Hi CFDIANS, why dont u try some machine learning algorithm for quick learning of new dynamics where there are too many input parameters. I dont work on these anymore but have a distinct feeling after going through the discussions that you could implement some neural network model than pure mathematical analysis. complexity can be reduced in estimation of the fluid dynamics if a fuzzy model is employed. get back to me if u have questions on the neural part and not if on the fluid part. i have used these in network analysis etc only. Bye Rajesh.S rajesh@au-kbc.org AU-KBC Research Centre www.au-kbc.org rajeshsweb@yahoo.com www.geocities.com/rajeshsweb/ hsejarm@yahoo.com www.geocities.com/hsejarm/ rajeshs17@rediffmail.com Rajesh Sundaram

Life, Nature, God, ... ask me not what they are. Ask me what Universe is not. If I answer it is part of this Universe.

Maybe you are right. As far as I know, somebody has used some kind of neural network model to detect the leakage of hydraulic (oil) pipelines. But as a stranger to neural network, I should say I wonder if it can work well when it comes to the mechanism of turbulence (especially to the process of transition), because machine learning is restricted to a lilmited number of samples, while the turbulence seems to be so different under different configurations.

Another queston is how to get the samples. Experiments can only measure the a few of the output parameters of turbulence at some discrete positions. Does it mean we must use the traditional CFD in order to get the input and output of a tubulence for the neural network?

Regards

Linfeng

I know people who have used fuzzy logic and Monte Carlo approaches to determine the impact of parameters uch as friction parameters and some turbulence effects (in the simplest case setting up a constant turbulent eddy viscosity) and propose the most likely solution for a particular flow situation based upon the "quality of the response" (I am not 100% sure that this is the right terminology, but basically they are looking at the evolution of the solution and evaluating its quality or efficiency).

This is particularly interesting for people looking at flood evaluation for example (usually 2D St Venant eqns) as it offers a probabilistic answer that insurers and decision makers can take into acount to evaluate the risk and give it a cost. One of the key advocate of his method is Prof. Keith Bevens at Lancaster University, UK, who devised the GLUE method for hydraulics and hydrology situations mostly. Another reason, which I should have quoted earlier maybe, is that at large scales there is so much uncertainty in the data (massflow, sinks and sources, surface level and roughness number, turbulence) that such an approach can prove more reliable in the certainty (the trust) one can place in a solution. This might be different for some much controlled CFD in the mechanical industry for example, but find an interest in chemical calculations (e.g. where PDF can be/are used).

Dear Rajesh, I don not know what you are talking about.But It sounds intersteing to me. Can you give sme more details. Please prasat

[CFD Lab]

Good day, I am from a CFD background. Want complete training that How I can implement Artificial intelligence techniques in Computational Fluid Dynamics. Looking for training on it. If you can provide it will be highly appreciated hamza.mec.edu@gmail.com