Available for R&D & Engineering Collaboration

Andino Börst

(Andino Boerst)

PhD Candidate in Computational Mechanics & Scientific Machine Learning

Barcelona, Spain•Universitat Politècnica de Catalunya (UPC) & CIMNE

I develop physics-derived neural surrogate architectures that accelerate high-dimensional transient structural dynamics by orders of magnitude. By coupling numerical mechanics with latent space learning, my research transforms multi-hour nonlinear finite element simulations into real-time predictive engineering tools for automotive, aerospace, and energy systems.

Download CV Email Me

Andino Börst (Andino Boerst) - Doctoral Researcher in Scientific Machine Learning & Computational Mechanics

Core Focus

Scientific ML & FEM

Surrogate Speedup

> 100x Real-time

Software

PyPI & PyTorch

Affiliation

CIMNE, UPC & SEAT S.A.

Research & Doctoral Thesis

Scientific Machine Learning & Structural Dynamics

My doctoral research at UPC Barcelona and CIMNE investigates how Hamiltonian and energy-consistent physics principles can be embedded inside deep neural networks to produce ultra-fast, numerically stable subdomain surrogates for nonlinear mechanics.

O(10²) inference acceleration

Physics-Derived Latent Surrogates

SPILS-Net Architecture

Developing specialized neural architectures whose internal latent spaces are conditioned on Hamiltonian and elastodynamic energy principles rather than treating dynamical systems as generic black-box regression.

Sub-millisecond resolution

Transient Structural Dynamics

High-Frequency & Shock Response

Formulating time-integration schemes and neural rollout formulations that suppress artificial numerical dissipation and eliminate compounding trajectory drift across thousands of continuous simulation steps.

Exact interface conservation

Subdomain Decomposition & Coupling

Hybrid FE-Neural Solvers

Constructing non-intrusive domain decomposition interfaces where high-fidelity neural surrogates interact seamlessly with legacy FE meshes (FEniCSx, Abaqus, LS-DYNA) under traction and displacement equilibrium.

Superior compression ratio

Data-Driven Model Reduction (ROM)

Beyond Classical POD-Galerkin

Overcoming the Kolmogorov n-width barrier for advection-dominated and localized shock problems where classical linear subspace methods (POD/SVD) struggle to provide compact representations.

Selected Publications

Peer-reviewed journal articles and international conference proceedings.

View all on Google Scholar

2026Computer Methods in Applied Mechanics and Engineering (CMAME)

DOI: 10.1016/j.cma.2026.119234

Accelerating Transient Structural Dynamics via SPILS-Net, a Physics-Derived Latent Space Subdomain Surrogate

Andino Börst, Pedro Díez, Sergio Zlotnik, Fabiola Cavaliere, Gabriel Curtosi, Xabier Larráyoz

Introduces SPILS-Net, an end-to-end physics-derived internal latent space neural network architecture for localized transient structural dynamics. The method isolates localized elastodynamic subdomains, maps their states into a compact latent space, and predicts multi-step transient evolutions with significant computational speedups while strictly respecting interface boundary conditions.

Scientific Machine LearningNeural SurrogatesStructural DynamicsCMAME 2026

PDF Code Repo PyPI

Engineering & Open Source

Industry & Applied Projects

Production scientific computing libraries, automated numerical solvers, and open-source tooling built for high performance and industrial research.

Lead Creator & Maintainer

spilsnet-torch

PyTorch Library for Physics-Derived Latent Space Surrogates

Production-grade Python package published on PyPI. Provides reusable PyTorch layers, temporal latent autoencoders, physics-informed penalty functions, and inference runners for transient structural mechanics.

Key Impact & Results

Packaged and distributed directly on PyPI (`pip install spilsnet-torch`).
Includes GPU-accelerated rollout loops with automated energy-conservation metrics.
Zero-dependency wrapper designed for drop-in coupling with FEM pipelines.

PyTorchPythonScientific MLPyPIAGPL-3.0

GitHub Repository Package / Archive

Lead Author

SPILS-Net Simulation & FEM Benchmark Suite

FEniCSx Multi-Physics Codebase & Results Reproduction

Complete scientific companion repository for the CMAME 2026 paper. Contains automated FEniCSx-based elastodynamic simulation pipelines, synthetic dataset generators, and neural predictor validation suites.

Key Impact & Results

Full one-command reproducibility of all publication figures and numerical benchmarks.
Generates multi-gigabyte high-fidelity transient dynamics datasets in parallel.
Automated Zenodo DOI integration for persistent scientific archival.

FEniCSxPythonMPIHDF5Docker

GitHub Repository Package / Archive

Beyond the Lab

Adventures, Outdoors & Craft

A balanced mindset fuels long-term scientific curiosity. When stepping away from code and compute clusters, you’ll find me on mountain trails, exploring coastal gravel roads, or working with mechanical rangefinders.

Dolomites, Pyrenees & Bavarian Alps

Alpine Ridge Trekking & Mountaineering

High-elevation ridgelines, multi-day self-supported crossings, and via ferratas. Navigating demanding terrain builds the same mental discipline and endurance needed for research.

MountaineeringAlpine RidgesEndurance

Coastal Climbs & Gravel Passes

Gravel & Long-Distance Cycling

Exploring remote terrain on two wheels. Long solo training rides, steep switchbacks, and an appreciation for the mechanical efficiency of lightweight bicycles.

Gravel BikingEnduranceOutdoor Routes

Mechanical Rangefinders & Technical Optics

Analog Photography & Precision Mechanics

Documenting travels on 35mm black & white film with vintage mechanical rangefinders. A deliberate creative outlet that celebrates precision engineering and timeless industrial design.

35mm FilmMechanical CamerasIndustrial Design