FASTFLO I - Fully automatic CFD system for inviscid flow simulations

FASTFLO I - Fully automatic CFD system for inviscid flow simulations

In the Brite-Euram project FASTFLO an automatic CFD (Computational Fluid Dynamics) system for three-dimensional flow simulations for complex aircraft configurations is developed (Ref. [1]). The objectives of this project are defined in terms of

1. CFD-problem-turnaround time, and
2. Accuracy of aerodynamic quantities.

The FASTFLO consortium consists of 6 partners: three research establishments (NLR, DLR and FFA), two aircraft manufacturing companies (Daimler Benz Aerospace and SAAB AB) and one engineering company (Ingenieur Büro Kretzschmar IBK; SME). The present paper provides a critical assessment of the automation level and the accuracy of the CFD system under development. Applications will be discussed to demonstrate the capabilities of the FASTFLO CFD system.

Introduction and rationale

Current forecasts as published by commercial airplane companies foresee a steady growth of air traffic and replacement of ageing aircraft over the next 20 years (Refs. [2,3]). To remain competitive on the international airliner market these airplane companies are under pressure to change continuously to more cost efficient development of new aircraft and derivatives. CFD-technology for improved aerodynamic design reducing development costs and allowing faster aircraft development cycles is one of a number of key technologies urgently needed by the European aerospace industry (Ref. [4]).

For CFD technology to have an impact on the aerodynamic design of airplanes the first requirement to be satisfied is that the CFD-problem-turnaround-time (incl. grid generation and aerodynamic post-processing) must be of the order of a day to a week, or less. Aerodynamic analysis is a process of looking at a significant number of flow conditions (lift coefficients, Mach numbers, Reynolds numbers) for more than one geometric variant such that a large number of calculations have to be made. If the application of CFD codes does not yield results at this industrial time scale the impact of CFD-technology on aerodynamic design will be reduced (Ref. [5]).

A second requirement which needs to be met by CFD tools for the development of commercial transport aircraft is high accuracy of predicted aerodynamic forces such that computed drag, pitching moment and lift can be relied upon to reduce the risks involved in airplane design. This second requirement translates for example into better turbulence models, and extreme grid resolution or automatic, adaptive grid generation if the first requirement (CFD-problem-turnaround-time) is also to be satisfied simultaneously.

Primary and secondary objectives

The primary objective of the FASTFLO project is to develop an automatic CFD system based on the three-dimensional Reynolds-averaged Navier-Stokes equations applicable to complete aircraft configurations e.g. aircraft with engines and high-lift systems; this CFD-system will have to satisfy the following primary requirements for industrial CFD:

The FASTFLO CFD system is based upon the hybrid (prismatic/tetrahedral) grid approach which has the potential to satisfy these two basic requirements on the level of the Reynolds-averaged Navier-Stokes equations (primary objective). At present the Euler equations are used as a stepping stone (secondary objective) towards the Navier-Stokes level. Technology ready multi-block methods do not satisfy the first requirement (with respect to turnaround time) for complex geometries. The technical limitation of the multi-block method lies in the multi-block grid generation process, which is internationally recognised as a time-consuming process in case of complex geometries despite significant efforts to shorten the turnaround time. For example in the ENIFAIR project a calendar time of one year is planned for the first generation of a viscous multi-block grid for the wing-body-pylon-engine configuration with high-lift devices. Although a second multi-block grid for the same configuration can be generated in a shorter period, a CFD-problem-turnaround time of a week remains a distant achievement for very complex configurations using multi-block methods. In the FASTFLO project this CFD-problem-turnaround time is to be reduced to a week at most by introducing an automated grid generation process.

Programmatic aspects, organisation, planning and costs

The automated CFD system is developed by the FASTFLO consortium which consists of 6 partners: three research establishments (NLR, DLR, FFA), two aircraft manufacturing companies (Daimler Benz Aerospace and SAAB AB) and one engineering company (Ingenieur Büro Kretzschmar IBK; SME). The small absolute size of this consortium is a necessity because the common code development is characterised by highly interdependent work packages. NLR is responsible for project co-ordination; the individual role of each partner in the FASTFLO consortium is presented in Table 1.

No. Organisation Function

1 Nationaal Lucht- en Ruimtevaartlaboratorium NLR Co-ordinator; task manager; developer

2 Deutsche Forschungsanstalt für Luft- und Raumfahrt e.V. Task manager; developer

3 Flygtekniska Försöksanstalten (FFA) Developer

4 SAAB AB End-user of results

5 Ingenieur Büro Kretzschmar Developer; end-user of results

6 Daimler Benz Aerospace AG End-user of results

Table 1 Companies and research establishments and their respective function in the project

The FASTFLO-project effectively started in April 1996. Duration of the project is two years and total costs of the FASTFLO-project amount to 1.5 MECU. The FASTFLO project is organised into three tasks, which are:

where the responsible partner for Task management is indicated. In Task I and II algorithmic components of the FASTFLO CFD system are developed which are integrated in Task III into one common code.

Functionality of the FASTFLO CFD system

An overview of the algorithmic components in the FASTFLO CFD system is shown in Figure 1. Starting point of the FASTFLO CFD system is the master geometry of an aircraft configuration. This master geometry is represented by either multi-block based curves and surfaces or IGES 5.1 curves and surface patches (CAD-format). Important is that the geometric representation of the aerodynamic aircraft configuration is sufficiently continuous, at least C0-continuous (airtight) at junction lines.

In the surface triangulation algorithm (see Figure 1) the surfaces of the geometric representation are triangulated. A distribution function in space controls the size of the edges in the surface triangulation. This distribution function is defined by means of source terms and a uniform mesh size.

Fig. 1 Algorithmic components of the CFDsystem FASTFLO

In the three-dimensional hybrid grid generation algorithm a prismatic grid is generated starting from the surface triangulation of the geometric representation. The tetrahedral grid is generated in the remaining part of the flow domain which is bounded by the triangulation of the farfield boundaries, the triangulation of symmetry plane(s) and the outer boundary of the prismatic grid. Subsequently the hybrid grid is formed by connecting the prismatic grid and tetrahedral grid.

A pre-processing algorithm is employed to achieve optimal vector and parallel performance in the flow calculation algorithm on two memory architectures: shared and distributed. In the flow calculation algorithm the three-dimensional Euler and Navier-Stokes equations are discretised based upon the vertex-based approach. Multigrid acceleration is accomplished using an agglomeration algorithm.

One of the advantages of using the hybrid grid approach is that it provides a natural framework for solution-adaptive refinement. Grid adaption is based on local grid refinement using a user-selected adaption indicator.

The post-processing algorithm allows a user to select and calculate aerodynamic quantities that are of interest to him in graphical form. Aerodynamic forces, moments and aerodynamic coefficients (drag and lift) are computed.

In order to improve the workflow using the FASTFLO CFD components, and to present the capabilities as a uniform system to the user, a system integration tool is used to integrate the system.

Results

Four applications are introduced to review the characteristics of the FASTFLO CFD system, see Table 2 and 3. Examples are shown for case 2 in Figure 2, for case 3 in Figure 3 and for case 4 in Figure 4. Starting point for all four cases is a multi-block based surface description of the aircraft configuration. The respective grids (tetrahedral and hybrid) are refined at locations of special interest such as the wing leading edge, wing trailing edge, wing tip and the nose region. The flow solution is obtained by taking a sufficient number of multigrid cycles.

No. test case cpu-time
grid-gen. cpu-time
flow. calc. Perc. of an
8 hour working day

1 wing-alone 8m 26s 35 m 9%

2 wing-body 52m 54s 2h 14m 38%

3 wing-body-pylon-nacelle 17 m 13 m 6.3%

4 generic fighter 5 m 8m 3.1%

Table 2 Computing time for grid generation and Euler flow calculation algorithms for each test case measured in terms of cpu-time and percentage of an 8 hour working day (m=minutes; s=seconds); cpu-time of grid generation measured on workstation (MIPS R10000); cpu-time of flow calculation (including pre-processing) measured on single processor of NEC SX4/16 supercomputer

case curves surfaces
nodes
aerod.
surface prism
layers nodes
prismatic
grid nodes
tetrahedral
grid

1 31 16 10970 10 120670 194579

2 39 14 65085 - - 694946

3 88 34 14252 5 85512 61452

4 298 130 11829 - - 38769

Table 2 Dimensions of the generated tetrahedral and hybrid grid for each test-case

Case Mach alpha Multigrid cycles Orders of convergence

1 0.84 3.06 250 3.1

2 0.8 2.2 300 4.6

3 0.8 2.2 200 2.7

4 0.9 4.12 500 4

Table 4 The flow-conditions, the number of multigrid cycles and the order of convergence of the flow calculation algorithm (using Jamesons's scheme) for each test-case

Fig 2 Surface triangulation for the AS28G wing-body configuration Fig 3 Pressure distribution calculated for the AS28G wing-body-pylon-nacelle configuration on the hybrid (prismatic/tetrahedral) grid; Mach = 0.8; alpha = 2.2 degrees Fig. 4 Surface triangulation for a generic fighter configuration; geometry from J.I. van den Berg (Ref. 6)

The computing time of the FASTFLO CFD system for these four cases is limited as can be observed in Table 2. The grid dimensions for each case can be found in Table 3. The Euler flow calculations using the classical Jameson scheme are performed on a single processor of the NEC SX4/16. In Table 4 it can be seen that the residuals have converged approximately 3 to 4 orders of magnitude for each case. The grid generation steps are run on a workstation (MIPS R10000). The results demonstrate that the computing time of the CFD system is relatively short; If the parallel version of the Euler flow solver is used the computing time does not seem to be the bottleneck to reach the first objective. The CFD-problem-turnaround-time however also depends on the automation level.

Achievements

Turnaround time
The results for the four applications presented in the previous section were obtained within one working day. From Table 2 it can be observed the FASTFLO CFD system has already a short CFD-problem-turnaround time. The computing time needed for the grid generation and flow calculation algorithms to generate a flow solution with engineering accuracy is in the range of 5 to 40 percent of an 8 hour working day. Since the flow calculation was performed on a single processor NEC SX4/16 a further reduction of the CFD-problem-turnaround time is to be expected by using the parallel capability of the FASTFLO CFD system.

Fig. 5 Pressure coefficient distribution for the ONERA M6 wing-alone at y=0.65; Comparison with multi-block result and wind tunnel experiment; results from J. E.J. Maseland (Ref. 7) Fig. 6 Skin friction distribution for the ONERA M6 wing-alone at y = 0.65.; Comparison with multi-block result; results from J.E.J. Maseland (Ref. 7)

Accuracy
For the wing-alone configuration (case 1) a hybrid (prismatic/tetrahedral) grid is generated in the framework of a DLR-NLR co-operation (Ref. [7]). In the prismatic part of the grid 25 prism layers are generated in order to accurately capture the boundary layer (wall-normal distance of first grid point is 5.0 x 10-6 based on the wing root chord). The hybrid grid consists of 1.031.683 nodes, 951.930 prismatic elements and 3431.524 tetrahedral elements. Figures 5 and 6 show the pressure and skin friction distribution at spanwise station y=0.65 for the ONERA M6 wing-alone configuration. It can be observed that calculated pressure distribution is close to the wind tunnel result and the numerical result obtained with a technology-ready multi-block structured method.

Conclusions

The objectives of the FASTFLO-project are defined in terms of CFD-problem-turnaround time, and accuracy. The present results show that the computing time of the CFD system is relatively short (at Euler level); if the parallel version of the Euler flow solver is used the computing time is not a bottleneck to reach the first objective.

The CFD-problem-turnaround time however also depends on the automation level. Critical sections are identified that determine the automation level. From these results it is concluded that a sufficiently high automation level is within reach.

The accuracy requirement is studied at the level of the Reynolds-averaged Navier Stokes equations in the framework of a DLR-NLR co-operation. The results for a wing-alone configuration demonstrate a good potential for accuracy.

References

[1] J.W. van der Burg (NLR), B. Oskam (NLR), T. Gerhold (DLR), M. Galle (DLR), T. Berglind (FFA), B. Arlinger (SAAB), G. Kretzschmar (IBK), W. Haase (DASA-LM), "FASTFLO contract no. BRPR-CT96-1084, Annex I Project Programme”, NLR CR 96275 L, February 23, 1996.

[2] Press release: "Airbus Industry 1997 global market forecast, confirming a very large demand”, March 1997.

[3] Press release: "Boeing projects healthy airplane demand over the next twenty years”, March 1997.

[4] IMG3: "The aeronautical industries technology needs for the fifth framework programme and beyond”, December 12, 1997.

[5] P.E. Rubbert, "CFD and the changing world of airplane design”, In ICAS Proceedings, 19th Congress of the International Council of the Aeronautical Sciences, September 1994.

[6] J.I. van den Berg, H.A. Sytsma and H. Schippers, "Computation of the flow about an F16-like configuration for several flow conditions”, NLR TP 95226 U, April 4, 1995.

[7] J.E.J. Maseland, "Contributions to the development of turbulent Navier-Stokes flow modelling capabilities using unstructured grids”, NLR TR 97390 L, June 1997.