COMSAC: Computational Methods for Stability and Control

NASA/CP-2004-213028/PT2

COMSAC: Computational Methods for

Stability and Control

Compiled by C. Michael Fremaux and Robert M. Hall Langley Research Center, Hampton, Virginia

April 2004

The NASA STI Program Office . . . in Profile Since its founding, NASA has been dedicated to the • CONFERENCE PUBLICATION. Collected advancement of aeronautics and space science. The papers from scientific and technical NASA Scientific and Technical Information (STI) conferences, symposia, seminars, or other Program Office plays a key part in helping NASA meetings sponsored or co-sponsored by NASA.

maintain this important role.

• SPECIAL PUBLICATION. Scientific, The NASA STI Program Office is operated by technical, or historical information from NASA Langley Research Center, the lead center for NASA’s programs, projects, and missions, often scientific and technical information. The NASA STI concerned with subjects having substantial Program Office provides access to the NASA STI public interest.

Database, the largest collection of aeronautical and space science STI in the world. The Program Office is • TECHNICAL TRANSLATION. English- also NASA’s institutional mechanism for language translations of foreign scientific and disseminating the results of its research and technical material pertinent to NASA’s mission.

development activities. These results are published by NASA in the NASA STI Report Series, which Specialized services that complement the STI includes the following report types: Program Office’s diverse offerings include creating custom thesauri, building customized databases, • TECHNICAL PUBLICATION. Reports of organizing and publishing research results ... even completed research or a major significant phase providing videos.

of research that present the results of NASA programs and include extensive data or For more information about the NASA STI Program theoretical analysis. Includes compilations of Office, see the following: significant scientific and technical data and information deemed to be of continuing • Access the NASA STI Program Home Page at reference value. NASA counterpart of peer- http://www.sti.nasa.gov reviewed formal professional papers, but having less stringent limitations on manuscript length • E-mail your question via the Internet to and extent of graphic presentations.

help@sti.nasa.gov • TECHNICAL MEMORANDUM. Scientific • Fax your question to the NASA STI Help Desk and technical findings that are preliminary or of at (301) 621-0134 specialized interest, e.g., quick release reports, working papers, and bibliographies that contain • Phone the NASA STI Help Desk at minimal annotation. Does not contain extensive (301) 621-0390 analysis.

• Write to: • CONTRACTOR REPORT. Scientific and NASA STI Help Desk technical findings by NASA-sponsored NASA Center for AeroSpace Information contractors and grantees.

7121 Standard Drive Hanover, MD 21076-1320

NASA/CP-2004-213028/PT2

COMSAC: Computational Methods for

Stability and Control

Compiled by C. Michael Fremaux and Robert M. Hall Langley Research Center, Hampton, Virginia Proceedings of a symposium sponsored by the National Aeronautics and Space Administration, Washington, DC, and held at the Holiday Inn and Conference Center, Hampton, Virginia September 23-25, 2003 National Aeronautics and Space Administration Langley Research Center Hampton, Virginia 23681-2199

April 2004

Acknowledgments The organizing committee acknowledges the highly professional administrative contributions of Ms. Susan N. Price of the NASA Langley Management Support Office. In addition, the committee thanks Mr. Douglas N. Ball (Boeing Commercial Airplanes), Dr. Pradeep Raj (Lockheed Martin), and Mr. John W. Clark, Jr. (NAVAIR) for their participation in the COMSAC planning sessions, and for sharing their personal observations and opinions in the symposium.

The use of trademarks or names of manufacturers in the report is for accurate reporting and does not constitute an official endorsement, either expressed or implied, of such products or manufacturers by the National Aeronautics and Space Administration.

Available from: NASA Center for AeroSpace Information (CASI) National Technical Information Service (NTIS) 7121 Standard Drive 5285 Port Royal Road Hanover, MD 21076-1320 Springfield, VA 22161-2171 (301) 621-0390 (703) 605-6000 Executive Summary A group of nearly 100 technical professionals from government, industry, and academia met in Hampton, Virginia on September 23-25, 2003, for a NASA-sponsored symposium on Computational Methods for Stability and Control (COMSAC) to discuss the status, opportunities, and challenges of applying Computational Fluid Dynamics (CFD) methodology to current and future issues in the field of aircraft stability and control (S&C). The unprecedented advances now being made in CFD technology have demonstrated the powerful capabilities of codes in applications to civil and military vehicles. Used in conjunction with wind-tunnel and flight investigations, many codes are now routinely used by designers in diverse applications such as aerodynamic performance predictions and propulsion integration. Typically, these codes are most reliable for attached, steady, and predominantly turbulent flows. As a result of increasing reliability and confidence in CFD, wind-tunnel testing for some new configurations has been substantially reduced in key areas, such as wing trade studies for mission performance guarantees.

Interest is now growing in the application of CFD methods to other critical design challenges.

One of the most important disciplinary elements for civil and military aircraft is S&C.

Experience has shown that predictions and analyses of aerodynamic S&C characteristics for full- scale aircraft can be in serious error because of Reynolds number effects, configuration sensitivities, dynamic motion effects, and other issues. Existing experimental facilities may not even be capable of replicating the motions required for aerodynamic measurements. As a result of these shortcomings, a major portion of aircraft development wind-tunnel time (about 60-70%) is typically devoted to S&C testing, especially for various off-design conditions ranging from takeoff and landing to cruise and maneuver. Even with an enormous amount of experimental work, pre-flight aerodynamic prediction errors result in unacceptable increases in program costs, “fly and try” approaches to fixing deficiencies, and extensive developmental delays.

Unfortunately, applications of current and emerging CFD codes to engineering analysis in the field of aircraft S&C have been extremely limited. Although isolated examples of success have been demonstrated for certain configurations, the more global issues in S&C – which may involve massive flow separation, unsteady and nonlinear phenomena, dynamic effects, and other extremely complex factors – have not yet been significantly addressed by the CFD community.

The current lack of COMSAC-related activities has been further aggravated by the fact that, in contrast to the areas of CFD and performance, very little cross-cultural interaction and communication appears to occur between participants in the areas of CFD and S&C. Within the aerospace community, it is generally agreed that the field of CFD has rapidly matured to the point that the next high payoff applications could occur in S&C. In particular, CFD offers the potential for significantly increasing the basic understanding, prediction, and control of flow phenomena associated with requirements for satisfactory aircraft handling characteristics.

The objectives of the 3-day symposium were to: 1. Discuss the unique aerodynamic phenomena and issues of S&C 2. Define the current characteristics, capabilities, and limitations of CFD codes 3. Define additional or new code requirements for S&C applications iii 4. Identify potential approaches to develop validated codes 5. Discuss the potential contents and funding opportunities for a COMSAC program The scope of technical discussions covered civil and military aircraft, including commercial transports, business jets, fighter and attack aircraft, military transports, and bombers. Discussions were limited to fixed-wing aircraft. All sessions were unclassified, and all non-proprietary presentations were collated in the form of PowerPoint presentations with note pages for post- meeting distribution to attendees.

Presentations by speakers described numerous examples of severe impacts of erroneous aerodynamic predictions on the stability and control characteristics of civil and military aircraft.

Typically, resolving and mitigating unexpected aerodynamic behavior involved laborious “cut and fly” approaches required during critical flight test programs. These shortcomings resulted in significant program delays, costs, mission limitations, non-optimum configurations (weight, capabilities, etc.), and severe scrutiny by stakeholders and customers.

In-depth discussions of specific experiences with actual applications of various levels of computational methods to S&C indicated a wide range of success and an overriding sense of skepticism by the attendees. After individual presentations were made to provide organizational and individual perspectives on CFD for S&C, the attendees were briefed on NASA’s vision of a COMSAC program. Comments were solicited to identify and prioritize technology areas for such a program. Finally, the Director of the NASA-Langley Aerospace Vehicle Systems Technology Office shared his view of a potential strategy to augment funding and program priority in this area.

The general findings of the workshop were: 1. Inaccurate prediction of aerodynamic stability and control parameters continues to have major cost and programmatic impacts in virtually every vehicle class. These impacts include unacceptable increases in program costs, “fly and try” approaches to fixing deficiencies, extensive developmental delays and profit losses due to delayed deliveries.

2. Prediction of the character of separated flows across the speed range (with the attendant issues of transition prediction, turbulence modeling, unsteady flows, etc.)

and the impact of separated flow on aircraft S&C should receive priority in a COMSAC program.

3. A pervasive attitude of skepticism regarding the success of CFD applications to aircraft S&C issues (especially for preliminary and conceptual design) exists within the CFD community, as well as the S&C community.

4. The application of advanced and emerging CFD methods as design tools will be dependent on the accumulation and demonstrated success of experiences for both generic and specific aircraft configurations.

iv 5. Issues regarding the CFD process (cost, time required, adaptive gridding requirements, error quantification, etc.) should be high priority targets for COMSAC efforts.

6. One of the most valuable contributions of the symposium was the mechanism to share perspectives and experiences between the diverse CFD specialists and S&C specialists. Prior to this meeting, communication between these two groups was extremely poor, resulting in a major barrier to the acceleration and acceptance of CFD methods for S&C applications.

Joseph R. Chambers ViGYAN, Inc.

Hampton, Virginia v Contents Executive Summary …………………………………………………………………………….. iii Attendees …………………………………………………………………………………………. x Part 1 Introductory Remarks …………………………………………………………………………... 1 Darrel R. Tenney Aerospace Vehicle Systems Technology Office, NASA Langley Research Center, Hampton, Virginia Introduction to Computational Methods for Stability and Control (COMSAC) …………… 7 Robert M. Hall and C. Michael Fremaux NASA Langley Research Center, Hampton, Virginia Joseph R. Chambers ViGYAN, Inc., Hampton, Virginia Stability & Control Challenges for COMSAC: a NASA Langley Perspective ……………. 28 C. Michael Fremaux NASA Langley Research Center, Hampton, Virginia Emerging CFD Capabilities and Outlook – A NASA Langley Perspective ………………... 48 Robert T. Biedron, S. Paul Pao, and James L. Thomas NASA Langley Research Center, Hampton, Virginia The Role for Computational Fluid Dynamics for Stability and Control – Is it Time? ……. 69 Douglas N. Ball Boeing Commercial Airplanes, Renton, Washington Northrop Grumman Perspective on COMSAC ……………………………………………… 98 Dale Lorincz Northrop Grumman Air Combat Systems, El Segundo, California Boeing Integrated Defense Systems Perspective on COMSAC …………………………… 118 David M. Evans Boeing Integrated Defense Systems – Tactical Aircraft and Weapons, St. Louis, Missouri Computational Methods in Stability and Control – WPAFB Perspective ………………... 124 William Blake AFRL Air Vehicles Directorate, Wright Patterson Air Force Base, Ohio William Thomas Aeronautical Systems Center, Wright Patterson Air Force Base, Ohio vi Perspective: Raytheon Aircraft Company ………………………………………………….. 144 Neal J. Pfeiffer and Dana C. Herring Raytheon Aircraft Company, Wichita, Kansas A Greybeard’s View of the State of Aerodynamic Prediction ……………………………... 181 Tom Lawrence Aeromechanics Division, Naval Air Systems Command, Patuxent River, Maryland Computational Methods for Stability and Control: A Perspective ……………………….. 214 Pradeep Raj Lockheed Martin Aeronautics Company, Marietta, Georgia Boeing TacAir Stability and Control Issues for Computational Fluid Dynamics ………... 227 William B. Hollingsworth Boeing TacAir, St. Louis, Missouri NAVAIR S&C Issues for CFD ……………………………………………………………….. 239 Steve Donaldson Flight Dynamics Branch, Naval Air Systems Command, Patuxent River, Maryland An S&C Perspective on CFD ………………………………………………………………… 248 Russ D. Killingsworth JSF Stability and Control, Lockheed Martin Aeronautics, Fort Worth, Texas Issues, Challenges & Payoffs: A Boeing User’s Perspective on CFD for S&C …………... 273 D. R. Bogue, T. R. Lines, and R. D. Doll Boeing Commercial Airplanes, Seattle, Washington Stability and Control in Computational Simulations for Conceptual and Preliminary Design: the Past, Today, and Future? ………………………………………... 309 William H. Mason Department of Aerospace and Ocean Engineering, Virginia Tech, Blacksburg, Virginia Computational Methods for Dynamic S&C Derivatives …………………………………… 341 Lawrence L. Green and Patrick C. Murphy NASA Langley Research Center, Hampton, Virginia Angela M. Spence Department of Aerospace Engineering, Mississippi State University, Starkville, Mississippi Part 2 Boeing TacAir CFD Capabilities/Issues …………………………………………………….. 365 David Stookesberry and Frank C. Berrier Boeing TacAir, St. Louis, Missouri vii TetrUSS Capabilities for S&C Applications ………………………………………………... 378 Neal T. Frink and Paresh C. Parikh NASA Langley Research Center, Hampton, Virginia Computational Simulations for Stability and Control – BCA State-of-the-Art …………. 396 N. J. Yu, T. J. Kao, D. R. Bogue, and T. R. Lines Boeing Commercial Airplanes, Seattle, Washington Time-Accurate Computational Simulation …………………………………………………. 417 S. Paul Pao and Pieter G. Buning NASA Langley Research Center, Hampton, Virginia Application of CFD to Abrupt Wing Stall Using RANS and DES ………………………… 433 James R. Forsythe Cobalt Solutions, LLC, Springfield, Ohio Application of Computational Stability and Control Techniques Including Unsteady Aerodynamics and Aeroelastic Effects …………………………………………… 489 David M. Schuster and John W. Edwards NASA Langley Research Center, Hampton, Virginia Hinge Moment Predictions Using CFD ……………………………………………………… 511 M. J. Grismer, D. Kinsey, and D. Grismer Air Vehicles Directorate, Air Force Research Laboratory, Wright Patterson Air Force Base, Ohio CFD Simulation of Aircraft in Coning Motion ……………………………………………... 533 Syta Saephan and C. P. van Dam Department of Mechanical and Aeronautical Engineering, University of California, Davis, California Use of CFD-Generated Aerodynamic Data for an F-15E/SLV Flying/Handling Qualities Analysis ……………………………………………………………………………... 565 Jeffery A. Batte and W. Shawn Westmoreland Jacobs Sverdrup, Eglin Air Force Base, Florida Rapid Euler CFD for High-Performance Aircraft Design …………………………………. 593 Eric F. Charlton Lockheed Martin Aeronautics Company, Fort Worth, Texas Quantitative Prediction of Computational Quality ………………………………………… 638 Michael J. Hemsch, James M. Luckring, and Joseph H. Morrison NASA Langley Research Center, Hampton, Virginia Best Practices System to Enhance CFD Use in Stability and Control Applications ……… 656 Michael R. Mendenhall Nielsen Engineering & Research, Inc., Mountain View, California viii Dynamic Water Tunnel Testing for Code Benchmarking …………………………………. 671 Brooke C. Smith and John Hodgkinson AeroArts LLC, Palos Verdes Peninsula, California COMSAC: Vision and Potential Program Planning ………………………………………. 692 Robert M. Hall and C. Michael Fremaux NASA Langley Research Center, Hampton, Virginia Joseph R. Chambers ViGYAN, Inc., Hampton, Virginia COMSAC Feedback ………………………………………………………………………….. 718 Pradeep Raj Lockheed Martin Aeronautics Company, Marietta, Georgia COMSAC Plan: NAVAIR Comments ……………………………………………………… 722 John W. Clark Naval Air Systems Command, Patuxent River, Maryland COMSAC Feedback – Boeing Commercial Airplanes ……………………………………... 730 Douglas N. Ball Boeing Commercial Airplanes, Renton, Washington ix NASA Langley Symposium on Computational Methods for Stability and Control Hampton Holiday Inn and Conference Center Hampton, Virginia September 23 – 25, 2003 Attendee List 1. Anderson, Dr. William K. 6. Berrier, Mr. Frank C.

UT SimCenter at Chattanooga Boeing TacAir Two Union Square, Suite 300 PO Box 516, MC S1066450 Chattanooga, TN 37402 St. Louis, MO 63166 kyle-anderson@utc.edu frank.berrier@boeing.com 423-648-0343 314-232-5774 2. Arabshahi, Dr. Abdollah 7. Biedron, Dr. Robert T.

UT SimCenter at Chattanooga M.S. 128 Two Union Square, Suite 300 NASA Langley Research Center Chattanooga, TN 37402 Hampton, VA 23681 Abi-Arabshahi@utc.edu robert.t.biedron@nasa.gov 423-648-0342 757-864-2156 3. Ashbaugh, Mr. William H. 8. Blake, Mr. William ASC/AAAV Air Force Research Laboratory 2145 Monahan Way, Bldg 28 AFRL/VACA Wright-Patterson AFB, OH 45433 WPAFB, OH 45433 william.ashbaugh@wpafb.af.mil william.blake2@wpafb.af.mil 937-255-7210,x-3864 937-255-6764 4. Ball, Mr. Douglas N. 9. Bogue, Mr. David Boeing Commercial Airplanes Boeing Commercial Airplanes 535 Garden Avenue N., Mail Stop 15530 Bothell Way NE #111 67-LH Renton, WA 98055 Seattle, WA 98155 douglas.n.ball@boeing.com david.r.bogue@boeing.com 425-234-1016 425-234-1078 5. Batte, Mr. Jeffery A. 10. Bower, Dr. Daniel R.

Jacobs Sverdrup National Transportation Safety 308 West D Ave, Suite 1 Board P.O. Box 1935 490 L'Enfant Plaza East Eglin AFB, FL 32542 Washington, DC 20594 batte@eglin.af.mil bowerd@ntsb.gov 850-882-0398 202-314-6562 x 11. Chambers, Mr. Joseph R. 17. Dreyer, Dr. James J.

ViGYAN, Inc. Applied Research Laboratory 205 Old Dominion Rd. Penn State University Yorktown, VA 23692 P.O. Box 30 jrchambers@cox.net State College, PA 16804 757-898-6080 jjd@wt.arl.psu.edu 814-863-3018 12. Chung, Dr. James J.

NAVAIR 18. Edwards, Dr. John W.

Bldg. 2187, Suite 1320B, Unit5 MS 340 48110 Shaw Road NASA Langley Research Center Patuxent River, MD 20670 Hampton, VA 23681 chungjj@navair.navy.mil john.w.edwards@nasa.gov 301-342-8547 757-864-2273 13. Clark, Jr., Mr. John W. 19. Evans, Mr. David M.

NAVAIR Boeing 3040 Blackberry Ln. Box 516, MS 2703760 Prince Frederick, MD 20678 St. Louis, MO 63166 clarkjw1@navair.navy.mil david.m.evans@boeing.com 301-342-8550 314-233-6290 14. Craft, Mr. John W. 20. Forsythe, Dr. James R.

Boeing, Huntington Beach Cobalt Solutions, LLC 5301 Bolsa Ave., MC H013-B318 4636 New Carlisle Pike Huntington Beach, CA 92647 Springfield, OH 45504 john.w.craft@boeing.com forsythe@cobaltcfd.com 714-235-8261 937-620-5938 15. Crider, Mr. Dennis A. 21. Fremaux, Mr. Charles M.

National Transportation Safety MS 153 Board NASA Langley Research Center 490 L'Enfant Plaza East, SW Hampton, VA 23681 RE-60 charles.m.fremaux@nasa.gov Washington, D.C. 20594 757-864-1193 criderd@ntsb.gov 202-314-6564 22. Frink, Dr. Neal T.

MS 499 16. Donaldson, Mr. Steve A. NASA Langley Research Center NAVAIR Hampton, VA 23681 B2187, Suite 1390A neal.t.frink@nasa.gov 48110 Shaw Rd., Unit 5 757-864-2864 Patuxent River, MD 20670 donaldsonsa@navair.navy.mil 301-342-0282 xi 23. Garb, Mr. Slava Z. 29. Grismer, Dr. Matthew J.

Gulfstream Aerospace AFRL/VAAC, B146 R225 PO Box 2206, M/S D-04 2210 Eighth Street Savannah, GA 31402 Wright-Patterson AFB, OH 45433 slava.garb@gulfaero.com matthew.grismer@wpafb.af.mil 912-965-3527 937-255-3876 24. Garcia, Mr. Garrett M. 30. Grove, Mr. Darren V.

U,S, Air Force NAVAIR, 75 Vandenburg Drive, Bldg 1630 Bldg. 2187, Suite 1320-D4 ESC/MAV 48110 Shaw Rd.

Hanscom AFB, MA 01731 Patuxent River, MD 20670 garrett.garcia@hanscom.af.mil grovedv@navair.navy.mil 719-964-7377 301-342-8562 25. Ghaffari, Mr. Farhad 31. Guruswamy, Dr. Guru MS 286 T27B-1 NASA Langley Research Center NASA Ames Research Center Hampton, VA 23681 Moffett Field, CA 94035 Farhad.Ghaffari-1@nasa.gov Guru.p.guruswamy@nasa.gov 757-864-2856 650-604-6329 26. Goble, Dr. Brian D. 32. Hall, Dr. Robert M.

Lockheed Martin Aeronautics Co. MS 499 101 Academy Blvd., Mail Zone 8656 NASA Langley Research Center P.O. Box 748 Hampton, VA 23681 Fort Worth, TX 76101 robert.m.hall@nasa.gov brian.d.goble@lmco.com 757-864-2883 817-935-4707 33. Heim, Mr. Eugene H.

27. Gopalarathnam, Dr. Ashok MS 153 North Carolina State University NASA Langley Research Center Mech. and Aerospace Engineering Hampton, VA 23681 Box 7910 eugene.h.heim@nasa.gov Raleigh, NC 27695 757-864-9638 ashok_g@ncsu.edu 919-515-5669 34. Hemsch, Dr. Michael J.

MS 286 28. Green, Mr. Lawrence L. NASA Langley Research Center MS 159 Hampton, VA 23681 NASA Langley Research Center michael.j.hemsch@nasa.gov Hampton, VA 23681 757-864-2882 Lawrence.L.Green@nasa.gov 757-864-2228 xii 35. Herring, Mr. Dana C. 41. Kerho, Dr. Michael F.

Raytheon Aircraft Company Rolling Hills Research Corporation MS 971-B6, PO Box 85 3425 Lomita Blvd.

9709 E Central Torrance, CA 90505 Wichita, KS 67201 MKerho@RollingHillsResearch.com dana_herring@rac.ray.com 310-257-9578 316-676-6718 42. Killingsworth, Mr. Russ D.

36. Hickey, Mr. Herbert J. (Skip) Lockheed Martin Aeronautics WJN Consulting PO Box 748, MZ 6468 5137 Croftshire Drive Fort Worth, TX 76101 Kettering, OH 45440 russ.d.killingsworth@lmco.com skiphickey@aol.com 817-7632915 937-433-5391 43. Kokolios, Mr. Alex 37. Hollingsworth, Mr.William B. NAVAIR Boeing TacAir B2187, Suite 1390A PO Box 516, MC S1066450 48110 Shaw Rd., Unit 5 St. Louis, MO 63166 Patuxent River, MD 20670 william.b.hollingsworth- kokoliosa@navair.navy.mil iii@boeing.com 301-342-8574 314-234-2221 44. Kolly, Dr. Joseph 38. Hoyle, Mr. David L. National Transportation Safety Lockheed Martin Board M/C 0685 490 L'Enfant Plaza East, SW 86 S. Cobb Drive Washington, D.C. 20594 Marietta, GA 30063 kollyj@ntsb.gov david.l.hoyle@lmco.com 202-3146622 770-494-8279 45. Komerath, Dr. Narayanan M.

39. Jiang, Dr. Minyee J. Georgia Institute of Technology U.S. Navy, NSWCCD School of Aerospace Engineering 9500 MacArthur Blvd. Atlanta, GA 30332 West Bethesda, MD 20817 narayanan.komerath@ae.gatech.edu jiangm@nswccd.navy.mil 404-894-3017 301-227-6090 46. Kramer, Mr. Brian R.

40. Karman, Dr. Steve L. Rolling Hills Research Corporation UT SimCenter at Chattanooga 3425 Lomita Blvd.

Two Union Square, Suite 300 Torrance, CA 90505 BKramer@RollingHillsResearch.com Chattanooga, TN 37402 310-257-9578 Steve-Karman@utc.edu 423-648-0595 xiii 47. Kumar, Dr. Ajay 53. Luckring, Dr. James M.

MS 285 MS 286 NASA Langley Research Center NASA Langley Research Center Hampton, VA 23681 Hampton, VA 23681 Ajay.Kumar-1@nasa.gov James.M.Luckring@nasa.gov 757-864-3520 757-864-2869 48. Laiosa, Mr. Joseph P. 54. Malone, Dr. John B.

Naval Air Warfare Center MS 285 Aircraft Division NASA Langley Research Center Bldg 2187, Suite 1320-E2 Hampton, VA 23681 48110 Shaw Road, Unit #5 John.B.Malone@nasa.gov Patuxent River, MD 20670 757-864-8988 LaiosaJP@navair.navy.mil 301-342-5723 55. Mangalam, Dr. Siva M.

Tao Systems 49. Lawrence, Mr. Tom 471 McLaws Circle NAVAIR Williamsburg, VA 23185 Aeromechanics Division siva@taosystem.com Patuxent River, MD 20670 757-220-5040 lawrencejt@navair.navy.mil 301-342-8551 56. Mason, Dr. William H.

Virginia Tech 50. Leavitt, Mr. Laurence D. 215 Randolph Hall, MC 0203 MS 499 Blacksburg, VA 24061 NASA Langley Research Center whmason@vt.edu Hampton, VA 23681 540-231-6740 Laurence.D.Leavitt@nasa.gov 757-864-3017 57. Masters, Mr. James Jacobs Sverdrup, AEDC Group 51. Lee-Rausch, Ms. Elizabeth M. Arnold AFB, TN 37389 MS 128 james.masters@arnold.af.mil NASA Langley Research Center 931-454-7739 Hampton, VA 23681 Elizabeth.M.Lee-Rausch@nasa.gov 58. McNamara, Mr. William G.

757-864-8422 NAVAIR/Naval Air Warfare Center Bldg 2035 52. Lorincz, Mr. Dale 48183 Switzer Road Northrop Grumman Corp. Patuxent River, MD 20670 One Hornet Way mcnamaraWG@navair.navy.mil 9L30/W6 301-757-0848 El Segundo, CA 90245 dale.lorincz@ngc.com 310-332-9966 xiv 59. Mehrotra, Dr. Sudhir C. 65. O'Callaghan, Mr. John ViGYAN, Inc. National Transportation Safety 30 Research Drive Board Hampton, VA 23666 490 L'Enfant Plaza E., S.W.

mehrotra@vigyan.com Washington, D.C. 20594 757-865-1400 ocallaj@ntsb.gov 202-314-6560 60. Mendenhall, Mr. Michael R.

Nielsen Engineering & Research 66. Om, Dr. Deepak 605 Ellis St., Suite 200 Boeing Commercial Airplanes Mountain View, CA 94043 Mail Code 67-LF, P. O. Box 3707 mrm@nearinc.com Seattle, WA 98124 650-968-9457 deepak.om@boeing.com 425-234-1116 61. Morrison, Mr. Joseph H.

MS 128 67. Ozoroski, Ms. Lori P.

NASA Langley Research Center MS 348 Hampton, VA 23681 NASA Langley Research Center joseph.h.morrison@nasa.gov Hampton, VA 23681 757-864-2294 Lori.P.Ozoroski@nasa.gov 757-8645992 62. Murphy, Dr. Patrick C.

MS 132 68. Pao, Dr. S. Paul NASA Langley Research Center MS 499 Hampton, VA 23681 NASA Langley Research Center Patrick.C.Murphy@nasa.gov Hampton, VA 23681 757-864-4071 s.p.pao@nasa.gov 757-864-3044 63. Murri, Mr. Daniel G.

MS 153 69. Parikh, Dr. Paresh C.

NASA Langley Research Center MS 499 Hampton, VA 23681 NASA Langley Research Center Daniel.G.Murri@nasa.gov Hampton, VA 23681 757-864-1160 paresh.c.parikh@nasa.gov 757-864-2244 64. Nelson, Dr. Robert C.

University of Notre Dame 70. Park, Mr. Michael A.

Department of Aerospace and MS 128 Mechanical Engineering NASA Langley Research Center Notre Dame, IN 46556 Hampton, VA 23681 robert.c.nelson.1@nd.edu Michael.A.Park@nasa.gov 574-631-4733 757-864-6604 xv 71. Pfeiffer, Dr. Neal J. 77. Roberts, Mr. Michael Raytheon Aircraft Company University of California, Davis MS 999-B12 , PO Box 85 Dept. of Mechanical & Aeronautical 9709 E. Central Engineering Wichita, KS 67201 1 Shields Ave., neal_pfeiffer@rac.ray.com Davis, CA 95616 316-676-6775 mike_m_roberts@yahoo.com 530-752-2261 72. Pirzadeh, Dr. Shahyar Z.

MS 499 78. Saephan, Mr. Syta NASA Langley Research Center University of California, Davis Hampton, VA 23681 Dept. of Mechanical & Aeronautical Shahyar.Z.Pirzadeh@nasa.gov Engineering 757-864-2245 1 Shields Ave., Davis, CA 95616 73. Player, Ms. Jennifer ssaephan@ucdavis.edu Bihrle Applied Research, Inc. 530-752-2261 18 Research Drive Hampton, VA 23666 79. Schuster, Dr. David M.

jplayer@bihrle.com MS 340 757-766-2416 NASA Langley Research Center Hampton, VA 23681 74. Raj, Dr. Pradeep david.m.schuster@nasa.gov Lockheed Martin Aeronautics Co. 757-864-2259 86 S. Cobb Dr.

D/1AEM, Z/0660 80. Shah, Mr. Gautam H.

Marietta, GA 30063 MS 153 pradeep.raj@lmco.com NASA Langley Research Center 770-494-3801 Hampton, VA 23681 gautam.h.shah@nasa.gov 75. Ralston, Mr. John 757-864-1163 Bihrle Applied Research, Inc.

18 Research Drive 81. Skujins, Dr. Ojars Hampton, VA 23666 Aeronautical Systems Center jralston@bihrle.com ASC/ENFT 757-766-2416 2530 Loop Road West Wright-Patterson AFB, OH 45433 76. Ratvasky, Mr. Thomas P. ojars.skujins@wpafb.af.mil 21000 Brookpark Rd, Mail Stop 11-2 937-255-4166 NASA Glenn Research Center Cleveland, OH 44135 82. Slomski, Dr. Joe thomas.p.ratvasky@nasa.gov NSWC Carderock Division 216-433-3905 9500 MacArthur Boulevard West Bethesda, MD 20817 slomskijf@nswccd.navy.mil 301-227-1935 xvi 83. Smith, Mr. Brooke C. 89. Tenney, Dr. Darrel R.

AeroArts LLC MS 208 P.O. Box 2909 NASA Langley Research Center Palos Verde Peninsula, CA 90274 Hampton, VA 23681 Brooke.Smith@AeroArts.com Darrel.R.Tenney@nasa.gov 310-547-0927 757-864-6033 84. Stookesberry, Mr. David C. 90. Thacker, Mr. Michael Boeing Cessna Aircraft Company P.O. Box 516, MC S106-6420 5701 E. Pawnee, D363P St. Louis, MO 63166 Wichita, KS 67218 david.c.stookesberry@boeing.com mthacker@cessna.textron.com 314-233-6346 316-831-2807 85. Stuckert, Dr. Gregory K. 91. Thomas, Dr. James L.

Fluent Inc. MS 128 10 Cavendish Court NASA Langley Research Center Lebanon, NH 03766 Hampton, VA 23681 gks@fluent.com James.L.Thomas@nasa.gov 603-643-2600, x243 757-8645578 86. Stuever, Mr. Robert A. 92. Van Dam, Dr. Case The Boeing Company University of California, Davis Wichita Development & Dept. of Mechanical & Aeronautical Modification Center Engineering P. O. Box 7730, MC K05-14 1 Shields Ave., Wichita, KS 67277 Davis, CA 95616 Robert.A.Stuever@boeing.com cpvandam@ucdavis.edu 316-523-4826 530-752-7741 87. Sung, Dr. Chao Ho 93. Vassberg, Dr. John C.

Dept of the Navy The Boeing Company 9500 MacArthur Blvd 5301 Bolsa Avenue West Bethesda, MD 20817 MC H013-B318 sungch@nswccd.navy.mil Huntington Beach, CA 92647 301-227-1865 john.c.vassberg@boeing.com 714-896-1607 88. Tello, Mr. Richard J.

The MITRE Corporation 94. Wahls, Dr. Richard A.

75 Vandenberg Drive MS 499 Attn: Joint Stars NASA Langley Research Center Hanscom AFB, MA 01731 Hampton, VA 23681 rtello@hansom.af.mil richard.a.wahls@nasa.gov 781-377-9919 757-864-5108 xvii 95. Westmoreland, Mr. William S.

Jacobs Sverdrup 98. Yip, Mr. Long P.

205 West D Ave. MS 254 Eglin AFB, FL 32542 NASA Langley Research Center westmore@eglin.af.mil Hampton, VA 23681 850-8820918 Long.P.Yip@nasa.gov 757-864-1700 96. Wong, Dr. Tin-Chee U. S. Army, AMSAM-RD-AE-A 99. Yu, Dr. Neng J.

Bldg 4488 Boeing Commercial Airplanes Redstone Arsenal, AL 35898 Enabling Technology & Research TinChee.Wong@us.army.mil M/S 67-LF, P.O.Box 3707 256-705-9605 Seattle, WA 98124 neng.j.yu@pss.boeing.com 97. Yang, Dr. Cheng I. 425-234-1192 Code 5400, Bldg 16 Naval Surface Warfare Center Bethesda, MD 20084 yangci@nswccd.navy.mil 301-227-4658 xviii TetrUSS is very much a team effort. This slide lists the current LaRC team members, but many have contributed in various capacities as well. We will soon launch a new and updated web site that lists many of the contributors.

TetrUSS is a suite of loosely coupled computational fluid dynamics software that is packaged into a complete flow analysis system. The system components consist of tools for geometry setup, grid generation, flow solution, visualization, and various utilities tools. Development began in 1990 and it has evolved into a proven and stable system for Euler and Navier-Stokes analysis and design of unconventional configurations.

It is 1) well developed and validated, 2) has a broad base of support, and 3) is presently is a workhorse code because of the level of confidence that has been established through wide use.

The entire system can now run on linux or mac architectures. In the following slides, I will highlight more of the features of the VGRID and USM3D codes.

The primary features of the VGRID code are listed here. This list will mean more to the CFD savvy individuals in the audience, but the bottom line is that it will generate high-quality Navier- Stokes grids on complex geometries with a nominal amount of training. It is now quite robust and fairly easy to use.

Here is an example of a full Navier-Stokes grid generated by a U.S. Air Force Academy student (albeit a sharp student). It is a C-130 with propellers with a slotted cargo release parachute.

The USM3D code is a cell-centered tetrahedral flow solver, in contrast to a “node-centered” solver for the CFD audience. It produces very accurate Navier-Stokes solutions on relatively coarse grids.

Turbulence is modeled by several models, i.e. the Spalart-Allmaras 1-eqn model, and the k- epsilon, Menter SST, and Algebraic Reynolds Stress Model (ARSM) 2-eqn models.

We can run both steady state with Local Time Stepping convergence acceleration, as well as nd unsteady flows with 2 order time accuracy.

We have a range of established upwind schemes, but have some very useful and heavily used special boundary conditions, such as propeller model, engine intakes and jet exhausts, porous surfaces, and wall functions.

The code is very fast by current standards and runs on multiple types of parallel machine clusters.

I’ll just highlight a few sample applications in industry and NASA in the next couple of slides.

Here we see a range of applications ranging from civil aircraft to fighter jets.

Back in 1992, MDA was unable to get FAA certification on the MD-11 due to a range shortfall traced to a higher than expected drag. We were requested by MDA to help eliminate an outboard pylon separation that was identified during a flight test with the expectation that it would eliminate enough drag to meet range requirements. We had a 3-month window to resolve the problem, have hardware built and flight tested. MDA sent an engineer here and we formed a tiger-team to accomplish this. The 3-month target was met and the drag was successfully reduced to permit FAA certification for range.

The JSF Design Team was another “Tiger Team” exercise with a short time scale. Several key LaRC code experts worked together to develop a new Passive Porosity BC for two structured and two unstructured flow solvers. While I cannot show any details, the this new PassPort BC was used as an S&C tool to reduce a high-alpha pitch-up problem encountered during landing.

We were recently involved in providing some computational support to the Airbus accident investigation.

LMAC is a heavy user with many large-scale applications. Here is a sample of generating a loads database for the P-3 Orion, which required over 250 Navier-Stokes solutions on the full configuration with four co-rotating propellers.

Here is an example where Piper Aircraft has used TetrUSS to certify a reengineering of one of its aircraft.

And more recently, Raytheon is using it to perform concept studies of a supersonic civil transport.

This just highlights a few of the NASA program applications.

TetrUSS was used in the mid-90’s in the Pegasus Return-to-Flight effort to assess the effect of some geometry modifications on the lateral-directional stability of the new flight vehicle.

It is heavily used in the HyPER-X mishap investigation and Return-to-Flight effort. I’ll show more on this in the next two slides.

TetrUSS was used in some Mars activities shown on the right.

It was heavily used in the Abrupt Wing Stall program investing the “wing drop” phenomenon encountered on the F/A-18EF. Here massively separated flows were routinely computed and accurate results obtained.

st The VGRID code was used to generate some of the unstructured grids used in the AIAA 1 and nd 2 Drag Prediction Workshops, the latter held this past summer just before the Orlando Applied Aero conference. These workshops drew many participants from many countries around the world.

This illustrates a large-scale application of TetrUSS in response to an urgent problem.

TetrUSS was the primary CFD code used in the HyPER-X Mishap Investigation and for Return- to-Flight support.

All related date is proprietary and the details cannot be discussed here, but this slide summarizes the investigation.

The presentation will show some of our recent work in using the NASA-built Navier-Stokes solvers for various applications including airplane control surface effectiveness study, Reynolds number scaling, and high lift configuration analysis.

Our first application of Navier-Stokes technology to S&C problem is the use of TLNS3D code to predict the high speed spoiler reversal phenomenon. At high transonic Mach with given spoiler deflection, the wing lift decreases at low angle of attack, as expected. However, as alpha increases beyond a certain value, lift increases; a phenomenon known as spoiler reversal observed in the wind tunnel test. CFD not only gives correct prediction, it also provides flow field details which explain the cause of spoiler reversal.

More recently, we have used both TLNS3D and CFL3D code to evaluate outboard aileron effectiveness with respect to Reynolds numbers. Multiblock grid approach was used to provide detailed representation of the geometry including aileron gaps. Both flow through and powered nacelles can be simulated in the analysis. Preliminary results show that effects of Reynolds number on aileron effectiveness were predicted correctly. More detailed study is in progress.

Applications of CFD for stabilizer or elevator effectiveness prediction are more challenging than the aileron analysis. One needs to resolve and capture the wakes generated by the wing, fuselage and nacelle, as they can affect the flow field in the horizontal tail region. With the use of denser grid in the wake regions and in the horizontal tail region, the CFD results correlate well with wind tunnel tail pressure data, and the tail lift slope curve also correlates with NTF data reasonably well.

Similar analyses were carried out for elevator effectiveness prediction. Here the elevator deflection up to -30 deg. was analyzed. Even with fairly massive flow separation downstream of the elevator hinge line, the CFD analysis with Menter’s two equation SST turbulence model provides good correlation with test data.

More extensive analyses were carried for low speed elevator Reynolds number effects. Boundary layer on the elevator becomes healthier at higher Reynolds number for a given elevator deflection. The flow separation downstream of the elevator hinge line at low Reynolds number reduces significantly at higher Reynolds number, and thus increases elevator effectiveness. The CFD prediction on elevator Reynolds number effects correlates reasonably well with the data.

For structured, multiblock grid approach, the simulation of a cruise airplane configuration including all vortex generators on the upper surface of the wing is a significant challenge to both the grid generation and flow analysis tasks. To capture flow field details generated by the vortex generators, one must use fine grid in the vortex generator regions, and to capture and preserve the vortical flows downstream of the vortex generators, accurate Navier-Stokes solver with minimum numerical dissipation is needed. A patched grid system was used in the present study.

Results showed that the effects of vortex generators on shock movement, and on airplane pitch characteristics were correctly predicted. The computing resources for this analysis are fairly reasonable, using 56 CPUs of an SGI Origin machine, one can get the solution of one flow condition (with 25 million grid points) within about 11 hour flow time.

The most extensive use of OVERFLOW within Boeing Commercial Airplanes is the high lift analysis. Due to geometry complexity, overset grid is a preferred method for high lift predictions. At angle of attack lower than 13 deg., OVERFLOW results agreed quite well with test data. Above 13 deg., significant flow separation occurs on the flap elements, CFD results showed premature drop in lift. The causes of such discrepancy could be grid resolution, or turbulence model effects, which will be addressed in the later slide. More recently, OVERFLOW was also used heavily for high speed configuration analysis and design.

For complex geometry analysis, grid generation is one of the most time consuming part of the analysis process. Grid spacing and grid quality affect numerical solution, especially for Navier- Stokes analysis. We have spent significant efforts in the past few years to develop both surface grid and flow field grid generation capability for either structured multiblock grid or overset grid.

For the Navier-Stokes codes we are using at the present time, good quality grid is essential to get accurate and reliably converged solutions. The slides show the patched grid system used in the vortex generator simulation.

For cruise configuration analysis at near design conditions, most turbulence models can provide results which correlate well with test data. However, problems related to S&C or Loads applications usually have significant separations in the flow fields, where different turbulence models could give rather different results. The plots showed that at large elevator deflection (-30 deg.) where massive separation occurred downstream of elevator hinge line, the two-equation SST model provides more realistic results than the one-equation S-A model.

Reliable convergence for N-S analysis is essential for S&C or Loads applications, where large number of cases are needed within a short period of time. The plots show typical convergence of CFL3D code at transonic speed. It requires many hundreds of multigrid cycles to converge the solution to acceptable level of numerical accuracy. For some cases, the convergence history could be worse, which deserve further research and improvement.

Abstract: Time accurate CFD may offer a faster approach to S&C aerodynamic database population than the conventional point by point steady state CFD. We would directly simulate α -, β -sweeps or other configuration movements typically of measurement sequence in wind tunnels. A second objective is to demonstrate potential applications to assessment of S&C dynamic derivatives by simulating vehicle motions such as free to roll, and nonlinearity such as the trends of aerodynamic forces near CL-max or flow hysteresis.

Moving grid algorithm must be efficient and robust for the user. The specifics include simple I/O, maintenance of good grid quality throughout thousands of computational cycles, and fast grid transformation such that grid motion time is a fraction of the time required for a single Navier-Stokes iterative step.

For obvious reasons, the helicopter aerodynamics research community had produced some of the most comprehensive time-accurate wind tunnel measurements and CFD analysis. We have chosen the original data by Piziali and the associated CFD studies as our reference for comparison.

Although the lift and pitching moment comparison are close, there are distinct differences between CFD and measurement. For example, the lift coefficient is higher than the measured data and the hysteresis loop is narrower for the CFD solution. The loop shape of the pitching moment are also different. However, this is typical for comparisons between CFD results using several other codes and the experimental data by Piziali.

Three dimensional unsteady CFD simulation is rarely available in the literature on account of the computational expenses and the difficulties in obtaining converged solutions. These are not fundamental obstacles and we should see much more often applications of 3D time accurate to practical problems in aerodynamics and S&C in the near future. This example demonstrates the distinct change in aerodynamic behavior in the spanwise direction. The pitching moment sign and slope near the wing tip are different from those for all the inboard stations.

Although we don’t have data to compare with, this is to demonstrate feasibility a continuous alpha sweep at a high subsonic Mach number. The lift and drag coefficient information is also plotted as a drag polar. The nondimensional time of T=210 for the angle of attack to go from 0 to 6 deg and back is relatively fast: less than one second in terms of wind tunnel length and time scales.

The subject of interest for this finite rate beta sweep simulation is the rolling moment coefficient.

The figure shows a narrow hysteresis loop with nonlinear slopes of Cl-beta for beta greater than about +/- 8 degrees.

The rolling moment response of the ONERA-M6 wing to a periodic body axis roll motion is classical. From this Cl versus Φ locus, we can determine the roll damping dynamic derivative for the given roll rate from the slopes of the curve at Φ = 0.

The nondimensional cycle time needs to be sufficiently large for steady-state simulations such that hysteresis due to motion dynamics is not present. A sequence of calculations at different cycle time ranging from 100 to 4000 indicated that cycle time above 1000 will do well. On the other hand, the time step size is governed by flow physics. The nondimensional time unit is the time required for a fluid particle to travel the distance of one chord. At DT=0.2, we are tracking a fluid particle only five times over the airfoil at it moves nominally according to the prescribed free stream velocity. As a result, a complete cycle required 20000 time steps. From a CFD point of view, it is equivalent to computing 10 independent steady state solutions for the same wing configuration.

The computed lift and moment coefficient curve shapes are now practically the same. One of the explanations could be that the CFD flow separation is delayed by 2 degrees in alpha versus the wind tunnel experiment. At this time, it is not entirely clear what is the cause of this discrepancy. The experiment was done with an infinite span wing, not a true two dimensional representation. Further investigation could repeat this simulation using an infinite span CFD configuration. Instead of URANS, we may have to use DES or other hybrid techniques.

This preliminary example is to demonstrate the unsteady behavior of CFD in the stalled regime.

An intriguing feature is the repeatability of the large amplitude oscillation in lift and drag as we repeated the time cycles. The lift and drag data from the experiment did not come from a balance but instead was integrated values from an array of pressure transducers. Assuming that the suction peak on the airfoil may not be perfectly captured by the fixed position pressure transducers, the lift would be lower and the drag would be higher than their respective actual values.

Previous work published at AIAA meeting in Reno 2003, and to be published in AIAA Journal of aircraft (FOM=Figure of Merit).

For DES, RANS is responsible for predicting boundary layer growth and separation. LES is responsible for predicting the geometry dependant turbulent flow features. Grid adaptation done using NASA Langley’s RefineMesh program. Adaptation on time average of vorticity DES results are time-averaged coefficients. Left axis removed to protect proprietary data.

These DES projects represent a cross section of those done over the past few years using Cobalt.

Delta wing vortex breakdown on a delta wing and the F-18C done by Major Scott Morton of the USAF Academy (Scott.morton@usafa.af.mil).

2-D forebody geometry by Kyle Squires (squires@asu.edu).

Prescribed spin of the F-15E by James Forsythe.

Prisms created using “Blacksmith” to recombine the tets in the boundary layer into prisms.

Blacksmith is a Cobalt grid utility.

The following are non-moving cases – but can be unsteady (for DES) CPU hours based on a Compaq ES45. Timestep for DES non-dimensionalized by chord and freestream velocity.

Model was set to a given pitch angle (theta), then rolled about the longitudinal axis (phi). This resulted in a decrease of alpha, and an increase in beta as phi increased. The CFD was performed at the given alphas and betas, which were corrected in the wind tunnel data for wall effects.

Note reversal of rolling moment for phi=30 using SST.

Yawing moment well predicted – as with all cases.

Side force well predicted – as with all cases.

Shock retreating off trailing edge of leading edge flap.

DES isosurface looks like separation is at trailing of leading edge flap. But it moves back from there unsteadily. This leads to the blue low pressure in the separation bubble (since it is not always separated).

The separation moving forward on the right wing is the cause for the roll moment reversal.

At this high phi, the alpha is reduced so much that the flow remains attached until the trailing edge of the wing.

Note asymmetries in wind tunnel data. Decrease in lateral stability derivative picked up with DES.

Good agreement for yawing moment, as with all cases – this is likely due to the attached flow at the tail, which is easily predicted.

Good agreement for side force, as with all cases – this is likely due to the attached flow at the tail, which is easily predicted.

Separation is making it onto the leading edge of the leading edge flap.

Large asymmetries in wind tunnel data. Around this angle there was difficulty in testing, since model dynamics became significant.

Good agreement for yawing moment, as with all cases.

Good agreement for side force, as with all cases.

Rolling moment offset predicted by DES – is the sample size large enough?

Looks like enough samples have been taken to well define rolling moment. However more might change the time-averaged rolling moment some.

Unsteadiness now is due to separation moving from leading to trailing edge of the leading edge flap.

ALE = Arbitrary Eulerian/Lagrangian.

Linear and well behaved. Stable roll damping.

Separation at trailing edge – flow well behaved.

Large rolling moment offset. Several cycles run with varied timestep, but offset remained.

Offset due to differences in separation location. Hysterisis?

Slightly chaotic behavior, but linear and stable roll damping.

Positive roll damping. Note lowered slope – due to lower lift curve slope once shock moves forward on the wing.

Study still underway – not enough samples. Looking at dependence of roll damping on roll rate.

This presentation discusses the requirements for and the ramifications of including unsteady aerodynamics and structural flexibility in the computation of stability and control derivatives for modern flight vehicles.

The motivation behind the inclusion of unsteady aerodynamics and aeroelastic effects in the computation of stability and control (S&C) derivatives will be discussed as they pertain to aeroelastic and aeroservoelastic analysis. This topic will be addressed in the context of two applications, the first being the estimation of S&C derivatives for a cable-mounted aeroservoelastic wind tunnel model tested in the NASA Langley Research Center (LaRC) Transonic Dynamics Tunnel (TDT). The second application will be the prediction of the nonlinear aeroservoelastic phenomenon known as Residual Pitch Oscillation (RPO) on the B-2 Bomber. Techniques and strategies used in these applications to compute S&C derivatives and perform flight simulations will be reviewed, and computational results will be presented.

Within the LaRC Aeroelasticity Branch (AB), there are two primary objectives supporting the computation of stability and control derivatives. The first is to support free-flying cable- mounted aeroelastic and aeroservoelastic wind tunnel investigations in the TDT. The second is to support full-scale aeroelastic and aeroservoelastic analyses of modern flight vehicles. In the former case, since wind tunnel models are often conceptual in nature, a large database describing the model’s flight characteristics, as is usually assembled for full-scale aircraft, is not available.

Therefore, we rely virtually exclusively on empirical, analytical, and computational methods to predict the S&C performance of the model. This includes a requirement to predict both static and dynamic derivatives as well as the impact of structural flexibility on the model’s performance. Since the TDT is a transonic facility, nonlinear aerodynamics is also an important contributor to the analysis. The widespread use of automated flight controls on virtually all modern commercial and military aircraft has introduced a new class of problems where the vehicle control system can interact with the aerodynamics and structural flexibility of the system.

The discipline investigating these interactions is known as aeroservoelasticity and is rapidly growing in importance for prediction of on- and off-design vehicle performance. To effectively predict aeroservoelastic problems, accurate computation of control effectiveness is a must.

The first application to be discussed is the prediction of S&C derivatives for a SST wind tunnel model tested in the LaRC TDT. The model is an aeroelastically scaled model of a 1970’s SST concept. It was developed to investigate control laws for the aircraft. The cable system employed in the TDT provides a five-degree-of-freedom mount for the model. A single vertical cable runs from the wind tunnel ceiling to the wind tunnel floor through a pair of vertically- mounted pulleys installed in the model just forward of the center of gravity. Similarly a single cable runs between the sidewalls of the wind tunnel through a horizontally mounted pair of pulleys aft of the center of gravity. In addition, four snubber cables run from the corners of the tunnel to the model near the C.G. These four cables can be interactively tightened and loosened.

In the tight configuration, they are used to hold the model at the center of the tunnel during wind off conditions, and they are slack during “free-flight” testing. They can also be rapidly tightened when the vehicle encounters an instability to attempt to stabilize the aircraft. The model also includes hydraulically actuated wing and horizontal tail control surfaces. Control effectiveness derivatives, including flexibility effects are required to design the flutter-suppression control laws that are the subject of the test. In addition, the precise position and tension on the cables is defined using a computer program known as GRUMCBL, which requires S&C derivatives for the model.

This slide discusses the wind tunnel test objectives and requirements for S&C derivatives to support the test, emphasizing the importance of COMSAC techniques for this type of testing.

This slide shows a video clip of the SST testing in which an aggressive flutter suppression control law is activated on the model. The model experiences a severe upset and the tunnel bypass valves and snubber support cables are activated. Unfortunately the model upset is too severe, and nonlinearity in the cable mount system and/or aerodynamics slowly drive the model to destruction. While it is unreasonable to blame the destruction of the model on poor predictions of S&C derivatives, this is a stark example of the importance of accurate predictions of these types of derivatives for this type of testing.

S&C derivative predictions for the SST model came from four primary sources, Stability and Control DATCOM, linear doublet lattice, transonic small disturbance potential flow (CAP-TSD), and wind tunnel balance data. Static, dynamic, rigid and flexible derivatives were developed for this configuration. Analyses using the Computational Aeroelasticity Program – Transonic Small Disturbance (CAP-TSD) are the focus of this presentation. Using this methodology, static derivatives were computed using a finite difference technique, but are not the main focus of this discussion. Dynamic derivatives were estimated by pulsing the configuration in pitch, plunge, yaw, and spanwise translation. Roll rate derivatives were computed using a steady analysis and imposing specialized boundary conditions to the lifting surfaces which represent the rolling motion of the aircraft.

This slide describes the essential features of the inviscid and viscous/inviscid interaction versions of CAP-TSD.

Longitudinal and lateral rate derivatives were computed using a pulse analysis. This slide represents a configuration plunge pulse, the lift coefficient response to the pulse, and the transfer function computed from the input and response. The transfer function is derived by dividing the complex Fourier transform of the response by the transform of the input. The character of the transfer function at zero frequency defines the static lift curve slope and the dynamic S&C derivative due to angle-of-attack rate. A pitch pulse of the configuration results in a combined pitch rate and angle-of-attack rate derivative, which in conjunction with the plunge pulse can be used to extract the pitch rate derivative. A similar procedure is used to compute the lateral derivatives due to yaw rate and sideslip rate.

This slide shows the longitudinal and lateral rate derivatives computed by CAP-TSD and compared with results from doublet lattice and DATCOM. While the various method show a general agreement in magnitude and sign between the methods, one is hard-pressed to say the correlation for this case is good. In general, CAP-TSD tends to over predict the magnitude of the rate derivatives, with the exception of pitching moment. There are several modeling assumptions inherent in each of the methods which could have a profound impact on the results, but given the time constraints and objectives of the analysis, it was impossible to investigate these issues. Certainly, further investigation of techniques for computing these derivatives is warranted before widespread acceptance of the methodology can be anticipated.

Roll rate derivatives were computed using CAP-TSD by modifying the lifting surface boundary conditions used in the code to represent a steady rolling motion of the vehicle. Incorporation of the rolling motion in this manner allows roll rate derivatives to be computed using a steady analysis as opposed to a time-accurate computation.

The roll rate derivatives computed by CAP-TSD using this technique are in much better agreement with doublet lattice and DATCOM than were the previous longitudinal and lateral rate derivatives. The exception being yawing moment due to roll rate, which is a historically- difficult derivative to estimate.

Structural flexibility effects were also investigated for the aircraft by adding structural modes to the CAP-TSD model and performing an aeroelastic analysis of the SST configuration. This slide shows the first six structural modes and frequencies included in the aeroelastic analysis.

Lift curve slope, elevator effectiveness and outboard aileron effectiveness as computed by CAP- TSD are compared with balance data on the model acquired in the TDT prior to cable-mount testing. Due to aeroelastic deformations, these derivatives are a function of the dynamic pressure. Since the wind tunnel model is inherently flexible, no rigid data for the model on the balance is available. In general, the magnitude and trends in the data as compared to experiment are very good with the exception of the lift curve slope. CAP-TSD does not compute the wing and horizontal tail carry-over lift across the fuselage making the CAP-TSD lift cure slope lower than that of the experiment. An important feature to note is the loss in elevator and aileron control effectiveness with increasing dynamic pressure predicted by the theory and supported by the experimental data. Both the theory and experiment indicate an elevator reversal for this aircraft at a dynamic pressure between 20 and 30 psf.

In summary, this analysis represents a pure application of the available methodology with minimal opportunity to effectively research the methods employed or the results obtained. All of the data were used to establish bounds for the input data to the GRUMCBL cable-mount stability program to determine cable positioning and tensions for the free-flying test. There is considerable scatter in the derivatives produced by the various methods, particularly for the longitudinal and lateral dynamic derivatives. Structural flexibility was a significant player in this analysis, and both CAP-TSD and the experimental data indicated an elevator reversal at a relatively low dynamic pressure.

This presentation is based upon AIAA Paper 2000-4325. The companion paper documenting the experimental portion of the study is AIAA Paper 2000-4016.

The model was held very securely by the mount and cables, so aeroelastic effects would be negligible. The wing was about 4 feet long with a 2 foot chord, and the tunnel cross section is 7 by 10 feet.

The line of pressure taps in the figure above got displaced somewhat, they should begin at the leading edge of the wing and end about the trailing edge of the flap.

This work was done with the government version of the code, at the time there only was one version… - Parallel computing framework - Required files: input,bc,grid - Submit to machine: choosing the number of processors we want/require - Automatically decomposes grid: ParMetis - ParMetis developed at the Univ. of Minnesota funded by the Army - Recombines into 1 zone when finished - Can restart on a different number of processors (just change 1 input) The difficulty in creating the digital model was defining the airfoil cross section in a way that was sufficiently smooth. Simply digitizing points of the wing was not sufficient as this lead to slope discontinuities between points. Tracing the model and fitting the trace with bezier cubic splines resulted in a continuous smooth digital representation. The main remaining difference between the digital model and the physical model is the fact that the actual wing is hollow, while the digital representation is solid. There are a number of holes and gaps, particularly in the flap area, that allow air to circulate inside the real wing.

For the cases that included the tunnel walls and mount, viscous layers were only included around the wing, flap and end plates.

The lack of agreement in 2D was not unexpected, as the flap was only partial span. Euler completely misses the trailing edge of the flap, laminar is unsteady, and the two turbulent solutions have unexpectedly separated partway along the flap.

Going the 3D really brings the solutions much closer to the experimental data. Unfortunately the separation on the flap is still evident in the turbulent solutions. Hinge moment results are much improved over the 2D solution, but still off by about 50%.

Including the tunnel walls improved the solutions, resulting in a good match along the lower surface of the wing for the viscous solutions. All the solutions moved closer to the experimental data along the upper surface of the wing. The turbulent solutions are still separated on the flap, and laminar is still unsteady.

The medium grid with 2nd order advection for the turbulence quantities is the best match to the experimental results.

Considering the uncertainty in the flap angle due play in the flap and flexing in the moment- measuring torsion cell leads to better agreement with the moment measurements.

Because there is now separation in both the experimental and numerical data, the hinge moment agreement between the two is now much better. Numerical solutions are still missing the suction peak in the experimental data, suggesting there may still be a problem with the digital representation of the model at the leading edge.

Considering the uncertainty in the flap, in this case broken up into the flex in the torsion cell and the play in the flap, lead to very little difference in the pressure coefficients. The hinge moment does indicate a slight improvement.

Qualitatively the computed solutions are very similar to the wind tunnel data, showing the same types of features as the oil flow on the model.

Separation on the flap is evident in the streamlines and velocity vectors of the computed solution.

We learned after the test that there were also maintenance issues with the tunnel. A number of the screens in the inlet of the tunnel are torn or have fallen down, leading to nonuniform flow in the test section.

This presentation is structured in a timeline format. Some introductory material and sample results for this project are presented. Scattered throughout the presentation are some “lessons learned” slides which convey insights that may be of some usefulness to the attendees of this symposium.

Accurate determination of the stability and control characteristics of an aircraft is critical to the safety of that aircraft. The flight envelope of some aircraft often pushes the limits on angle of attack and rapid motion maneuvers. At these flight conditions the flow about the aircraft tends to be unsteady and separated and the variation in the forces and moments with angle of attack or motion rate are often nonlinear.

During the development process of advanced aircraft, numerous hours are spent testing scaled models in wind tunnels to determine the stability and control characteristics. Much of this testing is conducted using models of simplified geometry at Reynolds numbers much lower than typical flight Reynolds number. However, the sensitivity of the forces and moments to Reynolds number makes these empirical approaches to predict dynamic characteristics at full scale high angle of attack conditions a challenging task.

Much of the wind tunnel testing is conducted with special equipment designed for this purpose.

Rotary balance apparatuses, similar to the one shown in the slide, were developed to provide information on the effects of angular rates on the aerodynamic forces and moments acting on the aircraft in flight. The apparatuses are typically complex because they must be capable of measuring forces and moments for the range of rotation rates experienced by high-performance aircraft. Major problems encountered in the application of this test technique include test equipment interference, wind tunnel wall effects, equipment blockage ratio, and Reynolds number scaling effects in addition to the difficulty of conducting detailed flow measurements such as surface pressures.

This study is motivated by the need of industry to quickly and inexpensively determine stability and control characteristics of new aircraft without solely having to resort to wind tunnel experiments. In particular, rotary balance testing at high Reynolds numbers is complex because of the high structural loads imposed on the model and the rotary test apparatus. The high loading is the result of the combined effects of high dynamic pressure of the wind tunnel flow and the high angular velocities required to match the flow velocities for a given spin coefficient. The potentially cheaper and faster aspects of CFD make it a helpful addition to experiments, if it can be proven to be reliable under dynamic conditions.

This project began prior to year 2000 with Teryn DalBello and Case van Dam. The computational grids were created from physical measurements of the actually test models. Some preliminary results were computed on NAS’ CRAY machines using the rotorcraft version of Overflow. Syta Saephan joined the project in year 2000 and continued running some additional cases.

The ultimate goal of the project was to assess the capability of CFD (and Overflow in particular) in predicting the flow about aircraft forebodies at high angle of attack rotary conditions. The plan was to start off with something simple and manageable to gain experience and familiarity with the grid generation tools and flow solver. The ogive geometry was chosen for two reasons.

First, the geometry resembled that of advanced aircraft and would produce similar vortical flow structures. Second, experimental data are available to validate the computational results.

In the mid 1990s, rotary balance experiments were conducted in Britain on isolated circular and square cross section ogive models at angles of attack of 60º and 90º over a range of Reynolds numbers from 80,000 to 2,250,000 based on the maximum body diameter.[1],[2] The purpose of these experiments was to determine the effects of Reynolds number, angular velocity, and nose shape on the aerodynamic characteristics of the models. These tests were unique in so far that this was the first time that surface pressure distributions were measured under rotary conditions in a pressurized wind tunnel. A second objective of these experiments was to provide a database for the development and validation of high angle of attack computational methods. This database forms the basis for the computational investigations of this report.

[1] Dunham, D.M., “Rotary Results on Forebody Models,” Cooperative Programme on Dynamic Wind Tunnel Experiments for Maneuvering Aircraft, AGARD AR-305, October 1996, pp.7-1–7-14.

[2] Pauley, H., Ralston, J., and Dickes, E., “Experimental Study of the Effects of Reynolds Number on High Angle of Attack Aerodynamic Characteristics of Forebodies During Rotary Motion,” NASA CR 195033, January 1995.

These two plots show the side force and yawing moment coefficients as a function of spin rate at various Reynolds numbers as measured in a spin tunnel. Notice that the forces and moments are highly nonlinear, even changing signs, over the range of spin rates and Reynolds numbers.

The computational grids were created from detailed physical measurements of the test models.

Several geometries were used in the wind tunnel experiments, but only two geometries are used for this project. Both ogives are 36 inches in length and 6 inches in diameter. The difference between the two geometries is their cross sectional shape, with one being circular and the other being a square with rounded corners. Both computational grids are single grids with 130 points distributed in the axial direction, 181 points in the circumferential direction, and 54 points in the normal direction.

Surface pressure data were measured with pressure taps on the test models. The measurements were taken at the axial locations labeled as stations on the slide. The station number refers to its location from the nosecone tip. Station 1 is one inch from the nosecone tip. Station 29 is 29 inches from the nosecone tip. All stations had 32 equally spaced pressure taps except for Station 1 which only accommodated 30 taps.

Note that the ogive is divided into three regions, the forebody, midbody, and aftbody. The forebody region is bounded by Stations 1 and 11 and will be referenced as “forebody”. The complete configuration comprising of all three regions will be referenced as “ogive”.

The flow solver used for this project is OVERFLOW. There are several versions of OVERFLOW in circulation, each having been specially modified to serve a particular need. We have used different versions of OVERFLOW for this project. The results shown in this presentation were obtained using the rotorcraft version of OVERFLOW 1.6.

Despite the numerous versions afloat, all derivative versions of OVERFLOW share some commonality. OVERFLOW is a Reynolds averaged Navier Stokes flow solver for structured grids. It can solve problems on single or chimera overset grids, with the more recent versions having more overset grid capability. Users can choose which turbulence model to use from a selection that includes Baldwin-Lomax, Baldwin-Barth, Spalart-Allmaras, and other higher order models as well.

One feature that is incorporated into the rotorcraft version of OVERFLOW is the use of source terms for constant rotary motion. The source terms allow the simulation of rotary motion with a static grid. Rotary motion can also be simulated by physically rotating the grid at each timestep in a time accurate mode. In this case, the source terms would not be needed.

The simulation parameters for the result presented in this presentation are shown in the slide.

The Reynolds number based on the ogive diameter is just over 2 million and the flow is assumed to be fully turbulent. These cases represent the highest Reynolds number cases tested in the wind tunnel. Three different spin coefficients were tested for each geometry. The spin coefficient is proportional to the ratio of the angular velocity and the freestream velocity.

The cases will be simulated using a static grid and source terms for the rotational effects.

Baldwin-Lomax algebraic turbulence model is used because other researchers have achieved better results with this model in these high-alpha flow problems.

Numerous cases were computed on the CRAY machines. The results need to be validated against the wind tunnel data.

The experiments measured the side force and yawing moment acting on the model via a vertically mounted sting at the model’s center of gravity. Surface pressures were also measured at the six forebody stations and two aftbody stations shown in an earlier slide. The surface pressure measurement allowed for a more detail comparison of the forces and moments acting on the forebody region of the model. The measured surface pressures are integrated over the forebody region to obtain the forces and moments acting on the forebody region.

These three plots show the drop in the residual and the convergence of the side force and yawing moment. The residual in terms of the L2-norm is reduced by three orders of magnitude and is approaching machine zero. The side force for the ogive (shown in blue) starts out very oscillatory but does dampen out and does converge in less than 10,000 iterations for this case.

The side force for the forebody region (shown in green) converges much faster than the overall ogive side force. In fact, the forebody forces and moment typically converge 60-80% faster. A similar convergence pattern is seen for the yawing moment.

The color contours compare the computational (left) and experimental (right) surface pressure distribution for the circular ogive case at a spin coefficient of -0.2. The experiments did not make any measurements at the very tip of the forebody and hence the white hole in the middle of the picture. The blue region is on the windward side.

Qualitatively, the comparison is not bad with the skewness and gradient patterns being captured in the simulation. To compare the results on a more quantitative level, the surface pressure is integrated to get the forces and moments acting on the forebody. The results of the surface pressure integration is shown in the table. The normal and axial components of the force along with the rolling and pitching components of the moment agreed to within a few percent of the experimental values. The side force and yawing moment components are the more difficult components to predict and the discrepancy is larger.

Whereas the last slide compared the surface pressure distribution in a colorful and qualitative manner, these pressure plots compare the results in a more quantitative manner. These three plots show the pressure distribution on the forebody at stations 2, 6, and 11. The trend predicted by the flow solver agrees with that measured in the tunnel, although pressures on the windward side are predicted with a higher degree of accuracy than values on the leeward side. The computed pressures are slightly yet noticeably shifted upwards when compared to the measured values. This small shift may be attributed to the known reference pressure error in the experimental data.

These two plots compare the surface pressures at the two stations on the aftbody. Again, the solver captures the trend of the surface pressure.

The circular ogive geometry was simulated at three spin coefficients. Rather than show line plots and color contours for each case, the results have been consolidated into this table comparing forebody forces and moments. Except for the side force and yawing moment, all of the other components have less than a 5% error relative to the experimental results. The errors in the side forces are generally smaller than errors in yawing moments as expected. Moments are much more difficult to predict as they are more sensitive to variations in surface pressure far from the moment center.

No one really gives reference frames a second thought and this is okay 90% of the time when all the different frames are aligned and identical. However, there may be instances where multiple reference frames exists and one needs to pay special attention to make sure all comparisons are with respect to the same reference frame.

For our problem, there are three different reference frames in use. The wind tunnel experiments measured forces and moments in body-fixed axes with the x-axis aligned with the axis of the model. Ideally, the CFD computational domain should employ the same axis orientation for a simple and direct comparison of results. However, there may be additional constraints imposed by the gridding process or flow solver that dictate a certain axis orientation as was the situation for this project. The source term formulation restricts the rotary motion about the z-axis only.

This single restriction led to the use of three separate reference frames. As mentioned, the experimental data referenced the body-fixed system. OVERFLOW reported forces in the wind axis system and moments in the body axis system. To directly compare the computational results with the experimental data, all computational forces and moments will be transformed into the body-fixed axes system.

Another lesson learned is that the forebody region is much more accurately predicted by the solver than the ogive as a whole. One possible reason why the forebody results tend to be better than ogive results is that the sting is not modeled in the computations. The forebody is upstream of the vertically mounted test model sting and the flow in that region is less affected by the sting.

Flow in regions downstream of the sting are undoubted affected by the sting. Hence, the lack of the sting in the computational model means that its effects are not captured and is a reason for the ogive result discrepancy.

Also, other wind tunnel interferences such as tunnel wall and the presence of the rotary rig (none of which are modeled in the simulations) can affect the solution as well.

Can the ogive force and moment discrepancy be completely blamed on wind tunnel interferences or are there other culprits?

The color contours compare the computational (left) and experimental (right) surface pressure distribution for the square ogive case at a spin coefficient of -0.1. The scale for the surface pressure is the same as that of the circular ogive case.

Qualitatively, the comparison is fair with the major flow features captured. Quantitatively, the computed forebody forces and moments differ the experimental values by varying amounts. The normal force, axial force, and pitching moment agree with the experimental values to within 11%. However, the other components of forces and moments differ by significant percentages.

It should be noted that some of these force and moment coefficients are very small and hard to predict and even small differences in value can result in large percentage deviations.

These three plots show the pressure distribution on the square ogive’s forebody section at stations 2, 6, and 11. The flow solver predicts the trend of the surface pressure although there are some discrepancies when comparing absolute values. The pressures on the windward side are predicted with a higher degree of accuracy than values on the leeward side, as was the case for the circular ogive geometry.

These two plots compare the surface pressures at the two stations on the aftbody.

The square ogive geometry was simulated at three spin coefficients. This table compares the predicted forebody forces and moments to the experimental values. Except for the side force and yawing moment, all of the other components have less than a 5% error relative to the experimental results. The errors in the side forces are generally smaller than errors in yawing moments as expected. Moments are much more difficult to predict as they are more sensitive to variations in surface pressure far from the moment center.

This table (and the one for the circular ogive) show nonzero side force, rolling moment, and yawing moment coefficients for the case with zero rotation. There are several reasons why they should not be zero as confirmed by the nonzero values measured in the wind tunnels. Neither of the two wind tunnel models is perfectly symmetric and the slight asymmetry is replicated in the surface grids. Even for nominally symmetric bodies the flow tends to become asymmetric in the angle of attack range from approximately 20˚ to 70˚. At lower angles of attack, the flow remains symmetric whereas at higher angles unsteady vortex shedding occurs. Hence, it is not unexpected that the side force, rolling moment, and yawing moment are nonzero at the stationary, condition for the bodies considered here.

By year 2002, NAS has replaced its older CRAY machine with a newer one. This newer CRAY will be operational for one to two years, after which time NAS will no longer use CRAYs.

Rather than transition over to the new CRAY, this was an appropriate time to transition to parallel computing (on NAS’ SGI clusters) and our own Linux clusters. The transition to new hardware was also an appropriate time to transition to a newer version of OVERFLOW.

OVERFLOW-D has major improvements over its predecessor. OVERFLOW-D is ideal for chimera overset grids with a built-in ability to fill the computational domain with Cartesian volume grids. Also, the flow solver can better load balance the problem between parallel processors by splitting large grids into smaller grids. OVERFLOW-D retained the source term capability, but has improved dynamic grid capabilities.

Several lessons were learned when we transitioned to OVERFLOW-D. We restricted ourselves to single grid topologies when we used the older rotorcraft version. When we transitioned to OVERFLOW-D, we tried to simulate the same case with overset grids. Even though the ogive geometry is very simple compared to aircraft geometries, we had difficulty making a good grid without any orphan points. We also ran into some problems with the hole cutting and domain connectivity steps. That problem was fixed with a small modification to a hole-cutting subroutine. Much of these problems and hardships can be avoided or minimized by using as few grids as possible.

An additional incentive to use single grid topologies whenever possible is that solutions on the a single grid domain converges much faster than solutions on chimera grids domains. For the ogive geometry, single grid cases converge 75% faster than chimera grid cases.

These rotary problems can be simulated in two ways. The “fast” way is to simulate the problem with a static grid and use rotational source terms. The “slower” way is to simulate the problem with a dynamic grid that rotates at each time step. Because the grid is physically rotating, the source terms are not needed.

There are several things that need to be considered when deciding on which approach to take.

First and foremost, can the problem be simulated with source terms. The use of source terms is only valid if the entire body is undergoing constant rotary motion and there is no relative motion between grid components. Secondly, does the flow solver support source term formulation?

The use of source terms will result in significant computational resource savings. Dynamic simulations often require very small time steps for solution stability. For the ogive geometry and flow conditions described for this project, a time step equivalent to 0.01 ° grid rotation was needed for solution stability. At each time step, at least three subiterations are needed for proper solution development. For one complete grid revolution, the solution would have computed 100,000+ iterations.

At the current time, we are attempting to simulate the F16XL under coning motion.

We will be simulating an F16XL under coning motion. The first task is to create a CFD suitable grid from the CAD grid we obtained. That task will be difficult and time consuming for a complex geometry such as this. Once the gridding step is complete, we plan to simulate the problem on our homogenous network of Linux computers.

In recent years, computing power has increased significantly while prices have dropped just as significantly. We transitioned away from the CRAY and into parallel computing. Currently, NAS operates several massively parallel SGI machines with 512 processors and 1028 processors.

Assess to these machines is charged by the hour. Unless you need the massive computing power that these SGI computers provide, PC computer prices have dropped so much that its economical to purchase a bunch of PCs and network them into a “mini” cluster.

We have assembled a homogenous network of 10 PC running Linux. An Athlon XP 2500+ with 512MB of memory and a 80GB hard drive can be purchased for about $500 each. These Linux clusters are highly scalable. The factor retarding further scalability (for us at least) is network connection speed. We are using a 100 MBit ethernet cards and network switch. 1000 Mbit cards and switches are available but a bit expensive at the moment. They should become more affordable as they gain wider acceptance and a larger user base. With gigabit network connections, these Linux cluster can be scaled to any number of processors without any noticeable performance degradation.

We had two objectives in mind when putting together this presentation. First, we wanted to let people know what we were doing in terms of rotary flow problems. Ogives were simulated under coning motion to determine the feasibility and capability of OVERFLOW. Sample results were presented for cases simulated. The results for the forebody show reasonable agreement to experimental data.

Our second objective is to convey some of the lessons we learned during the course of the project. We hope that these lessons will be useful to the attendees of this workshop.

The goal here was to present one approach to rapid CFD for S&C using an unstructured inviscid method, in order to eventually assess S&C properties as early in the design process as possible.

Specific results are presented regarding time, accuracy (as compared to a baseline wind tunnel database) and simplicity for the user. For COMSAC, it’s more important to talk about the “specifications” required by Advanced Design and S&C, as well as how the CFD results can be combined for envelope evaluation.

Two configurations were considered, the tailless delta wing (“ICE”) configuration and the MTVI configuration. Each configuration actually has two vortices, with the second vortex on the ICE model starting at the change in camber and thickness where the fuselage and the wing blend.

Accurate CFD analysis on vortex-dominated flows requires resolving the vortex core; adaptive methods can focus in on the core, but may need to be setup to do so. Vortex breakdown will also be significant in a vortex-dominated flow.

Many CFD cases were run on the two configurations to complete the run matrices.

Two versions of Splitflow were used; the parallel version does not support the hybrid grids made from prisms extruded from the surface triangulation. The parallel version is still used today on engineering workstations, SGI Origins, and parallel clusters.

The contract effort was divided into two parts, CFD analysis and metrics. The CFD analysis used solution-based adaptation, checked for grid convergence as part of the data comparison, and considered grid resolution as part of the study. Metrics were specified by representatives from two groups, Preliminary Design and Stability & Controls, for the time required, accuracy required, and some measure of the ease of use.

“You cannot solve what you do not resolve.”—Steve Karman. Solution-based grid adaptation gives the grid the opportunity to adapt to the flow solution as the solution progresses. The grid itself helps set what kinds of flow solutions can be modeled, so it is critical to have an appropriate grid. The three plots here show an example of adapting to vorticity, helicity, or not at all, and they show dramatically different results. Traditionally, helicity was used by Splitflow to adapt on vortical structures—the concern was that vorticity would simply add cells to the boundary layer. Unfortunately, helicity does not highlight the vortex core, which is critical for modeling a vortex, and the field value of helicity itself fall to near-noise levels after a vortex burst. Together, those effects make considering raw vorticity critical to adapting to vortex- dominated flowfields.

Sufficient leading-edge resolution is often critical to setting up the proper flowfield and accurately computing the aerodynamic characteristics on the suction-side of the vehicle.

However, the gradients of classic analyzed variables (e.g. pressure, Mach number, density) are not very large compared to those inside of vortices or near shocks. This results in most of the solution-based adaptation going into those areas, not to the leading edge.

Metrics are one of the real topics of discussion here. Metrics were developed by consulting with specific colleagues in the areas of Advanced Design and Stability & Control. The metrics presented here could be considered a start for a discussion of more general standards.

Criteria for evaluating CFD for Preliminary Design were produced by discussions with individuals involved in Preliminary Design. At this point, the data presented only represents a few opinions on the matter, but it’s a start.

Preliminary design is focused on generating the necessary lift at the minimum drag.

CFD tools are very useful already, but there are still needed improvements. The user interface is the weak spot, both in getting control information into the code, and in getting configuration information in. These weaknesses affect other codes, too, not just Splitflow, which is very dependent on a “good” surface mesh.

S&C requirements are a little different than Advanced Design, which really focuses on performance. The really tough one is the “within two degrees of zeroes.” When plotted, that band really necks down, and when evaluated, it’s very difficult to accomplish with the methods in this presentation.

Originally, results were shown for each sweep listed above. Those results are included at the end for completeness but will not be presented here.

This condition is shown to highlight some of the success and some of the difficulty in this kind of CFD for S&C. Note that the results look “pretty good” until alpha>20. The error bars indicate how much change there was in the CFD results near the end of the run—note that some of these cases converged very well, just not to something that compares to the wind tunnel data.

This group of results shows just how squirrelly some of the CFD analyses were. Note the pitching moment at alpha > 10 in particular, even though the general comparison to the wind tunnel data looks better than it did for the beta=10. Roll Pitch Yaw This is one example of the kind of convergence experienced on these CFD runs. Note the yawing moment, whose sign is questionable. The value itself is small, however.

Some good results, some bad. At lower AOA, where 24 degrees isn’t all that low, the results could be pretty useful. Above 24 degrees, with massive separation, vortex breakdown, and maybe other effects, the results were unpredictable.

This shows some of the emphasis and difficulty in reproducing tunnel results. The error bars on the CFD values show how much the data was changing as the run “converged.” Note that the for the left case, all of the runs converged quite well; at 24 degrees, it converged to something completely different from what was measured in the wind tunnel. At 20 degrees sideslip, beyond 14 degrees, the CFD really did not converge very well at all. One thing to keep in mind is that we do not know what the degree of variation of the wind tunnel data was.

This test of the same ICE model shows that some of the wind tunnel data may not be as certain as it is credited. Note particularly that the 20 and 30 degree sideslip data lie between the 0 and the +/-10 degree sideslip curves, indicating that something happens to change the pitching moment, and then it comes back.

Falling Leaf is an extreme flight condition where the airplane rapidly transitions between high beta and low alpha, to high alpha and zero beta, on to high alpha and high negative beta, and then back. The Sustained Roll-Yaw Parameter measures the susceptibility to Falling Leaf, compared to simple departure or a stable Dutch roll.

These are two definitions and the numeric values for the ICE configuration.

Then, defining those two parameters, we can check for susceptibility to Falling Leaf.

Now we can take both the wind tunnel dataset and the Splitflow dataset, compute the Dutch-roll parameter and the “syrup” parameter, plot them together, and determine the envelope susceptible to Falling Leaf. Here, the first important thing to note is that the Splitflow envelope prediction was pretty close to the wind tunnel; the second is that an active control system would be required for all angles of attack for this aircraft. Note that the x-scales are different between Splitflow and SARL data These methods can produce useful results, and in fact these and similar methods are already in use in the LM Aero Advanced Design and S&C community. It may not be routine yet, and each needs large run matrices that are not yet supported by our CFD tools and computer resources.

Another problem is that while this project set out to find a “black box” configuration, no such thing was determined, and in many cases each flight point CFD run required individual attention.

Some of the performance applies to five years ago, too. Now, a lot of this analysis can be (and is) performed, opening up these requirements to apply to complete aircraft instead of the simpler configurations (ICE & MTVI) presented herein.

The remainder of the presentation consists of support slides.

The MTVI model was run in an alpha sweep at zero sideslip and small (2 degrees) sideslip.

The ICE model was run for an alpha sweep at several sideslip angles (0,5,10,20 degrees). A beta-sweep was also done at 20 degrees AOA.

A few viscous cases were also added to investigate the region near AOA=24 at sideslip, an area where the results compared poorly.

Most of this is old news. Our J90 isn’t even used to heat the room anymore. These cases would require about 4 hours on 16 CPU of our Pentium4 cluster, which is very inexpensive.

5 years ago, this is how this project concluded—we needed faster solutions and more of them.

Minimum drag, of course, is now the subject of its own AIAA committee and annual meetings, and it doesn’t seem that anyone has a good handle on it yet.

Our choice of title may seem strange but we mean each word. In this talk, we are not going to be concerned with computations made “after the fact”, i.e. those for which data are available and which are being conducted for explanation and insight.

Here we are interested in preventing S&C design problems by finding them through computation before data are available. For such a computation to have any credibility with those who absorb the risk, it is necessary to quantitatively PREDICT the quality of the computational results.

Please note two things: There are a large number of people at Langley Research Center who are working on these issues, but we got tasked with presenting this talk.

We do not claim that these notions are original to us, but the application and emphasis may be.

We want to make two points here: 1. No answer or a qualitative answer to the question “How good is my answer?” is not good enough for assessing risk. We will address this point in more detail later 2. Where insightful and accurate S&C predictions are most desperately needed is in the design environment. Making a computation after data have been obtained is not a prediction --- it is an explanatory effort. Explanatory efforts can be very useful but they do not require prediction of uncertainty. Note that attempts at calibration do, however, require uncertainty assessments of both the prediction and the experimental data. Otherwise, one has no idea how “fuzzy” the calibration/validation is.

This chart is designed to illustrate the relationship of uncertainty quantification to risk.

At the left end of the chart, there is no defined and managed process in place and no uncertainty quantification is possible. For this state of affairs, the decision maker, i.e. the person or group that uses the computational results, necessarily assumes all of the risk associated with any inaccuracy of the prediction.

At the other end of the chart, the computationalist predicts the uncertainty following protocols and certification procedures suitable for use in a Court of Law. For this state of affairs, the risk is assumed entirely by the computationalist and he or she can be sued.

The other stages progress from the left to right, but please note that even the very first stage beyond the state of no quantification requires definition of a process and some sort of management system for verifying that the process is being followed.

Most (or virtually all) CFD is performed today without quantifying the consequences of uncertainty to an outcome metric. When uncertainty has been considered, it is usually restricted to a limited assessment of grid effects; other sources (turbulence model, algorithm, parameters, user practices, …) are generally left unaddressed.

The chart contrasts two customer requirements for a CFD computation of pitching moment. On the left is a cruise transport trim application where the required accuracy of the Cm prediction is +/- 0.001. On the right is a high angle of attack S&C application where the performance requirement is to have at least -0.1 nose down authority. The scales are set accordingly for each customer requirement, and the chart also provides for some grid sensitivity information to be added.

In the absence of quantified uncertainty, all that known is the deterministic result that Cm = - 0.1351. It is not know to any level of confidence if this calculation meets either customer’s requirement. Under such circumstances, the prediction is of limited use.

This figure has the identical format to the previous one. However, the results from a fairly extensive uncertainty quantification process are included. Forty-five computations were performed at three different grid density levels. Simple statistics now tell us that Cm = -0.137 +/- 0.017 at 95% confidence. This outcome includes a variety of uncertainty sources (different grids, different turbulence models, different flow solvers, etc.)

The individual results are also shown on both the left and right sides of the figure to put them in the context of the two customer requirements. It is clear that the variation is (1) completely unacceptable to the cruise trim requirement on the left and (2) completely acceptable to the high- a S&C objective on the right.

Simply put, uncertainty quantification entails determination of conventional terms (average, standard deviation, and confidence) subject to certain process requirements. However, practical techniques will be required to quantify computational uncertainty within available resources and on a timescale consistent to project requirements.

Looking at individual pieces of the quality problem shines more light on them and actually recognizes that they each require different processes and ways of thinking. They end up being separate disciplines which develop on their own and co-evolve as well.

The question “How good does my answer have to be?” can only be answered by the customer of the computational results, i.e. the risk taker. Of course, the customer should always be informed of the likely quality of the process before he/she commissions the work. Furthermore, the resources provided by the customer can have a significant impact on the possible quality of the computational results.

Unfortunately, the general absence of quantitative predictions of computational uncertainty has led to a typical customer demand of “Do the best you can.” However, recent efforts at several institutions to establish wind tunnel data quality assurance programs have encouraged some customers, most notably performance groups, to revisit their quality needs and to develop well- defined processes for establishing defendable uncertainty requirements. These uncertainty requirements are usually traceable to some design or regulatory requirement that must be met for the airframe program to succeed. Some of these requirements are not even technical in nature, but nevertheless must be met.

Answering this question is the present focus of the computational uncertainty quantification work at Langley Research Center. It is impossible in this short talk to address anything more than the general notions. We recommend the following references: P. J. Roache, “Verification and Validation in Computational Science and Engineering”, Hermosa, 1998.

W. L. Oberkampf, T. G. Trucano, “Verification and Validation in Computational Fluid Dynamics”, Progress in Aerospace Sciences, Vol. 38, No. 3, 2002, pp. 209-272.

J. M. Luckring, M. J. Hemsch, J. H. Morrison, “Uncertainty in Computational Aerodynamics”, AIAA-2003-0409, January 2003.

M. R. Mendenhall, R. Childs, “Best Practices for Reduction of Uncertainty in CFD Results”, AIAA-2003-0411, January 2003.

M. J. Hemsch, “Statistical Analysis of CFD Solutions from the Drag Prediction Workshop”, AIAA-2002-0842.

M. J. Hemsch, “Statistical Analysis of CFD Solutions from 2nd Drag Prediction Workshop”, AIAA-2004-0556, January 2004.

This question really addresses the issue of what process do I need?

It is possible to use lower-order-physics codes for S&C problems as long as the domain of uncertainty predictability is known in advance. This means that the problem of interest would have to be pretty close to a previously-quantified domain.

For true prediction, when such a previously-quantified domain does not exist, quantified uncertainty prediction does not seem possible.

Note that this notion is especially important when novel configurations are being considered.

We recommend the following reference for further reading on process quality assurance: M. C. Paulk, et al, “Capability Maturity Model for Software, Version 1.1”, Technical Report CMU/SEI-93-TR-024 (also ESC-TR-93-177), Software Engineering Institute, Carnegie Mellon University, February 1993 (download from SEI website).

The above referenced document shows how to create and manage such a process. Paulk, et al have applied the approach to the software development process, but it applies just as well to any process, including uncertainty prediction/quantification.

We would like to note that often the very act of measuring the outcome of a process (Evaluation) will lead to improvement in the process result. This was evident in the improvement of the Second Drag Prediction Workshop results over those of the First. We have also seen this in our development of statistical control of wind tunnel measurement processes.

If we think of prediction as a manufacturing process, then we have the situation described schematically above. We would never expect every widget coming off the line to have identical dimensions and, similarly, we should not expect every prediction to have no variation across, codes, grid types, users, turbulence models, etc.

We do want to emphasize that to realize the full benefit of thinking this way and making it happen, it will be necessary to be fairly proficient at some basic statistical methods. The methods of greatest interest are the same ones used extensively in metrology and experimentation, particularly statistical quality/process control. Fortunately, these methods are not complicated. They do, however, require the user to get into a “statistical frame of mind” in order to use them effectively and correctly.

In this presentation, we talk a lot about processes because the notion is fundamental to quality assurance, especially quantitative quality assurance.

The best way that we know of to enable determination of quality is to think of computation as a process for manufacturing numbers. One of the biggest advantages of thinking this way is that we can borrow most of the methods and strategies of the manufacturing quality assurance community that have been developed over the last 80 years. In addition, we can borrow the extensions of those ideas to precision experimental work that have been developed at the National Bureau of Standards over the last 40 years.

The quality assurance levels listed in the slide have been implicit in the quality literature but they were first promoted heavily by the Software Engineering Institute. (see reference on slide 11).

These aspects are crucial for the credible prediction of computational uncertainty. The DoD actually has a process for certifying the quality assurance level attained on a sustained basis by a contractor’s software development process, ranging from Level 1 (no process) to Level 5 (all of the above attributes are included).

This breakdown of tasks was established by the National Bureau of Standard over 40 years ago for precision measurement in standards and calibration labs. It makes a seemingly impossible task not only tractable but controllable and credible.

The first task, Calibration of Instruments, is done offline and provides a common reference state for all facilities which are traceable to national standards.

The second task involves periodic offline testing of the measurement system using standard artifacts which are called check standards. This task is done solely for the purposes of tracking any possible drift in the mean or dispersion of the measurement output of the system. It also allows the credible characterization of that dispersion.

The third task involves the off-line determination of systematic errors in the measurement system. For a wind tunnel, some examples would be imperfect knowledge of the test section calibration coefficient and imperfect correction of wall effects.

The fourth task involves those quality checks to be conducted during a test. Those checks are conducted by comparing data taken during the test for that purpose against historical data. There are many such checks that need to be done in a wind tunnel test.

The first task again provides referenceability by being able to prove that the code is doing what it is purported to do. This task is usually called “Code Verification” (Roache) The second task requires that the output variation of the computational process be controlled and evaluated. This can be effected through best practices and comparing the results of multiple codes, grid types, turbulence models, users, etc. There is a belief that attempts at grid convergence will be helpful here with part of this variation but preliminary results are not encouraging. This task is part of what is usually called “Solution Verification” (Roache).

The third task involves parameter and model form uncertainty. There are a variety of ways to propagate parameter uncertainty into the code output and we are encouraged that these methods not only work but can be reasonably implemented. Model form uncertainty is another story and much work needs to be done here. The most promising notion that we’ve seen is the idea from statistics of “severe testing” in which one attempts to find both the portions of the envelope where the predictions are reliable and the accuracy can be evaluated and the boundaries of those portions where the predictions become less reliable and accuracy becomes more difficult to predict. This task is usually called “Validation” (Roache).

The fourth task involves checks to be made when a prediction is being made for a customer. Here it will be necessary (1) to assure that the best practice system is being followed so that predictions of process uncertainty have credibility and (2) to estimate the locations of the envelope boundaries where the credibility of the predicted systematic uncertainty becomes more problematical. This task is the on-line part of Solution Verification.

See http://aaac.larc.nasa.gov/tsab/cfdlarc/aiaa-dpw/.

We do not want, with the emphasis of this slide, to inadvertently give the impression that only on-line work counts. To the contrary, Slide 15 shows that we consider the off-line work described therein to be essential for a tractable and accurate process. By “local”, we simply mean local in the physical inference space (right physics).

The best practices work is being accomplished under NASA Langley Research Center contract number NAS1-03053.

Technical Monitor: J. H. Morrison 757-864-2294 Joseph.H.Morrison@NASA.gov Successful use of CFD to provide aerodynamics for stability and control (S&C) applications will require that the traditional time and costs associated with CFD be reduced and that the errors and uncertainties currently associated with CFD be better understood. CFD will be required to work under a wide range of flow conditions and provide fast and reliable aerodynamics if it is to contribute to this next generation of S&C analyses. CFD solutions have errors and uncertainties due to poorly converged solutions, solution anomalies caused by grids, turbulence models, and parameter selection, and other manifold reasons.

In addition to the above problems, there will be a requirement for communications between the CFD expert and the S&C expert and possibly experts from other related disciplines. The CFD expert may not understand the technical problems associated with S&C, and it is almost certain the converse is true.

Some problems to be anticipated in using CFD for stability and control involve the need for aerodynamic characteristics for a broad range of flight conditions. These flight conditions may include flow regimes that are very difficult to handle with CFD; for example, post stall flight and unsteady flow conditions. It may be necessary for the CFD engineer to use solutions that are not converged and glean the best possible aerodynamic characteristics from this information. It will be essential to know and understand the quality of the solution and the uncertainties that must be associated with the aerodynamic coefficients.

In addition, the large number of points required to complete an aerodynamic database will put strong demands on the stability and control budget. It will be necessary to set up and complete runs faster and more efficiently to keep costs down. It will be important to eliminate incorrect runs and avoid unnecessary repeat runs. A typical aerodynamic database may require hundreds if not thousands of runs. This large quantity of data may require automated techniques to evaluate the results.

The CFD user has a critical role in current CFD processes, and the quality of computational aerodynamic results is subject to significant variability. Many problems can be traced to inexperienced users producing results with software they do not understand.

1.

Lee, J. R., “Certainty in Stockpile Computing: Recommending a Verification and Validation Program for Scientific Software,” SAND98-2420, Sandia National Laboratories, November 1998.

NEAR is currently working to develop a system of best practices for CFD. The purpose of this work is to provide a set of user guidelines for running CFD codes to assist all users in obtaining high-quality solutions with reduced uncertainty and at lower cost. The system includes specific guidelines for problem definition, input preparation, grid generation, code selection, parameter specification, and results interpretation. The objective of the best practices system is to ensure that all reasonable steps are taken to achieve the most accurate and reliable CFD solutions possible.

Mendenhall, M. R., Childs, R. E., and Morrison, J. H., “Best Practices for Reduction of Uncertainty in CFD Results,” AIAA 2003-0411, January 2003.

This is a diagram of a proposed approach to integrate Best Practices and COMSAC. The user will specify the flight conditions of interest, the objectives of the design, and the geometry of the configuration. Best practices will use this information, along with the expert CFD knowledge in the system, to specify a number of guidelines for setting up the CFD runs to provide the best possible aerodynamics information. These aerodynamic characteristics will be available to the stability & control area as needed for design purposes. In addition, the up-to-date aerodynamics can be made available to other disciplines in a similar manner.

In the diagram, the dashed box represents COMSAC. The aerodynamic information generated for COMSAC can be shared with other disciplines.

The objective is to provide a state-of-the-art capability for CFD analysis which can be broadly applied by users interested in (1) improving the accuracy and reducing the uncertainty of CFD results, and (2) reducing the time and cost associated with CFD applications. Best Practices must include the following: • Provide an intuitive process for general acceptance by the CFD community.

• Demonstrate ease of use for all levels of users.

• Provide information appropriate for all CFD users to achieve more reliable CFD results with less effort.

• Provide a comprehensive compendium of procedures and expertise that should be followed to get the required accuracy from CFD.

• Evolve with advancements in CFD algorithms and codes.

• Choose a framework for best practices applicable to all algorithms and solvers that developers and/or users are willing to support.

• Permit individual users to customize details of best practices to support specific needs and provide for proprietary versions of the system.

• Provide a self-critical system by noting the relative confidence in specific guidelines and characterizing the aspects of CFD practices that are poorly understood.

• Provide the flexibility to evolve into a future system which may require highly automated CFD quality assurance algorithms, including automatic grid generation and solution interrogation algorithms.

Since different CFD codes have strengths and weaknesses in different areas, the approach to best practices is to include information and guidelines tailored for specific codes. The knowledge database will be obtained from experienced and successful code developers and users. Links to references will provide the user with sources of additional information. A number of standard runs which can be used for validation purposes will be included. The system will be easily updatable since knowledge and experience with the codes is always changing.

The expert knowledge for best practices is obtained through personal interviews with expert users and developers. The recorded interviews are transcribed, checked for accuracy, and edited before the information is distributed in the knowledge database. The information is linked by a framework of keywords. The knowledge database is also linked to a references database containing citations to published information. When a keyword is selected by the user or automatically chosen by the best practices system, a search will identify the appropriate expert knowledge and related technical references.

The initial best practices system will be a public version for unlimited distribution. Future systems developed for commercial or government organizations will include proprietary information. These systems will contain the corporate memory of the organization, and it may also include sensitive and competitive information for restricted use.

A hierarchical keyword structure is used to organize the knowledge stored in the databases.

Each node in the hierarchy is represented by a keyword that describes a topic. This model allows the information stored to be linked in a logical fashion, which assists in both the knowledge acquisition and the knowledge retrieval mechanism. The user can add additional keywords as needed.

For purposes of this discussion on best practices, a high-level hierarchy of keywords is shown above. This list is just the major topic areas, and the order shown is not important. The user can enter the system at any location in the hierarchy.

The hierarchy is a way to organize the information in the databases so that it is easily accessible for editing and maintenance purposes. It is important that the keyword list be comprehensive and as complete as possible; however, the actual position or location of the topics in the hierarchy is not critical to the best-practices process. There are many links and connections between keywords at all levels in the hierarchy so that the interdependence between topics is maintained without regard to their physical position in the hierarchy.

Although the best practices system currently under development is aimed at the CFD user, it will be equally usable by technical managers or the engineers in S&C (or other discipline) to better understand the problems and difficulties associated with achieving good CFD results. It is conceivable that a future system which includes expert knowledge and experiences from S&C or other discipline could be equally useful to the CFD engineer. A future COMSAC capability could include a coupled system of best practices and expert knowledge from S&C, CFD, and other related disciplines.

A future BPX/COMSAC system could include an expert knowledge database with stability and control knowledge and prediction methods. This would be a way to preserve the corporate memory in stability and control for use by future generations of engineers. The advantages of having this information available for training purposes and for use in future designs are manifold. First, it is a way to reduce cost and risk on future programs by eliminating the mistakes that have already been made. Second, it provides a way to train the new engineers who may not have access to the senior engineers who did the original work. Finally, it will maintain the organization experience base as engineers retire or otherwise become unavailable to the technical discipline.

This chart illustrates a future integrated design and analysis system with several related disciplines shown for example. The computational aerodynamics discipline will be run by a best practices system described previously. Similarly, the stability and control discipline could have its own best practices system as suggested on the previous chart. The two disciplines could be linked by an overall best practices framework that will permit them to work together efficiently.

The complete system could be expanded incrementally to include the other disciplines to provide a multidisciplinary expert design system.

The approach of coupling expert knowledge and prediction capability has been demonstrated successfully by NEAR in other technical areas. LVX is a system for the aerodynamics of launch vehicles which couples expert knowledge, corporate memory, design experience, and aerodynamic prediction methods. RSX is a similar system based on the same computational framework for the aerodynamic design and analysis of rocket sled test vehicles.

Mendenhall, M. R. and Hegedus, M. C., “LVX – An Integrated Aerodynamic Design and Analysis Method,” ICAS 2002-0234, September 2002.

Mendenhall, M. R. and Hegedus, M. C., “An Engineering Analysis Tool for Rocket Sled Aerodynamics,” NEAR TR 582, February 2003.

This picture not only shows the first flight of an aircraft, it also shows the first flight test. The Wright brothers’ experience in this flight test has been repeated in almost every new aircraft program to this day. That is, the brothers discovered aerodynamic/handling qualities/flying qualities anomalies that were not predicted by ground based tests and calculations.

However, discovery of such problems in the flight test phase of an aircraft development is too late.

-Research is difficult and expensive, -Experiments are challenging, -Budget and Schedule pressures are high, -Configuration changes are costly, False color flow visualization of a generic transport model at static flight condition.

At 1 foot/sec in water, Reynolds number is ~100,000 per foot. Skin friction drag will be incorrect. Boundary layers will separate prematurely on curved surfaces. Tripping is not practical. Don’t use a water tunnel for designing airfoils or determining transport performance.

However, for many of the “interesting” parts of the envelope where non-linear or unsteady aerodynamic phenomena exist, even the full scale aircraft will be experiencing massive separation. Sharp edge configurations will be even less sensitive to Re. Vortex effects, pressure footprint, trajectory, and burst points are relatively insensitive to Reynolds number. Mach number will be ~zero. However, most aircraft will not be at high speeds for very long if maneuvering.

Made a few improvements to traditional water tunnel testing – moving beyond pretty pictures.

Added a computer-controlled model support, a submersible force balance, image analysis programs, and a computer to coordinate it all.

We think it is an ideal environment for addressing the issues of developing better predictive models. Not only as a primary research facility, but also in a supporting roll for development, validation, and verification of other predictive methods, such as CFD.

Here’s why...

Dynamic scaling of motion requires that the non-dimensional angular rates or frequencies are matched.

This example compares a full scale fighter type aircraft with a typical sub-scale models tested in a wind tunnel. The fighter, with a flight speed of 300 feet/sec, and a roll rate of 180 degrees/second will have a non-dimensional roll rate of 0.2.

To match this rate, a typical sub-scale model in a typical low speed wind tunnel will need to roll at 1800 degrees/second, or 10 times the full scale rate. Compare this to the same test conducted in a water tunnel. Typical model scales and test velocities result in the water tunnel model rotating at 18 degrees/second.

It is convenient to think of this as time going 10x faster than real time in a wind tunnel, and 1/10x in the water tunnel.

Water tunnels are renowned for outstanding flow visualization, usually at static conditions.

Here we have added a moderate amplitude motion. Simple single-axis harmonic motion.

Good, but yields limited information – questions are still unanswered – need more data – more experiments.

Time runs slower in the water tunnel test. Accelerations and inertia forces scale with time squared.

As a result, inertia forces developed by a maneuvering model are very small (~1%) compared to the aerodynamic forces of interest, and can be lumped into an error budget.

This is where it all comes together, the three pillars: Advantages: • Outstanding flow visualization • Force measurement • “Unlimited” motions possible • Self-consistent tests -- Assumption Checking • Separation of structural and aerodynamic frequencies • Simple dynamic tares Possible to test any motion or maneuver.

As a benefit of the stiff structure and low mass, the system is not limited to harmonic motions (sine, cosine). Here is a demonstration of pure angle of attack steps, by performing triangular wave z translations with increasing rates. The final reversal in this example executes a 40 degree AOA step.

AOA initial 15°, delta AOA = ±5°, ±10°, ±15°, ±20° In addition to standard pitch motions that have alphadot=pitch rate (q), pure alphadot, pure q, or any combination thereof can be performed. Likewise, betadot can be specified independent of r (yaw rate).

A recent test performed by AeroArts used the Scorpio data acquisition and image capture system to grab flow visualization photos concurrently with force and moment data. Custom image analysis algorithms automatically isolated the left and right LEX vortices, identified the vortex burst points as shown in this screen capture. The software recorded the images, the burst points, and plotted the burst point migration in real time as the test was being performed.

Multi-axis motion – a rolling pitch-up.

±30° body axis roll with a pitch ramp from 0° to 60°.

A falling leaf-type maneuver idealized for study from an in-flight incident.

The falling leaf incident began with a roll reversal at a low angle of attack and developed into a roll and yaw oscillation with a sinking flight path.

This motion requires rotations in all axes and coordinated translations in all three axes.

Here’s the tool. What can it do for you?

Adaptable Any motion (that doesn’t crash into something) is possible. Powered rotorcraft models have been run with full 3-axis swashplate control under computer control.

Intent of this talk is to present the S&C priorities as seen by the Langley team. No roadmaps or 5 year plans will be presented. We are actively soliciting your feedback, your ideas, and your help in building and executing this program.

A program like COMSAC will have to have a balance of generic and real configurations.

The list is our best shot of what we view as important while we understand that there are a myriad of candidates. If an issue that is important to you is not reflected in the list, let us know!

Form Approved REPORT DOCUMENTATION PAGE OMB No. 0704-0188 The public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number.

PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.

1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To) 04 - 2004 01- Conference Publication 4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER COMSAC: Computational Methods for Stability and Control 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6. AUTHOR(S) 5d. PROJECT NUMBER Fremaux, C. Michael and Hall, Robert M. (Compilers) 5e. TASK NUMBER 5f. WORK UNIT NUMBER 23-762-45-AF 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NUMBER NASA Langley Research Center Hampton, VA 23681-2199 L-18378B 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR'S ACRONYM(S) National Aeronautics and Space Administration NASA Washington, DC 20546-0001 11. SPONSOR/MONITOR'S REPORT NUMBER(S) NASA/CP-2004-213028/PT2 12. DISTRIBUTION/AVAILABILITY STATEMENT Unclassified - Unlimited Subject Category 08 Availability: NASA CASI (301) 621-0390 Distribution: Nonstandard 13. SUPPLEMENTARY NOTES An electronic version can be found at http://techreports.larc.nasa.gov/ltrs/ or http://ntrs.nasa.gov 14. ABSTRACT The unprecedented advances being made in computational fluid dynamic (CFD) technology have demonstrated the powerful capabilities of codes in applications to civil and military aircraft. Used in conjunction with wind-tunnel and flight investigations, many codes are now routinely used by designers in diverse applications such as aerodynamic performance predictions and propulsion integration. Typically, these codes are most reliable for attached, steady, and predominantly turbulent flows. As a result of increasing reliability and confidence in CFD, wind-tunnel testing for some new configurations has been substantially reduced in key areas, such as wing trade studies for mission performance guarantees. Interest is now growing in the application of computational methods to other critical design challenges. One of the most important disciplinary elements for civil and military aircraft is prediction of stability and control characteristics. CFD offers the potential for significantly increasing the basic understanding, prediction, and control of flow phenomena associated with requirements for satisfactory aircraft handling characteristics.

15. SUBJECT TERMS Stability and Control; Computational Fluid Dynamics; Stability Derivatives; Aerodynamics 19a. NAME OF RESPONSIBLE PERSON 18. NUMBER 17. LIMITATION OF 16. SECURITY CLASSIFICATION OF: OF ABSTRACT STI Help Desk (email: help@sti.nasa.gov) a. REPORT c. THIS PAGE b. ABSTRACT PAGES 19b. TELEPHONE NUMBER (Include area code) (301) 621-0390 U U U UU Standard Form 298 (Rev. 8-98) Prescribed by ANSI Std. Z39.18