First, the basics.  Fatigue-crack *initiation* comes first, obviously.  There are a few phases of this, but the entire process can be many times longer than the following "propagation" phase.

You can polish the suspect surface and make a plastic replica.  Study this in a SEM or a TEM, and you should find some answers.

I'm not a materials specialist.

How could you determine the number of stress cycles remining till failure by a test of the member? I can't see how that could be possible.

Wouldn't you look in the opposite direction - ie, how many cycles have occurred in the past? Then, if you have a recommended life, you know where you are?

Can you normally see any evidence of fatigue before cracking?

What is SEM and TEM ? What do they tell you from the plastic replica?

Cheers,
John.

SEM - scanning electron microscope - uses reflected energy
TEM - transmission electron microscope - uses transmitted energy

You cannot "predict" failure point for a single given unit.  Failure and fatigue values are based on ensemble statistical behavior.

TTFN

I have had some success with identifying the fatigue life of a combination of various loads by stressing to failure at the worst load. Knowing when failure occurs at the highest load leads to relations as follows:

S1 x Nf1 = K

This is the stress x cycles to failure at that load. The K represents the product of total damage of the material at the test temp. Now:

S1 x N1/Nf1 is the fraction of damage at max load. If N1 is half the cycles to failure, then the material has been damaged 50%. Further:

S1 x N1/Nf1 + S2 x N2/Nf2 +....= K

This is the summation of damages at various loads. The cycles to failure at various loading should be tested.

I am skeptical that micro examination of material will lead to conclusions of life remaining.

Can your current yield strength state be ploted on the fatique curve to estimate life?

Boo1,

Fatigue strength doesn't correlate well with YS like it does with UTS.

Tsukuba,

Is the fatigue eval. stressing enough to eventually cause cracking?  If you know the stresses and cycles pretty well you might be able to calculate an estimate-assuming the service cond. are also well known.

Otherwise you can use the replica method.  It is commonly done on turbine shafts etc. if there is a known area where cracks will start-such as a keyway, etc.

Metalguy,
I suppose I don't know the previous record of the fatigued structure.
Dose it mean we cannot estimate its remaining fatigue life by a nondestruction method in this case?

No, but I think it pretty much eliminates trying to calculate it.  Is your structure steel or Ti (or Ti alloy), which have an infinite cyclic life provided the stresses are below some value, or Al etc. which do not have such a limit?

I think the only practical way to use a NDE method is the plastic replica.  They can provide amazing detail, and identify very tiny cracks.  You'd have to know where the stresses were 9and will be) highest in order to avoid having to replicate the whole thing.

What sort of structure do you have?

Metalguy,

The fatigued structure is stainless steel.

The plastic replica method can provide information about the tiny cracks only on the surface. Even we know these tuny cracks exist, we still cannot say how long the remaining service life of the structure has.

If you make the assumption that you already have some tiny fatigue cracks, then you must deal with the propagation phase-much easier than the initiation.

Periodic NDE, using methods such as UT, eddy current, penetrant testing are what you need to do, unless you are very sure of the future stresses and cycles.  Even then it's still a good idea to perform the NDE.

If you don't know much about the stresses except that they do not change (level and cyclic nature), you will be able to predict the usefull life AFTER you build a database via NDE for your particular application.

If you aren't sure the stresses will stay the same during use, you don't have much choice except to perform periodic NDE.

Thanks Metalguy,
The plastic replica may be the only NDE method to evaluate the remaining fatigue life.

An update:

A more correct relationship of stress and cycles would be:

S^3 x N = const

If this is evaluated at failure, then

S1^3 x N1 + S2^3 x N2 + ... = const for several stresses and associated cycles.

(My previous post showed S raised to unity power.)