Fluid flow simulations in the undergraduate fluids class
Transcription
Fluid flow simulations in the undergraduate fluids class
Introducing CFD in Undergraduate Fluid Mechanics John Cimbala, Mechanical Engr., Penn State Univ. with collaboration from: Shane Moeykens, Strategic Partnerships Manager, ANSYS. Ajay Parihar, FlowLab Support Engineer, ANSYS. Sujith Sukumaran, FlowLab Support Engineer, ANSYS Satyanarayana Kondle, FlowLab Support Engineer, ANSYS. ISTEC Meeting, Cornell University July 25-26, 2008, Ithaca, NY Introduction z It has become important in recent years to introduce the fundamentals of CFD in introlevel undergraduate fluid mechanics classes due to the changing requirements of the job market for graduating engineers z At a minimum, it is desirable to teach the Many of them will use CFD in their fundamental steps required to obtain a useful whether they know anything CFD jobs, solution about CFD or not! z Many instructors want to include CFD in their undergrad fluids course, but don’t know how and/or think they can’t afford the class time Our first attempt to introduce CFD to undergrads z Undergraduate fluid mechanics textbook, Fluid Mechanics: Fundamentals and Applications, by Y. A. Çengel and J. M. Cimbala, McGraw-Hill, 2006 z Chapter 15: Introduction to CFD Çengel-Cimbala textbook zThe CFD chapter introduces: – grids – boundary conditions – residuals – etc. Çengel-Cimbala textbook zThe CFD chapter introduces: – grids – boundary conditions – residuals – etc. Çengel-Cimbala textbook zThe CFD chapter introduces: – grids – boundary conditions – residuals – etc. Çengel-Cimbala textbook zThe CFD chapter introduces: – grids – boundary conditions – residuals – etc. – just the basics, not anything about numerical algorithms, stability, etc. – how to use CFD as a tool. Intro to CFD using FlowLab z The Çengel-Cimbala book includes FlowLab as a textbook companion, where CFD exercises are employed to convey important concepts to the student z 46 FlowLab end-of-chapter problems are included in Ed. 1, Chapter 15 (Intro to CFD) z FlowLab exercises jointly developed by John Cimbala and Fluent Inc. (now part of ANSYS). z FlowLab & these FlowLab templates are free to students who use the Çengel-Cimbala book What is FlowLab? zA virtual (CFD) fluids laboratory z Simple to use with a very fast learning curve z Runs pre-defined exercises (templates) z Setup, solution, and post-processing are all performed in the same interface z Students vary only one or two parameters in each template (to look at trends, compare boundary conditions, grid resolution, etc.) FlowLab Templates zEach homework problem, along with its corresponding FlowLab template, has been carefully designed with two major learning objectives in mind: – Enhance the student’s understanding of a specific fluid mechanics concept – Introduce the student to a specific capability and/or limitation of CFD through hands-on practice Original Templates for Ed. 1 z FlowLab HW problems only in CFD chapter z Most templates are too complex to compare with analytical calculations (e.g., flow over cylinders, flow through diffusers, etc.) z Emphasis mostly on CFD – grid resolution, extent of computational domain, BCs, etc. z In the first edition, the primary emphasis of the FlowLab templates was as a CFD learning tool, with only a secondary emphasis on learning fluid mechanics New Templates for Ed. 2 z New FlowLab templates in almost all chapters – goal is to introduce students to CFD early on z Most new templates compare CFD calculations with analytical calculations z The primary emphasis is learning fluid mechanics, with a secondary emphasis on CFD z New templates are intentionally more simple z Homework problems show a progression in difficulty and level of sophistication, often based on the same base problem or theme Examples: New homework & templates, Ed. 2 z End-of-chapter homework problem, Chap. 2 2-89 A rotating viscometer consists of two concentric cylinders – an inner cylinder of radius Ri rotating at angular velocity (rotation This is a astandard analytical problem rate) ωi, and stationary outer cylinder of as found in most undergraduate fluids books. the tiny gap between the inside radius Ro. In two cylinders is the fluid of viscosity μ. The length of the theanpage in the They arecylinders able to (into obtain analytical sketch) is L. L is large such that effects are (approximate) solution forend a small gap. negligible (we can treat this as a twodimensional problem). Torque (T) is required to rotate the inner cylinder at constant speed. Ro Ri ωi Fluid: ρ, μ Rotating inner cylinder Stationary outer cylinder (a) Showing all your work and algebra, generate an approximate expression for T as a function of the other variables. (b) Explain why your solution is only an approximation. In particular, do you expect the velocity profile in the gap to remain linear as the gap becomes larger and larger (i.e., if the outer radius Ro were to increase, all else staying the same)? z Solution (from solutions manual) 2-89 (a) We assume a linear velocity profile between the two walls as sketched – the inner wall is moving at speed V = ωiRi and the outer wall is stationary. The thickness of the gap is h, and we let y be the distance from the outer wall into the fluid Inner cylinder (towards the inner wall). Thus, y du V V u =V and τ = μ =μ h dy h h u y where h = R - R and V = ω R o i i i Outer cylinder Since shear stress τ has dimensions of force/area, the clockwise (mathematically negative) tangential force acting along the surface of the inner cylinder by the fluid is μωi Ri V 2π Ri L F = −τ A = − μ 2π Ri L = − h Ro − Ri But the torque is the tangential force times the moment arm Ri. Also, we are asked for the torque required to turn the inner cylinder. This applied torque is counterclockwise (mathematically positive). Thus, Analytical solution for a small gap 2π Lμωi Ri 3 2π Lμωi Ri 3 = T = − FRi = Ro − Ri h z Another end-of-chapter homework problem, Chap. 2 2-90 This is one of their first exposures to CFD through FlowLab Consider the rotating viscometer of the previous problem. We make an approximation that the gap (distance between the inner and outer cylinders) is very small. Consider an experiment in which the inner cylinder radius is Ri = 0.0600 m, the outer cylinder radius is Ro = 0.0602 m, the fluid viscosity is 0.799 kg/m⋅s, and the length L of the viscometer is 1.00 m. Everything is held constant in the experiment except that the rotation rate of the inner cylinder varies. (a) Calculate the torque in N⋅m for several rotation rates in the range -700 to 700 rpm. Discuss the relationship between T and ωi (is the relationship linear, quadratic, etc.?). (b) Run FlowLab with the template Concentric_inner. Set the rotation rate to the same values as in Part (a), and calculate the torque on the inner cylinder for all cases. Compare to the approximate values of Part (a), and calculate a percentage error for each case, assuming that the CFD results are “exact”. Discuss. In particular, how good is the small-gap approximation? Note: Be careful with the sign (+ or -) of the torque. They calculate torque as a function of rpm z Solution (from solutions manual) 2-90 (a) Note that we must convert the rotation rate from rpm to radians per second so that the units are proper. When ωi is -700 rpm, we get rot ⎞⎛ 2π rad ⎞⎛ 1 min ⎞ rad ⎛ ωi = ⎜ −700 73.304 = − ⎟⎜ ⎟⎜ ⎟ min ⎠⎝ rot ⎠⎝ 60 s ⎠ s ⎝ For h = 0.0002 m, the torque is calculated (using the equation derived in the previous problem). Note, however, that since we are calculating the torque of the fluid acting on the cylinder, the sign is opposite to that of the previous problem, 2π Lμωi Ri 3 2π Lμωi Ri 3 T=− =− Ro − Ri h =− We run various rpm cases, both manually and with CFD rad ⎞ 3 ⎛ 2π (1.00 m )( 0.799 kg/m ⋅ s ) ⎜ −73.304 0.0600 m ( ) ⎟ ⎛ ⎞ N ⎝ 0.0002 m s ⎠ ⎜ 2 ⎟ kg m/s ⋅ ⎝ ⎠ = 397.445 N ⋅ m ≅ 397. N ⋅ m where we have rounded to three significant digits. We repeat for various other values of rotation rate, and summarize the results in the table below. (b) The FlowLab template was run with the same values of ωi. The results are compared with the manual calculations in the table. The agreement between manual and CFD calculations is excellent for all rotation rates. The relationship between torque and rotation rate is linear, as predicted by theory. z Solution (from solutions manual - continued) Analytical Agreement between analytical and CFD results is excellent FlowLab Discussion Since the gap here is very small compared to the radii of the cylinders, the linear velocity profile approximation is actually quite good, yielding excellent agreement between theory and CFD. However, if the gap were much larger, the agreement would not be so good. z Another end-of-chapter homework problem, Chap. 2 2-91 This is the next problem in this series Consider the rotating viscometer of the previous problem. We make an approximation that the gap (distance between the inner and outer cylinders) is very small. Consider an experiment in which the inner cylinder radius is Ri = 0.0600 m, rotating at a constant angular rotation rate of 300 rpm. The fluid viscosity is 0.799 kg/m⋅s, and the length L of the viscometer is 1.00 m. Everything is held constant in the experiment except that different diameter outer cylinders are used. The gap distance between inner and outer cylinders is h = Ro – Ri. (a) Calculate the torque in N⋅m for the following gaps: 0.0002, 0.0015, 0.0075, 0.02, and 0.04 m. (b) Run FlowLab with the template Concentric_gap. Set the gap to the same values as in Part (a), and calculate the torque on the inner cylinder for all cases. Compare to the approximate values of Part (a), and calculate a percentage error for each case, assuming that the CFD results are “exact”. Discuss. In particular, how good is the small-gap approximation? Note: Use absolute value of torque to avoid sign inconsistencies. This time we vary gap size at a fixed rpm z Solution (from solutions manual) 2-91 (a) First we convert the rotation rate from rpm to radians per second so that the units are proper, rot ⎞⎛ 2π rad ⎞⎛ 1 min ⎞ rad ⎛ = ωi = ⎜ 300 31.416 ⎟⎜ ⎟⎜ ⎟ min ⎠⎝ rot ⎠⎝ 60 s ⎠ s ⎝ When h = 0.0002 m, the torque is calculated (using the equation derived in the previous problem). Note, however, that since we are calculating the torque of the fluid acting on the cylinder, the sign is opposite to that of the previous problem, 2π Lμωi Ri 3 2π Lμωi Ri 3 We run various gap size cases, T=− Ro − Ri =− h both manually and with CFD rad ⎞ 3 ⎛ 2π (1.00 m )( 0.799 kg/m ⋅ s ) ⎜ 31.416 0.0600 m ( ) ⎟ ⎛ ⎞ N s ⎠ ⎝ =− ⎜ 2 ⎟ ⋅ 0.0002 m kg m/s ⎝ ⎠ = −170.336 N ⋅ m ≅ −170. N ⋅ m where we have rounded to three significant digits. We repeat for various other values of gap distance h, and summarize the results in the table below. (b) The FlowLab template was run with the same values of h. The results are compared with the manual calculations in the table and plot shown below. Note: We use absolute value of torque for comparison without worrying about the sign. z Solution (from solutions manual - continued) The agreement between analysis and CFD is great for small gap sizes But the agreement is not so good for the larger gap sizes z Solution (from solutions manual - continued) The agreement between manual and CFD calculations is excellent for very small gaps (the percentage error is less than half a percent for the smallest gap). However, as the gap thickness increases, the agreement gets worse. By the time the gap is 0.04 m, the agreement is worse than 50%. Why such disagreement? Remember that we are assuming that the gap is very small and are approximating the velocity profile in the gap as linear. Apparently, the linear approximation breaks down as the gap gets larger. Discussion We used a log scale for torque so that the differences between manual calculations and CFD could be more clearly seen. Students realize that their simple small-gap approximation breaks down as the gap gets larger. At this point in their study of fluid mechanics, they do not know how to calculate this flow exactly for arbitrary gap size and rpm – that is not learned until Chapter 9, the differential equations chapter. Examples: New homework & templates, Ed. 2 z End-of-chapter homework problem, Chap. 9 This is a follow-up problem to those of Chapter 2 just discussed 9-92 An incompressible Newtonian liquid is confined between two concentric circular cylinders of infinite length – a solid inner cylinder of radius Ri and a hollow, stationary outer cylinder of radius Ro. (see figure; the z axis is out of the page.) The inner cylinder rotates at angular velocity ωi. The flow is steady, laminar, and twodimensional in the r-θ plane. The flow is also rotationally symmetric, meaning that nothing is a function of coordinate θ (uθ and P are functions of radius r only). The flow is also circular, meaning that velocity component ur = 0 everywhere. Generate an exact expression for velocity component uθ as a function of radius r and the other parameters in the problem. You may ignore gravity. Hint: The result of Problem 9-91 is useful. Now we advance to Chapter 9 problems Ro Ri ωi Fluid: ρ, μ First, an analytical solution Rotating inner cylinder Stationary outer cylinder z Solution (from solutions manual) 9-92 The solution is fairly long and not repeated in its entirety here. The NavierStokes equations are solved analytically for this simple geometry, and the boundary conditions are applied. Here are the last few lines of the solution: The solution: uθ = C1 r C2 + 2 r Apply one boundary condition: R C 0 = C1 o + 2 2 Ro Apply another boundary condition: or Ri C2 Ri Ro 2 Riωi = C1 + = C1 − C1 2 Ri 2 2 Ri Solve for the constants of integration: −2 Ri 2ωi C1 = 2 Ro − Ri 2 The final equation is ⎛ Ro 2 ⎞ −r⎟ ⎜ ⎝ r ⎠ Ri 2ωi uθ = 2 Ro − Ri 2 Ro 2 C2 = −C1 2 Ro 2 Ri 2ωi C2 = 2 Ro − Ri 2 Analytical solution for any gap with This is a closed-form analytical solution. In the next problem, wesize compare CFD. z Another end-of-chapter homework problem, Chap. 9 9-93 This is the next problem in this series Glycerin (ρ = 1259.9 kg/m3, and μ = 0.799 kg/m·s) flows between two concentric cylinders as in the previous problem. The inner radius is 0.060 m, and the inner cylinder rotates at 300 rpm. The outer cylinder is stationary. Recall from Chapter 2 that when the gap between the cylinders is small, the tangential velocity of the fluid in the gap is nearly linear. When the gap is large, however, we expect the linear approximation to fail. Run FlowLab with the template Concentric_gap. Run two cases: (a) a small gap of 0.001 m and (b) a large gap of 0.060 m. For each case, plot and save the velocity profile data. Compare to the analytical prediction for both cases. Is there good agreement? How good is the linear approximation? Discuss. Now we use FlowLab (actually the same template as in Chapter 2) to compare CFD-generated velocity profiles to those generated analytically. We do this for two gaps, a small gap and a large gap. z Solution (from solutions manual) 9-93 (a) Small gap (gap = 0.001 m): We apply the equation from the previous problem to calculate the tangential velocity as a function of radius, ⎞ Ri 2ωi ⎛ Ro 2 uθ = 2 −r⎟ 2 ⎜ Ro − Ri ⎝ r ⎠ and we plot the velocity profile, uθ as a function of r, in the plot below. We run FlowLab for the same geometry and conditions, and plot the velocity profile on the same plot for comparison. The agreement is excellent (less than 0.02% error at any radius). This is not surprising since the flow is laminar, steady, etc. CFD does a very good job in this kind of situation. The small errors are due to lack of complete convergence and a mesh that could be a little finer. The profile is nearly linear as expected since the gap is small. Students compare the analytical (exact) solutions to those obtained by FlowLab for both cases, small gap and large gap. z Solution (from solutions manual - continued) The small gap results show excellent agreement as expected, and the velocity profile is nearly linear since the gap is so small. z Solution (from solutions manual - continued) (b) Large gap (gap = 0.06 m): We repeat for the larger gap case. The plot is shown below. Again the agreement is excellent, with errors less than 0.1% for all radii, but the profile is not linear – the linear approximation breaks down when the gap can no longer be considered small. This time, the large gap results show excellent agreement as well since we have not made a small-gap approximation; but the profile is not linear. z Solution (from solutions manual - continued) Discussion Problems such as this in which a known analytical solution exists are great for testing CFD codes. The fluid properties did not enter into the calculations – viscosity affects only the transient solution, not the final flow field. • Note how this one simple problem yields several homework problems – even across chapters. • Students get a feel for using CFD and compare the results with analytical analysis. • They see where their simplified analysis works well and where it breaks down (e.g., small gap approximation breaks down when the gap is too large). • These types of analytical/FlowLab problems have been added to nearly all the chapters in Ed. 2. FlowLab Details for this problem Here is the mesh that FlowLab generates for the same geometry as in the exact analysis. Residual plot (iteration takes only a couple minutes) They look at velocity magnitude contours They plot velocity magnitude vs. radial position. They save these data points to an Excel file. Live FlowLab Demonstration We will demonstrate the templates called “Concentric_gap” and “Submerged_plate_angle” [These templates will have corresponding endof-chapter homework problems in Ed. 2 of the Çengel-Cimbala undergraduate fluids textbook.] If time, also show some other templates “live”. Summary z It is possible to introduce the fundamentals of CFD into an undergraduate fluids course (I do it in only one class period, plus homework) z FlowLab software enables students to experience thedetails key – can CFD without getting boggedHomework down inisthe introduce students to CFD z Each FlowLab exercise has two objectives: without taking much class time – Enhance understanding of fluid mechanics – Teach the capabilities and limitations of CFD z Most of the new templates in Ed. 2 of the fluids textbook by Çengel and Cimbala compare analytical solutions to those obtained with CFD for enhanced learning and good exposure to CFD How to Integrate CFD into an Undergraduate Fluids Course zDevelop the continuity and Navier-Stokes equations for fluid flow, as usual zShow how to solve simple problems analytically (solve N-S equations): – Couette flow between plates This Thisisiswhat whatisis – Fully developed pipe flow new normally – added done to in the an course. introductory – Etc. fluid mechanics waycourse. to do the zThen, introduce CFD as a same thing, but with a computer. How to Integrate CFD into an Undergraduate Fluids Course zThe CFD lecture takes only about one class period, where we briefly explain: – Computational domain and types of grids – Boundary conditions and initial guesses – The concept of residuals and iteration – Post-processing (contour plots, etc.) zIn-class “live” demonstration of FlowLab zAssign homework requiring FlowLab The homework is where students get hands-on CFD practice Sample lecture notes from Fall 2005, Penn State (the lecture where CFD was presented for the first time) These notes are directly from my lecture notes, given using a tablet PC, and posted on the Internet for students to download Still-Slide Back-Up to Live Demonstration of FlowLab template “Diffuser_angle” Example: Flow through a conical diffuser z Fluid Mechanics Learning Objective: Compare pressure recovery in conical diffusers of half-angle 5° to 90° z CFD Objective: Observe streamline patterns and flow separation as diffuser half-angle increases; compute pressure recovery for all cases Flow through a conical diffuser θ V D2 D1 x L1 L2 Wall Wall Pout Pin V Geometry and dimensions x Axis Computational domain, assuming axisymmetric flow User Interface for FlowLab Operation options Main working window Result table Overview window Graphical display window Display options Flow through a conical diffuser Diffuser section x Hybrid mesh for the 5o half-angle conical diffuser Flow through a conical diffuser (continued) X-Y plot of residuals for the conical diffuser case, θ = 5o Flow through a conical diffuser (a) θ = 5o x (g) θ = 20o x (b) θ = 7.5o x (h) θ = 25o x (c) θ = 10o x (i) θ = 30o x (d) θ = 12.5o x (j) θ = 45o x (e) θ = 15o x (k) θ = 60o x (f) θ = 17.5o x (l) θ = 90o x Streamlines through conical diffusers of various half-angles Flow through a conical diffuser Pressure difference from inlet to outlet of a conical diffuser as a function of diffuser half-angle θ ΔP 5 7.5 10 12.5 15 17.5 20 25 30 32.5 35 37.5 45 60 75 90 -49.1371 -47.7787 -44.9927 -42.4013 -39.6981 -37.6431 -36.0981 -32.7173 -29.9919 -23.2118 -21.6434 -21.0490 -19.6571 -18.7252 -18.1364 -18.3018 Flow through a conical diffuser -10.0 Pressure difference from inlet to outlet of a conical diffuser as a function of diffuser half-angle -20.0 ΔP (Pa) -30.0 -40.0 -50.0 0 50 θ (degrees) 100 Flow through a conical diffuser (a) θ = 5o (b) θ = 30o (c) θ = 45o Pressure contours through a conical diffuser of three different half-angles. Colors range from dark blue at -60 Pa to bright red at 0 Pa gage pressure. Flow through a conical diffuser (a) θ = 5o (b) θ = 30o (c) θ = 45o Contours of turbulent kinetic energy through a conical diffuser of three different half-angles. Colors range from dark blue at 0 m2/s2 to bright red at 3.5 m2/s2 Still-Slide Back-Up to Live Demonstration of FlowLab template “Block_mesh” Example: Flow over a rectangular block z Fluid Mechanics Learning Objective: Compare drag coefficient with empirical results z CFD Objective: Learn to refine a mesh until grid independence is achieved