Surrogates and Approximate reliability methods

Good Morning All,

Thank you so much for the useful software Uqlab. It is well known that surrogates like PCE or Kriging can be used with reliability methods like MCS, IS…I wish to know if these methods (PCE, Kriging, SVM…) can be implemented with approximate methods like SORM or FORM using uqlab. In this case, I mean for example PCE-SORM or Kriging-FORM . Appreciate your thoughts on this.

Dear Sir

surrogates allow you to replace a computationally costly model by an analytical function so that 10^6 samples or more can be evaluated in no time. In contrast FORM/SORM methods are approximation methods for reliability analysis that aim at reducing to the very minimum the number of calls to the (possibly costly) model/limit state function.
Once you have a surrogate, say an accurate polynomial chaos expansions, it does not make any sense to use FORM which would replace it by a linearized function in the standard normal space! Some simulation method, e.g. crude Monte Carlo if the probability of failure is not too small, or subset simulation in general, should be used with the surrogate.

So: you can in theory combine any surrogate modelling technique with any reliability method, but certain combinations do not really make sense, e.g. FORM or SORM after building a Kriging or a PCE model.
Our recent active learning module allows you to do these general combinations, also including various enrichment (a.k.a. infill) criteria. We recommend PC-Kriging and subset simulation as the default setup.

Best regards

1 Like

Thank you very much Prof for the clarification.