Comments

• MikeMcMullen

  February 11

  Try a lower load then run again with higher loads till you get the convergance problem. Back off to the last load that worked and look at the strains and deflections. It only takes one node not converging in force or displacement. When Calculix does stop, It tells you the node number with the greatest displacement error and the one with the greatest force error. You may see something. Calculix does not converge very well when near peak strength strains. The problem may be with how the analysis settings are set, but I have not figured anything better. One problem is potentially the tolerance for convergance in calculix which is very tight when you consider the units are very small. Compression only does do tension, just not with the curve you would expect. In your concrete model the concrete would be limited to about 3Mpa tension, but have a slowly increasing tension with added strain after that to avoid negative stiffness.
  
  I have not tried it yet but Radioss may work better, for one thing, I think it has a better concrete model available.
• Aftab

  February 11

  Thanks, Mike, for the heads up. I will give it a try once more.
• MikeMcMullen

  February 11

  I had a similar problem. The issue is multiple. Shear between the steel and concrete induces tension in the immediately adjacent concrete. It stops increasing in capacity, sheding load to adjacent concrete elements on the next calculation cycle. They then fail in tension in the next cycle etc. (crack propagation). My problem was rebar in concrete. A movie of the attempted convergance had the rebar danceing.
  Actual behavior is not a fixed tensile strength, but one that is larger and larger as the element under consideration gets smaller. Actual behavior has the tension vector rotating as the concrete fails in tension, putting compression across the crack, and the concrete expanding (poisson ratio > .19). In your case this expansion can cause the tube to restrain the expansion putting a confining compression on the concrete. Increasing the tensile strength limit up to about 14 Mpa can arrest this unziping in shear behavior, but the issue is more that the lack of variable poissons ratio, the sudden kink in behavior of the concrete, and the lack of an accurate shear model make composite/steel plastic behavior unable to be modeled with the compression only concrete model, which does not even include poissons ratio...
• Aftab

  February 12

  Thanks Mike for your insight — that was really helpful.
  
  I’ve been dealing with a bit of a conundrum, as several of these challenging issues seem to be present in my case. I was exploring an alternative approach by defining contact between the concrete and both the sleeve and the pile, with the idea that the concrete would remain primarily under compression. In this setup, no direct tensile load would be applied from either the pile or the main leg acting as the sleeve.
  
  This would allow me to define the concrete using plastic material properties. I realize this may be somewhat of a simplification, since it essentially treats the concrete response more like a von Mises-type behavior, which is not truly representative of concrete’s directional characteristics. However, since the concrete is externally confined and mainly subjected to compression, it seemed like a reasonable approximation — although the internal stress state under loading is still an open question.
  
  That said, I’ve been struggling with convergence when modeling the contact interaction between the plastic concrete and the steel components, so it’s not an ideal solution. I may look into trying the OpenRadioss solver as an alternative.
• MikeMcMullen

  February 12

  Tell me how that works! For me the compression only worked OK if I did not model the steel and just used it to increase the effective tensile capacity of the concrete. It converged OK since there were no shears between concrete elements and steel ones, but when I attempted to code in the steel bars explicitly, no luck. Part of this as I mentioned is that for this shear on a small scale, in reality, the tensile strength is much higher than the value you used which is appropriate for gross features. I have followed reasearch on concrete filled tubes for decades. They behave really well, but it took a long time for codes to catch up. I am not sure they have caught up to the point of being useful on a plastic FEM basis yet. Things have gone pretty silent...
• Aftab

  February 22

  Hi Mike,
  
  I hope you’re well.
  
  I’ve managed to run several analyses to convergence — some using compression-only concrete, and others using a Mohr-Coulomb model. In these models, I defined gap elements between the sleeve and the concrete, and between the concrete and the pile.
  
  I’ve also created another version using contact with a small tangential stiffness, allowing a small amount of slip in case the concrete needs to “breathe.”
  
  One strange observation is that all the models seem to stop converging around the 0.17–0.18 mark. I have scaled the loads by a factor of five, so 0.2t represents the full design load. As you know, the joint was already struggling under the classical API joint check approach, which is why I moved to nonlinear analysis.
  
  I’ve attached a few example cases for your review. The models are quite large, so I’ve included a download link instead.
  
  I would appreciate it if you could take a look and let me know whether the approach and results seem reasonable.
  
  https://www.dropbox.com/t/UQ3ibRoDO5eMzggN
  
• MikeMcMullen

  February 23

  I think you are well beyond me at this point. Mohr coulomb is already an improvement. Allowing a small slip should approximate actual behavior better at discontinuities or load application points by not triggering an unzipping starting at a small stress concentration, often due to the discrete nature of FEM modeling (does a high stress at an infinitismally small volume mean anything in reality?). In reality interface shear between as produced steel and concrete does not fully slip until the strain is above 0.005... at least as far as I have read. Rough surfaces do better. Realistic poissons ration are important though. These tend to increase after linearity is passed. I think mohr coulomb can have positive or negative dillantancy (something akin to poissons ratio), This reflects if the volume shrinks on shearing or increases. Most concretes expand on shearing, but if the void ratio is high enough they can collapse. Depends on a lot of material factors including aggregate strength (compared to hydrated cement) and size. Those who research these topics (for filled tubes) usually are modeling a test item to calibrate their model.
• MikeMcMullen

  February 24

  Note this issue is not in my typical design bailiwick. I just follow it because filled tubular structures have a huge energy absorbing potential that eventually could have applications in crash structures, prefabricated pier columns, and rockfall control, things I have been roped into over the last 3.5 decades. My usual wotk in is fairly standard bridges with post tensioning or prestressing (sometimes unique construction methods or bad skews or curvature). Most of this has been done without FEM. Codes usually proscribe the sorts of things allowed there. Note sections 6.9.5 and 6.9.6 of AASHTO LRFD Bridge Design code covers concrete filled tubes, and the commentary says the formulas do not assume composite action. There are references to the research in the commentary.
  
  I.e. I think your problem is on the bleeding edge, but with appropriate modeling you may get somewhere, but full composite behavior will require work on how you model the interface and concrete, and possible modification of your section (roughening, dimpling, internal ribs, expansive concrete... etc.). Note rebar have ribs, the configuration of which was developed over about a century to get well behaved composite behavior. Concrete shear is another similar problem. Explicit modeling with a penalty model is usually used (see RaDIOSS OR lsDYNA), but the parameters usually need to be chosen from what fits test data rather than code recommended values.
  
  https://maps.app.goo.gl/k7rnooejCvnYWaxC7
  
  Picture of a crash barrier to deal with overheight trucks before they hit fuel line on the RR bridge a hundred yeads further on. Triangular high strength thick wall steel tube filled with concrete. Not designed composite, but probably is at least partially.
• Aftab

  February 27

  Hi Mike,
  
  Thank you for your valuable insight. I was able to run the compression-only material model and obtained reasonable results, whereas the other model failed to converge.
  
  I understand that both API RP 2A and ISO 19902 address grouted joints and double-skin tubular members where the annulus is filled with grout (concrete). I have therefore adopted those provisions for the time being.
  
  Perhaps in the future I may attempt to develop a more robust numerical model; however, given the current challenges associated with modelling concrete behaviour and the reluctance of CalculiX to handle it effectively, this seems to be the most practical approach for now.