Uncertainty Quantification in Computational Mechanics

Innovative, non-intrusive measurement technologies provide richer data at increasing rates, whilst the complexity of mechanical models, the number of material parameters and our demands increase by the day. The Computational Mechanics community is in the process of developing new approaches to incorporate measurements in predictive simulations. This three-day EUROMECH colloquium focuses on the exploitation of data for increasing confidence in computational mechanics simulations. Key to this endeavor is the process of Uncertainty Quantification (UQ), which defines our potential in building accurate simulations of physical systems. The colloquium aims to facilitate in-depth knowledge transfer and cross-pollination between the domains of UQ and Computational Mechanics.

Uncertainty quantification aims to quantify the trust allocated into the precision of predictive models, or in the accuracy of inference of associated parameters. Since datasets are typically corrupted by noise and may be incomplete, uncertainty comprises an inherent characteristic of the inference procedure. UQ’s relevance becomes apparent when insufficient measurement data is available. This is not limited to niche applications governed by high velocities or aggressive environments. It is even relevant in the simplest of engineering problems, such as for a standard tensile test, if one desires to identify parameter fields instead of a single set of parameters, as well as in material characterization under availability of a limited number of specimens.

When UQ is coupled with an existing model structure, it is typically tied to an identification problem, which ought to be properly formulated in a probabilistic setting. The uncertainty in this case is typically expressed in terms of random variables of which the joint distribution is to be identified. Identifying distributions comes with another issue: efficient numerical frameworks are required since the model needs to be consulted numerous times. The problem formulation in a probabilistic setting and the numerical reduction of the model, often referred to as a surrogate, go hand in hand, since the discrepancy between the original and the reduced model is assumed to be characterized by a stochastic error. Under availability of measurement data, the full or reduced order models may be updated to fit the observed response, either via identification methods or model updating schemes, such as Bayesian frameworks.

This EUROMECH Colloquium aims at overviewing methods and techniques dealing with different aspects of quantifying uncertainty in computational mechanics simulations. Topics relevant to the scope of this colloquium include, but are not limited to, statistical methods for uncertainty quantification, propagation and multiscale inverse problems, Bayesian inference and regularization, regression, projection and extrapolation, real-time data fusion, acceleration methods for large-scale (industrial) applications (model order reduction), model updating and model selection, virtualization of physical systems, and more.

-----------------------------------------------------------------------------------------------------------------------------------------------

In case of questions, do not hesitate to contact: euromechcolloquium618@uni.lu

-----------------------------------------------------------------------------------------------------------------------------------------------

Besides financial support of EUROMECH, the colloquium also receives financial support from the DRIVEN project funded by the European Union's Horizon 2020 research and innovation programme under grant agreement No. 811099.