Engineering Answers / Prototyping & Validation
Most custom silicone parts do not start with a 3D file. They start with a shape someone can describe, a sample in a bag, or a photo of a part that broke. The sequence from there to a mold is more defined than it looks.
You do not need CAD to start, but you do need three things: the function, the interfaces, and a rough form. Function and interfaces can be written and dimensioned; the form can come from a sketch, a foam or clay model, a 3D print, or a physical sample we measure or scan. The supplier generates the 3D model and the drawing, you approve a prototype, then the tool is cut. What breaks the process is not the missing file — it is a missing tolerance, an unnamed sealing surface, or a sample whose material nobody knows.
It rarely means no information. It usually means the information exists in a form that is not a 3D file: a sketch on paper, a photo of a broken part, a sample in a bag, or a verbal description of what the part has to do.
That is enough to start. What it is not enough for is going straight to a tool. The gap between "a shape someone described" and "a cavity that molds at 99% yield" is closed by three specific pieces of information, and only one of them is geometric.
1. Function. What the part does, in one sentence. Seals a 12 mm port against 0.5 bar. Covers a USB-C connector and stays on through 5,000 insertions. Cushions a 200 g module in a 1.2 m drop. Function decides material, hardness, wall thickness and the tolerance that matters.
2. Interfaces. The dimensions of whatever the part touches. A bore diameter, a housing pocket, a connector shell, a screw pitch. These are the dimensions that have to be right, and they are the ones a photo cannot give. If the part seals, the sealing surface has to be named explicitly — not "the outside", but which face, and against what.
3. A rough form. Anything that communicates shape: a sketch with three views, a foam or clay carving, a 3D print from a hobby printer, a physical sample, or a photo with a ruler in frame.
| Step | Who does it | Output | Time |
|---|---|---|---|
| Function and interface brief | You | One page, dimensions where known | 1–2 days |
| Rough form | You, or us from your description | Sketch, carving, print or sample | 1–5 days |
| 3D model | Us | STEP file, modelled for molding | 2–4 days |
| Drawing for approval | Us | Dimensioned drawing with tolerances and finish | 1–2 days |
| Prototype tool and samples | Us | Printed mold, 10–50 parts | 5–10 days |
| Review and iterate | Both | Revised model or approval to tool | 3–10 days |
| Production tool | Us | Steel cavity, first article | 3–8 weeks |
The two steps people try to skip are the drawing for approval and the prototype. Both exist to move a decision earlier, where it is cheap.
A missing tolerance. "About 12 mm" becomes a modelled 12.00 mm with a ±0.1 mm tolerance that nobody agreed to, and the first article is rejected against a number that was invented by the supplier. Name the dimensions that matter and give them a tolerance, even a loose one.
An unnamed sealing surface. If the drawing does not say which face seals, the parting line will land wherever it is easiest to machine, and it will land on the seal. This is the single most common cause of a second tool.
A sample of unknown material. A part in the bag might be silicone, TPE or a soft PVC, and the three behave differently under heat, compression and cleaning. If the sample's material is unknown, we identify it before modelling — a burn test and a hardness reading settle it in minutes.
A design that ignores molding. A shape can be modelled perfectly and still be unmoldable: an undercut on the wrong axis, a wall at 0.3 mm, a zero-draft deep pocket. Modelling for molding is a different skill from modelling for a render, and it is the reason the model should be built by the party that will cut the tool.
The mating part. A sketch plus the housing it fits into removes more uncertainty than a perfect drawing of the part alone, because most of the tolerance that matters lives at the interface, not in the body of the part.
Measuring or scanning a sample gives geometry quickly and accurately. What it does not give is intent.
A sample might have a 0.8 mm wall because the original material needed it, or because the original process could not go thinner, or because it was designed for a load that no longer applies. Copying that wall into a different compound at a different hardness may be right, and may be three times heavier than necessary.
The way we handle it: scan or measure for shape, then rebuild the model against the function and the interfaces you describe. The sample becomes a starting point that saves a week of modelling, not a specification that overrides the design conversation.
Send those and the modelling conversation starts the same day. Send a photo alone and the first two weeks go into questions that a ruler would have answered.
Three things: what the part does, what it fits against, and roughly how big it is. From that we can name a material and a process and give a range. For a firm quote we need the interfaces dimensioned — the bore it seals against, the housing it sits in, the connector it covers — because those dimensions set the tolerance that matters. A photo alone gets you a conversation, not a price.
Yes, by measuring or scanning it and rebuilding the model, with one caveat: we can reproduce the geometry but not tell you why it was designed that way. If the sample is a competitor's part, the copy may carry a feature that only works with their mating component. We build the model from what you tell us the part must do, using the sample as a starting shape rather than as a specification.
Typically two to three weeks for a simple part: a few days to model and agree the drawing, a week for a printed mold and a first cast set, and a few days for your review. Complex geometry with several iterations takes longer, and the iteration is usually about the fit rather than the material. If the schedule is tight, the fastest thing you can do is send the mating part as well as the sketch.