PDA

View Full Version : Monte Carlo Simulation Needed



joeb33050
01-05-2006, 08:30 AM
I'm working on powder measure accuracy for the book, have tested some and have got tests from other guys. Wrote to RCBS, LEE and Redding asking to borrow measures to test.
I'm at the point where I need to define what the accuracy means, so I need a Monte Carlo Simulation, 2 really.
For a normal distribution with mean and S.D. = 1, sample 5 at random and calculate max-min = difference. Repeat 1000X. Need the 1000 differences or the distribution.
Then, sample 10 at random.
If you've got the software up and running, please?
I don't want anyone to have to write code.
Thanks;
joe b.

Buckshot
01-06-2006, 03:08 AM
.............Joe, maybe 6 years ago Precision Shooting magazine had an article about powder measure accuracy. I sure don't recall the testing perameters, but I do remember they tested measures from the then fairly new Lee Perfect Powder Measure, to the Culver. IIRC there were several.

In this case he found all the measures to be admirably accurate, if opperated consistantly. The plastic cheap Lee was no better or worse then any other he tested. The price of the unit didn't have any bearing on accuracy, just possibly longevity.

..............Buckshot

Maven
01-06-2006, 12:57 PM
joeb, Silly me: I thought you were asking about a gun stock! Please explain what you mean by a Monte Carlo Simulation and what software, if any is required to perform it. Thanks!

StarMetal
01-06-2006, 01:01 PM
Maven


http://www.decisioneering.com/monte-carlo-simulation.html

Joe

wills
01-06-2006, 01:52 PM
I dont know about monte carlo, but you might get carpetman into a game of dominoes.

Maven
01-06-2006, 02:07 PM
Starmetal, I looked at the site and read its contents, but still haven't a clue about how to perform the simulation. Coincidentally, I got a chance (meaning I got off my keister) to test a "new" Redding powder measure I purchased along wi. a Rockchucker press & Redding powder scale (made by Pelouze; in the original box) last month. The powder measure has a crude micrometer dial on the bottom of the operating mechanism. I tested it with the micrometer set at "20" rotations/clicks for the following powders: IMR 4198, IMR 4064, IMR 4350 and IMR 7383. I threw 10 -15 charges/powder to "set" the measure and then 20 charges/powder to establish the mean and SD, particularly the latter. The results were impressive in that 4198, 4064, and 7383 all gave a SD of.2gr. [Means were 27.4 for 4198; 27.6 for 4064, and 25.3 for 7383.] Only 4350 showed a slightly higher SD, namely .3gr. wi. a Mean of 29.2.

Just for grins, I then set the measure to "30" rotations/clicks, but tested IMR 7383 only as it was too tedious to retest, record and recalculate the other powders. Again, I was more interested in the SD than the mean. This time I got the same .2gr. variation (SD) with a Mean of 36.6gr.

What's the coincidence? After I entered the raw data and calculated the means & SD's, I went to this site and saw joeb's post.

joeb, if you read this, I still have the raw data. Let me know if they're of any use to you.

scrapcan
01-06-2006, 03:34 PM
Hey guys Joe is talking about running a set of Monte Carlo Simulations to address what can be expected from a powder measure. The idea is to do a very large simulation and based on the parameters for the simulations you can veryify the model fit based on a much smaller actual dataset.

I think the real idea is not to wear out the tester or the equipment just testing it.

One common use of Monte Carlo simulation is risk assessment in environmental arena. Example is modelling exposure to a toxin or chemical and predicting health or ecological effect, thus the term risk assessment.

Some people are wiling to be the guinea pig, but if you miss the indicators your pool of willing testers has just been decreased. Then you are looking for more testers. Simulation or modelling precludes the use of "real data" excep to verify the simulation or model.

Other than the above I do not know the actual mechanics, err mathematics, to do Monte Carlo Simulation.

My suggestion would be to contact your nearest university and stop in. You would be surprised at how many shooters are in the academic world, despite what most people think. Not all professors are on the side of gun control.

Jeremy

grumble
01-06-2006, 04:11 PM
Google returned about 2 million hits on "Monte Carlo Simulation."

Shouldn't be too hard to learn more for those really interested. I remember doing some of that stuff in a statistics class, but forgot how as soon as I possibly could. That was in the pre-personal computer days, and everything had to be calculated by hand. Ug.

StarMetal
01-06-2006, 04:24 PM
Paul,

That Redding powder scale you spoke of, is it old? I have an old Redding powder scale that is about..hhhhhmmmm...30 almost 40 years old. I just retired it last year and got a new RCBS. I'm kind of leary of the new scale if you know what I mean. Got so use to that old Redding.

As far as powder measure I don't think I'll ever use anything but my Bellding & Mull. I toyed with the idea of getting a new Redding but couldn't bring myself to do it. I just rebuilt my Belding & Mull. Carpetman will tell you too how good they are.

Joe

scrapcan
01-06-2006, 07:08 PM
Grumble,

I too had the pleasure of wanting to forget ASAP. I was successful. It would be easier to find someone who thinks they have to know how this is done and that does it on a regular basis. Also let them do the analysis.

I recall that it can be done in excel. don't remeber all the details (remember it is a forget or you have to put other useless tuff in the trash bin). Here is a link that may spark someones interest.

http://office.microsoft.com/en-us/assistance/HA011118931033.aspx

joeb33050
01-07-2006, 07:04 AM
What's the coincidence? After I entered the raw data and calculated the means & SD's, I went to this site and saw joeb's post.

joeb, if you read this, I still have the raw data. Let me know if they're of any use to you.

Maven;
I have a test procedure measuring 4 elements of accuracy of powder measure/powder combinations, and need volunteers to measure accuracy this way. In particular, we need a minimum of 30 measurements of each variable set of circumstances.
We have standard deviations of repeatability from 0 = ZERO to .15 grain.
All the results are in an Excel spreadsheet in Files on the CBA forum.
I'm looing for volunteers!! A test of a given powder/measure combo takes me a little over an hour.
Thanks;
joe b.

joeb33050
01-07-2006, 07:24 AM
joeb, Silly me: I thought you were asking about a gun stock! Please explain what you mean by a Monte Carlo Simulation and what software, if any is required to perform it. Thanks!

Let's say we have a measure/powder, EX Lyman 55/AA#9 combination that repeats with standard deviation of .15 gr. Then +3SD = +.45 grain, and - 3 SD = - .45 grain, and the difference = .9 grains. In my 30/30 load of 12.5 grains, .9 grains is a lot. BUT, a difference of +/- 3 SD is going to happen very seldom.
So, say we make a normal distribution with mean = ?10? and SD = ?1?, why not, and from it we select 5 numbers, weights, at random. Most of the time the difference, high to low, will be small. If we select 5 samples many times and calculate the difference = hi-low, we'll get a distribution of differences. We can easily fiddle with that and get to where we can make statements such as:
"With a SD of .15 grain;
40% of the time hi-low = 0
60% of the time hi-low = .1 grain or less
75% of the time hi-low = .2 grain or less
85% of the time hi-low = .3 grain or less
90% of the time hi-low = .4 grain or less
.
.
99% of the time hi-low = .6 grain or less
All this for N=5, a 5 shot group.

But we don't know the distribution, is why I need somebody to do the Monte Carlo simulation = plugging in the mean and SD and distribution and selecting the set of ?1000? samples of 5.
If you've got the program and use it, it's easy to do. I don't have it.
Somebody out there does.
This is like ANSI Y 1415M, that changed how we dimensioned drawings. We finally realized that a round pin of 1" +/- .005" will go in a round hole 1.005" +/-.005" almost all the time, if .005" = 3 SD. Everything we knew about tolerance stack up went out the window.
Enough;
Thanks;
joe b.

joeb33050
01-13-2006, 07:17 AM
Jeroen Hogema of the Netherlands came through again, I've got it and am all set.
Thanks;
joe b.