column buckling check 2

[Lion06]

I tried following the procedure in this thread (suggested by 271828) to get a column buckling load out of a program, but it just isn't working. It keeps upping the lateral displacement, but it never becomes unstable near (or above) the elastic critical buckling load. I could double the euller buckling load, but the lateral displacement is always appropriately proportional.


thread507-211026

 

[haynewp]

It doesn't look like the program knows in the analysis that the column is unbraced (whether x or y direction) and buckles, the steel design check is run separately from the analysis. It is just keeps going through the displacements without considering a buckling point.

 

[Lion06]

haynewp-

I actually thought that until it buckled in a different mode than I originally assigned (which is making no sense to me whatsoever).

 

[haynewp]

I see where you said that above now. It looks about 196 kips it buckles and at 197 kips it is going the opposite direction. At 196.5 kips I get 124" of displacement.

I tried modeling it without the initial curvature and it never buckles.

 

[haynewp]

Try 196.29 kips, I get 2754 inches in the positive direction then it switches at 196.3 kips to negative deflection.

I don't know why it never represents buckling if you take out the initial curvature.

 

[haynewp]

If you run your model without the second order option checked, it doesn't buckle. I think what we are seeing is the p-delta blowing out without considering column buckling.

 

[JAE]

I used the same 20' model, with W8x31 weak axis bending on RISA 2D. Had to put in an initial offset of 0.06" before it would register any second order buckling (might be a round off of cooordinates within the solver).

Here's my results:

Load Defl at midpoint
w/o Shear Def. w/ Shear Def.
100 k .058" .058"
150 k .207" .210"
175 k .808" .842"
180 k 1.561" 1.695"
185 k 13.86" 42.30"
186 k buckles buckles

Per AISC 13th Edition - Equation E3-4, Fe x Ag = 184.9 kips

 

[JAE]

I forgot to mention for the above numbers - I also divided up the member into 1 ft. segments with P[Δ] turned on.

 

[haynewp]

JAE, what buckling value does Risa give if you run the same model with P-Delta turned off?

 

[271828]

Without an initial curvature, this method will not help to find the buckling load. {F} only contains one element--the one corresponding to axial load. There's no way for this force to cause moment and corresponding transverse deflection. Therefore, {Delta} only contains deflections corresponding to axial deformation.

First-order analysis won't give the buckling load using this method because the following system is being solved: [Ke]{Delta}={F}. Because [Ke] only has material props, section props, and lengths in it, if you increase {F}, then {Delta} MUST increase proportinally.

Without iterating and moving nodes, the following must be solved: [Ke+Kg]{Delta}={F} [Kg] has P/L terms if it's linearized and higher order terms if it's geometrically consistent. This allows non-proportionate {Delta} growth with tiny {F} increases. This is very, very similar to how AISC's B1 grows as Pr approaches Pe1 in Eq. C2-2. Pr is analogous to [Kg] and Pe1 is analogous to [Ke].

StrlEIT, try breaking the members up into shorter members. If Ram uses a linearized Kg, then it's internally overly-restrained which would make it a little stiffer than it should be. SAP uses a geometrically consistent Kg, so this doesn't happen as much. I seem to remember that RAM used the goofy linearized Kg. It's about 10 min. extra effort to use the geometrically consistent one.

 

[JLNJ]

Any ideas?

Don't use RAM Advanse? I had a bad experience with an oddly shaped truss where RAM Ad was telling me certain web members were zero force. I didn't believe it. Risa3D and RAM Ad were giving very different answers and the Risa3D results seemed to make sense to me. (It wasn't P-delta related.) To this day I still don't know why the difference.

 

[haynewp]

I got it to buckle below 185kips in Advanse by using the Modified Newton Raphson option and adjusting the convergence tolerance to 1.5E-4.

 

[JoshPlumSE]

All -

I'm curious if the RAM guys were ever contacted and asked to explain the differences between their P-Delta results and the other programs.

It would be interesting to hear what they said. Oftentimes, tech support guys can give a simple explanation for what seems like a very complex difference.

I find it hard to believe that the average user can be expected to play with convergence tolerances and solution algorithms (Modified Newton Raphson vs what?). That sounds more like what is expected from the users of a true non-linear analysis program.

Josh

PS Disclaimer: My interest is really related to the strengths and weaknesses of various methods of accounting for the P-Delta effect. I work for another analysis program. We are considering adding an alternate method of P-Delta analysis. Hence my interest in better understanding the strengths and weakness of the various methods.

 

[271828]

The Ram Advanse user's manual is at

http://docs.bentley.com/product.php?prod=204

After reading it, I have no idea why StrlEIT's analysis didn't work or work without tweaking the parameters. They're using the geometrically consistent [Kg] so that's not it. They're using element geometric stiffness matrices instead of story-wide matrices, so that should be OK. I don't see the difference between their description and what SAP2000 does. Maybe SAP2000's default second-order parameters are better. I had forgotten that iteration is still required because P in each element might change from iteration to iteration, changing each element's [Kg].

My only guess is that there's something basic wrong with the model, making it stiffer than it should be.

 

[271828]

Josh,

How 'bout adding response history analysis?!

100000e

 

[271828]

Sorry for the repeated posts, but I just thought of something in the Advanse manual that seems like at least a small booby-trap.

They do not form [Kg] for shell elements. Not saying that I know it's wrong, but I'd sure do some careful checking if I had a shearwall structure or was creating a member out of shells to do an analysis similar to StrlEIT's.