|INSIGHT: In CFD We Trust (Or Not)
CFD is only as good as its ability to accurately predict flow behavior. Without it, you've got to design conservatively and lean on costly prototypes and testing; with it, you're more agile and better able to innovate. In this month's issue of The Flow, we sit down with Steven Allmaras and Bob Ni to discuss predictive accuracy, when you know you've got it, and how best to get there in a commercial setting. Steve is the Vice President of CFD at ADS. Prior to joining ADS, Steve spent 22 years at the Boeing Company, where he was a key member of the CFD methods group and a co-author of the Spalart-Allmaras turbulence model. Bob is the Chairman and CTO of ADS. Prior to founding the company, Bob spent 28 years at Pratt & Whitney leading turbomachinery CFD efforts.
FLOW: How do you define predictive accuracy?
BOB: Strictly speaking, it's a measure of how well CFD predicts the true performance of a design, as measured by experimental data.
FLOW: And how do you know when your CFD has achieved good predictive accuracy?
BOB: As we all know, it's impractical to expect a perfect match between predictions and data in commercial design—there just simply isn't enough time, resource or budget to get there. From a practical point of view, you know you're in good shape when your CFD: (1) consistently reveals fow insights that help to discern design improvements from design mistakes; (2) demonstrates proper trend predictions over a body of work, not just a single design; (3) demonstrates good (but not necessarily exact) agreement with experimental results; (4) engenders a sense of trust with the designer.
FLOW: What's interesting is that you emphasize directional accuracy over precise matching to data. Is this correct?
BOB: Yes. Our collective experience suggests that the first priority should be to get to the point where your CFD consistently helps you to make good design decisions for the range of cases of interest. Only after this is in place will most organizations focus on tightening tolerances.
STEVE: I'd also add that tight tolerances are usually not necessary early in the design cycle since you are looking for directional guidance. When it does become more necessary later in the cycle, most organizations utilize validation databases to help them "map" the CFD predictions back to reality.
FLOW: Makes sense. What are the key drivers of predictive accuracy?
STEVE: There are typically four sources of error impacting predictive accuracy: physical modeling error, geometry error, discretization error and solution error.
Physical modeling error relates to the choice of governing equations and how well it models actual flow behavior. These types of errors can stem from the limitations of a particular turbulence or transition model, for example, or from the incorrect specification of boundary conditions. Geometry error relates to differences between physical and modeled geometries, such as the discrepancies between hot geometries typically used for CFD simulation and cold geometries coming out of the CAD system. Discretization error relates to errors stemming from representing the governing equations in discrete space and time. Finally, solution error results from insufficient solution convergence of the nonlinear algebraic equations due to discretization of the governing equations.
FLOW: How do you suggest minimizing these errors to improve predictive accuracy?
STEVE: To minimize these errors we recommend incorporating the following steps into your standard work process if you haven't already:
- Account for hot geometry. A major source of geometry error is due to the difference between an airfoil geometry in its stationary (or "cold") state versus its running (or "hot") state. Be wary of using CAD geometry data as input for CFD simulation unless it properly reflects the hot geometry. This will in turn ensure a higher fidelity representation of the flow field and help to improve predictive accuracy.
- Verify your boundary conditions and selection of analysis methods with those in the know. To minimize physical modeling error, seek out your aerodynamicist or experimentalist to ensure the proper specification of boundary conditions and selection of analysis methods and turbulence/transition models.
- Conduct a mesh refinement study. To minimize spatial discretization error for fixed grid methods, always conduct a mesh refinement study to ensure predictions will be independent of mesh density. Based on your findings, standardize all meshing for the case using this mesh density setting.
- Conduct a solution convergence study. To minimize temporal discretization error, conduct a mesh refinement study in time (i.e. rerun the case at 1/2 the time step). For periodic flows, run the simulation long enough to ensure the time-averaged results are invariant. The actual number of iterations required will depend on the length of the relevant period and how quickly initial transients are eliminated. Once quantified, standardize all of your runs using this iteration count. Note that tighter convergence tolerances will also help to reduce solution error but at the cost of longer run times.
BOB: I'd like to add that it's equally as important to ensure the experimental data is properly reviewed and interpreted before comparing with predictions. We've definitely seen situations where calculations of certain parameters are carried out incorrectly or differently than the corresponding CFD calculation, leading to perceived mismatch where in fact there may be none at all.
FLOW: What techniques are on the horizon that have the potential to materially improve predictive accuracy?
BOB: There are several promising approaches, but as always, one needs to balance predictive accuracy against turnaround time and cost. For example, we think direct numerical simulation (DNS) is very promising for low Reynolds number laminar flow, but it is extremely computationally expensive and unlikely to be of practical value during design for awhile. In contrast, one area where we do see a near term opportunity to improve predictive accuracy is with 3D multistage unsteady simulation. As turbomachinery designs have continued to shrink and bear heavier loads for improved efficiency and performance, it's become much more challenging to rely solely on traditional 3D steady analysis for good directional guidance. 3D unsteady improves predictive accuracy by providing time-varying insight during design to the inherently unsteady flows that can adversely impact durability and performance. This type of early warning system will prove invaluable to designers as it gives them the means to anticipate and mitigate problems before committing to hardware.
FLOW: Thanks Steve. Thanks Bob.
STEVE, BOB: You're welcome.
CASE STUDY: Computational Assessment of the Aerodynamic Performance of a Variable-Speed Power Turbine for Large Civil Tilt-Rotor Application In this paper from the American Helicopter Society 67th Annual Forum, Dr. Jerry Welch from NASA Glenn Research Center describes the design and aerodynamic analysis of a variable-speed power turbine for large civil tilt-rotor application. <more>
TECHTIPS: Meshing Highly Pitched Airfoils Using Code Wand Highly pitched airfoils such as aircraft propellers or wind turbine blades can be meshed easily using the IFUTURE parameter in Code Wand. <more>
Welcome to The Flow
Welcome to The Flow, a newsletter for monthly insights on turbomachinery CFD published by AeroDynamic Solutions, Inc.
Each month we'll spotlight a topic of interest, discuss a case study and/or provide useful pointers about how to get the most out of the ADS CFD system.
You are receiving this email because you or someone else you know thought you would be interested. To unsubscribe please click here. We value your privacy.