cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Showing results for 
Search instead for 
Did you mean: 

Community Tip - New to the community? Learn how to post a question and get help from PTC and industry experts! X

Weibull fitting problem

Raiko
16-Pearl

Weibull fitting problem

Can anybody point me to my error. I can't get a decent fit for my data; see attached MC15 file.

 

Thanks in advance

 

Raiko

 

P.S. For some reason the website does not accept the MC15 file claiming that "...the contents of the attachement doesn't match its file type." henec, I've added a pdf.

1 ACCEPTED SOLUTION

Accepted Solutions

I am not sure about the Gaussian.

Here is a direct comparison side by side (worksheet attached):

Pic.png

 

View solution in original post

12 REPLIES 12
LucMeekes
23-Emerald III
(To:Raiko)

Attachments is a known problem that is being worked on. You can attach if you first put the worksheet in an archive (.zip).
Luc

OK Luc, here we go

 

Raiko

-MFra-
21-Topaz II
(To:Raiko)

Hi,

The data aren't distributed according to the Weibull distribution that you have traced in your graph (for s<=1). So you will never be able to adapt them to Weibull's distribution. The sample is small. How do you say it is distributed according to Weibull? With a larger sample it may turn out to be Gaussian.

weibull.jpg

Werner_E
24-Ruby V
(To:Raiko)

Maybe a scaling problem

Pic.png

mvenich
12-Amethyst
(To:Raiko)

Here's a PDF of what Maple gives you

wb.png

Mathcads weibull unfortunately is standardized to scale parameter (in German: charakteristische Lebensdauer) T=1.

To fit we need a function with two parameters, scale T and shape b.

Guess thats the way to achieve that in Mathcad:

Pic.png

 

 

Fred_Kohlhepp
23-Emerald I
(To:Raiko)

F M is (I believe) correct, This should be a normal fit; if you're measuring hardness that's what I'd expect.  You can get a normal fit as shown--the same basic method as Werner did for the Weibull.

I am not sure about the Gaussian.

Here is a direct comparison side by side (worksheet attached):

Pic.png

 

Raiko
16-Pearl
(To:Werner_E)

Hello Werner,

 

sorry for the late reply but I wasn't in office for a few days. Thanks a lot to all of you guys for your effort and thank you Werner for the solution.

 

Raiko

LucMeekes
23-Emerald III
(To:Raiko)

Your data is skewed, hence a Weibull distribution is more plausible than a normal distribution.

WeibullFit.png

Success!
Luc

Not sure that I agree.

 

Using the probability plotting of the Data Analysis Extension for both Normal and Weibull:

comparison of distribution fitscomparison of distribution fits

There is no realistic way to know what the distribution should be, and 25 data points is surely not going to allow you to determine it statistically (although I have not done any statistical calculations to prove that Smiley Wink). Maybe there is some systematic error that skews the hardness measurements, maybe not. If not, it's probably Gaussian. If there is such an error, it probably does not follow any statistical distribution. So what distribution to use in the fit likely depends on the end goal more than anything else.

Top Tags