Dimensional Analysis and the Pi Theorem
Dimensional Analysis and the Pi Theorem: Theoretical Foundations
Fundamental Principles of Dimensional Analysis
Professor, I thought dimensional analysis was just about "checking units," but is it really that important in CFD?
It's more than that. Dimensional analysis is a powerful technique for minimizing the number of independent parameters governing a physical problem. Buckingham's π theorem is at its core.
The π theorem states this: If a physical phenomenon is described by $n$ physical quantities containing $k$ fundamental dimensions (like mass M, length L, time T, etc.), then the phenomenon can be completely described by $n - k$ independent dimensionless numbers.
What does that mean concretely?
For example, consider the pressure loss $\Delta p$ in a circular pipe. The relevant physical quantities are $\Delta p, \rho, U, D, \mu, L, \varepsilon$ — 7 in total. The fundamental dimensions are M, L, T — 3 in total, so it can be expressed by $7 - 3 = 4$ dimensionless numbers.
Here, $\text{Re} = \rho U D / \mu$ is the Reynolds number, $L/D$ is the aspect ratio, and $\varepsilon/D$ is the relative roughness.
Reducing 7 variables to 4 is a big deal!
Key Dimensionless Numbers and Their Physical Meaning
Let's organize the dimensionless numbers frequently used in CFD. Each represents the "ratio of two forces or effects."
| Dimensionless Number | Definition | Physical Meaning |
|---|---|---|
| Reynolds Number Re | $\rho UL/\mu$ | Inertial Force / Viscous Force |
| Mach Number Ma | $U/c$ | Flow Velocity / Speed of Sound (Indicator of Compressibility) |
| Froude Number Fr | $U/\sqrt{gL}$ | Inertial Force / Gravitational Force |
| Weber Number We | $\rho U^2 L/\sigma$ | Inertial Force / Surface Tension |
| Strouhal Number St | $fL/U$ | Characteristic of Unsteady Oscillation |
| Prandtl Number Pr | $\nu/\alpha = c_p\mu/k$ | Momentum Diffusivity / Thermal Diffusivity |
| Nusselt Number Nu | $hL/k$ | Convective Heat Transfer / Conductive Heat Transfer |
| Grashof Number Gr | $g\beta\Delta T L^3/\nu^2$ | Buoyancy / Viscous Force |
I knew about the Reynolds number, but I didn't realize there were so many.
Choosing the appropriate dimensionless numbers for the problem is where the skill in dimensional analysis comes in. For example, in natural convection, the Rayleigh number $\text{Ra} = \text{Gr} \cdot \text{Pr}$, the product of $\text{Gr}$ and $\text{Pr}$, becomes the governing parameter.
Non-dimensionalization of the Navier-Stokes Equations
What happens when you apply dimensional analysis to the Navier-Stokes equations?
Non-dimensionalizing with characteristic length $L$, characteristic velocity $U$, characteristic time $L/U$, and characteristic pressure $\rho U^2$, the incompressible N-S equations become:
From this form, we can see that a larger Re means a smaller contribution from the viscous term, and a larger Fr means a smaller influence from the gravity term. Ultimately, the CFD solution is determined by these dimensionless numbers.
So that's why matching the Re number is so important in wind tunnel testing!
Exactly. If the geometry is similar and Re matches, the dimensionless solution of the flow field will be identical. This is the dynamic similarity law.
Froude's Ship Model Experiments — The Discovery of Scaling Laws
In the 19th century, naval architect William Froude tackled the problem of "whether a ship's performance could be predicted from resistance measurements on a model." What he discovered was the scaling law: "If the Froude number (Fr = V/√(gL)) is equal, the wave patterns are geometrically similar." This was one of the prototypes of the π theorem. Even today, ship design fundamentally combines model tank testing and CFD, and matching the Froude number is essential. Without dimensional analysis, there would be no valid basis for extrapolating results from a 1/100 scale model test to a full-scale ship.
Computational Methods for Dimensional Analysis and the Pi Theorem
Implementation of Non-dimensionalization in CFD
How is dimensional analysis actually used in CFD? When you put it into the solver, it's dimensional, right?
Good question. There are two implementation approaches.
Approach 1: Calculate in dimensional form, then non-dimensionalize in post-processing
- Input physical quantities directly in SI units
- Divide results by characteristic quantities for non-dimensionalization (e.g., $C_p = \Delta p / (\frac{1}{2}\rho U_\infty^2)$)
- This is the mainstream approach in Ansys Fluent or STAR-CCM+
Approach 2: Solve non-dimensionalized equations from the start
- Input variables normalized by characteristic quantities
- Used in research codes or custom solvers in OpenFOAM
- Advantage: Only the Re number, for example, can be set as a parameter
So dimensional is standard in commercial software, huh.
Scaling Techniques Using Similarity Laws
The biggest application of dimensional analysis is scaling using similarity laws. Identify the dimensionless numbers that must match between the actual machine and the model to determine experimental conditions.
For example, consider using a 1/5 scale model for automotive wind tunnel testing.
If using the same air (same $\rho, \mu$), then $U_m = 5 \times U_f$ is required. For a real car at 100 km/h, the model would need 500 km/h, making compressibility effects non-negligible.
So perfect similarity can be difficult sometimes.
Right. When perfect similarity is impossible, we use partial similarity. Match only the dominant dimensionless numbers and compromise on those with minor influence. A typical example is ships: match the Fr number (for wave-making resistance) and handle the Re number mismatch with correction formulas.
Calculation and Monitoring of Dimensionless Numbers in CFD
Are there any points to note when calculating dimensionless numbers from CFD results?
The choice of characteristic quantities influences the interpretation of results. Pay attention to the following points.
| Dimensionless Quantity | Selection of Characteristic Quantity | Points to Note |
|---|---|---|
| Drag Coefficient $C_D$ | Frontal Projected Area $A$, $U_\infty$ | Unify the definition of the area |
| Pressure Coefficient $C_p$ | Free-stream Velocity $U_\infty$, Static Pressure $p_\infty$ | Choice of reference pressure |
| Friction Coefficient $C_f$ | Wall Shear Stress $\tau_w$ | Distinguish between local value vs. average value |
| $y^+$ | $u_\tau = \sqrt{\tau_w/\rho}$ | Quality indicator for the first layer of wall mesh |
Especially $y^+$ is an important dimensionless number for judging the applicability of turbulence models. For wall function models, $30 < y^+ < 300$ is a guideline; for wall-resolved models, $y^+ \approx 1$ is the target.
I didn't realize $y^+$ connected to dimensional analysis here.
Dimensionless Numbers are Tools that "Drastically Reduce the Number of Calculations"
The practical power of dimensional analysis lies in "reducing the number of parameters." For example, pressure loss in a pipe seems to depend on 6 variables: "flow velocity V, density ρ, viscosity μ, pipe diameter D, length L, roughness ε." But using the π theorem, it compresses to a function of just 2 variables: "Re number and ε/D (relative roughness)." This is the Moody chart. An experimental plan varying 6 variables independently would require 6^3=216 tests, but with dimensionless organization, a few dozen suffice. The same applies to parameter studies in CFD; organizing with dimensionless numbers first can drastically reduce the number of calculation cases.