RAM concept zero tension solver: slower than expected convergence

We are using RAM concept to model a base slab, subject to heave pressures, and piles, which are subject to loads from the superstructure. The mat close to the piles needs compressive area springs to help carry 'late loads' from the superstructure that travel down the piles and spread into the mat, the mat away from piles uplifts under the heave pressures, so the area springs there need to be removed.

Having had issues modelling the whole slab, we have moved to a simple example model where every area spring should be in tension, but still aren't sure how it is iterating to remove or soften the area springs when they are under tension. I believe the solver should analyse the structure, determine which area springs are in tension, remove those springs (i.e. all of them), re-analyse with only the point springs that represent the piles, and terminate. However, the results I am seeing don't seem to show this, with much slower convergence (rather than remaining unchanged for any number of iterations beyond 2 or terminating at 2 iterations).

There are some strange factors that seem to affect the rate of convergance: when the slab on its point supports is quite unstiff (i.e. a central deflection of 200mm) and the slab with an UDL on the area springs (i.e. typical deflection on area springs <<1mm [UDL/area spring stiffness]) the convergance is really slow (I gave in at 1000 iterations, and it was about 2/3 of the way to the true solution, this is just unworkable for a real model). When I make the deflections between the slab (no area springs) and the area springs more comparable, by reducing the area spring stiffness, the convergance speeds up, but never to convergance in 2-3 iterations, the best I can manage is always about 5-10.

We have noticed that most of the literature about how the zero-tension solver seems to refer to overturning, so only a few area springs going into tension, rather than heave, where many might go into tension, and that might have affected how the solver has been designed? We have also read Seth's response to a previous question where he recommends just omitting the area springs entirely to speed up the model - unfortunately this isn't an option for us here - we need the area springs around the settling pile to help distribute the late loads from the super-structure, and don't know where that boundary is. We could remove springs manually, but that is pretty labour intensive.

Also, we are curious, what approach does the solver take? The description given in the manual under 30.1.6 doesn't define a 'tension force offset' anywhere, and refers to 'reducing tension' which could mean either removing springs, or reducing their stiffness, or moving their 'fixed end' downwards with the slab?

thanks,

David

Parents
  • Hi David,

    RAM Concept handles tension in soil (area springs), by adding an offsetting load for the tension, and re-running the analysis. Area springs are not released in the process. For example, if an element of the slab during the first iteration has an area spring reaction (pulling downward) of 100 psf, the program will add an upward load (counteracting the spring) and re-run the analysis. This leads to iteration as the next analysis will have a greater upward deflection and hence need a larger offsetting load.

    The iterations will converge if the structure is stable. When nearing convergence, the deflections in iteration i+1 will be nearly identical to the deflections in iteration i, so the required offsetting load for the iteration i+1 will be nearly identical to the offsetting load for iteration i.

    The accelerator factors can be used to speed convergence as they will attempt to anticipate the final convergence offsetting load. The accelerators work well provided you don't set them so high that you no longer get monotonic convergence.

    You can determine the level of convergence by looking at the soil bearing plots. They will show the effective area spring tension (the tension minus the offsetting load). For practical purposes, you should be happy with convergence if the effective soil bearing pressure is less than the accuracy to which you know your slab weight. I would be very comfortable with 1-2% of slab weight.

    One caution - if 100% of the slab is in tension, the program will stop iterating as it assumes the structure is unstable. This is a correct assumption for standard mats, but not for mixed pile & soil supported mats.

    Feel free to post further questions, or send the file to support (to my attention) if my comments don't answer your questions.

    -Allan

    Answer Verified By: David Hewlett 

  • thanks Allan, this is the best explanation I have come across, and helps my understanding a lot!
    On your second to last paragraph:

    1) is there a way to make the solver continue iterating, even if all the area springs are in tension, to get to the correct result?

    2) On the other hand, if the slab lifts off entirely, then I could save a copy of the model, removing the area spring, and run it again, as we know that the area spring does not contribute. Does the error "ERROR: An error has occurred while trying to eliminate area spring tension.  There is net tension on the slab. Investigate test zero tension.'" indicate that the entire area spring is in tension and the zero tension analysis has stopped iterating?

    thanks,

    David

  • Thanks, David.

    There is a way to trick the program into continuing despite the net tension condition. You need to add a small isolated slab adjacent to your structure and put area springs under it (ensure to provide lateral stiffness also so it is stable). As all of your zero-tension load combinations will include some factor of self-dead load, this small slab will always have spring compression and you will avoid the net tension error.

    A few additional notes on acceleration factors:

    Internally, the program will linearly taper the maximum acceleration factor down to 1.0 as the iterations progress. This allows you to specify a higher max acceleration factor safely.

    You can view the acceleration factors that are used in the Calc Log. If you see a factor go below 1.0 you are overshooting the answer (the accelerator factors for the previous iteration were too large). If you see factors just less than 1.0 (e.g. 0.998) your settings are probably just over optimum (one caution with overshooting is that the reported soil pressures will be positive, perhaps hiding the fact that iterations were did not exactly zero out the tension (they more than zeroed out the tension)).

    The default accelerator settings generally work well for the overturning case, but you may want to use larger parameters for your net tension case. While the perfect factors are slab-dependent, I ran a net-tension test model with Accelerator Power of 5.0 and a Max Acceleration of 3.0 that seemed to work fairly well. I suggestion you experiment a little using a small number of iterations, such as 10.

    Hope this helps,

    -Allan

Reply
  • Thanks, David.

    There is a way to trick the program into continuing despite the net tension condition. You need to add a small isolated slab adjacent to your structure and put area springs under it (ensure to provide lateral stiffness also so it is stable). As all of your zero-tension load combinations will include some factor of self-dead load, this small slab will always have spring compression and you will avoid the net tension error.

    A few additional notes on acceleration factors:

    Internally, the program will linearly taper the maximum acceleration factor down to 1.0 as the iterations progress. This allows you to specify a higher max acceleration factor safely.

    You can view the acceleration factors that are used in the Calc Log. If you see a factor go below 1.0 you are overshooting the answer (the accelerator factors for the previous iteration were too large). If you see factors just less than 1.0 (e.g. 0.998) your settings are probably just over optimum (one caution with overshooting is that the reported soil pressures will be positive, perhaps hiding the fact that iterations were did not exactly zero out the tension (they more than zeroed out the tension)).

    The default accelerator settings generally work well for the overturning case, but you may want to use larger parameters for your net tension case. While the perfect factors are slab-dependent, I ran a net-tension test model with Accelerator Power of 5.0 and a Max Acceleration of 3.0 that seemed to work fairly well. I suggestion you experiment a little using a small number of iterations, such as 10.

    Hope this helps,

    -Allan

Children
No Data