headshot

Raphaël Gautier

Research Engineer II
Advanced Methods Division

Dr. Raphaël Gautier is a Research Engineer II within the School of Aerospace Engineering at the Georgia Institute of Technology. His interest lies in the development and implementation of computational methods that help solving real-life engineering problems. This includes supervised learning methods to create surrogate models of expensive simulations or physical experiments, Bayesian methods to quantify epistemic uncertainty and create more robust models in the presence of limited observations, as well as more general probabilistic models, uncertainty propagation, engineering optimization, designs of experiments, and adaptive sampling. Dr. Gautier also has experience with modeling and multi-disciplinary analysis.

  • Doctor of Philosophy, Aerospace Engineering, 2016-2022, Georgia Institute of Technology, Atlanta, GA - USA
    • Focused on the creation of surrogate models for expensive analyses with high-dimensional inputs when only few observations are affordable
    • Developed a fully Bayesian method for supervised dimension reduction
    • Extended it to the multi-fidelity context using deep multi-fidelity Gaussian processes
    • Investigated sampling strategies leveraging a low-dimensional feature space
  • Master of Science, Aerospace Engineering, 2014-2016, Georgia Institute of Technology, Atlanta, GA - USA
    • Courses in advanced design methods, optimization, controls, mathematics, management, fluid dynamics, and aerodynamics
  • Engineering Degree, Computer Science & Engineering, 2012-2014, Supélec, Rennes - France
    • French top-tier engineering school in computer science, electronics, and controls
  • Preparatory School, Mathematics, Physics & Chemistry, 2010-2012, Lycée Privé Sainte-Geneviève, Versailles - France
    • Two-year intensive program preparing to nationwide competitive entrance exams

The earlier projects I took part in dealt with the automated design and rapid prototyping of small drones. Starting late 2017 onwards, I joined the Airbus/Georgia Tech Center for MBSE-enabled Overall Aircraft Design and worked on applying surrogate modeling and uncertainty quantification to challenging contexts, such as in the presence of high-dimensional inputs and outputs, expensive analysis costs, or multi-disciplinary analyses.

Strategies for calibrating a low-fidelity multi-disciplinary analysis (MDA) using limited standalone evaluations of a high-fidelity disciplinary analysis, sponsored by Airbus Commercial Aircraft, 2021

  • Context  This effort was part of the years-long research partnership between Airbus and ASDL that aims at demonstrating the value of an MBSE-enabled approach to Overall Aircraft Design. That year, my group focused on topics related to calibration and uncertainty quantification.
  • Task  We were tasked with proposing and comparing options for a flexible approach to calibrating a low-fidelity MDA when only few observations of a high-fidelity version of a single of its constituting analyses are available.
  • Approach  The first step was to develop a benchmark use case meaningful to the sponsor’s stakeholders: in collaboration with a second researcher, I designed and implemented a simplified version of the sponsor’s internal physics-based MDA process using the open-source OpenMDAO-based OpenAeroStruct. Data generation involved the use of an HPC cluster to save time on computations. Then, I used this use case to compare different calibration approaches that we proposed and presented the results in a Jupyter notebook.
  • Outcome  The deliverable was accepted and selected for presentation to the head of Airbus’ flight physics department as part of the annual’s review.

Integration of multi-disciplary analysis and optimization (MDAO) approaches within a model-based systems engineering (MBSE) framework, sponsored by Airbus Commercial Aircraft, 2020

  • Context  This effort was part of the years-long research partnership between Airbus and ASDL that aims at demonstrating the value of an MBSE-enabled approach to Overall Aircraft Design.
  • Task  There were two main components to that year’s objectives. First, further demonstrate the linkage between MBSE and MDAO, including the opportunity for model-based automation. Second, propose an approach for exploring trade-offs encountered when designing analysis processes that support decision-making.
  • Approach  I developed an ontology for MDAO processes, such as optimization or designs of experiments. I implemented the ontology in MagicDraw (for data entry and linkage to MBSE model) and Python (for programatically enforcing relationship between entities), and demonstrated its usage on two use cases. First, I developed a Python software for demonstrating the automatic assembly and execution of MDAO processes using OpenMDAO based on their MBSE specification. Second, I created a dashboard for exploring trade-offs between different MDAO processes based on their specification created using the ontology. Estimates for the resulting uncertainty, accuracy, and total analysis cost were used as trade-off metrics.
  • Outcome  The deliverables were presented and accepted by the sponsor. This work was selected to be presented to the head of Airbus’ flight physics as a highlight of the year’s achievements and was later featureed in multiple workshops held with various Airbus groups.

Research Internship, at GE Research, late 2019

  • Context  I joined the probabilistic design group at the GE research center from August to December 2019 as a research intern.
  • Task  I was tasked with various activities throughout the internship: conducting literature reviews, investigating numerical methods to improve in-house tools, assisting team members to implement probabilistic methods, participating in the development of physics-based probabilistic models, investigating a fully Bayesian approach to supervised dimension reduction methods.
  • Approach  I conducted numerical experiments to assess the potential of preconditioned conjugate gradient method (PCGM) for speeding up the training of Gaussian process regression models. I directly contributed to the development of physics-based probabilistic models for estimating the failure rate of mechanical parts. I developed and implemented a proof of concept for a fully Bayesian approach to supervised dimension reduction relying on approximation by ridge functions.
  • Outcome  The conclusions of the PCGM study were used to inform future developments of in-house tools. The alternate approach and Cython implementation that I proposed for the failure rate models led to dramatic reductions in training times (from minutes to seconds). The Bayesian approach to supervised dimension reduction that I developed enabled the refinement of my PhD topic and also led to further research efforts in this direction.

Uncertainty quantification in the context of MDAO and MBSE-MDAO demonstration, sponsored by Airbus Commercial Aircraft, early 2019

  • Context  This effort was part of the years-long research partnership between Airbus and ASDL that aims at demonstrating the value of an MBSE-enabled approach to Overall Aircraft Design.
  • Task  1) I was tasked with notionally demontrating the impact of uncertainty in analyses’ outputs in the context of an MDA. 2) I was tasked with supporting the demonstration of the parametrization of an MDA driven by MBSE.
  • Approach  1) I implemented a notional MDA using analytical functions in OpenMDAO and prepared the scripts needed for propagating uncertainty using Monte-Carlo. 2) I developed generic wrappers and OpenMDAO blocks in order to easily integrate the analysis codes provided by the sponsor. I designed and implemented the interfaces necessary to automatically assemble and run the codes based on parameters obtained from MagicDraw.
  • Outcome  The deliverables were presented and accepted by the sponsor. The demonstration was the basis for the MBSE/MDAO linkage work undertaken the following year.

Surrogate modeling of analyses with high-dimensional inputs and outputs, sponsored by Airbus Commercial Aircraft, late 2017 and 2018

  • Context  This effort was part of the years-long research partnership between Airbus and ASDL that aims at demonstrating the value of an MBSE-enabled approach to Overall Aircraft Design.
  • Task  Our objective was to investigate alternate surrogate modeling methods suitable to analyses with high-dimensional inputs and outputs in the context of the coupled aerostructural problem.
  • Approach  A literature review directed us to the investigation of the class of High-Dimensional Model Representation (HDMR) methods. I implemented ANOVA-HDMR and Cut-HDMR and applied them to the prediction of an aircraft wing’s static aeroelastic response to changes in shape, internal structure, and flight conditions (~30 input variables). Responses of interest included scalars (lift and drag coefficient) and fields (spanwise distributions for the aerodynamic loads and structural deformations). Proper Orthogonal Decomposition (POD) was used to handle field outputs.
  • Outcome  The performance of the different methods was compared using out-of-sample predictive error metrics. The deliverables were presented and accepted by the sponsor.

Design and prototyping of a novel architecture VTOL UAV, sponsored by the US Special Operations Command (SOCOM), mid-2016 - mid-2017

  • Context  This ambitious one-year project aimed at designing and prototyping a small recoverable unmanned aerial vehicle to perform reconnaissance and payload cargo missions for special operation forces.
  • Task  I was responsible for the majority of the development of the analysis capabilities, as well as part of the vehicle prototyping.
  • Approach  In collaboration with a second research, I developed a Python-based configuration-independent modeling and analysis toolset for small-scale electric-powered UAVs featuring modular disciplinary analysis codes for UAV components (lifting surfaces, motors, propellers, electrical systems).
  • Outcome  The developed framework was used to model the novel coaxial-rotor fixed-wing VTOL UAV configuration under study. It was used to carry out multi-objective optimization using NSGA-II. The optimized configuration was prototyped and partially flight-tested. This project led to two AIAA conference publications.

Development of an Android application for the visualization of Georgia Tech’s campus sustainability metrics (Smart Campus), sponsored by Georgia Tech, mid-2015 - mid-2016

  • Context  As part of its broader sustainability initiative, Georgia Tech fostered an effort to optimize the operation of energy systems on campus by leveraging its own facilities data and teams of researchers.
  • Task  I was tasked with developing an app to 1) enable the visualization of building-level metrics, and 2) enable interactive what-if scenarios to understand the impact of building-level changes to campus-level metrics. The target operating system was Android.
  • Approach  I developed a native Android app displaying the Georgia Tech campus in 3D, which involved handling the drawing of 3D shapes and custom touch actions for interactivity. The first step consisted in displaying building-related information obtained from a database of historical information. The second step consisted in implementing an “edit mode” to modify building-level metrics and automatically propagate these changes.
  • Outcome  The app was deployed on a coffee-table-shaped touchscreen used by Georgia Tech’s vice president for sustainability.

Automated design and rapid prototyping of multi-rotor UAVs (Micro-Autonomous Systems Research), sponsored by the Army Research Laboratory, mid-2014 - mid-2015

  • Context  The Army Research Lab funded the MASR research initiative to investigate ways to rapidly design and build small-scale UAVs tailored to a specific mission on the battlefield. This particular project was carried out in the context of ASDL’s Grand Challenges where a student team is supervised by a research engineer.
  • Task  We were tasked with developing and implementing a process to automate the design of quadrotor UAVs, starting from end-user requirements such as range and endurance, and resulting in sized components (motors, propellers…) and CAD geometry.
  • Approach  I developed the logic used to design the vehicle’s geometry and components and implemented it in Microsoft Excel using VBA. The Excel tool was integrated with Dassault Systèmes’ Catia such that the resulting design parameters were used as inputs to a parameterized quadrotor CAD model developed by other team members.
  • Outcome  Our tool was used to 3D-print functional quadrotors and served as the basis for a continuing research collaboration with the Army Research Lab. This project led to a publication at the AHS forum.