Required Mesh for Plate Analysis

[NoName1]

I have modelled a 10 ft by 10 ft concrete plate with a hydrostatic load to verify the output against plate formulas in "Roark's Formulas for Stess and Strain" So I can apply to a concrete tank model. Using the quadmesh tool, I need about a 3" mesh before results seem to converge.

Is this reasonable? I am new to Risa. The tank I need to model is 212 ft by 50 ft x 25 ft, so a 3" mesh makes a huge model.

Thanks in advance for any help.

 

[miecz]

I generally get reasonable results with a 1 ft by 1 ft mesh. Anything finer will make the solution take forever. Also, this makes for easy load input and interpretation of results. Consider modeling a quarter tank, if everything is symmetrical.

 

[NoName1]

Thanks for the response.

When you say "reasonable results" do you mean the results are close to published plate results?

 

[miecz]

Yes. I do it both ways, PCA Tank tables and finite element, and use the worst result at each location. I'm not confident enough to do it one way. Haven't done one in a while, so I forget how close is close. If I did one now, I'd be happy with agreement within 10-15%.

 

[DWHA]

Make sure you width to thickness ratio of each plate is 3:1 or smaller. Otherwise your results will not be accurate.

 

[NoName1]

I will have to adjust my mesh to match the 3:1 ratio. I was using a 33" wall with the 3" mesh

Thanks for the help.

 

[JoshPlumSE]

I should point out that the 3:1 ratio is not a magic number. That's the recommended limit. But, it not like a 4:1 ratio is going to produce 100% inaccurate results.

Instead, you will get a gradual degradation of the results as you get larger than the 3 to 1 ratio. But, it should be gradual. Not like in the old days (1970's elements) where you'd get "shear locking" and your results immediately turned to garbage when you exceed their thickness limit.

That beins said an 11 to 1 ratio certainly exceeds the limit by a significant amount! Hence, I'd be very, very cautious about using a ratio that large.

 

[JoshPlumSE]

Another thing: What results were you looking at to converge? Frequently, you will get localized stress risers around a point support or a point applied load. I do not generally look at these stresses to determine convergence.

Instead, I like to look at deflection to determine convergence. I usually use about a 10% convergence rule. If I can sub-mesh 2x2 and the max deflection changes by less than 10% then I feel like I have converged.