Several Parameters

Itterative procedures are fine...especially if you program your computer to do it...

But seeing as the average PC no longer has Quick Basic on it and many languages are bogged down with cumbersome GUI front ends this isn't such an easy proces...

Whoops, sorry, this is called progress...I must mail mr Gates anther cheque asap.

Del

THank you Del the cat for your reply.

But while working on any parameter the way is: I analyse the heat storage tank via CFD program. then I use the result of CFD Prog. (a temperature at a node) then I calculate the efficiency with the information that I get from CFD Prog. (the temperature)

So it is hard to make iteration I think (iteration codes must contain the CFD Program study). Or I can't think like you. Could you explain your way a bit more please?

I can't help with any specifics as I'm not familiar with your CFD program...but I'd have thought a clever program would be able to allow you to input results from one run straight back into another run. Or at least it should have a simple way of saving results as a file and reading those data back as inputs to the next test...

Failing that...try and see which are the most significant variables and concentrate on them, simplify by eliminating the least significant variables.

Good luck...

Del

You are righ Del the cat. If I can write a program containing the numerical analysis of the (Navier Stokes Equations) heat storage tank, then iteration could be used. But what I do; contain manual things: I manually take some result value from the result of software package program analyss and then use it in a formulation.

I wanted to do the analysis in numerical way but I do not want to get lost in differentual equations and I have lack of time. :(

A very good question. There are mathematicians who specialize in this sort of thing and you might ask if you know one, but your case seems simple enough.

I presume you are doing this on a computer and not manually. Generally the "straight-through" approach works OK for most optimizations, but the thing that will catch you up is local maxima/minima. So, some people will go through a second time, varying each optimization parameter enough to see that it is robust. If any turn out to change significantly, go through again. Three or four passes should do for a nice "seat of the pants" design. This WILL NOT catch problems where two or more parameters significantly interact.

Good luck.

you are right TVP45 I am doind this on a computer. But also I use computer there is manual study on my work.

Thank you TVP that you explain the way I got curious about. But which one is better or suitable for my situation I don't know. For me finding the local maxima would be enaugh for my design, but for my teacher analyzing combination of parameters (81 analysis :( ) must be done?

So TVP45: you mean that the two options are the solution ways. First one gives course result second one gives as you named "seat of the pants" result. So I can force my teacher for the first solution way for my thesis??

Oh, and by the way, I remembered another way to check your work. Just run the same values again, but vary the order. If you get the same results, your optimizations were pretty independent.

Sorry I couldn't understand you well at this subject. Would you please inform me more about this? It sounds good for me.

OK, I'll give a very simple analogy. This isn't a very exact analogy, but it'll give you an idea of what I mean.

This shows how to go through several times:

Suppose you had a piece of property and you wanted to find the lowest spot. You could start out walking downhill in a North-South direction till you got to the lowest point that way. Then you could walk downhill in the East-West direction till you got to the lowest spot that way. Then you walk downhill N-S again. If you don't go far, you're at a good solution; if you do go far, continue doing E-W, N-S, E-W, etc until you settle near one point.

But, you might have just found a small depression. To check for that, reverse the order, going back to the original starting point and this time walking downhill in the E-W direction, then N-S, etc. If you wind up at the same spot as before, you've got a good solution. If not, the parameters are not independent.

Again, there is a very precise mathematical process for optimization that is better than my simple methods and you might ask about that. But, for many optimizations (for example minimum weight beam design), my rough-and-ready way works well.

In a multi-variable optimization there is a way called "of the most important slope". I only mention it since it is possible it will not to work in your case.

You start from a parameter combination chosen by a logical qualitative analysis, maintain one parameter constant and vary all other one by one all the rest keeping their initial value, you consider the parameter which gives the most important slope in the direction you want to go. You make a step in this direction with the old parameter values and a new value for the most active parameter. You repeat the cycle.

It is iterative and accepts automation which makes it apt for programming.

If you start from an other starting combination and get same final result then your solution is a general optimum and not a local.

thanks for you all.

I define a way to solve my problem (design the storage tank)

I first examine the first parameter for 5 values then for 3 of the first values I examine the second parameter for 5 values then. It is a time burglar way I know but my professor is little crazy: if I do not do what she wants my studies for master education would be trashed.

But as I guessed before I started this discussion here: examining one parameter and then choosing the best solution and then according to the solution of first parameter (I fix the solution) I examine the second parameter and find a result. And then fix the solution of second parameter.

Because when I look at my graphs, it is clearly understood that solution of fisrt parameter is always efficient solution for all values of second parameter according to the solution of other first parameter for all values of second parameter.

Again thank you friends as you were with me when I'm in confusion.