Van den BroekComputational Mechanics

Computational mechanics · AI · Engineering data

Rigorous mechanics for engineering teams and the AI they rely on.

Finite element analysis, uncertainty quantification and surrogate modelling for engineering teams, plus verified engineering-mechanics data for training and evaluating language models.

Discuss a project Explore services
  • Nonlinear FEA
  • Buckling & stability
  • Random fields
  • Surrogate models
  • LLM training data
Fig. 1. Random-field imperfection on a thin cylindrical shell. A stochastic buckling analysis evaluates thousands of these.
Dr.-Ing.Doctorate in structural analysis, Leibniz Universität Hannover
MSc, TU DelftMechanical engineering, Delft University of Technology
University of BristolPostdoctoral research; Honorary Senior Research Associate
Peer-reviewedAIAA Journal, Composite Structures, Thin-Walled Structures and more

01 / Services

Three ways to put rigorous mechanics to work

From a quick second opinion to a full stochastic design study, or a steady supply of verified training data.

01

Computational mechanics

Linear and nonlinear finite element analysis of thin-walled, composite and additively manufactured structures.

  • Buckling, post-buckling & imperfection sensitivity
  • Composite and fibre-steered structures
  • Random fields, Monte Carlo & robust design
  • Independent review of analyses
Engineering services
02

AI for simulation

Surrogate models built on sound simulation data, with honest uncertainty estimates. Explore a design space in seconds instead of days.

  • Gaussian-process & neural surrogates
  • Dimensionality reduction of fields and time histories
  • Sobol sensitivity & optimisation under uncertainty
  • Automated simulation data pipelines
AI for computational mechanics
03

LLM training data

Expert engineering-mechanics data for training and evaluating language models, written and verified by a computational mechanics PhD.

  • Original problems with verified answers
  • Reasoning traces, rubrics & preference data
  • Benchmarks & held-out evaluation sets
  • Simulation-in-the-loop agent tasks
Training data services

For AI labs & data teams

Teaching language models to get mechanics right.

Models quote textbook formulas but still miss boundary conditions, sign conventions and knock-down factors. The errors look plausible, and in engineering they matter. Every task I deliver has an answer verified by derivation and, where needed, by simulation.

  • 01Original, contamination-resistant problems with auto-gradable answers
  • 02Step-by-step reference solutions, rubrics and preference judgements
  • 03Agentic tasks that require writing and running simulation code
LLM training data services
task_0142.json answer verified 
{
  "domain": "structural stability · thin shells",
  "level": "graduate",
  "prompt": "Estimate a design buckling load for an axially compressed aluminium cylinder: R = 500 mm, t = 1 mm, L = 1 m, E = 70 GPa, ν = 0.33 …",
  "answer": { "value": 86.5, "unit": "kN", "rel_tol": 0.02 },
  "verified_by": ["closed form", "independent recomputation"],
  "rubric_points": 10,
  "known_failure": "stops at the classical load ≈ 269 kN (unconservative, ×3.1)"
}
Illustrative task record. See the full worked example on the LLM training data page.

02 / Track record

Research that turns variability into better designs

Ten years of peer-reviewed work on stochastic and nonlinear structural mechanics, from method development to experimental validation, with partners in aerospace and wind energy.

stiffness–buckling correlation map

AIAA Journal · 2020

Imperfection statistics as a design tool

Monte Carlo random-field analyses show where a panel is most sensitive to local variation. Redistributing stiffness or thickness along that pattern, at the same average, raised the linear buckling load by up to 29%.

+29% buckling load doi:10.2514/1.J058962
perturbed fibre paths P w stable unstable

Composite Structures · 2021

Nudging a composite panel onto its stable path

Small, targeted changes to the fibre paths steer a panel with asymmetric post-buckling behaviour onto its stable branch and make it less sensitive to fibre misalignment.

more robust post-buckling doi:10.1016/j.compstruct.2021.114011
≈ 4–5× life baseline tailored σ cycles to failure (log)

Progress in Additive Manufacturing · 2022

Longer fatigue life for 3D-printed parts

The same approach tailored the local thickness of 3D-printed open-hole specimens. In fatigue tests they lasted four to five times longer than the baseline design.

≈4–5× fatigue life, tested open access
GP mean ± 2σ simulations design parameter

Aerospace Science and Technology · 2026

Landing-gear loads at surrogate speed

Gaussian-process surrogates with principal component analysis predict nonlinear landing-gear dynamics in early aircraft design, enabling Sobol sensitivity analysis and optimisation under uncertainty. Developed with Airbus engineers.

GP + PCA surrogate doi:10.1016/j.ast.2025.111475

Also: geodesic random fields on arbitrary finite element meshes (Thin-Walled Structures, 2022) and higher-order finite elements for wind turbine blades (AIAA SciTech, 2023).

All publications

03 / Approach

Every number traceable

The same discipline applies to a design recommendation and to a reference answer in a training set.

  1. 01

    Frame

    Pin down the question: loads, boundary conditions, acceptance criteria and the decision the result must support.

  2. 02

    Model

    Use the simplest model that answers it: hand calculation, reduced-order model, nonlinear FEA or surrogate.

  3. 03

    Verify

    Convergence studies, benchmarks and independent checks, so every number can be traced.

  4. 04

    Deliver

    A clear report with the models and scripts, so you can reproduce the results.

Next step

Have a structure to analyse or a model to teach?

Tell me what you are working on. A short call is usually enough to see if I can help.

Get in touch About Sander