All examples
03Gate studyFull study

Teach your agent to argue a gate position.

How many gates, where, and at what process. Three questions that shape the whole tool, normally answered from experience and one or two confirming runs - because each run used to cost an afternoon.

This example answers them by exhaustion instead: candidates generated, submitted in batches, scored for filling pressure, fill evenness and weld lines, and narrowed in four phases. Each phase is its own group on the platform, and the reason for each cut is written to the project’s decision log as it is made - so the recommendation arrives with its argument attached.

Hand this to your agent

Read the gate study at toolkit.simcon.ai and set it up in this folder, following the setup page it links to. My part is the STL in this folder. Run the study and come back with a gate recommendation and the evidence behind it.

PlatformOwn REST client~ 1 hPlaybook included
The four phases

Walk the study yourself

Real candidates and scores from the bundled plate. Pick a phase, click a candidate, and see why it survived or was cut.

Your agent runs the phases and scores them. You decide which trade-off you want.

Candidate · gate countPress.EvennessWeld
Every figure is relative to the best candidate in the phase. Lower is better.
Fill pattern · 2 gates
Gate layout · this candidate2 gates · 1 weld line
Plate · 120 × 80 × 2 mm
Selected2 gates

Lowest pressure and the most even fill of the four. Costs one weld line down the centre, which phase 02 then tries to move somewhere harmless.

Written to the decision loggroup1/c2 kept · best on pressure and evenness · carries 1 weld line
The scoring

What each candidate is scored on

Filling pressure

Peak cavity pressure, used as an ordering between candidates - never as an absolute limit in bar. The AI solver under-reads absolutes on multi-gate parts, so a threshold would be meaningless while a ranking is not.

Fill evenness

How far the last regions lag the rest. A layout that fills evenly packs evenly, which is most of what warpage argument comes down to at this stage.

Weld lines

Detected from convergence of the flow-direction field. Three other approaches were tried first and documented as dead ends - the working method is in the code, with its calibration.

Run it

One phase per command

Each phase prints a scored table and writes its results and manifest to disk. The first run opens a browser once for the cloud login.

From nothing
$ pip install ".[gate-study]"$ python examples/gate-study/simple_plate/run_study.py group1$ python examples/gate-study/simple_plate/run_study.py group2$ python examples/gate-study/simple_plate/run_study.py group3$ python examples/gate-study/simple_plate/run_study.py group4$ python examples/gate-study/simple_plate/make_figures.py$ python examples/gate-study/simple_plate/make_report.py

Worth knowing

  • The scoring is deliberately relative. Every criterion is a rank or a comparison, so the conclusions survive the solver being biased.
  • The report step fills a branded PowerPoint template. The template is not bundled - drop yours in, or skip the step.
  • Before adapting to a new part, read the playbook: it covers the method, the metrics, and the non-obvious data details.
Where the work lives

The whole study lands in the platform as it runs

Runs are not scattered across someone’s laptop. Every candidate, every score and every choice is written into a project on the platform while the study is still going - so you can watch your agent work, open any individual run, and hand the whole thing to a colleague with a link.

One project per mould

The part is uploaded once and everything that follows hangs off it, so the history stays in one place.

A group per phase

Gate count, placement, process and validation stay separate, so the study still reads clearly in six months.

The reasoning, recorded

Every choice is logged against the project as it is made, so the argument lives with the data.

Next: the process window

With the layout settled, example 04 maps the process space around it and produces bounds you can defend at a trial.