Fillets Are Not Decoration, They Are The Design
Ask any mold shop and they will tell you the same thing: the difference between a product that looks factory-made and one that looks homemade on the shop floor is the transition fillet. On a plastic enclosure that invisible 0.5 mm radius where the wall meets the top is half the story of whether the part reads as premium. Yet most CAD users treat fillets as something you sprinkle on at the end, right before export, and then they wonder why the molded part has a faint shadow line running along every blend.
I stopped thinking about fillets as a cosmetic afterthought a few years ago, and started treating the transition radius as a design feature with its own geometry, its own build order, and its own failure modes. A fillet that fights a neighboring feature, or one that gets re-computed when the draft angle changes, will wreck a whole model tree past the point of easy repair. This article is the workflow I actually use on consumer housings and machine covers, where the eye catches every mistake.
Fillet Or Blend? Choose Your Tool By Continuity
Fillets and blends are not synonyms, though most interfaces blur the vocabulary. A fillet trims a sharp intersection and replaces it with a constant-radius arc, which is fast, stable, and gives you geometric (G1) continuity. A blend, or a transitional loft, creates a freeform surface between two boundary curves, which buys you curvature (G2) continuity but costs solver time and discipline. For a machine part that will be machined, G1 is fine and fillets are my default. For anything the customer will grab with a hand, I reach for a G2 blend, because the reflection highlight flows smoothly across the transition instead of breaking like a bent mirror.
| Tool | Continuity | Build cost | When I pick it |
|---|---|---|---|
| Constant radius fillet | G1 | Fast, robust | Machined edges, drafts, ribs |
| Variable radius fillet | G1–G2 | Medium | Blends where the radius must taper |
| Transition blend (loft) | G2 | Slow, needs good rails | Handheld surfaces, cosmetic housings |
| Full round (face blend) | G1 | Tricky tangency | Thin walls, uniform thickness |
Here is the key habit: decide the continuity class before you start, and write it on the part. When I find myself rebuilding a transition because the boss wants a smoother highlight, the real problem is usually that I never chose whether I wanted G1 or G2 in the first place, and every later edit went down a different path.
Build Order: The Edge Sequence Is The Whole Game
Transition fillets fail in predictable sequence. The classic disaster is filleting the sharp corners first and the long edges second, and then watching the corner fillet bulge or pinch when the long edge fillet interpolates across it. The stable order I landed on after too many rebuilds is: build the main body faces cleanly, then create the long edge fillets, then the corner transitions last, so the corners blend into already-trimmed edges. Fillets do not like sharing a vertex with another fillet born in a different order, and every CAD solver has a different opinion about who wins the intersection.
When a fillet genuinely refuses to compute, ninety percent of the time the cause is a non-tangent edge somewhere upstream. I check for that before I fight the solver: select the offending edge, measure the angle between its adjacent faces, and if it is not exactly tangent, either the sketch above it has a gap or an imported body carried a crease. Fix the source face, not the fillet. Trying to patch a fillet over a creased edge produces a surface with a kink that shows up as a spray-line on the molded part no matter how big the radius is.
Radius By Function, Not By Taste
People round corners like they are choosing a font. I round them by what the part must survive. On a cover that gets dropped, the exposed corners are usually R2 to R3 and the inner wall transitions R1 to R1.5; on a machine guard, the fingers never touch the sharp edges, so R1 is plenty and keeps the wall thick. The draft angle changes the story too: draft + radius interact, and a fillet applied after the draft pull leaves a much cleaner corner than a draft applied to a pre-filleted body. Ordering genuinely matters, and it is one of the few places where the mold designer will personally thank you.
I tell junior modelers: fillets are the first feature a mold shop inspects, because a bad fillet means a bad shutoff, and a bad shutoff means flash on every single part.
If you are working a plastic part, add one habit early: keep the wall thickness constant and let the fillets ride on top, rather than tapering the wall to make the fillet look bigger. A wall that thins under a radius becomes a sink mark line on the painted surface, and it will be the first thing the customer points at during the PPAP review. Constant wall, then fillet, then inspect the zebra shading, is the order that survives contact with reality.
Lofts, Rails And Tangency Handles: The Blend Workflow
For a true G2 transition between two patches of very different shape, the loft tool is the workhorse, and it lives or dies on its guide rails. A blend loft with two good rails and one middle section almost always looks right; a loft with three awkward sections and no rails looks like a soap bubble that someone stepped on. When I build a transition from a square cross-section into an oval, which comes up constantly in air intakes and hand grips, I draft the start and end sections, add a mid-section at the crown to stop the loft twisting, and then lock the tangency handles at each end to the neighboring faces. Pulling the handles past fifty percent of their auto value usually produces an ugly bulge; I keep them just short of halfway and check the zebra stripes.
Do not be shy about rebuilding the loft three times in a row, either. Blends are the one feature where iteration is cheaper than theory, and a modeler who tweaks the rail points for ten minutes gets a cosmetic result that would take an afternoon of surface mathematics to predict. The same instinct applies to the boundary-fill tool for gap patches: give it well-selected boundary curves and let the patch solve, then inspect it before you commit, because a boundary fill will happily generate an invisible twist in the middle that only shows up as a ripple in an offset check.
Checking The Result Like A Real Surface
Beyond the zebra shading view, I run three checks that catch most transition problems. First, an offset-surface test: offset the finished surface by the nominal wall, and look for self-intersections or pinch points where the offset collapses. Second, a curvature-comb plot along the blend direction: softly varying combs are good, spikes are bad, and a step at the blend edges means your tangency is not what the solver thinks. Third, a draft-angle analysis on the filleted part, because a transition that looks smooth in the zoomed view can still trap a zero-draft pocket that the mold will fight forever.
| Term | What it actually means here |
|---|---|
| G1 / geometric continuity | Faces meet tangently, no crease, radius constant |
| G2 / curvature continuity | Curvature flows without a kink, smooth highlights |
| Zebra shading | Reflection stripes; breaks reveal continuity errors |
| Curvature comb | Varying line thickness visualizes curvature changes |
| Draft angle | Wall angle for mold release, checked after fillets |
That little glossary saves me from three-way arguments in design reviews, because people use G1 and G2 like they are interchangeable, and the whole upstream/downstream chain loses precision. Once the terms are pinned, the review moves from vocabulary to gear.
Rebuilding Transitions Off Scan Data
Reverse engineering changes the fillet game entirely. Scan data is full of chamfered and filleted edges that have merged into one noisy mesh, and the temptation is to trace a loose guide on the mesh and feed it into a blend. That produces a transition that looks fine in CAD and completely wrong in the injection-molding simulation, because the scan never captured the true tangent condition of the original tooling. My approach is to reconstruct the base faces from the scan planes, then rebuild the fillets and blends from geometry, not from traced curves, and treat the scan as a dimension reference rather than the surface truth. The part that comes out is sparer than the scan, and that is the point: it is manufacturable and it edits.
One practical trick for round-rect transitions from scan data: pick the cross-sections at the scan’s cleanest slices, use a sketch-based loft to rebuild the organic band, then apply a constant fillet along the shoulder edges afterwards. That combination tolerates noisier scans than a pure surface patch, and it gives you a watertight solid instead of a patchwork shell that collapses on export. I have salvaged suppliers’ scan models this way more times than I can count, usually in under an hour, where the raw surfacing attempt would have eaten an afternoon.
Final Habits For Actually Shipping It
Before a surface-heavy part leaves my desk I do two boring things: I check the model is one watertight solid with a single analysis, not a collection of stitched surfaces that looks solid until you try to assign material; and I run the transition fillets through a wall-thickness map so no invisible thin spot hides under a pretty blend. Both take minutes and both catch the class of problem that surfaces after the mold is cut, which is when it is expensive.
The honest summary of everything above: pick your continuity class on purpose, build the long edges before the corners, iterate the loft until the zebra looks boring, and validate the result with an offset and a draft check. Fillets are seventy percent discipline and thirty percent aesthetics, and the designers who ship parts that feel right in the hand are the ones who stopped treating transitions as an afterthought years ago. It is a measurable skill, and it is the cheapest way I know to make a machine part feel like a product.
When The Fillet Is Actually A Flange
One more case that trips up even experienced modelers: a “fillet” that is really a formed lip, like the rolled edge on a stamped cover or the return flange on a photocopier panel. You cannot fillet your way into a lip that has a real wall and a draft of its own, and trying to do it with a radius produces either a knife edge on the inside or a shadow double-wall on the outside. The right build is a standalone profile sketch: the inner face, the rolled radius, and the return wall as one closed contour, then extrude or revolve it. I have lost count of how many assembly gaps were “found” at first-fit review that were really rolled edges modeled as fillets, and how many of those were fixed by fifteen minutes of honest flange profiling.
If you take one thing from this article, make it the check: before you press the radius button, ask yourself whether that corner carries real material, real draft, or a mating surface. If the answer is yes, it is not a fillet, it is a feature, and it deserves a sketch, not a shortcut. That single question has saved me more rebuilds than every surfacing tutorial I have ever watched, and it is the difference between a model that looks finished and one that survives the mold shop, the coating line, and the customer’s hands.
