AI shape and topology optimization
Algorithms driven by deep neural networks that explore thousands of variants, place material where it is needed and find the best trade-offs between conflicting goals such as cost and performance.
NUORKS srl
Artificial intelligence for engineering: neural networks that design, optimize and generate CAD-ready geometry, speed up simulations and reconstruct images from imperfect data.
Services
We bring deep learning into design and analysis, from concept to CAD file and from simulation to inspection, integrated with the tools your engineering team already uses.
Generative CAD/STEP geometry
Automatic generation of parametric geometry exported to STEP, ready to open in your CAD and hand off to simulation or manufacturing.
Surrogate models for CFD, FEA and energy
Neural networks trained on your fluid, structural or energy simulations that predict performance in milliseconds instead of hours, so you can iterate and optimize faster.
Custom numerical simulation
Lattice Boltzmann CFD solvers, FEA and dynamic building energy models, written to run fast on the GPU, even directly in the browser.
Imaging and inverse problems
Tomographic reconstruction and correction of artifacts from noise, metal and incomplete data with neural networks, for non-destructive testing and image-based measurement.
Industrial computer vision
Visual inspection, defect detection and automatic measurement from images, including on edge devices.
Case study: STEP geometry generated by DNN optimization
For a high-end car manufacturer we built a system that generates STEP geometry from an optimization algorithm based on deep neural networks.
- Problem: exploring the space of possible shapes by hand takes many iterations between CAD and simulation.
- Solution: a deep neural network drives the optimization of geometric parameters; the best solutions are rebuilt as solids and exported to STEP.
- Outcome: the design team receives geometry that is ready for CAD and validation, instead of starting from scratch at every iteration.
- Constraints and targets
design requirements and metrics - DNN optimizer
exploring the shape space - Geometry reconstruction
from parameters to B-rep solid - STEP export
ready for CAD and simulation
Demos
Small working tools that run entirely in your browser: nothing to install, no data sent to any server. They show, on simple problems, the same techniques we use in our projects: energy simulation and multi-objective optimization, topology optimization, fluid dynamics, deep-learning image reconstruction and computer vision.
Home energy retrofit
Hourly simulation of a house in 8 Italian cities and a neural surrogate: the best trade-off between investment and energy bills for insulation, windows, heat pumps and solar panels.
Open the demo →Topology optimization
Supports, loads and keep-out zones on a 2D domain: optimal material layout with DXF, SVG and STEP export.
Open the demo →2D wind tunnel
Real-time Lattice Boltzmann CFD on the GPU: draw an obstacle and measure drag, lift and vortex shedding.
Open the demo →CT reconstruction with deep learning
Simulate a CT scan with noise, metal and few projections: filtered back-projection compared with a U-Net that corrects the artifacts, right on the GPU.
Open the demo →Face recognition
From the webcam, every face gets a numeric fingerprint and a made-up name and is recognized when it comes back into view. Images never leave the device.
Open the demo →Contact
Tell us about your design or simulation problem: we reply within one business day.