Physics-AI · LPBF · Metal additive

Melt pool
to part.One chain.

The cloud Physics-AI platform for metal AM. SolidNetics predicts distortion, residual stress and microstructure for laser powder-bed fusion — melt-pool physics and grain structure feeding the part-scale answer directly, in a browser, in minutes.

StressForge · σ_vM + distortion · SS316L · layer-by-layer
6
Connected solvers
5M
Element ceiling
4
CAD formats in
0
Local installs
§ 01

Three scales. One prediction.

Molten → solid

One physics chain spanning five orders of magnitude — from the laser spot to the part on the plate.

~200 µm

Melt pool

Transient thermal solve of the laser–powder interaction, per parameter set — not a fitted heat input.

~50 µm

Microstructure

Grain growth through solidification — size, texture, and the anisotropic stiffness and yield that follow.

~100 mm

Whole part

Layer-by-layer solve over the full geometry, driven by the two scales before it.

The difference

Everywhere else, these are separate products. Parameter development sits in one tool, part distortion in another, and the microstructure never reaches the mechanical answer. We build them as a single chain — each stage consumes what the last one emits, with no re-export and no manual handoff.

§ 02

Learn once. Predict at part scale.

High-fidelity physics is too slow to run over a whole part — so we learn it once, then predict in minutes.

01 · Simulate

High-fidelity physics

Melt-pool thermal history and solidification microstructure — the ground truth.

FusionCoreGrainPath
02 · Learn

Train the surrogate

Thousands of runs swept across the process window, fit to a Physics-AI model.

FusionMapGrainMap
03 · Predict

Part scale, in minutes

Residual stress and distortion across the whole part — no meso-scale solve in the loop.

StressForge
04 · Optimize

Close the loop

Search the process window to cut stress and distortion while holding build rate.

ProcessPilot
Zone segmentation · +Z build
S1 · PathWeaver

Scan path, resolved to every voxel

An impeller voxelised and segmented into manufacturing zones — bulk, contour, up- and down-skin — the per-region plan the scan generator hands downstream.

Voxels266,499
Voxel size0.5mm
Contour zone25%
Up / down-skin17%
Build layers84
Temperature · layer by layer
S2 · FusionCore

Melt-pool thermal history

A three-layer LPBF build solved on a graded mesh — the laser melt pool sweeping each track while fresh powder layers deposit above it.

Melt-pool width208µm
Melt-pool depth173µm
Laser power200W
Scan speed1000mm/s
MaterialSS316L
RVE grains · 360°
S3 · GrainPath

Solidification microstructure

Five representative volumes sampled across the part's thermal clusters, each solved cell by cell into a full grain structure — coloured by thermal cluster, grain size and columnar fraction.

RVEs sampled5
Mean grain size53.5µm
Grains / RVE~830
CA cells375,000
MaterialSS316L
Progressive layer activation
Stage 03 · StressForge prediction

Residual stress & distortion, whole part

One SS316L bracket, solved layer by layer on the cloud — the part-scale field you'd otherwise wait on a meso-scale simulation to see.

Max distortion171µm
Peak von Mises611MPa
Yielded volume23%
Build layers62
MaterialSS316L
§ 03

One connected pipeline

Geometry → verdict

Each stage consumes the contract the stage before it emits. No manual handoff, no re-exporting between tools.

CLOSED-LOOP PROCESS OPTIMIZATION trajectory G · R fields stress trains trains surrogates PathWeaver S1 · Scan path FusionCore S2 · Melt pool GrainPath S3 · Microstructure StressForge S4 · Residual stress CertifyAM S5 · Qualification FusionMap Thermal surrogate GrainMap Grain surrogate ProcessPilot S6 · Optimize
Solvers S1–S6 · Internal surrogate engines · Select a stage to open it
§ 04

Minutes, not days

What changes when the whole chain lives in one cloud platform and the expensive physics is learned up front.

Residual stress
Minutes
Part-scale via surrogates — vs. days of chained solves.
Deployment
Zero installs
Runs in a browser tab — vs. a workstation and a licence server.
Qualification
Before you print
Defect maps and pass/fail gates — vs. trial builds.
§ 05

What makes it different

Six pillars · 01–06

Why the chain is built this way — and what it means for the shop floor.

01

Solve once. Predict every part.

We run the melt-pool and microstructure physics once across the process window and train surrogate models on it. Part-scale predictions then evaluate the surrogate instead of re-solving — so a whole part takes minutes, and changing parameters doesn't mean starting over.

02

Microstructure that changes the answer

Predicted grain size and texture set the anisotropic stiffness and yield used in the part-scale solve. Not a separate report — an input. The same chain carries process conditions through to properties.

03

A browser tab, not a licence server

Nothing to install, no workstation, no CAE specialist to keep it running. Elastic cloud compute, pay for what you run, and a price a production shop can approve without a procurement cycle.

04

Your geometry stays yours

Runs execute in a single named region, encrypted and isolated per account, and are never used to train models shared with anyone else. Deleted on request, NDA before first upload. Private-VPC deployment on the roadmap.

05 · You upload

Four things. No meshing.

  • CAD — STEP, IGES or STL
  • Alloy — from the library, or we characterise yours
  • Parameters — power, speed, hatch, layer thickness
  • Build setup — orientation and supports
You get back

A build verdict. In minutes.

  • Distortion field in µm, with recoater-collision risk
  • Residual stress across the whole part
  • Grain size and texture by region
  • Where it will go wrong — before the powder
06 · Next

From prediction to prescription

ProcessPilot searches the process window for the parameter set that minimises distortion at a held build rate. Thousands of trial evaluations are only affordable because the surrogates answer instantly — the same architecture that makes one part fast makes optimisation possible.

§ 06

Where it has to be right first time

Aerospace

Flight-critical parts, qualified before the build.

Medical

One-off titanium implants, right without trial prints.

Tooling

Pre-compensated distortion in conformal-cooling tools.

Energy

Integrity confirmed under extreme thermal load.

§ 07

CAD to qualified build

In the browser
Step 01
Upload & set the process

Import CAD, define the LPBF strategy with surface-based selection.

Step 02
Generate on the cloud

Scan path and simulation domain built on elastic infrastructure.

Step 03
Run & qualify

Solve, then explore defect maps and qualification results online.

Also in SolidNetics

The Core physics module

Beyond additive — general solid mechanics, heat transfer and design optimization on the same cloud: finite elements, physics-informed neural networks and peridynamics.

Metal AM, simulated end to end

Stop qualifying by trial and error.

Predict melt-pool, microstructure, residual stress and defect risk before you commit powder and machine time.