Bayesian Inversion with different inputs to the experiments


I would like to know if there is an option in UQLab to perform Bayesian inversion when the inputs to the experiments are different.

For example, in the beam problem provided in the user manual, if we have five experimental data of the mid span deflection for five different loads ‘P’ then how do we go about calibrating the Young’s modulus in UQLab.

Thanks in advance for the help.

Hi Thiagarajan

Welcome to UQWorld! If the data you have at your disposal comes from 5 different experiments that differ only in the applied load p^{(i)}, i=1,\dots,5, you will have to set up your problem slightly differently than described in uq_Example_Inversion_01_Beam.

Assuming you want to calibrate the Young’s modulus E, you should supply 5 different forward models that all take different applied loads p^{(i)} as an input and return the deflection under this load. You can then assign each datum to the corresponding forward model and calibrate for the Young’s modulus across all 5 experiments.

Please have a look at the multiple forward models feature of the Bayesian inversion module (https://www.uqlab.com/inversion-user-manual, 2.5).

Let me know how it goes :smiley:

Hi Wagner,

Thank you so much for the instant reply. I think this answers my question