CAREER: Fast Surrogate Modeling for Design under Uncertainty of Complex Engineering Systems

University Of Colorado At Boulder, Boulder CO

Investigators

Abstract

This Faculty Early Career Development (CAREER) Program grant will establish an integrated research and education program motivated by the challenges associated with the development of predictive, simulation-based methods for the design and optimization of complex engineering system. Complex engineering systems often involve multiple physical phenomena at multiple scales. Such systems dominate current engineering interests, for example, as in the design and manufacturing of materials and energy storage systems. The dynamics of these systems are intrinsically variable due to, for instance, material properties or manufacturing imperfections, so that there is an imperative need to quantify the impact of such uncertainties for accurate performance prediction and design optimization. To this end and with the objective of advancing the current simulation technologies available for design and optimization, this award supports the development of a set of novel theories, algorithms, and software tools for fast characterization and propagation of uncertainty. The increasing significance and societal impact of predictive simulation capabilities in present and future design of engineering systems will additionally form the basis of an outreach and education effort to attract and engage future generations of engineers, especially from women and underrepresented minorities, entering and studying at the University of Colorado, Boulder. The approach for uncertainty characterization and propagation is based on new and scalable surrogate modeling 
schemes, together with effective computational tools for robust design and optimization of complex engineering systems. The idea of surrogate modeling is to construct an approximate (but inexpensive to evaluate) representation of the mapping between the parameters of a model and the performance objectives. This surrogate model is then used for robust design, optimization, sensitivity analysis, or decision making. To enable fast construction of surrogate models, novel deterministic and random sampling schemes along with model reduction approaches will be developed, in the context of sparse and low-rank approximations. The key to the scalability of the new algorithms is that they effectively and automatically identify the lower-dimensional manifold on which the possibly high-dimensional or non-smooth system solution exists. The surrogate modeling tools will be employed to better predict the reliability of lithium ion battery cells, and to facilitate the uncertainty-aware design of their electrodes.