Surrogate-assisted RBDO-No Feasible Point Problem

Dear UQLab

Greeting on your new release: V 1.4
I have one question, in a sentence:
How to achieve convergence in an arbitrary RBDO problem?
In detail, I have a surrogate-assisted RBDO problem with the following configuration:

  • Both of the Two-Level and QMC options are applied separately.
  • HGA used as the optimization algorithm
  • Target beta=4
  • Max optimization iteration: 1000
  • Reliability methods: Both of the MCS and Subset methods are applied separately
  • Both of the PCE and Kriging methods are applied separately.
  • Diffrenet initial samples are defined: 20, 40, 50, 60, 100, 200
  • Different added samples are defined: 10, 20, 30, 40, 60, 100

This RBDO problem consists of 21 hard constraints. I also record the values of constraints which are mostly positive!.
Unfortunately, the “No Feasible Point” is the output.
Is there any opportunity for me to take the advantage of having the guidance of @moustapha or other experts?

Best regards

Dear @ali ,

Sorry for my late reply.

Feasibility is assumed when all the constraints are satisfied (series system). With 21 limit-states, I wouldn’t be surprised if you can’t find a state where all of them are positive. I would suggest to do the following checks to diagnose the issue:

  1. Have you checked convergence in a deterministic setting? This would give you an idea of the relative size of the feasible domain and the margin to achieve this.
  2. I would also start with a smaller reliability index (hence increasing the size of the feasible domain) - that is if your computational model is not too expensive and you can afford wasting some calls to it.
  3. Finally, what about the surrogate models? Have you checked their accuracy ? The enrichment process devised in building the surrogate is unfortunately very basic, especially when it comes to multiple limit-states functions. Does the algorithm stop because the maximum number of samples is reached ? If so I would increase the limit.

Please let me know if any of these checks lead to you a better understanding of the issue (if there is any actually, as you may also be possible that there is no feasible domain under the probabilistic constraints).



Dear @moustapha

I really appreciate your fruitful suggestions and thanks again for your UQLab. To address my problem, I tried to incorporate the “Kriging-assisted RBDO” configuration into the Bracket Structure example of UQLab. Here is my finding:

  • By adding the Bounds to the Environmental Variables via .Bounds command, it seems the inputs of the surrogate-assisted module of UQLab doesn’t compatible with the applied Bound. For instance, please see the:

As can be seen, I assumed InputOpts.Marginals(5).Bounds=[4.75 5.25]; %%%Added by @ali, some of the generated samples of this example for the InputOpts.Marginals(5).Name = 'L'; break the applied bound:
5.47 5.11 4.68 5.29 4.71 5.31 4.91 4.54 4.98 5.38 5.22 4.80 5.07 4.59 5.04 4.89 4.65 4.83 5.41 5.18
The same problem can be seen in other Environmental Variables. Would you mind please see the attached file for further investigation?
uq_Example_RBDO_02_BracketStructure_AA.zip (4.3 KB)
Best regards

1 Like

Hi @ali

This is a bug, thanks a lot for noticing. I did not anticipate the use of environmental variables with bounds. I just made the necessary changes. This should be available on the next UQLab release.
However, since the changes are confined to a single file, I am attaching the file to the end of this post. You can simply download it and replace it with the one in your current installation. You should find the file under the path: UQLABROOTPATH\modules\uq_analysis\builtin\uq_rbdo\Metamodels

A quick note: if you would like to specify the type of augmented space the command should be:
RBDO_KrgOpts.AugmentedSpace.Method= 'hybrid';
RBDO_KrgOpts.AugmentedSpace.Method= 'hypercube';
(which is not the command you are using in the file you sent.)


uq_buildAugmentedSpace.m (6.6 KB)


Dear @moustapha

I would like to extend my sincere gratitude to you and other RSUQ members for your sustainable supports.
Problem solved!

Best regards