TY - UNPB

T1 - Uncertainty Quantification for Damage Assessment Using Repeated Deterministic Model Updating

AU - Wolniak, Marlene Theresa

AU - Hofmeister, Benedikt

AU - Dierksen, Niklas Paul

AU - Ragnitz, Jasper

AU - Jonscher, Clemens

AU - Hübler, Clemens

AU - Rolfes, Raimund

PY - 2024/10/17

Y1 - 2024/10/17

N2 - Uncertainty exists in every model updating application in structural dynamics and many recent studies point out and aim to overcome the associated challenges. In this context, the measured data always represent one of the major sources of uncertainty, which can be minimized but never fully eliminated. Established examples for uncertainty quantification in model updating include probabilistic Bayesian approaches and non-probabilistic interval or fuzzy approaches. Naturally, these have their own advantages and disadvantages and the focus of this work is to present an alternative to these two approaches.To this end, a repeated deterministic model updating (RDMU) approach based on a sampling method combined with a numerical optimization algorithm is proposed. The idea is to decouple the incorporation of uncertainty from the actual (deterministic) model updating procedure, making the approach proposed independent of previous assumptions about the distribution of the input data. This is done by generating samples of the (distribution of) uncertain input data and, subsequently, using an optimization algorithm for the deterministic model updating procedure based on each of these samples. In this proposed procedure, both the sampling method and the optimization algorithm are freely selectable. Thus, the approach is very adaptable with respect to specialized problem formulations, which often occur in model updating.To demonstrate the RDMU approach, a specific realization is chosen within this work. Therefore, the well-known Monte Carlo sampling method in combination with the global pattern search optimization algorithm are selected. This specific realization of the RDMU approach is referred to as Monte Carlo global pattern search, or MCGPS. To demonstrate its validity, it is compared to the transitional Markov chain Monte Carlo (TMCMC) method, chosen as an established representative of Bayesian model updating methods. The two specific methods are first verified using an analytical system with two degrees of freedom. Subsequently, they are comprehensively validated using an experimental laboratory steel cantilever beam with multiple damage scenarios.In this study, the selected Bayesian benchmark method and the considered realization of the RDMU approach yield similar results for the analytical and the experimental application presented in this work, confirming the validity of the proposed approach. Furthermore, the straightforward adaptability and the independence of previous assumptions about the input data of the presented RDMU approach open up great opportunities for individual problem formulations as well as further modifications and enhancements.

AB - Uncertainty exists in every model updating application in structural dynamics and many recent studies point out and aim to overcome the associated challenges. In this context, the measured data always represent one of the major sources of uncertainty, which can be minimized but never fully eliminated. Established examples for uncertainty quantification in model updating include probabilistic Bayesian approaches and non-probabilistic interval or fuzzy approaches. Naturally, these have their own advantages and disadvantages and the focus of this work is to present an alternative to these two approaches.To this end, a repeated deterministic model updating (RDMU) approach based on a sampling method combined with a numerical optimization algorithm is proposed. The idea is to decouple the incorporation of uncertainty from the actual (deterministic) model updating procedure, making the approach proposed independent of previous assumptions about the distribution of the input data. This is done by generating samples of the (distribution of) uncertain input data and, subsequently, using an optimization algorithm for the deterministic model updating procedure based on each of these samples. In this proposed procedure, both the sampling method and the optimization algorithm are freely selectable. Thus, the approach is very adaptable with respect to specialized problem formulations, which often occur in model updating.To demonstrate the RDMU approach, a specific realization is chosen within this work. Therefore, the well-known Monte Carlo sampling method in combination with the global pattern search optimization algorithm are selected. This specific realization of the RDMU approach is referred to as Monte Carlo global pattern search, or MCGPS. To demonstrate its validity, it is compared to the transitional Markov chain Monte Carlo (TMCMC) method, chosen as an established representative of Bayesian model updating methods. The two specific methods are first verified using an analytical system with two degrees of freedom. Subsequently, they are comprehensively validated using an experimental laboratory steel cantilever beam with multiple damage scenarios.In this study, the selected Bayesian benchmark method and the considered realization of the RDMU approach yield similar results for the analytical and the experimental application presented in this work, confirming the validity of the proposed approach. Furthermore, the straightforward adaptability and the independence of previous assumptions about the input data of the presented RDMU approach open up great opportunities for individual problem formulations as well as further modifications and enhancements.

UR - https://ssrn.com/abstract=4991201

U2 - 10.2139/ssrn.4991201

DO - 10.2139/ssrn.4991201

M3 - Preprint

BT - Uncertainty Quantification for Damage Assessment Using Repeated Deterministic Model Updating

ER -