Stress-Life (S-N) Fatigue Data Analysis 4

[mjmcondor]

Hello everyone. I have some questions pertaining to analysis of fatigue data. To be honest I haven't done this sort of thing since college and I've spent countless hours researching the topic online. I've also been asking around at work and can't get a consistent response so I've decided to consult the forums.

[li]How should I perform the analysis?[/li]I've used MMPDS (ver. 07) and ASTM E739. A colleague said they use TableCurve to fit an equation to their S-N data. Which way is the proper way?

[li]Is there a standard equation used to fit fatigue data?[/li]MMPDS and ASTM both use a log base 10 approach: [tt]log(N) = A + B * log(S)[/tt]. MMPDS allows for additional parameters than [tt]A[/tt] and [tt]B[/tt] (and actually uses [tt]A[/tt]₁ and [tt]A[/tt]₂, not [tt]A[/tt] and [tt]B[/tt]) to account for different stress ratios and non-linearity. A colleage has used [tt]S[/tt]² [tt]= A + B(N)[/tt]^-0.5 and [tt]S = A + B(N)[/tt]^-0.5 based upon the fit from TableCurve.

[li]How should I be plotting my data?[/li]Should I be using log-linear or log-log plots? MMPDS and ASTM both use log-log, but my colleagues all suggest log-linear.

I've already searched the forum for similar topics but didn't find much. This was the most relevant one but still didn't help much.
[URL unfurl="true"]http://www.eng-tips.com/viewthread.cfm?qid=253671[/url]

Any help is greatly appreciated.

 

[mjmcondor]

MMPDS stands for Metallic Materials Properties Development and Standardization handbook. It was formerly known as MIL-HDBK-5 (or Mil-Handbook-5).

 

[dwallace1971]

Get yourself a copy of "Fatigue of Structures and Materials" by Schijve and read it.

"On the human scale, the laws of Newtonian Physics are non-negotiable"

 

[EdStainless]

If this is for aerospace you had better use the methodology of MMPDS.
B-Basic ( and other design data) data is derived by a very specific method.
If this is anything else then read up on fatigue theory and do it however fits your needs.

= = = = = = = = = = = = = = = = = = = =
Plymouth Tube

 

[mjmcondor]

Thank you dwallace1971 and EdStainless for your help. This analysis is aerospace related.

I spoke with a representative from MMPDS and they said their methodology is more complex than ASTM's. MMPDS accounts for multiple stress ratios, non-linearity, non-uniform variance, outliers, model assessment, combining different datasets and runouts. Most of this doesn't apply to my data which is why I created this post.

I've done my analysis according to MMPDS because it was suggested to me by others at work, but I was never told the why behind it. I think I understand the reasoning now: aerospace application and a more thorough methodology.

I will get a copy of "Fatigue of Structures and Materials" to review.

 

[Dougt115]

MMPDS yes.
Log linear plot yes.

You said you are analyzing fatigue data, that implies that you have the data already? If so you only need to fit the lines to the data.

If this is the case you should be more concerned with how the failure occurred. MMPDS may be able to give you some insight into the failure modes or how to test in order to validate or eliminate suspected failure modes.

MMPDS test methods can also help you evaluate your test setup, that you are not under testing it. Depending on the complexity of the application of the part it could be that your testing is not insightful enough, MMPDS will help with this.

 

[jagad5]

I highly recommend that you review chapter 9 of MMPDS and try to find a copy of Analysis and Representation of Fatigue data by Conway and Sjodahl. Those two sources are by far the most complete, in my opinion.

The standard equation in MMPDS is log(N) = A + B*Log(Seq - C) models only the typical behavior. You can find A and B using least squares and you find C with the Solver in Excel; Conway's description of how the stats is very good. Seq might be Smax (if you have one R-Ratio) or Smax*(1-R)^n or there is one other that I can't recall, and have never used. Seq can be any mean stress corrected equivalent so you can use Goodman, Smith-Watson-Topper, or others to collapse multiple R-Ratio data to one line.

To get the design curve, I recommend (and MMPDS will soon include) the following two approaches:
1. Compute the standard error of the estimate, notated as see below (STEXY function in Excel) and compute the one-sided lower tolerance bound using the factors in table 9.6.4.1. The design curve equation becomes log(N) = A + B*Log(Seq - C) - kA * see. This estimates the 99/95 design curve. You need a tolerance bound, not a confidence interval.
2. Compute the typical curve and then divide life by a scatter factor. Some aerospace product lines swear by this approach, using scatter factor of 4.0

If you want to get fancy, research the Random Fatigue Limit model developed by Pascual and Meeker in the late 1990's. It handles several flaws in the method outlined above. Runouts are handled with statistical rigor (correctly) and it estimates the scatter of the endurance limit of the specimens tested. See Technometrics, 1999, volume 41, No.4

I don't think Schijve's book discusses this topic. But Fatigue Testing and Analysis by Lee, Pan, Hathaway and Barkey covers this and several other uncommon topics very well. The only thing wrong with that book is an error filled index.

I participate in the twice annual MMPDS Industrial Steering Group. See

https://projects.battelle.org/mmpds/index.htm

for more info.

Doug