A generator for pipe reducers that outputs real, weldable geometry.

This wasn't written in one pass. It went through gated phases — prove the toolchain, one hardcoded part, real dimension data, the generator, validation against real parts, a pressure rating, geometry sanity checks — each one checked before moving to the next. A few of the real mistakes caught along the way, because they say more about how the kernel actually works than the finished code does:

• Local vs. world coordinates. CadQuery's revolve() takes its axis in the workplane's local frame, not global coordinates. The first attempt silently revolved around the wrong axis — caught by checking the bounding box against what it should have been, not by trusting the code compiled.
• Booleans don't hand back what you'd expect. fuse() and cut() return a Compound wrapper even for a single solid, unlike revolve()'s bare Solid. A validity check written for one didn't work for the other until it unwrapped and checked what was actually inside.
• The flush side of an eccentric reducer isn't flat. It's tangent along one line and curves away immediately on either side — confirmed by probing the actual solid, not assumed from the construction. Real fittings work the same way; it's not a modeling shortcut.
• Hub length isn't one number. Three real manufacturer parts measured 7.6 mm, 10.2 mm, and 33.9 mm of straight land before the taper starts — for the same nominal size range. No pattern found, so none is claimed.

Concentric: velocity climbs as the bore narrows

Continuity says it has to: as the flow area shrinks, velocity rises to match. This is exactly that — a clean, smooth gradient from ~1 m/s at the large end to ~11 m/s at the small end, no surprises.

Eccentric: the flush side changes the picture

Same idea, different shape. The eccentric offset that makes one side "flush" (see the limitations below) also means the flow doesn't accelerate evenly across the cross-section — velocity concentrates toward one corner instead of the smooth, centered gradient the concentric part shows.

Worth being honest about what these are: default-mesh, first-pass runs meant to show the shape of the physics, not a validated CFD study. Good for the picture, not for citing a specific number.

Every combination here is one of the 128 real parts from the library above — not a live, unchecked computation. The viewer mesh is for looking; the actual CAD deliverable is the STEP file the download button gets you.

Worth being precise about what this section is and isn't. Within what the generator actually models, nothing here has been found wrong — every OD, wall, and offset checked against a real manufacturer part matched, and volume/mass landed inside the tolerance agreed before looking. What follows is scope: things deliberately left out because no reliable source exists for them yet, and a statement about fabrication-readiness — not a list of mistakes.

These models are for layout and reference only. Verify against the governing standard (ASME B16.9, B31.3, and the applicable material spec) before using any of this for fabrication or pressure service. This is student work, not a certified engineering deliverable.

• Hub (straight-land) length isn't reliably sourced across the size range. Three real parts measured 7.6 mm, 10.2 mm, and 33.9 mm — no size or type pattern found. One shared default is used everywhere; treat hub-sensitive numbers (volume, mass, center of mass) as approximate.
• No weld bevel or transition fillet modeling. Real parts have both — confirmed by inspecting the manufacturer STEP files directly — but neither is sourced precisely enough to model without guessing a number.
• The pressure rating (ASME B31.3, ASTM A106 Grade B) is a straight-pipe-equivalent value per end, inherited via the code's acceptance of B16.9 fittings — not a derived stress analysis of the cone itself. Single material only.
• No standard, drawing, or datasheet found anywhere treats the transition's cone angle as a controlled dimension. Real manufactured parts span roughly 3° to the upper 30s° with no documented ceiling — flagged for review where a generated part sits well outside that range, not auto-rejected against an invented number.
• As of this writing: the dimensioned/toleranced drawing for one size hasn't been completed, and the generator hasn't been used on an actual design yet. Both are in progress, not done.