Benchmark against AVL

Updated October 21, 2026

------------------ WORK IN PROGRESS ------------------



Contents


Back to main table of contents



Purpose

The goal of this study is to

Note: A previous version of this page presented a comparison of flow5 v7.24 results to AVL 3.36. This updated benchmark presents a comparison of flow5 v7.58 to AVL 3.36. This latest version of flow5 corrects several bugs and issues:




Back to top


Axes and frames of reference

The axes in which the forces, moments and aerodynamic coefficients are expressed are detailed here.
Lift and drag are expressed in wind axes.
The pitching moment is around the wind y-axis which is the same as the geometric y--axis.
The rolling and yawing moments are around respectively the wind's x and z axes.

The axes are set in AVL as follows:
A xis orient. : Geometric axis orientation, X aft, Z up
R ate,mom axes: Rates,moments about Stability Axes, X along Vinf




Derivative calculation

The methods to calculate the derivatives differ in AVL and flow5. To the author's understanding, the methods implemented in AVL are the ones described in "Flight Vehicle Aerodynamics", Mark Drela 2014.

Stability derivatives

AVL

The method takes advantage of the linear dependency of the panel forces with the 6 degrees of freedom in motion. This allows to express the stability derivatives explicitly, and to and solve them by direct back-substitution and linear combination which makes their calculation fast and accurate.

flow5 v7.58

The derivatives are evaluated using finite differences.

The method used by AVL could be implemented for flow5's VLM. However it is not transposable to the thick surface panel methods in which the pressure forces are derived from the surface gradient of the doublet densities and do not depend linearly on the freestream's angle of attack and sideslip.

Control derivatives

The rotations of the control surfaces change both the linear system's matrix and the right hand side (RHS) vector. Strictly speaking, the problem should therefore be built and solved entirely for each derivative. This however would be computationally very expensive. A standard procedure is to assume that the change to the influence matrix is small, so that the LU-factorized matrix can be reused as-is, and that only the change to the RHS needs to be implemented. This simplification reduces the computations to the construction of the RHS and to its back-substitution.
This is how AVL, flow5 and other panel codes proceed. A numerical experiment was carried out with flow5 to evaluate the error implicit in this approximation and is presented hereafter.

AVL

The boundary condition are linearized since only the first order development is required, and the derivatives are calculated explicitly and solved by direct back-substitution and linear combination of deflections.

flow5 v7.58

The problem cannot be linearized for the same reason as in the case of the stability derivatives. The derivatives are calculated by finite differences, with an RHS vector being constructed and back-substituted for each derivative.




Back to top


Test case

Description

The test case is the simple 1.5 m span fictitious plane illustrated in the image on the right.
To keep the comparisons with AVL relevant:

  • The fuselage was not included in the model.
  • All flow5 analyses are inviscid.
  • The first set of calculations in flow5 was done with the VLM2.

The corresponding flow5 and AVL project files can be downloaded here.

Model validation: linear analyses

The purpose of these first runs is to check that the analyses are set up identically in flow5 and AVL.

Type 1 comparisons at β=0°

Both AVL and flow5 predict that the pitching moment is zero at α = 1.07°.

Type 5 comparisons at α=0°

Type 5 comparisons at α=7°

Stability derivatives

VLM2, no fuselage

The intent of this first run is to compare the stability derivatives using flow5's VLM2 quad vortices method.

The stability derivatives are compared at the zero pitching moment angle, i.e. α = 1.08° for both programs.

AVL 3.36 flow5 7.24 dev./AVL flow5 7.58 dev./AVL
CLa 5.469969 5.4932 0.4% 5.49312 0.4%
Cma -1.389971 -1.4483 -4.0% -1.4483 -4.0%
AVL 3.36 flow5 7.24 dev./AVL flow5 7.58 dev./AVL
CYb -0.111854 -0.129153 -13.4% -0.129152 -13.4%
Clb -0.064076 -0.056346 13.7% -0.056345 13.7%
Cnb 0.035702 0.042822 -16.6% 0.042821 -16.6%

AVL 3.36 flow5 7.24 dev./AVL flow5 7.58 dev./AVL
CYp -0.073157 -0.093627 -21.9% -0.092333 -20.8%
Clp -0.555007 -0.55689 -0.3% -0.55689 -0.3%
Cnp -0.016401 -0.002399 89.4% -0.00239861 85.4%
AVL 3.36 flow5 7.24 dev./AVL flow5 7.58 dev./AVL
CLq 9.345922 9.489867 -1.5% 9.5355 -2.0%
Cmq -15.269398 -15.855615 -3.7% -15.9639 -4.4%
AVL 3.36 flow5 7.24 dev./AVL flow5 7.58 dev./AVL
CYr 0.091689 0.096878 -5.4% 0.10604 -15.7%
Clr 0.056059 0.007483 649.2% 0.055192 1.57%
Cnr -0.031223 -0.035711 -12.6% -0.03601 -13.3%

The derivatives are in reasonable agreement except for Cnp. To understand the discrepancy, the derivatives have been plotted as a function of the angle of attack.

As it appears, the value of the derivative Cnp estimated by AVL depend strongly on the angle of attack, which makes them very sensitive to the point of evaluation. No obvious explanation could be found for this difference of sensitivity exhibited by both programs.

The other lateral derivatives are sensitive to the angle of attack to various degrees.

On the other hand, the longitudinal derivatives are not sensitive to the angle of attack.

Influence of the panel method


AVL (ref) VLM2 deviation VLM1 deviation Tri-uni thick deviation Tri-uni thin deviation
CXa
0.1221
0.1734
0.1292
0.1247
CZa -5.4700 -5.4931 0.42% -5.4943 0.4% -5.7684 5.5% -5.6370 3.1%
Cma -1.3900 -1.4483 4.20% -1.4497 4.3% -1.5145 9.0% -1.2228 -12.0%
CXq 0.1195 0.1525 27.63% 0.0093 -92.2% 0.1584 32.6% 0.1613 35.0%
CZq -9.3459 -9.5355 2.03% -9.5366 2.0% -9.8515 5.4% -9.6236 3.0%
Cmq -15.2694 -15.9639 4.55% -15.9903 4.7% -16.6044 8.7% -10.8727 -28.8%
CYb -0.1119 -0.1292 15.46% -0.1283 14.7% -0.1209 8.1% -0.1190 6.4%
Clb -0.0641 -0.0563 -12.07% -0.0564 -12.0% -0.0680 6.1% -0.0485 -24.4%
Cnb 0.0357 0.0428 19.94% 0.0430 20.5% 0.0443 24.0% 0.0278 -22.2%
CYp -0.0732 -0.0923 26.21% -0.0759 3.8% -0.0987 34.9% -0.0980 33.9%
Clp -0.5550 -0.5569 0.34% -0.5569 0.3% -0.5694 2.6% -0.3859 -30.5%
Cnp -0.0164 -0.0024 -85.38% -0.0024 -85.2% -0.0148 -9.9% 0.0012 -107.3%
CYr 0.0917 0.1060 15.66% 0.1054 15.0% 0.0998 8.8% 0.0977 6.6%
Clr 0.0561 0.0552 -1.55% 0.0552 -1.5% 0.0592 5.5% 0.0490 -12.6%
Cnr -0.0312 -0.0360 15.32% -0.0362 15.9% -0.0362 15.9% -0.0246 -21.2%

The scattering of the predictions can be quite large depending on the derivative, but is not easy to interpret. Cnp is the standout derivative.






Back to top


Control derivatives

Result comparison

The derivatives have been evaluated in flow5 using the VLM2 method.

The control derivatives are calculated at the zero-pitching moment angle.

Aileron

AVL flow5 deviation
CXd
CYd -0.000725 0.00076154 5.0%
CZd
Cld -0.005344 0.0054668 2.3%
Cmd
Cnd -0.000032 -0.00000049 -101.5%

Elevator

AVL flow5 deviation
CXd
CYd
CZd 0.005236 -0.0057114 9.1%
Cld
Cmd -0.02009 -0.021882 8.9%
Cnd

Rudder

AVL flow5 deviation
CXd
CYd 0.001242 0.0014878 19.8%
CZd
Cld 0.000067 0.000075661 12.9%
Cmd
Cnd -0.000489 -0.00058663 20.0%

The derivative which stands out is again Cnd, for the same reason as in the case of the stability derivatives. This is illustrated by a plot of Cnd vs. the angle of attack.

Not only is the absolute value of the derivative small, but it also exhibits in AVL an important variation with the angle of attack. This makes its estimation very sensitive to the setup of the model and of the analysis.


Sanity check: influence matrix

A numerical experiment was made in flow5 to evaluate the error implicit in the assumption that the linear system's matrix is not modified by the deflection of the control surfaces. The influence matrix was rebuilt for each of the aileron, elevator and rudder controls, and the problem was entirely solved for the derivatives in each case. The results were compared to those presented above in the VLM case.

Aileron

Ref. Mod. matrix deviation
CXd
CYd 0.00076154 0.0007525 1.2%
CZd
Cld 0.0054668 0.0054548 0.2%
Cmd
Cnd -0.00000049 -0.00000046 6.6%

Elevator

Ref. Mod. matrix deviation
CXd
CYd
CZd -0.0057114 -0.0057132 0%
Cld
Cmd -0.021882 -0.021889 0%
Cnd

Rudder

Ref. Mod. matrix deviation
CXd
CYd 0.0014878 0.0014955 -0.5%
CZd
Cld 0.000075661 0.000076071 -0.5%
Cmd
Cnd -0.00058663 -0.00058967 -0.5%

It appears that the differences are minor and no larger than the variations seen between the VLM2 and the tri-uniform method, which justifies the approximation.




Back to top


Conclusion and recommendations

Conclusions

The main conclusions of this sensitivity study are that

Recommendations


Back to top