Download as pdf
Download as pdf
You are on page 1of 7
CFD Solution Analysis: Essentials Chapter | 6 245 error. The increasing mesh resolution demonstrates two important outcomes. sly, as the spatial discretization error becomes smaller, a grid-independent solution is achieved. Secondly, the excellent agreement between the approxi- ‘mate and analytical solutions verifies the aumerical algorithm that is adopted and provides credibility of the computational solution 6.7.2. Test Case B: Flow over a 90° Bend This test case was calculated using a commercial finite-volume CFD computer code ANSYS FLUENT, version 6.1 Model description: The geometry of the test problem is a three-dimensional turbulent flow over a 90° bend. The schematic view of the experimental setup ‘comprising an open-circuit suction wind-tunnel system for the 90° duct bend is shown in Fig. 6.5, which comprises a 3.5 m long horizontal duct, a 90° bend with a radius ratio of 1.5, and a 1.8m long vertical straight duct. Air flows through a 10 mm thick Perspex square test section with the bulk gas velocity Uy adjusted through the aid of a variable frequency controller. Experimental Laser Doppler mnemometry system Boll mouth entry Converging duct Horzontal duct “) 90° ben 35m .am|! | vertical uct osm Centrfugal fan Cross-sectional ‘ow area si FIG. 6.5. Schematic view of the experimental rig ofthe 90" bend 0 246 Computational Fluid Dynamics data are obtained on this experimental setup using flow visualization and laser Doppler anemometry (LDA) system Grid: For the 90° square-section bend, the computational domain begins at a distance of 2D upstream from the bend entrance and extends to 20D down stream from the bend exit. A structured mesh of 325x43%41 control volumes in the respective directions along the streamwise, width, and height is generated for the whole computational domain. Features of the simulation: This test case illustrates the importance of eval uating the choice of the turbulence models, standard ke (Launder and Spalding, 1974), and Reynolds stress (Launder et al., 1975; Launder, 1989), for comput ing the flow separation around the 90° bend and validating the computational solutions against experimental data The algorithm for the solution of the Navier-Stokes equations relies on the implicit segregated velocity-pressure formulation such as the SIMPLE scheme, This leads to a Poisson's equation for the pressure correction that is solved through the default iterative solver, normally the multigrid solver in the ANSYS FLUENT computer code. Finite-volume discretization is employed to approx- imate the governing equations. To avoid nonphysical oscillations of the pres- sure field and the associated difficulties in obtaining a converged solution on collocated grid arrangement, the Rhie and Chow (1983) interpolation scheme is employed. ‘The working fluid, ar, is taken to be incompressible, and the default initial conditions implemented in the computer code are used for the simulations, Boundary conditions: At the inlet, Dirichlet conditions are used for all vari- ables, The bulk velocity U, was taken as constant with a value of 10 m/s. With the density and viscosity of air having the values of 1.2 kg/m’ and 2.x 10-* kg 'm s, respectively, the corresponding Reynolds number based on ths inlet veloc- ity and height of the channel is 90,000, The turbulent kinetic energy k and dissipation ¢ are determined from the measured turbulence intensity J of about 14% atthe centre of the duct cross-section, For the Reynolds stresses, the diagonal ‘components are taken to be equal to 2/3 k, whereas the extra-diagonal compo- nents are set to zero (assuming isotropic turbulence). At the outlet, Neumann boundary conditions are applied forall the transported variables. The nonequili ‘rium wall function (as will be described in the next chapter) is employed for the airflow at solid walls because of its capability in better handling complex flows ‘where the mean flow and turbulence are subjected to severe pressure gradients and rapid change, such as flow separation, reattachment, and impingement. Results: The compatison between the measured and calculated longitudinal ‘mean velocities normalized by the bulk velocity U, at the bend exit (9=90°) and 0.5D after the bend exit is illustrated in Fig. 6.6. The prediction of the streamwise velocities using the Reynolds stress model is observed to yield bet- ter agreement in contrast to the standard ke model, which is due to the capa- bility of the Reynolds stress model capturing the anisotropy behaviour of the flow separation region around the 90° bend. CFD Solution Analysis: Essentials Chapter |6 247 08 064 [ Experimental daa ‘Standard kee model aD 0.44 | — Reynolds stress model osf | osf | g 1 5 i oa i 02 ! ° d “0300308 «08 «1215 6) ay FIG. 6.6 Comparison between measured and calculated longitudinal mean velocities nonalized bythe hulk velocity U: (A steams velocity at bend exit (6=90") and (B)sreamwise velocity 1.3 afer the ben exit (ae igure for more desrison). The predicted longitudinal mean velocity normalized by the bulk velocity ‘Uy using the Reynolds stress model is further compared with the measure data at different locations in the duct centre plane, as represented in Fig. 6.7. At the bend entrance (@=0°), the turbulence model successfully predicts the acceler- ation of the airflow near the inner wall. The fluid deceleration caused by the unfavourable pressure gradient is also captured near the outer wall. More impor- tantly, the turbulence model adequately reproduces the distorted longitudinal velocity profiles at the angles of @=30°, @=45°, and @=60° after the bend entrance. 248 Computational Fluid Dynamics cter wall ap 8=15" 0=30" 0=45" 6=60" inner wall (A) Outer wal aD (807080 10-290 4D SD 7D 8 merya FIG.6.7 Comparison between messured and calculated longitudinal mean velocities normalized byte lk velocity Uy at diferent locations the duct centre plane (A) hetwoen bend entrance and cit and (B) downstream of bend exit (sce again Vig. 6.6 for moe description To better understand the flow characteristics around the 90° bend, Fig. 6.8 presents the calculated velocity vectors and pressure distribution of the airflow ‘obtained through the Reynolds stress model at different cross-sections of the duct flow. Fig. 6.8 and B shows the calculated air velocity vector and pressure distribution at the cross-sectional middle plane of the duct flow. Favourable (positive) and unfavourable (negative) longitudinal pressure gradients persist near the inner and outer walls of the bend entrance, The presence of the favour- able and unfavourable pressure gradients is caused by the balance of centrifugal force and radial pressure gradient in the bend (Humphrey et al., 1981). This physical phenomenon is typical of curved duct flows, Secondary flows ate developed due to the direct consequence of the cross stream pressure gradient, ‘The predicted secondary flow vectors are clearly depicted by the three square CFD Solution Analysis: Essentials Chapter |6 249 to. 22.7649 9 125976 8 245039 7 199684 8 178447 5 279913 4 3 2 1 738.1386 28-2858 “38.433, “68.5803 Inner wall Outer wall, (©) Outer wall Inner wall “| outer wat (E) FIG, 6.8 Caleuated low fil (A) airlow in the duct mde plane, (B) state presse distin on in the duct ide plane, (C) secondary flow at cross-sectional flow arca of location 9 = 45", (secondary flow at cross-sectional flow area of location 890 (bend exit), and (2) secondary flow at erostsectional flow ares of location S/D ~ 3D after the bend exit. 250 Computational Fluid Dynamics cross-sectional flow areas at angles of @=45° and @=90° and distance of 3D after the bend exit (S/D=3) in Fig. 6.8C-E, respectively. Conclusion: This test case focuses on the use of approximate models such as the turbulence models to predict the physical characteristics of the turbulent flow around a 90° bend. Owing to the absence of analytical solutions, validation of the computational solutions is performed by comparing the predictions with the experimental data in order to address the simulation model uncertainty and the degree to which models correspond to an accurate representation of the real physical flow, The flow around a 90° bend though geometrically simple exhibits ‘complex flow structures duc to the existence of secondary flows in the vicinity of the bend region, which are generally anisotropic in nature. The results from the more sophisticated Reynolds stress model are shown to better capture the anisotropy behaviour of the flow in contrast to the standard ke model that assumes isotropy in its original mode! formulation. 6.8 SUMMARY ‘The credibility of a computational solution is analysed and assessed in this cchapter through the consideration of the various aspects of consistency, stabil- ity, convergence, and accuracy. tis usually possible to demonstrate whether a discretized form of the governing fluid flow equations is consistent and also if the algebraic form of these equations is stable. Convergence requires, however, implications from consistency and stability, that is consistency+tstabili- 1y=convergence. In any numerical calculations, errors and uncertainties affect the accuracy of the computational solution. Itis imperative that the errors and "uncertainties are systematically reduced so that the computational solution bet ter represents the real physical flow problem that is being solved. The concep- tual framework linking the various aspects of consistency, stability, convergence, and accuracy beginning from the governing partial differential ‘equations as considered in Chapter 3 and arriving at the approximate solution of the algebraic equations as described in Chapter 4 can be seen in Fig, 6.9. The application of a Taylor series expansion to the discrete approximation of the governing equations results in an explicit expression for the truncation error (Section 6.2). This error measures the accuracy of the finite-difference or finite- volume approximation and determines the rate at which the error decreases as the finite quantities, time step and/or mesh spacing, diminish. It is noted that because of the close correlation between truncation error and solution error (Section 6.5.1), reducing the truncation error has beneficial consequences in the likelihood of also reducing the solution error. Numerous worked examples and test cases presented throughout this chap. ter have aimed to better illustrate the conceptual properties of consistency, sta- bility, convergence, and accuracy whilst solving the discretized form of the partial differential equations. In addition, itis important that the computational solution should be subjected to the rigorous process of verification and CFD Solution Analysis: Essentials Chapter | 6 251 Diseretistion Chapter 3 Governing equations [Numerical methods partial afferental ‘equations (PDEs) Chapter 4 Conver into systom of Consistency ‘algebraic equations asin y Az AIO * ‘Nurerica! tabi ‘Stabilty| | iechniques x - , Convergence . Exact solution ny Chapter 4 ° ‘Approximate solution ‘a8 ax ay, Az, at 90 “e ‘Accuracy = 6-"3.\_ J FIG. 6.9 A conceptual framework linking the various aspects of consistency, stability, conver- gence, and accuracy in avin a a solution for the transport equations. validation, which was demonstrated through two test cases. For a simple two-dimensional channel flow problem in test case A, it is feasible to verify the computational solution against an analytical relationship, However, for a more complicated flow problem that employs a computational solution in a three-dimensional domain, the absence of an analytical solution requires the dependency of benchmark and/or experimental data to validate the computa- tional solution. This is evidenced by test case B for a three-dimensional air- flow around a 90° bend. In this example, we demonstrated the use of appropriate turbulence models to better capture the physical flow behaviour in a 90° bend. This aspect highlights the uncertainty that arises through the use of approximate models for solving turbulent flow that will be further investigated in the next chapter. More practical guidelines will be further dis- cussed in this chapter in better equipping the reader in handling and solving a range of CFD problems. REVIEW QUESTIONS 6.1 Why do the results obtained through numerical methods differ from the exact solutions solved analytically? What are some of the causes for this difference? 6.2 In the analysis of CFD results, what does consistency imply? 6.3 What are the key aspects of consistency’? 6.4 Ifa system of algebraic equations is equivalent to the partial differential equation as the grid spacing tends to zero, does this also mean the solution

You might also like