Machine learning and uncertainty quantification framework for predictive ab initio Hypersonics

Venturi, Simone

Permalink

https://hdl.handle.net/2142/110403 Copy

Description

Title

Machine learning and uncertainty quantification framework for predictive ab initio Hypersonics

Author(s)

Venturi, Simone

Issue Date

2021-01-04

Director of Research (if dissertation) or Advisor (if thesis)

Panesi, Marco

Doctoral Committee Chair(s)

Panesi, Marco

Committee Member(s)

• Dutton, Craig J.
• Stephani, Kelly
• Bemish, Raymond
• Jaffe, Richard L.

Department of Study

Aerospace Engineering

Discipline

Aerospace Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Hypersonics
• Nonequilibrium Flows
• Ab Initio Calculations
• Potential Energy Surfaces
• Dissociation Kinetics
• Thermal Protection Systems
• Uncertainty Quantification
• Reduced Order Models
• Machine Learning
• Data Science
• Bayesian Inference
• Bayesian Neural Networks
• Variational Inference

Abstract

Hypersonics represents one of the most challenging applications for predictive science. Due to the multi-scale and multi-physics characteristics, high-Mach phenomena are generally complex from both the computational and the experimental perspectives. Nevertheless, the related simulations typically require high accuracy, as their outcomes inform design and decision-making processes in safety-critical applications. Ab initio approaches aim to improve the predictive accuracy by making the calculations free from empiricism. In order to achieve this goal, these methodologies move the computational resolution down to the interatomic level by relying on first-principles quantum physics. As side effects, the increase in model complexity also results in: i) more physics that could be potentially misrepresented and ii) dramatic inflation of the computational cost. This thesis leverages machine learning (ML), uncertainty quantification (UQ), data science, and reduced order models (ROMs) for tackling these downsides and improving the predictive capabilities of ab initio Hypersonics. The first part of the manuscript focuses on formulating and testing a systematic approach to the reliability assessment of ML-based models based on their non-deterministic extensions. In particular, it introduces a novel methodology for the quantification of uncertainties associated with potential energy surfaces (PESs) computed from first-principles quantum mechanical calculations. The methodology relies on Bayesian inference and ML techniques to construct a stochastic PES and to express the inadequacies associated with the ab initio data points and their fit. The resulting stochastic surface is efficiently forward propagated via quasi-classical trajectory (QCT) and master equation calculations by combining high fidelity calculations and reduced order modeling. In this way, the PES contribution to the uncertainty on predefined quantities of interest (QoIs) is explicitly determined. This study is done at both microscopic (e.g., rovibrational-specific rate coefficients) and macroscopic (e.g., thermal and chemical relaxation properties) levels. A correlation analysis is finally applied to identify the PES regions that require further refinement, based on their effects on the QoI reliability. The methodology is applied to the study of singlet (11A') and quintet (25A') PESs describing the interaction between O2 molecules and O atoms in their ground electronic state. The investigation of the singlet surface reveals a negligible uncertainty on the kinetic properties and relaxation times, which are found to be in excellent agreement with the ones previously published in the literature. On the other hand, the methodology demonstrated significant uncertainty on the quintet surface due to inaccuracies in the description of the exchange barrier and the repulsive wall. When forward propagated, this uncertainty is responsible for the variability of one order of magnitude in the vibrational relaxation time and of factor four in the exchange reaction rate coefficient, both at 2,500 K. The second part of this thesis presents a data-informed and physics-driven coarse-graining strategy aimed to reduce the computational cost of ab initio simulations. At first, an in-depth discussion of the physics governing the non-equilibrium dissociation of O2 molecules colliding with O atoms is proposed. A rovibrationally-resolved database for all of the elementary collisional processes is constructed by including all nine adiabatic electronic states of O3 in the QCT calculations. A detailed analysis of the ab initio data set reveals that, for a rovibrational level, the probability of dissociating is mostly dictated by its deficit in internal energy compared to the centrifugal barrier. Due to the assumption of rotational equilibrium, the conventional vibrational-specific calculations fail to characterize such a dependence, and the new ROM strategy is proposed based on this observation. By relying on a hybrid technique made of rovibrationally-resolved excitation coupled to coarse-grained dissociation, the novel approach is compared to the vibrational-specific model and the direct solution of the rovibrational state-to-state master equation. Simulations are performed in a zero-dimensional isothermal and isochoric chemical reactor for a wide range of temperatures (1,500 - 20,000 K). The study shows that the main contribution to the model inadequacy of vibrational-specific approaches originates from the incapability of characterizing dissociation, rather than the energy transfers. Even when constructed with only twenty groups and only 20% of the original computational cost, the new reduced order model outperforms the vibrational-specific one in predicting all of the QoIs related to dissociation kinetics. At the highest temperature, the accuracy in the mole fraction is improved by 2,000%.

Graduation Semester

2021-05

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/110403

Copyright and License Information

2021 by Simone Venturi. All rights reserved

Owning Collections