System and method for modeling the flow performance features of an object

• US 5,649,064 A
• Filed: 05/19/1995
• Issued: 07/15/1997
• Est. Priority Date: 05/19/1995
• Status: Expired due to Fees

• Current Assignee: Administrator of The National Aeronautics and Space Administration
• Original Assignee: United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration
• Inventors: Ross, James, Jorgensen, Charles
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): SMITH, JEFFREY S
• Application Number:
  US08/446,071
• Time in Patent Office: 788 Days
• Field of Search: 395/21, 395/22, 395/23, 395/24, 395/902, 395/905, 395/906, 395/907, 364/148, 364/149, 364/152, 364/578, 73/147
• US Class Current: 706/17
• CPC Class Codes: G06F 30/15 Vehicle, aircraft or watercraft design
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30
• CPC Class Codes: Y02T 90/00 Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
• CPC Class Codes: Y10S 706/905 Vehicle or aerospace

Abstract

The method and apparatus includes a neural network for generating a model of an object in a wind tunnel from performance data on the object. The network is trained from test input signals (e.g., leading edge flap position, trailing edge flap position, angle of attack, and other geometric configurations, and power settings) and test output signals (e.g., lift, drag, pitching moment, or other performance features). In one embodiment, the neural network training method employs a modified Levenberg-Marquardt optimization technique. The model can be generated "real time" as wind tunnel testing proceeds. Once trained, the model is used to estimate performance features associated with the aircraft given geometric configuration and/or power setting input. The invention can also be applied in other similar static flow modeling applications in aerodynamics, hydrodynamics, fluid dynamics, and other such disciplines. For example, the static testing of cars, sails, and foils, propellers, keels, rudders, turbines, fins, and the like, in a wind tunnel, water trough, or other flowing medium.

Claims

1. A method for generating a model of one or more flow performance features of an object in a flowing medium having a plurality of geometric configurations, the method comprising the steps of:

(a) forming a neural network comprising;

a plurality of input nodes associated with the geometric configurations of an object,

a plurality of hidden nodes,

one or more output nodes associated with the flow performance features of the object, and

a weight matrix for defining connections between the input nodes and the hidden nodes, connections between the hidden nodes and the output nodes, and weight values associated with each connection,

(b) forming a test database, coupled to the neural network, comprising a plurality of test input signals and corresponding test output signals, the test input signals associated with the geometric configurations of the object and the test output signals associated with the flow performance features of the object, the test input signals associated with the input nodes and the test output signals associated with the output nodes, said weight values being modified as a function of said test input signals;

(c) solving for a search direction to obtain one of said flow performance features comprising the steps of;

(i) initializing a parameter vector representative of a weight value of said one of said flow performance features;

(ii) selecting a random predictor value for said one of said flow performance features and having an error cost function with a gradient; and

(iii) iterating the parameter vector and random predictor value to determine said search direction, said determination being satisfied when qualifying one of the conditions selected from the group comprising;

(1) the gradient of the error cost function is near zero, (2) the error cost function is below a predetermined value, and (3) a predetermined number of iterating of the parameter vector and random predictor value has been accomplished.

2. The method according to claim 1, wherein said error cost function is a mean square error cost function.

3. The method according to claim 2, wherein said mean square error cost function is based on all of the parameter vectors and random predictor values used for said iterating step.

4. The method according to claim 3, wherein said mean square error cost function is determined by an average generalization error technique.

5. The method according to claim 4, wherein said average generalization is established by a Akaike final prediction error (FPE) estimate.

6. The method according to claim 1, wherein said error cost function is a relative absolute error function.

7. The method according to claim 1, wherein said output nodes is selected to be of a number which exceeds the number of selected weight values.

8. The method according to claim 1, wherein the geometric configurations are selected from the group comprising leading edge flap, trailing edge flap, angle of attack, flap positions, slat positions, deflection angles, elevator deflections, rudder deflections, stabilizer position, roll, pitch, and yaw positions, angle of attack, mach number, and velocity.

9. The method according to claim 1, wherein the hidden nodes produce a tangent hyperbolic activation function signal.

10. The method according to claim 1, wherein the output nodes produce linear output function signals.

11. The method according to claim 1, wherein the flow performance features are selected from the group comprising of lift, drag, lift/drag ratio, side-force, pitching moment, yawing moment, rolling moment, and maximum lift.

12. The method according to claim 1, wherein a set of paired test input signals and corresponding test output signals are generated from wind tunnel testing of the object.

13. The method according to claim 1, wherein said iterating step is accomplished by a Levenberg-Marquard technique.

14. The method according to claim 1, wherein said iterating step is accomplished by a back-propagation technique.

15. The method according to claim 1, wherein said random predictor value is chosen from a Gaussian distribution.