Computational simulations are useful for predicting the behavior of complex engineering systems, but they are imperfect due to only representing reality as well as it is known. This limitation is embodied by uncertainty in model parameter values as well as model form. Bayesian calibration addresses model parameter uncertainty by estimating the posterior distributions of model parameters based on the likelihood of observing the given data. In some systems, the available experimental data is high-dimensional field data, which is also called spatio-temporal or functional-response data. Calibration is not straightforward with field data since the outputs may vary spatially and/or temporally and contain an underlying correlation structure; in particular, this is a challenge for surrogate modeling and the likelihood construction required for Bayesian calibration. A large field dataset must be distilled down to its true information content and the correlation be addressed before it can be useful in calibration. The current state of the art, termed manual feature selection, involves an analyst manually selecting representative features (e.g., the value of the response at particular points in space or time, or derived quantities like the maximum and mean value) from the large field dataset. Manual feature selection has several drawbacks including: analyst-to-analyst variability, no mathematical underpinning, not automatable, and not scalable as the amount of field data grows. These drawbacks are addressed by an algorithmic mapping based on singular value decomposition (SVD) recently proposed by Neal et al. [1]. In this method, both the simulation and experimental field data are mapped to a low dimensional, uncorrelated latent space where surrogate models are constructed and Bayesian calibration is performed. The current work compares manual feature selection and the SVD-based approach for extracting a subset of low-dimension features from field data to be used in Bayesian calibration. First, a simple analytical problem, with a known truth, is used to assess the accuracy of both approaches. Then, the two methods are demonstrated on a thermal battery application problem containing high-dimensional field responses.&nbsp;References:[1] Neal, K., Schroeder, B., Mullins, J., Subramanian, A., and Mahadevan, S., 2020. Robust Importance Sampling for Bayesian Model Calibration with Spatio-Temporal Data.&nbsp;International Journal for Uncertainty Quantification,&nbsp;(in production).&nbsp;This research is sponsored by Sandia National Laboratories through a Laboratory Directed Research and Development&nbsp; (LDRD) project. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology &amp; Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

Extracting Low-Dimensional Features From Field Data for Calibration

Paper Proceedings: https://asmedigitalcollection.asme.org/VVS/VVS2021/volume/84782

ASME 2021 Verification and Validation (VVS2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.