Cyclic loading-fatigue failure 6

[elogesh]

Hai,

I got good responses with earlier thread of nonlinear analysis permeanent deformation.
Now I have a practical problem in my industry to be resolved.

We have accelerated life testing for evalauting the life of the system and components for the specified warrenty period.

In our system, a pipe connection has failed during this acclerated life testing.

We want to simulate the similar conditions using FE technique.

The pipe at one end has a dynamic member(moving member) and other end connected to static member. Inside the tube there is high pressure fluid.Therefore the tube is subjected to 1) Internal pressures 2) End forces due to the moving member.

The internal pressure is increased in the following fashion
1) First pressure load p1 with number of cycles n1
2) Second pressure load p2 with number of cycles n2
3) Third pressure load p3 with number of cycles n3
4) Fourth pressure load p4 with number of cycles n4

p4>p3>p2>p1.

How to evaluate the life using FE technique?


I may be asking very broad question? But if you can provide some tips,websites,suggestions it will be helpful to my problem?

I will try to get s-n curves from our laboratory?

I am planning to do the analysis as mentioned below.

1) Applying P1 pressure load and finding out the static equivalent stresses.Then going back to the s-n chart and finding out the number of cycles allowed(N1) for this stress condition. Then finding out the remaining life period using cumulative fatigue damage.

That is n1/N1+n2/N2+n3/N3+ .. = 1

2) The above procedure is repetated for p2 and p3,p4.

I have few queries in this regard.

1) For static failure in tube, I used to follow von-mises stress failure theory, I think that is appropriate for my ductile material condition.Now for the fatigue case whether I can consider the von-mises stress as the equivalent stress or I have to consider principal stress.If principal stress then why have to consider principal stress?

2) I am going through the stuffs like miners rule,strain life based approach,rain flow ... counting,etc.Can I have any suggestions in this regard?

IF there is something technically wrong,please correct me.

I have good knowledge of FE and ANSYS,general purpose finite element analysis package.

I keen to study and learn any text books required for the analysis.

Looking forward for your suggestions/feedbacks.

Regards,
E.Logesh

 

[elogesh]

Hai,

I forget to add one more detail. I hope it may be important one. the detail is as follows,

The total number of cycles N1+N2+N3+N4+... is less than or equal to 1000 cycles. I think we can call it as low cycle fatigue.

Regards,
Logesh.E

 

[corus]

The procedure you describe sounds fine to me. You do mention that it is a pipe connection though and, as such, if it is a weld then you will need the S-N curves for welded connections using the appropriate classification for the type of weld and the direction of the principal stress, eg. whether the stress is in the direction of the weld or transverse. I use british standards, such as BS7608, for the assessment though there are other similar standards such as ASME and the International Institute of Welding, Fatigue design of welded joints and components. These can be obtained from the British Library of Lending, though if you're in another country the bus ride will be expensive.
The S-N curve for the material you have won't be necessary if it is a weld, though you will need the Young's Modulus if it's different from steel.
As for why you use principal stress. I'm not sure as to the exact reason as some codes refer to the stress range while others refer to the principal stress range. Which ever, you will need to know the direction of the principal/stress range for your assessment. The Stress Intensity, ie. von Mises is used for the static assessment, as you said, and is only required to check for failure by yielding in a single event. The stress intensity has no direction of course, which is one good reason for not using it in a fatigue assessment.
Care should be made in using the results from the FE model in that you don't use the stress concentration from the joint as the fatigue codes use the nominal stress away from the joint. Stress concentrations at geometric features such as holes are used though. For a weld the correct procedure is to plot the stresses up to the joint but use linear extrapolation from values away from the joint to obtain values at the joint rather than the peak stress value that will include a stress conctration of the joint. Another reference for Fatigue Life Assessment using FE is John Draper who sells the program FE-SAFE.
For low cycle fatigue failure, I believe the procedure is no different and you can interpolate the S-N curves back up to a limit of twice the yield stress of the material.
If you get results that agree with your testing by a factor of two, then I'd be happy with them given the statistical uncertainty of the test and S-N data.

 

[ifegermany]

here are some questions of mine - appart from the influence of size and surface finish, from my understanding S-N curves should have been extraced under sufficiently similar loading AND geometry compared to those modeled with FEA

1) I understand that, if a uniaxial specimen is the S-N source, FEA should have a clearly predominant uniaxial stress/strain field, i.e. one dominant principal stress/strain component. Otherwise the S-N data are not corresponding

In line with the above, if FEA model simulates e.g. bending or torsion, should S-N test curves come from a specimen/load which corresponds in load AND geometric shape? given that recommended fatigue strengths vary for different load types.

2) if a compound load, a superposition of e.g. bi-axial or other loading is at hand, would simple superposition of load events be o.k.? i.e. using the S-N data for each isolated load type, adding damages in sequence, to obtain total damage with accumulation rule S ? or do we need S-N curve extracted under COMBINED loading AND geometric similarity, to make live estimates turn out o.k.

Thanks,
Frank

 

[corus]

Frank,
I think you're right concerning fatigue of welds. As far as I know there is no allowance for multi-axial stress conditions though the direction of the principal stress governs which class of weld it is for fatigue assessment. Thus if a weld has a principal stress in the direction of the weld and a principal stress across the weld then two fatigue assessments can be made at that weld which may not necessarily imply that the higher stress produces the lower fatigue life. The stresses are not presently considered in combination though.
Different weld classifications are generally given on the geometry of the weld as well as the principal stress direction. Thus depending on the load and the geometry a fatigue life assessment can be made using the S-N data.
Superposition of loads can be made providing stresses remain within yield. The superposition of the loads would provide you with the stress for that particular cycle. The previous point about the effect of mulitaxial stress remains with no allownace for the combined effect on fatigue life remains however.
For plain material subject to multiaxial stress, there has been a lot of work carried out with various theories. Watson-Topper, I think, is one. Some software is available that can take the results of a FE analysis and apply these theories to give you the effect of the geometry and the combination of stresses at a point in the structure to calculate the fatigue life, shown as contours on the structure. One I know of is FESAFE which can use the output from various programs. For welds though, FESAFE uses the standard S-N approach with all its short-comings.

 

[ifegermany]

hello corus,

thanks for valuable comments. My question perhaps without explicit mention, were formulated with elogesh's posting status in mind. Namely refering to S-N data at the surface of standard parts, i.e. the general fatigue question instead of special case represented by welding - seen that elogesh did not respond so far with a statement wrt if he does have/not a welded situation.

That said, the core of my question boils down to the following - regarding validity of S-N curves for a part of given material, loading and geometric shape: namely, do we need test data which actually reflect geometric shape and size of the part (besides load) or are standardized shaped test specimen sufficient?

This because: if geometry (shape and size) of actual part are of significant influence (think about any irregularily shaped industrial part) on the actual S-N curve data obtained - what relevance can the numerics based (and analytic/hand calcs for that matter) fatigue evaluation achive, if they are based on data obtained from standard specimen tests

Thanks,
Frank

 

[corus]

Frank,
I know that the size and surface finish of the component is relevant, and that there are various factors employed that reflect that on the fatigue life of the component. This is particularly so in the case of rotating components where such data is available. I do not believe there is a need to test a particular geometry or shape unless you were concerned that the stresses were near a flame cut hole, for example, or were operating in a corrosive environment. Even in these cases there are factors available that apply to the standard specimin results to allow for these various cases to make your fatigue assessment relevant.
As for the case where there are significant stresses in more than one direction at any particular point you would have to refer to those theories I mentioned previously. I'll provide some references later but I suspect that these theories relate to standard specimin tests. If you can test a particular geometry and load against predicted fatigue life then that is obviously preferably as it would give you confidence in the methods employed.
In general the codes of practice are written such that a conservative estimate of the fatigue life is given to give a high level of confidence in the life of the component for design purposes. The conservatism that is built in to these codes is presumably to allow for those unkown effects you mention. So far they appear to give reasonable results.

 

[corus]

Frank,
Further to my note above, you may wish to refer to work done by McDiarmid, Brown-Millar, Socie-Bannantine, and Wang-Brown who all consider multi-axial stress and the effect on fatigue life. As far as I am aware their results are correlated from data obtained from simple tests.

 

[ifegermany]

hello chorus,

thanks for additional comments and also for those references. I take what you say as given, i.e. standard specimen S-N data being sufficient to evaluate fatigue at surface locations of arbitrarily shaped parts.

The question came up as some people from testing sustain that you need to test the actual geometry to get realistic values, be it fatigue, be it e.g. structural damping. Same line as with the argumentation that creep data need to be extracted testing over the whole real creep time instead of using extrapolation rules from short time creep tests.

 

[corus]

Frank,
My comment would be that if you can test the actual part then why do any calculations? The idea is that you test a specimin and then apply those tests to any loading/any shape using FE models. Where the loading is difficult to predict in service the method commonly used is to measure the strains in a component in service and then to model the component and correlate the results from the FE model to actual measured results at particular points. The benefit of the FE model is it can then be used to predict the stresses anywhere in the structure, and from those stresses predict the fatigue life at any point.
The point about creep data is valid but you cannot be expected to wait 10 years to see the results of measured creep data so I presume some other methods are employed to extrapolate from measured data. Perhaps they use accelerated results and then apply some creep law to them to obtain results for any temperature. I don't know.

 

[ifegermany]

Hello chorus,

clear, sense making, all points you make. I tried only to trow some light on the weight of the experimenters arguments mentioned, because FEA based fatigue evaluation sometime has an area of retroactivity instead of a predictive nature (i.e. confident prediction of life range ahead of prototyping)

Thanks for comments,
Frank

 

[elogesh]

Hai,

Eventhough, I don't have much knowledge about fatigue,being an original poster, I want to add few more points

The actual component, which I have mentioned was planned to subject for the fatigue analysis.The S-N data was planned to extracted from the actual component and also loading conditions similar to the one experienced in actual conditions.But unfortunately FE analysis only carried out with some old data of S-N values.
The points mentioned by both corus and frank are well valid one.
I heard about LMS-Falancs claimimg that they can carryout multi-axial fatigue compared to other commericial sotwares.
I hope,this this particular thread will be added with more valuable comments,which will guide me in understanding the fatigue in better manner.At the meantime,I will update myself by referring good books for fatigue.

Thanks,
Logesh.E

 

[AOktem]

Another point:for low cycle fatigue, strain based life estimation approaches can give more reliable results compared to stress based ones.However strain based FEA techniques are much harder to implement (IMHO).

 

[ifegermany]

hello elogesh,

can you indicate e.g. some web link wrt accelerated life
testing (some clear cut info on how it is done)?

wrt multi-axial enabled codes, I'd think also nCode and
probably some other have this capability.

Regards,
Frank

 

[elogesh]

Hai,

Unfortunately, I don't know much about accelerated life testing.

We have to warranty for our product for 5years.But We can test it for 5years.We subject it to severe conditions for reduced time period usually 3 to 6 months which equals the conditions of 5 years.How they arrived at all these severe conditions(loading conditions) and time period, I don't know.
But this accelerated life testing is one of discussion topics in our company.For new product with different specifications, how to arrive at the parameters for acclerated life testing.
We are doing extensive work in this regard, both analytically as well as experimentally.Currently I am not involved in the project in the company.
If I get anything in future, I will get back to this forum.

Regards,
Logesh.E

 

[ifegermany]

Hello elogesh,

thanks for the status.
I'll check if I find published sources on that theme, get back

Regards,
Frank

 

[drmh]

Dear all,

just a few comments to this problem. In principle a strain-life approach using the four events one after the other. Due to the low number of complete cycles one has to take the cycles that occur due to the consequtive loading of the different events into account, as well. Software codes like the above mentioned LMS FALANCS are able to take this into account, automatically.

Since in the mid of the thread seam welds were mentioned just have a look at the LMS group, were this was discussed already.

For accelerated fatigue testing using multiaxial rainflow methods the most advanced methods take into account both real events and target multiaxial rainfow count and optimise combinations of real and synthetic loads to get closest to the target set of rainflow counts. This is especially important since in this case the original phasing of the loads is not lost. There is also software available for this (sorry again from LMS) called TecWare.

Best Regards

Michael

 

[jagad5]

My comments may be way too late to be useful and I didn't study every single post but I think there may be a couple of things you haven't considered.

If you run an accelerated test, you must make sure you don't accelerate it so much that you change the failure mode. Do the parts actually see failures caused by the substantial plastic strains you are generating? If you are failing a ductile material that quickly, you might be setting the accelerating stress too high.

Someone mentioned surface finish as a factor. At these low cycles, surface finish reduction of the fatigue strength will be minimal because the high stresses quickly yield any sharpness at the bottom of the ridges (that sound technical, doesn't it? )

You're using Miner's rule for your test but make sure that in service you use it appropriately. For stresses causing failure at very different cycles, interactions start to cause failures at sums less than 1.0. See International Journal of Fatigue, Halford, Cumulative fatigue damage modeling - crack nucleation and early growth, Volume 19, Supplement No. 1

For material properties - take a look at the paper "Detailed evaluation of methods for estimation of fatigue properties" by Jun-Hyub Park and Ji-Ho Song, International Journal of Fatigue Volume 17, No 5. They compared several methods for estimating strain-life curves based on minimal tensile strength info and make suggestions on which fit best for different catagories of metals.

Hope it works or worked.

Doug

 

[Irwin]

I think Fatigue analysis is impossible in FE. You need an auxilary software like FEMFAT, FALANCS or nCode. The other way you write one. We use FEMFAT which is suitable for welds also!

Best regards

Irwin

 

[Irwin]

One more question. Are p1, p2 etc. constant pressures or amplitude pressure?
What I would do:
1. If the pressures are constant and the Force is changed during the process In this case you need 8 different load cases:
lc1: p1 pressure and Fmax
lc2: p1 pressure and Fmin
lc3: p2 pressure and Fmax
lc4: p2 pressure and Fmin
.
.
.
lc8: p4 pressure and Fmin

If the pressures are amplitude pressure and Force is constant you need also 8 load cases:
lc1: p0 base pressure + p1 amplitude + F
lc2: p0 base pressure - p1 amplitude + F
lc3: p0 base pressure + p2 amplitude + F
lc4: p0 base pressure - p2 amplitude + F
.
.
.
lc8: p0 base pressure - p4 amplitude + F

So that is the important thing that you need maximum and minimum stress distributions (or amplitude/mean stress distributions) for each cycles.

Second step: Define SN curves and Haigh diagram!
Haigh diagram helps you to calculate with the mean stress effect.
If you use shell model or solid model the analysis will be more complicated because you need to transform the nominal SN curves to local SN curves in each node! (FEMFAT does this.)
If you use beam elements (1D elements) the task is easier, because in this case you can use the SN curves form standards (like BS5400) or tests!

Third step: Calculate damage!

(2 and halfth step: Use some other additional factors, like size effect etc.)

That is all!

If you have more question write me an e-mail!

Best regards,

Irwin