Monday, April 22, 2024 11:00AM
Ph.D. Proposal


Mariam Emara

(Advisor: Prof. Dimitri Mavris)


A Multi-fidelity Modeling Approach for Vehicle Performance Tuning


 Monday, April 22nd 

11:00 a.m.

Collaborative Visualization Environment (CoVE), 

Weber Space Science and Technology Building (SST II) 


Microsoft Teams



Decision-making plays a critical role in engineering and research problems, particularly in vehicle mission studies where precise parameter tuning of the vehicle is essential for achieving performance targets. With the increasing reliance on digital models and virtual testing, it is essential to have reliable models for making decisions. Especially in complex systems with high consequences, balancing accuracy and cost in these digital models poses a significant challenge to allow for trustworthy findings within the limitations of computational and time budgets.

This challenge of accuracy and cost trade off motivates the overarching research question of the current work: What is the minimum acceptable fidelity level required that would enable reliable virtual testing and tuning of vehicles for defined mission studies? Within engineering teams, fidelity decisions are often subjective and non-systematic, or dictated by the software used. Striving for the highest fidelity can lead to unnecessary costs, while extrapolating from a low fidelity model beyond its reliable capabilities can result in misguided decisions.

The present work aims to develop a methodology for determining the required fidelity level for informed decision-making regarding vehicle parameter tuning. To achieve this, the combination of multifidelity modeling and simulation methods and uncertainty quantification techniques are explored as a potential solution for balancing and quantifying accuracy and cost in digital models. multifidelity methods combine results from low fidelity models and high fidelity models to trade off accuracy and cost. The lack of clear and comprehensive studies of cost vs. accuracy for multifidelity models makes it difficult to justify their use over single fidelity models. This work proposes leveraging uncertainty quantification concepts to provide quantifiable and comprehensive analysis of uncertainty in multifidelity models, towards the goal of finding optimum reliable models that balance accuracy and cost.

The presented methodology is implemented for the use case of an automotive vehicle, but the results from the car example can be extended to different fields towards the goal of knowledge transfer across industries. The first step of the proposed approach is creating a library of multifidelity model alternatives. The multifidelity methods explored are surrogate modeling methods, hierarchical methods, and multidimensional methods which involve simulating low fidelity models and high fidelity models in the same simulation environment. The uncertainty in each of the model alternatives is quantified, with focus on epistemic uncertainty. Finally, a cost-accuracy tradeoff is performed employing multi-attribute decision making and optimization techniques to determine the optimum model that ensures reliability in decision-making regarding vehicle tuning. The methods identified to be most effective throughout this approach will be used to establish a systematic and generalized methodology for determining the minimum acceptable fidelity for decision support.



·         Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)

·         Prof. Graeme Kennedy – School of Aerospace Engineering

·         Prof. Daniel Schrage– School of Aerospace Engineering

·         Dr.  Michael Balchanos – School of Aerospace Engineering