Fatigue Doubt 1

In conventional fatigue analysis there is no difference between crack initiation and fatigue failure. Given the millions of cycle to get a crack to initiate, discerning the few thousand cycles of crack growth is irrelevant and supposes way too much precision in the analysis (I mean safe life is fatigue life /5).

You're reading words and trying to put precise meaning to them. Fatigue is fatigue, yes you can call it "crack initiation" but that is looking at "fatigue life" from a damage tolerance perspective. Before damage tolerance, fatigue was life, not crack initiation life.

Read (or calculate) fatigue life from an S/N curve; calculate crack growth using DTA.

another day in paradise, or is paradise one day closer ?

You might consider the distinction to be the size at which a crack can be detected by various types of inspection methods. This means the "initiation life" will be a function of the type of inspection and how reliably it can be detected with the given method(s) (also a function of equipment, personal, etc.). So while there may be an approximate size for the crack initiation threshold size, it is not a universal number. After the "initiation life" (and assuming there were no initial flaws), you can consider DT as you stated, which requires that cracks be detectable (and where the threshold for detectability has been determined).

The objective is that the "initiation life" or "fatigue life", as sometimes called in the aircraft industry, be sufficient such the structure is durable and does not require frequent repair (not a safety concern at this point, but rather an economic concern). DT is a safety concern, which ensures that you detect and address cracks before they reach a critical threshold (obviously this requires a crack to be detectable before this can have meaning and hence the use of an "initiation life"). Note that DT can also be used to address manufacturing flaws as well (which may have detectable defects from day one and does not just at the end of the "initiation life").

Brian

OK, so in my opinion there are a lot of misnomers and confusion that gets perpetuated regarding this temrinology.

In the United States at least, this is in part due to how the continued airworthiness requirements per the FAA have evolved over time. Particularly FAR 25.571. Back in the day, when things were based only on safe-life criteria, there was no account taken for potentially rogue flaws and everything was just referred to as "fatigue". The remnant of this still exists in 25.571 (c).

However, we have evolved from a safe-life approach to a fail-safe approach, and finally to the current standard of "damage tolerant design" This basically just means we consider the potential for macroscopic initial cracks and use fracture mechanics principles to account for widespread fatigue damage (WFD) and multi-element damage (MED). We still use safe life approach for things where inspection methods or time between detectable and critical could be very small, like landing gear.

Now first misnomer - DTA is not synonymous with crack growth. Notice the language of 25.571(b), the DTA requirement includes determinations for damage due to "fatigue, corrosion, and accidental damage".

The bottom line:
1. DTA is a damage tolerance assessment.

2. Fatigue is defined as the accumulation of damage due to cyclic or repeated load. Period. What do you think is happening when "they" do a rotating beam test on a specimen? You have mircoscopic inclusions, grain boundary defects, etc which are growing. This is crack growth. All damage that results from cyclic load is crack growth. There are flaws that propagate in the structure and lead to failure. All damage that results from cyclic loading is fatigue, by definition. They are synonymous. This is why you usually see the terms "fatigue crack initiation" for the early stage and "fatigue crack growth". These in my opinion, are more correct terms.

3. The real distinction is how we deal with analysis to develop our continued airworthiness inspection programs. On one hand there is a requirement to check our inspection thresholds based on statistical fatigue data. That is, S-N curve data which has been statistically generated via test. The REAL distinction is that these specimen had no initially induced macroscopic flaw. That is, the life is from "pristine" condition, to fracture. [fracture is simply defined as the separation of a piece of material into parts]. So yes, you are correct they do not stop the test when they can detect a crack. The point is, if there is no initial "rogue" flaw, the total life is dominated by the time spent growing a grain-level flaw to the point where is is large enough to grow quickly. When people say fatigue is the number of cycles to get to the point of crack initiation, this is a bit of a misnomer as well. Remember, a crack is a crack, whether it is microscopic or not.

In reality, the period you refer to is the time taken for flaws in a structure to reach Region II of the da/dN vs. deltaK curve (and the flaw may or may not be detectable at that point. At this point, the growth happens much more quickly. "Fatigue" is not at all defined as this life. Nor is "damage tolerance" defined as the remaining life. The total statistical fatigue life, with a scatter factor applied, is one criteria for the inspection threshold.

4. Another criteria for the threshold, and the way we develop a repeat interval, is by using linear elastic fracture mechanics to determine the fatigue crack growth life of a piece of structure with an assumed initial macroscopic rogue flaw. So we more or less skip region I of the da/dN vs. deltaK curve (but no scatter factor here). We take to total fatigue life of this assumed initial configuration and apply factors to develop another potential threshold. Then we look at the detectable crack length we have and figure out the amount of time between detectable and critical. [Side note - critical is not always fracture, it might be NSY or some other criteria].

So in short, people like to say "fatigue" when referring to S-N data, and "crack growth" or "damage tolerance" when referring to LEFM. But in reality, they are the same principles, but different approaches. One approach is based on statistical data with no induced flaws. One is based on statistical data with an assumed flaw.

I think the confusion stems from the fact that the FAR used to only require a "fatigue" analysis, and this language was retained when another paragraph was added to include what we call damage tolerant design, which includes the account of rogue flaws.

All you really need to remember is that in cyclic load ie fatigue, structures (especially assembled structures) which are normally ductile, can eventually behave in a brittle manner. We want to encapsulate this and there are multiple approaches. One way is to simply use test data & statistics. Another way to is assume we have a flaw and treat it with the principles of FM.

ESPcomposites: "Initiation life" is NOT a function of inspection type, and a flaw may not be detectable before it reaches Region II.

Keep em' Flying
//Fight Corrosion!

Thank you rb1957, ESPcomposites and LiftDivergence!

Your explanation clear my doubt and I clearly understand it is two different methods used for different purposes. And yes for the DT, we used it for inspection purpose.

Another question which bothering me a while is Notched and Smooth in LCF. I understand from previous work that Notched Life Curve gives more life cycles than Smooth Life Curve because Notched area shows less debit and locally plastic stress distribution.

However, what I learned from the lecture is that S-N curve for Notched Specimen always show lesser life cycles than Unnotched Specimen.I wonder if it is terminology again?

Thank you everyone!

I cannot see a situation where a notched S/N curve will give a higher life than an unnotched curve. The unnotched specimen is the very best the material can give, and "must" give a life better than "any" notch geometry. Now there are things you can do to the specimen to improve the life, eg a shotpeened specimen will/should give a higher life than an un-shotpeened specimen notched and/or unnotched.

Can you share this "previous work" that led to your understanding "Notched Life Curve gives more life cycles than Smooth Life Curve".


another day in paradise, or is paradise one day closer ?

Hi rb1957, I don't think I can share the document but the content is pretty much like notched life curves are usually higher than smooth because it subjects to less defects debits.

seems very odd … a smooth polished specimen having a shorter life than a specimen with a (any?) Kt.

I always understand Kt is a measure of the local stress concentration with respect to the far field stress.
Higher Kt = higher stress = shorter life.

another day in paradise, or is paradise one day closer ?

Lee.Conti

First, remember that just because we see the term "notch" does not mean we are referring to a Kt. Kt is used a lot in industry (even in MMPDS, etc) so it can be confusing. Kt is generally a static parameter. For fatigue we normally have Kf (although as mentioned, sometimes Kt is used in fatigue data references).

Kf is related to fatigue through the formula Kf = 1 + q * (Kt - 1)

Where q is the notch sensitivity of the material. Make sure you always know when data is referring to Kf and when it is actually referring to Kt.

Second, I think your confusion may be arising from your application of the data.

I suspect you have some structural detail for which you have calculated a Kt and you are trying to find a statistical fatigue life. You have done this and noticed the life based on the notched data curve gives you more life than that based on the unnotched data curve.

To understand this, just think about your application. You have a Kt. Because your structure has a "notch". So does it make sense to look at the unnotched data? Not really.

The stress concentration at the notch is highly localized. When you enforce a Kt by calculating a life based on notched data, these local affects are accounted for.

But how does your calculation work if you use the unnotched data? You are still factoring your input stress by the Kt but with the unnotched data there is no local affect. So are you assuming the stress everywhere in the part is multiplied by Kt?

Basically, I'm asking you how the calculations in you document work.

Keep em' Flying
//Fight Corrosion!

rb1957,

I have the same thought about Kt. Hope to find out by asking some questions here.

LiftDivergence,

The Fatigue Life tool is an in-house tool from one of the OEM(s). The LCF life prediction is done by entering max/ min stress, strain range, mean stress, temperature and stress concentration factor (kt).

For instance, if I want to look at LCF where the peak stress at fillet, Kt factor is calculated at the peak stress location. It is not necessary the Kt at the fillet must use notched life curve, only if the Kt exceeds 1.75 (arbitrary), Notched life curve is used.

There is a gap in between smooth life curve and notched life curve. Smooth Life Curve is used if Kt <= 1.25 (arbitrary number) and Notched Life Curve is used if Kt => 1.75 (arbitrary number). So, there is a gap between 1.25 - 1.75. How do we get the life for where the Kt falls in this range? An interpolation (not linear) formula given to calculate the life at the Kt.

"How do we get the life for where the Kt falls in this range?" … Ask you people, I'd assume Kt = 1.75 covers 1.25<Kt<1.75.

I'd still question a smooth (Kt = 1) specimen having a fatigue life being less than an equivalent notched specimen.
That doesn't pass the smell test ...


another day in paradise, or is paradise one day closer ?

rb1957, I am thinking if it is something about the nominal stress?

if the nominal stress is really low, such that fatigue really isn't a problem ? if so your methodology should say "caution with stress levels below Xpsi" or truncate life to 10^8 cycles.

another day in paradise, or is paradise one day closer ?

rb1957, I contacted with my ex-colleague... to run a few run based on same temperature and max/min stresses but different Kt and Life Curve (smooth/ notched).

Both smooth/ notched curves show decreasing life cycles from Kt = 3.0,2.0,1.5,1.0....

But Notched curve show higher life cycles compare to Smooth curve regardless of which life curve should be (note that smooth curve is defined for lower Kt and notched curve is defined for higher Kt).

do you mean that a Kt of 3 had more life than a Kt of 1 ??

this is a very "odd" situation.

what material are we talking about ? metal ? non-metal ? other ??

another day in paradise, or is paradise one day closer ?

I mean the life decreases when Kt increases for both Notched and Smooth Life Curve. However, Notched Curve shows higher life than Smooth Curve at the same Kt. I am referring to metal material.

Sorry, that makes no sense to me.

Ok, Life decreases as Kt increases, as it is expected to.

But …
1) how can the notched S/N curve show a higher life than the unnotched specimen (if overall life is less) ??
2) what do you mean notched and smooth (unnotched) at the same Kt ? Unnotched is Kt = 1 ??

I am assuming that "smooth" is "unnotched".

Or are you talking about surface roughness ??

Can you show a couple examples ? leave off material, you can leave off the units (but leave the gridlines)

another day in paradise, or is paradise one day closer ?

I was also struggling to get it ... (I am checking with the people from OEM to find out)

Below are the input used for the fatigue life prediction:

Max/Min Stress Input
1. Max/ Min Stress
2. Max/ Min Temp
3. Max/ Min Kt

Material Option:
1. Life Curve: Smooth or Notched
2. Process Debit: Surface Finishing, etc

Few iterations are compared as below:

Material Option: Smooth Curve
Max/ Min Stress Input: Kt = 1.0
Kt = 2.0
Kt = 3.0

Material Option: Notched Curve
Max/ Min Stress Input: Kt = 1.0
Kt = 2.0
Kt = 3.0

Note other input remains same for Smooth and Notch curves. Both Smooth curve and Notched curve show life is decreasing when Kt is increasing. However, if comparing Notched Curve to Smooth Curve at the same Kt, Notch Curve shows higher life.

ok, it looks like you're extrapolating smooth life from a notched S/N curve …

ok, now the issue is clear ! extrapolating Kt = 1 life from notched data is IMO fraught with problems. Extrapolating Kt is always difficult to get good results, but you can see that extrapolating Kt = 5 to Kt = 6 is one thing (only a change in magnitude of an established response) BUT extrapolating Kt = 1 from Kt = 2 is such a large change that I wouldn't trust it. Similarly extrapolating Kt = 2 from Kt = 1 data is too different to be believed.

Look into how your company validated this extrapolation.

another day in paradise, or is paradise one day closer ?