~, r ~ o n the stream- ~ ~... vortex line to l n'oducetile additional i l ~inched- off effects ( c ) Tc ) vie~ Fig.lO Schematic of the vortical structure 28 Journal of Thermal Science, Vol. 7, No.l, 1998 reduced with downstream distance causing the mutual annihilation of the vortices. Analyzing this movement of vortices, based on inviscid vortex dynamics (as shown in Fig.10), suggested t h a t the normal vorticities shed at the trailing edge had actually provided a "pinched-off' effect to two adjacent streamwise vortex lines within a lobe forcing the streamwise vortices to interact with each other. However, the vector plots in Fig.8 did not show the expected movement of the vortex. It is advisable that further computational domain should encompass one full lobe rather than a half lobe as most of the calculations were being conducted. CONCLUDING REMARKS C o m p u t a t i o n a l studies has been conducted on the lobed mixer flows. In general, good agreement can be obtained for the streamwise mean velocity distribution at successive downstream station. However, the decay of the streamwise circulation is somehow underpredicted. The difference m a y be due to the fact that normal vorticities shed at the trailing edge actually provided a "pinched-off' effect to two adjacent streamwise vortex lines within a lobe forcing the streamwise vortices to interact with each other. This m a y provide a faster decay to the streamwise circulation. Future computational work should include one full lobe instead of a half lobe. REFERENCES [1] Presz, W.M., Blinn, R.F. and Morin, B. "Short Efficient Ejector Systems," AIAA paper No.87-1837, Jan., (1987). [2] Presz, W.M., Gousy, R. and Morin, B. "Forced Mixer Lobes in Ejector Design," AIAA Journal of Propulsion and Power, 4, No.4, p.350, July (1988). [3] Paterson, R.W., "Turbofan Mixer Nozzle Flowfielda Benchmark Experimental Study," ASME Journal of Engineering for Gas Turbines and Powers," 106, 692, July, (1984). [4] Barber, T., Paterson, R.W. and Skebe, S.A., "Turbofan Forced Mixer Lobe Flow Modeling Vol.l: Experimental and Analytical Assessment," NASA CR4147, (1988). [5] Werle, M.J., Paterson, R.W. and Presz, W.M. Jr., "Flow Structure in a Periodic Axial Vortex Array," AIAA Paper No.87-0610. [6] McCormick, D.C. and Bennett, Jr. J.C., "Vortical and Turbulent Structure of a Lobed Forced Mixer FreeShear Layer," AIAA Journal, 32, No.9, pp.1852-1859, (1994). [7] Yu, S.C.M. and Yip, T.H., "Measurements of Velocities in the Near Field of a Lobed Forced Mixer Trailing Edge," The Aeronautical Journal of the Royal Aeronautical Society, 101, pp.121-129, (1997). [8] Koutomos, P., and McGuirk, J.J.,"Velocity and Turbulence Characteristics of Isothermal Lobed Mixer Flows," ASME Journal of Fluids Engng., 117, pp.633-638, (1995). [9] Tsui, Y.Y. and Wu, P.W., "Investigation of the Mixing flow Structure in Multilobe Mixers," AIAA Journal, 34, No.7, pp.1386-1391, (1996). [10] Jameson A. and Baker T.J., "Solution of the Euler Equation for Complex Configuration," AIAA Paper No.83-1929, (1983). [11] Baldwin B.S. and Lomax, H., "Thin-Layer Approximation and Algebraic Model for Separate Turbulent Flows," AIAA Paper No.78-0257, (1978).