Efficient model identification using a PGD forward model - Influence of surrogate accuracy and converergence approach
- There is a rising attention of using numerical models for effcient structural monitoring and ensuring the structure's safety. Setting up virtual models as twin for real structures requires a model identification process calculating the unknown model parameters, which mostly are only indirectly measurable. This is a computationally very costly inverse optimization process, which often makes it unfeasible for real applications. Effcient surrogate models such as reduced order models can be used, to overcome this limitation. But the influence of the model accuracy on the identification process has then to be considered. The aim is to automatically control the influence of the model's accuracy on the identification. Here, a variational Bayesian inference approach[3] is coupled with a reduced forward model using the Proper Generalized Decomposition (PGD) method. The influence of the model accuracy on the inference result is studied and measured. Therefore, besides the commonly used BayesThere is a rising attention of using numerical models for effcient structural monitoring and ensuring the structure's safety. Setting up virtual models as twin for real structures requires a model identification process calculating the unknown model parameters, which mostly are only indirectly measurable. This is a computationally very costly inverse optimization process, which often makes it unfeasible for real applications. Effcient surrogate models such as reduced order models can be used, to overcome this limitation. But the influence of the model accuracy on the identification process has then to be considered. The aim is to automatically control the influence of the model's accuracy on the identification. Here, a variational Bayesian inference approach[3] is coupled with a reduced forward model using the Proper Generalized Decomposition (PGD) method. The influence of the model accuracy on the inference result is studied and measured. Therefore, besides the commonly used Bayes factor the Kullback-Leibler divergences between the predicted posterior pdfs are proposed. In an adaptive inference procedure, the surrogate's accuracy is iteratively increased, and the convergence of the posterior pdf is analysed. The proposed adaptive identification process is applied to the identification of spatially distributed damage modeled by a random eld for a simple example with synthetic data as well as a small, reinforced bridge with real measurement data. It is shown that the proposed criteria can mirror the influence of the model accuracy and can be used to automatically select a suffciently accurate surrogate model.…
Autor*innen: | Annika Robens-Radermacher |
---|---|
Koautor*innen: | Isabela Coelho Lima, Jörg F. Unger |
Dokumenttyp: | Vortrag |
Veröffentlichungsform: | Präsentation |
Sprache: | Englisch |
Jahr der Erstveröffentlichung: | 2022 |
Organisationseinheit der BAM: | 7 Bauwerkssicherheit |
7 Bauwerkssicherheit / 7.7 Modellierung und Simulation | |
DDC-Klassifikation: | Technik, Medizin, angewandte Wissenschaften / Ingenieurwissenschaften / Ingenieurbau |
Freie Schlagwörter: | Bayes factor; Kullback-Leibner divergence; Model identification; Model order reduction; PGD |
Themenfelder/Aktivitätsfelder der BAM: | Infrastruktur |
Veranstaltung: | The 8th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS) 2022 |
Veranstaltungsort: | Oslo, Norway |
Beginndatum der Veranstaltung: | 05.06.2022 |
Enddatum der Veranstaltung: | 09.06.2022 |
Verfügbarkeit des Dokuments: | Datei im Netzwerk der BAM verfügbar ("Closed Access") |
Datum der Freischaltung: | 27.06.2022 |
Referierte Publikation: | Nein |
Eingeladener Vortrag: | Nein |