Fatigue: FEA (fatigue analysis at every node) 3

[KnacKeN]

thread727-156008

forgive my english, regarding using principal streses for fatigue analysis.

If I have two FE cases, and suppose FEA based fatigue analysis will be caried out by cycling between these two cases.

How would principal stresses be used? Does it matter that the principal vectors may be in different directions between the two cases?

 

[corus]

Take the difference between the stress components Sxx, Syy, Sxy etc., and calculate the principal stress from that difference.

corus

 

[gwolf2]

Using principal stresses to calculate the maximum and minimum values for your stress range is a simple and conservative way to assess fatigue. If you want a more accurate answer then use method outlined by corus.

If you want a really accurate answer consider DeltaSnn and use multi-axial fatigue formulas.

You will find that for most geometries, for which fatigue is important, the stress field is 2D and often 1D because it is at a surface or edge. In these cases taking the stress range of the components Sxx, Syy etc often reveals only one stress component to be dominant anyway.

 

[feajob]

The most conservative choice is signed Tresca. Sign is defined by the sign of max or min principal stress. Signed Von Mises is not a good choice for your application.

Ref. MSC.Fatigue Manual

A.A.Y.

 

[KnacKeN]

The most conservative choice is signed Tresca. Sign is defined by the sign of max or min principal stress. Signed Von Mises is not a good choice for your application.

Thanks, however I was more concernd with the use of Principal streses.

I may also looked at a number of cases, like maybe 10 instead of just 2. That is, my spectrum might contain 10 different load case. I guess I would need to just use Principal stresses becasue taking the difference might be impractical.

 

[johnhors]

For a spectrum of cases, you should carry out a rainflow analysis at every node (using the dominant principal stress for the case), to get pairings of cases. There may be several pairings to consider (not just one). Damage can then be calculated using say Goodman's and a S/N curve or a Mil. Handbook formula and summed using Miner's rule.

 

[rb1957]

amazing how quickly a simple problem becomes complicated ...
use principal stresses;
after that you have to understand your loading spectrum; like john says above, you need to pair stresses together (don't forget the null stress, before the loading is applied; does the structure return to null state between repetitions of the spectrum ?)
wiki/google "rainflow fatigue" to get more info on rainflow analysis of a fatigue stress spectrum.

 

[KnacKeN]

For a spectrum of cases, you should carry out a rainflow analysis at every node (using the dominant principal stress for the case), to get pairings of cases.

dominat is largest magnitude principal stress? So for if a spectrum the dominant principal stress at node are 0,-10,0,25,65,0,-82,50,70,0 (10 cases) the vectors could all be different?

wiki/google "rainflow fatigue" to get more info on rainflow analysis of a fatigue stress spectrum.

I think I know how to rainflow to find the reversals, and add the damages together for each pair, but as above i was wondering how correct it would be if the streses is in different directions.

 

[TGS4]

The method that corus states is the most rigorous. Calculate your principal stresses based on the component stress ranges.

This is the methodology followed by many Codes and Standards, including the ASME Boiler and Pressure Vessel Code.

 

[rb1957]

if you've got one cycle of each loadcase, then i'd analyze your spectrum as 0-65, 0-70, 50-70 (delete the -ves, neglect different directions,)

 

[kellnerp]

See ASTM Standard E 1049-85; “Standard Practice for Cycle Counting in Fatigue Analysis.” for means to count cycles. I agree with Corus on which FEA results to use.

There are Matlab algorithms for doing RF.

TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[gurmeet2003]

Knacken in his original question indicated that the principal stresses might not be in the same direction for the two load cases. Corus does your approach apply when the principal stress are changing direction or are non-proportional? I have heard that for non-proportional stresses the critical plane method is the best. Though I do not know all the details of the method.

Gurmeet

 

[kellnerp]

Corus suggested calculating the principal stress from the DIFFERENCE in sxx, syy, sxy between two load cases. There can be no difference in principal stresses between load cases because this is a difference.



TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[rb1957]

IMHO, i've got no idea what the "principal stress based on the difference of stress components" means (practically, i know whatit is mathematically).

take the principal stress from each load case, conservatively neglect that the direction changes from case to case.

 

[KnacKeN]

take the principal stress from each load case, conservatively neglect that the direction changes from case to case.

Yes thats looking likely, I will take the "dominant" principal stress.

 

[kellnerp]

Let's do a mind experiment here:

Imagine a thin square plate.
Starting at the top edge label the edges A, B, C and D going clockwise.
The x axis is edge C, positive to the right.
The y axis is edge D, positive up.

Load Case 1
tensile load on edges A and C of 1.
no load on edges B and D

Load Case 2
tensile load on edges B and D of 1.
no load on edges A and C

Using the efunda principal stress calculator:

Taking the difference of Principal stresses:
LC1 P1 = 1, P2= 0 and theta = 0 tau.max=.5
LC2 P1 = 1, P2= 0 and theta = 90deg. tau.max=.5

delta P1 = 0, delta P2= 0, delta tau.max = 0 and there would be no fatigue predicted, clearly non-conservative.

Using the difference of stress components to calculate principal stresses:

LC1 syy = 1, sxx = 0, sxy = .5
LC2 syy = 0, sxx = 1, sxy = .5

delta syy = 1, delta sxx = -1, delta sxy = 0

P1 = 1, P2 = -1 and theta = 135deg tau.max = 1 so there would be fatigue predicted and Corus' answer would be more conservative. If a crack were present it would not be Mode I, but would probably grow by shearing.




TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[gwolf2]

KnacKeN,

I suggest you use Corus' and rb1957's methods which have a sound formal code-based methodology according to TGS4, or if you absolutely need a more accurate answer use my individual component range method (plus rainflow, now that you reveal multiple load cases) if you have biaxial fatigue data. I repeat, if you look at what all the components are doing it WILL condense to a 2D or 1D stress field because of the geometry.

The rest of these comments are IMHO just partially informed babble which just confuse you and do not solve your problem. Anyone who honestly doesn't know what they are talking about, please don't post.

Regards,

Gwolf

 

[gwolf2]

Just one last thing:

"it WILL condense to a 2D or 1D stress field because of the geometry"

You will probably have to transform the stresses using a button in your post-processor to the local geometry system - e.g. cylindrical at a hole, to make the off-axis stresses dissapear and make it 2D or 1D.

Have fun, I enjoy it.

gwolf

 

[rb1957]

kellnerp,

how does the loading change from case 1 to case 2 ? does it unload between them ? in which case the cycle would be 1-0-1. maybe you load up 2 whilst maintaining 1, then once load 2 is fully applied you relax 1; in which case the cycle would be 1-2-1.

 

[kellnerp]

The intention of the mind experiment was to depict a square plate with the load going from 1 to 0 in one direction while increasing simultaneously from 0 to 1 at right angles. This matches what KnacKeN has told us about his analysis, two load cases. If you assume the load goes to zero and then increases in the other direction you have a different situation. But

I have two FE cases

so there is no 1 0 1 situation.

TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."