Multidisciplinary design optimization · Blackbox optimization · Nonlinear optimization · Machine learning · Engineering design

Ahmed Bayoumy, PhD

I work on algorithms that find good solutions to expensive blackbox and multidisciplinary design optimization problems using as few evaluations as possible, and on AI methods that accelerate those algorithms without compromising the solutions they return.

Advanced Software Engineer and Optimization Specialist, Siemens Digital Industries Software
Course Lecturer, MECH 559 Engineering Systems Optimization, McGill University
Montreal, Canada

Figure 1. A simplified mesh adaptive direct search, the method implemented in OMADS, minimizing the Rosenbrock function. Each iteration polls the incumbent along a randomly rotated orthogonal basis; the frame shrinks after an unsuccessful poll. It runs live in your browser.

About

Portrait of Ahmed Bayoumy

Since 2020 I have been at Siemens Digital Industries Software in Montreal, working on the optimization engine of Simcenter HEEDS. My job there is to turn research in derivative-free optimization, distributed multidisciplinary design optimization (MDO), and AutoML into features that engineers use on real design problems. In Fall 2026 I am also the instructor of record for MECH 559, Engineering Systems Optimization, at McGill.

I completed my PhD in Mechanical Engineering at McGill in 2019, in the Systems Optimization Laboratory with Prof. Michael Kokkolaras. My thesis introduced the Relative Adequacy Framework, which decides which of several models of a problem to evaluate at each step of an optimization run. Before that I spent five years in industry in Cairo, at Bechtel's PGESCo and at EIE Group, doing piping stress analysis and writing finite element design software, and earned an MSc from Cairo University on the dynamics of large wind turbine blades.

Interests
Multidisciplinary design optimization (MDO) of complex engineering systems, derivative-free and blackbox optimization, Bayesian and surrogate-based optimization, AutoML and hyperparameter optimization, uncertainty-aware machine learning, distributed and multi-agent optimization, robust optimization of engineering systems
Service
Member, AIAA MDO Technical Committee (Education, Research, and Benchmarking subcommittees). Reviewer for Structural and Multidisciplinary Optimization, ASME Journal of Mechanical Design, Scientific Reports, Journal of Computational Science, and Journal of Multi-body Dynamics
Languages
English, Arabic (native), French (basic)

Research

How can learned models accelerate the optimization of expensive blackbox problems while still giving guarantees on the quality of the result?

Many decisions in engineering and science come down to optimizing a function that is expensive, noisy, and only available as a blackbox: a simulation that takes hours, a machine learning pipeline with dozens of hyperparameters, or a system whose parts are designed by separate teams with separate tools. My current work follows three connected directions.

  1. Uncertainty-aware surrogate and Bayesian optimization

    Criteria, built on the relative adequacy and trust-region ideas from my thesis, that decide when a surrogate (a Gaussian process, an ensemble, or a neural network with calibrated uncertainty) can stand in for a true evaluation, combined with direct search so the resulting methods keep their convergence guarantees.

  2. Automated machine learning and algorithm selection

    Choosing and tuning a model or an optimizer is itself an expensive blackbox problem. I study derivative-free and multi-fidelity methods for the combined algorithm selection and hyperparameter optimization (CASH) problem, backed by statistically sound benchmarking.

  3. Distributed and multi-agent optimization

    Large systems are designed by many agents, people or software, that each control part of the problem. I work on ADMM-type and analytical target cascading coordination in which learned models speed up the negotiation between agents, with parallel and asynchronous implementations.

Projects

2025 – present

AI-driven systems MDO

Siemens DISW and McGill University · Mitacs Accelerate

A funded project with Prof. Kokkolaras on data-driven design in distributed MDO architectures: Bayesian systems design, AI-based coordination accelerators, and automatic selection and tuning of models during the design process.

2023 – present

Uncertainty-aware AI accelerators for blackbox optimization

Siemens DISW

Deciding when the predictions of a learned model can be trusted in place of an expensive evaluation, so the optimizer keeps the speed-up without stalling on a plateau caused by over-trusting the model.

2023 – 2024

Systems multidisciplinary analysis and design optimization

Siemens DISW · Simcenter blog post

My open-source DMDO package became the back-end of HEEDS Connect: Systems MDO, which solves distributed problems with non-hierarchical analytical target cascading, sets up workflows through an interactive design structure matrix, and coordinates subproblems in parallel.

2020 – 2023

Solver performance and distributed AutoML

Siemens DISW

Improved HEEDS solvers on high-dimensional problems, integrated parts of OMADS to handle non-smooth design spaces and sharpen multi-objective Pareto fronts, and built a benchmarking suite that compares algorithms statistically across random seeds. Developed distributed CASH methods that cluster the configuration space to sample it more efficiently.

2017 – 2020

Distributed MDO of aero-derivative gas turbines

NSERC CRD · Siemens Energy, McGill, ÉTS

Developed a distributed optimization code based on non-hierarchical analytical target cascading solved with ADMM, used it to converge the coupling between engine performance and secondary air system analyses, built reduced order models for expensive dynamic simulations, and integrated the work into Siemens design automation tools.

2015 – 2019

Relative Adequacy Framework for multi-model management

PhD, McGill University · McGill Engineering Doctoral Award

A framework that estimates relative errors among the available models of a problem, regardless of their nominal fidelity, and uses trust-region principles with mesh adaptive direct search to decide when a cheaper model is good enough. Extended to time-invariant and time-dependent multidisciplinary problems.

Software

I write my research as C, C++, and Python packages that can be used on their own or together. They are all on GitHub.

  • Mesh adaptive direct search for constrained single- and multi-objective blackbox optimization. Under review at ACM Transactions on Mathematical Software.

    pip install OMADS
  • Distributed multidisciplinary design optimization with non-hierarchical analytical target cascading. Back-end of Simcenter HEEDS Connect: Systems MDO.

  • RAF

    Code

    The Relative Adequacy Framework for managing multiple models of different cost and accuracy during optimization.

  • UARAF

    Code

    An uncertainty-aware extension of RAF for multifidelity distributed MDO.

  • Data-driven models for optimization: Kriging, radial basis functions, response surfaces, and weighted ensembles behind one interface, with analytic gradients and uncertainty estimates. Benchmarked on seven engineering problems.

  • samplersLib

    Code

    Sampling and design of experiments.

  • MECH559

    Code

    Jupyter notebooks for Engineering Systems Optimization at McGill.

Publications

Journal articles

  1. 2020
    A. Bayoumy and M. Kokkolaras. A relative adequacy framework for multimodel management in multidisciplinary design optimization. Structural and Multidisciplinary Optimization, 62(4):1701–1720. DOI
  2. 2020
    A. Bayoumy and M. Kokkolaras. Multi-model management for time-dependent multidisciplinary design optimization problems. Structural and Multidisciplinary Optimization, 61(5):1821–1841. DOI
  3. 2019
    A. Bayoumy and M. Kokkolaras. A relative adequacy framework for multi-model management in design optimization. ASME Journal of Mechanical Design, 142(2). DOI
  4. 2014
    A. Bayoumy, A. Nada, and S. Megahed. Methods of modeling slope discontinuities in large size wind turbine blades using absolute nodal coordinate formulation. Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-body Dynamics, 228(3):314–329. DOI
  5. 2012
    A. Bayoumy, A. Nada, and S. Megahed. A continuum based three-dimensional modeling of wind turbine blades. Journal of Computational and Nonlinear Dynamics, 8(3). DOI

Conference papers

  1. 2019
    T. Peoc’h, A. Bayoumy, M. Staniszewski, H. Moustapha, M. Kokkolaras, and F. Garnier. Integration of secondary air system for multidisciplinary design optimization of gas turbines. AERO 2019, Canadian Aeronautics and Space Institute, Laval, QC.
  2. 2018
    A. Bayoumy and M. Kokkolaras. A relative adequacy framework for multimodel management in multidisciplinary design optimization. AIAA Multidisciplinary Analysis and Optimization Conference, Atlanta, GA. DOI
  3. 2017
    A. Bayoumy and M. Kokkolaras. A reference error formulation for multi-fidelity design optimization. ASME IDETC/CIE, Cleveland, OH. DOI
  4. 2015
    A. Papadopoulos, M. Ismail, and A. Bayoumy. Dynamic amplification factor for rigid and flexible piping system due to steam hammer transient load. ASME IMECE, Houston, TX. DOI
  5. 2014
    A. Bayoumy and A. Papadopoulos. Time history steam hammer analysis for critical hot lines in thermal power plants. ASME IMECE, Montreal, QC. DOI
  6. 2013
    A. Bayoumy, A. Nada, and S. Megahed. Use of forward dynamics model for designing large-size wind turbine blades. ASME IMECE, San Diego, CA. DOI
  7. 2012
    A. Bayoumy, A. Nada, and S. Megahed. Modeling slope discontinuity of large size wind-turbine blade using absolute nodal coordinate formulation. ASME IDETC/CIE, Chicago, IL, pp. 105–114. DOI

Under review

  1. 2026
    A. Bayoumy. OMADS: A Python implementation of the mesh adaptive direct search algorithm for constrained and multiobjective blackbox optimization. ACM Transactions on Mathematical Software, under review. Code
  2. 2026
    A. Bayoumy and M. Kokkolaras. Uncertainty-aware relative adequacy framework for multifidelity distributed multidisciplinary design optimization of wind turbine blades. ASME Turbo Expo 2027, GT2027-200150, under review.

In preparation

  1. –
    A. Bayoumy and M. Kokkolaras. A hierarchical design structure diagram for distributed multidisciplinary design optimization. For Structural and Multidisciplinary Optimization.
  2. –
    A. Bayoumy and M. Kokkolaras. Bayesian learning-based enhancement of augmented Lagrangian coordination for distributed multidisciplinary design optimization. For an AIAA journal.
  3. –
    A. Bayoumy and M. Kokkolaras. An adaptive hybrid of reduced-space tree-structured Parzen estimators and a model-free trust region for high-dimensional black-box optimization. For a SIAM journal.

Marked entries are the three papers I consider most representative of my work. A full list is on Google Scholar.

Talks and workshops

  • 2026 3rd AIAA Workshop on Multifidelity Methods for Design and Uncertainty Quantification, Chicago (agenda)
  • 2025 Interview on the role of strategic AI in the engineering design of complex systems, Siemens DISW, Lyon
  • 2024 Multidisciplinary design optimization, Learning Day session, Siemens DISW, Montreal
  • 2022 2nd AIAA Workshop for Multifidelity Modeling in Support of Design and Uncertainty Quantification, Chicago

Teaching

Fall 2026 · Course Lecturer

MECH 559: Engineering Systems Optimization

A graduate and senior undergraduate course at McGill that I designed and teach. It is built around computation: students implement and compare optimization algorithms in Python, then formulate and solve a realistic design problem as their project. Lectures cover optimal design principles, single- and multidisciplinary formulations, and derivative-free methods.

Course notebooks
  • 2022 – 2025
    Guest lecturer, MECH 559, McGill. Derivative-free optimization (slides) and multidisciplinary design optimization (notebook).
  • 2016 – 2019
    Teaching assistant, McGill. MECH 559 Engineering Systems Optimization; MECH 501/502 Analysis, Synthesis and Optimization of Engineering Systems; MECH 539 Computational Aerodynamics; MECH 290 Design Graphics (SolidWorks); MECH 292 Conceptual Design; MECH 400 Engineering Professional Practice.
  • 2023 – present
    Mentoring, Siemens DISW. Supervising research interns through problem formulation, implementation, testing, and technical writing.

How I teach

I want students to leave able to formulate a problem, pick a method for good reasons, and write code that solves it. In practice that means every topic comes with something to implement, projects use realistic problems, and course material stays public so students can keep using it after the term ends. I still work through derivations on the board, because some ideas only become clear when you watch them being built step by step.

Running Scrum for R&D teams taught me to value short feedback loops, and I bring that into the classroom: I ask for feedback during the term, not just at the end, and adjust while the students giving it can still benefit. I state expectations clearly from the first week, phrase instructions in terms of what to do, and keep office hours and email open.

Undergraduate courses I can teach

Introductory programming (Python), C/C++, algorithms and data structures, numerical methods and scientific computing, introduction to machine learning, data science, optimization, software engineering practice.

Graduate courses I can teach

Optimization for machine learning, derivative-free and blackbox optimization, Bayesian optimization and surrogate modeling, AutoML, uncertainty quantification, distributed optimization. I would like to develop a project-based course on AI-accelerated optimization.

Experience

  • 2026 – present
    Course Lecturer · McGill University, Montreal

    Engineering design optimization, AI, and MDO.

  • 2020 – present
    Advanced Software Engineer and Optimization Specialist · Siemens Digital Industries Software, Montreal

    Optimization and AI methods for Simcenter HEEDS: hierarchical and non-hierarchical MDO, AutoML, multifidelity and reduced order models, uncertainty quantification, derivative-free optimization, benchmarking. Supervisor of research interns and Scrum Master for R&D projects since 2023.

  • 2017 – 2020
    Research Assistant · Siemens Energy (NSERC CRD), Montreal

    Distributed MDO architectures for aero-derivative gas turbine design, integrated with AutoOpti, ACES, and NX Open. Research group lead from 2019.

  • 2012 – 2015
    Engineering Group Leader · PGESCo (Bechtel), Cairo

    Wrote PSD, a finite element package for pipe support design. Led design review, stress analysis, and hydraulic and transient analysis groups for combined cycle and thermal power plants.

  • 2010 – 2012
    Mechanical Design Engineer · EIE Group, Cairo

    Piping design and stress analysis; reverse engineering with FARO laser scanning and Geomagic.

Education

  • 2019
    PhD, Mechanical Engineering · McGill University

    A relative adequacy framework for multi-model management in single- and multidisciplinary design optimization. Supported by a McGill Engineering Doctoral Award (2015–2018).

  • 2014
    MSc, Mechanical Design and Production Engineering · Cairo University

    Modeling and simulation of large-size wind turbine blade geometry using the absolute nodal coordinate formulation.

  • 2008
    BEng, Mechanical Engineering · Benha University