Using neural networks for CDF response surfaces
 

Hi everybody. Here is my problem: Two disciplines are connected together by two common variables and each one has different number of variables these are my sub systems, in system level I have a function, which should be minimized. I want to use collaborative optimization method for this ultidisciplinary problem. Each of two subsystems has a response table with about 100s of CFD runs. To learn these response surfaces I want to use a neural network with back propagation or genetic algorithm. In each iteration, system level optimizer will give a suggestion for common variables I will go to subsystem with these constants and I will try to minimize the subsystem Function. For optimizing these subsystem I want to use simulated annealing or Genetic algorithm (I don't want to use a gradient base method which needs derivatives) They will ask for value of function in different points these values will come back from Already trained neural networks. Now if anybody has some ideas or advise about this scenario please let me know.I have never used neural networks before. any idea may save time. thank you.

Re: Using neural networks for CDF response surfaces
 

(1). S. Pierret,"Turbomachinery Blade Design Using a Navier-Stokes Solver and Artificial Neural Network" Journey of Turbomachinery,p-326/Vol. 121, April 1999. (2). There are several problems: I am not familiar with neural network, fluid dynamics problems are very complex, parametric models need to be created to simulate the fluid dynamic problem which may not be the best choice.

Re: Using neural networks for CDF response surfaces
 

a lot of people have used neural networks for optimisation. the reference John gave is good. also there is a company called Engineous software(www.engineous.com) that produces a general optimisation package called iSight. they've published articles using GA and NN for aero optimisation and someone there can give you references to their papers. perhaps the jornals like 'Engineering Optimization' or SIAM's optimization journals have references you can use. if you are at a university you'll find these kinds of references in your engineering or math/comp. sci. libraries

Re: Using neural networks for CFD response surfaces
 

Search the online database of AIAA paper by the keyword.

http://www2.aiaa.org/citations/mp-search.cfm

Z. Lei