Overlay of the demonstrator bridge (BLEIB) and the numerical model
Source: BAM, Modelling and Simulation Division
Numerical models are increasingly being used to monitor critical traffic structures such as bridges. The numerical models describe as virtual twin the real structure and can be used to assess damage or service life. The unknown model parameters, which are usually not directly measurable, must be estimated, to obtain the best possible predictions with the numerical model. Therefore, a stochastic model identification is required, which is computationally expensive due to its inverse character and often unfeasible in real applications. Efficient surrogate models, such as reduced order models, can be used to increase the efficiency and enable real-time model identification. Since the numerical accuracy of the reduced models influences the result of the model identification, the optimal surrogate model must not only be computationally efficient, but also accurate enough with regard to the identification parameters.
This contribution develops a method to automatically control the computational accuracy of a Proper Generalized Decomposition (PGD) model (a specific surrogate model) in a goal-oriented manner with respect to the parameter identification. In an iterative process, the surrogate’s numerical accuracy is successively increased and the influence on the result of the model identification is examined. The effects are measured by two stochastic metrics, the Bayes factor and a developed criterion based on the Kullback-Leibler divergence. The application of the developed method both to test examples and to a real demonstration bridge shows that the identification of spatially distributed damage with a PGD surrogate model (with a random field for the stiffness parameter) is possible in real time and that the required surrogate’s accuracy can be determined automatically.
The real application example (see picture) is the demonstrator bridge of the BLEIB project of the BAM financed by the Federal Ministry for Economic Affairs and Climate Action.
Isabela Coelho Lima, Annika Robens-Radermacher, Thomas Titscher, Daniel Kadoke, P.-S. Koutsourelakis, Jörg F. Unger (2022) Bayesian inference for random field parameters with a goal-oriented quality control of the PGD forwardmodel's accuracy. Computational Mechanics 70, pages 1189–1210. https://doi.org/10.1007/s00466-022-02214-6
BAM Department Safety of Structures
BAM Division Modelling and Simulation