Design for Manufacture and the Cost of a Bad Geometry

There is a curve familiar to anyone who has taken a product to tooling: the cost of changing a design rises by roughly an order of magnitude at each stage. A geometry change on a screen costs an hour. The same change after a mould is cut costs a five-figure sum and six to ten weeks. The same change after first production costs the tooling, the inventory and the launch window together. Nothing about the invention has altered; only the point at which somebody noticed.

What makes this trap so reliable is that the early decisions do not feel like cost decisions. A designer thickens a boss, adds a snap detail on an internal face, rounds a corner the other way. Each choice is defensible in isolation and invisible on a rendering. Collectively they determine how many cavities a tool can carry, how long each cycle takes, how many separate mouldings the product needs and how many seconds of human labour each unit consumes. That is unit cost, and it is largely locked in before anyone has spoken to a factory.

The Tool · Steel Rules

Draft, Wall Thickness and the Parting Line

Injection moulding imposes a small set of non-negotiable geometric constraints, and most first-time products violate at least two of them.

Vertical faces need draft — typically one to two degrees of taper — so the part can release from the steel. A face drawn perfectly vertical will either scuff on ejection or require additional mechanism to free it. Wall thickness should be uniform and, for common thermoplastics, sits somewhere between about 1.2 and 3 millimetres; thick sections cool slowly, shrink unevenly and leave sink marks visible on the show surface. Ribs are the standard fix, and the convention is to keep a rib at roughly 50 to 60 per cent of the adjoining wall so the junction does not become a thick spot of its own.

Then the parting line: the plane where the two halves of the mould meet. Any feature that projects across that plane — a side hole, a hook, an internal lip — is an undercut, and undercuts cannot be moulded by a simple two-plate tool. They need a side action, a lifter or a collapsible core, each of which adds tooling cost, adds moving parts that wear, lengthens the cycle and introduces a new witness mark on the surface. An undercut that could have been redesigned as a through-feature at concept stage is the single most expensive small decision in the field.

Cycle time follows from wall thickness, because cooling dominates it. A well-drafted thin-walled part might cycle in twenty to thirty seconds; a heavy one takes twice that. Multiply by hundreds of thousands of units and the difference is not a detail — it is whether the factory can quote a price that leaves room for a royalty at all.

Part Count · Assembly Seconds

Every Fastener Is a Line in the Cost Model

Serious design for manufacture work usually begins by attacking part count rather than part cost. Each additional component brings its own tool or purchase order, its own inspection, its own inventory line, its own failure mode, and — most consequentially — its own seconds on the assembly bench. Assembly is priced in seconds multiplied by a labour rate, and the arithmetic is unforgiving: eight screws at four seconds each is more than half a minute per unit before anything is tested.

The standard moves are well established. Replace screws with snap fits where the joint does not need to be serviceable. Replace two hinged parts with one part and a living hinge in a suitable polymer. Combine a housing and a bracket into a single moulding with an integral standoff. Use ultrasonic welding or heat staking where a permanent joint is acceptable. Design components so they can only be assembled one way, so that no operator judgement is required and no orientation error is possible.

Each of those moves reduces cost twice — once in materials, once in labour — and often improves reliability, because the most common field failures occur at joints. A product that goes from fourteen components to six is not eight parts cheaper; it is a different proposition to a manufacturer, because it changes the capital required to start and the volume needed to break even.

Tolerance · The Prototype That Lies

Why the Printed Model Fits and the Moulded Part Does Not

A printed prototype is built by adding material and is dimensionally stable as it comes off the machine. A moulded part is formed hot and shrinks as it cools — commonly by around half of one per cent, varying by polymer, by wall thickness and by fibre orientation in filled materials. A design validated only on printed parts can fail on first mouldings for reasons that have nothing to do with whether the concept works.

Tolerance stack-up compounds this. If four components each hold plus or minus a tenth of a millimetre and their variations align in the same direction, the assembly can be four-tenths out at the point where it matters. Practice is to identify which single dimension controls function, hold that one tightly, and deliberately loosen everything else — rather than specifying tight tolerances everywhere, which is the reflex of an inexperienced drawing package and a reliable way to double a quotation.

The corollary is that first-article inspection matters more than any earlier test. The first parts out of production tooling are the first honest information about whether the design works, and they arrive after the money has been committed. Anyone who has spent time in this part of the process describes the same discipline of imagining a form and then interrogating what it would take to produce it repeatably — a habit visible in the way one long-serving figure in the field describes how speculative fiction shapes what inventors think is possible to build, and in a broader profile of decades spent moving independent ideas toward production.

Rights · Moving Geometry After Filing

What Happens to the Patent When the Part Changes

Here is the coupling that catches people. Production optimisation almost always alters geometry, and a patent application describes a specific embodiment. You cannot add new matter to a pending application — if the manufacturable version relies on a structure the original filing never described, that structure is unprotected, and the filing date you have does not cover it.

Which is why good specifications are drafted with ranges rather than single values, describe alternative arrangements for each functional element, and state the function in terms broad enough to survive the redesign that production will demand. Where a change is genuinely new, the route is a further filing — a continuation or a fresh application — with its own later date, and any intervening public disclosure of the earlier version now sits in the record. Sequencing matters: filing after the geometry has stabilised, or filing broadly enough that stabilisation does not escape the claims, is a choice worth making deliberately rather than discovering. This is the same coupling that makes the rights position and the cost model inseparable in a buyer's reading of any submission.

None of this argues for delay. It argues for treating design for manufacture as work that starts at concept stage rather than after acceptance — the discipline that keeps unit cost, tooling capital and claim scope moving together instead of in sequence. Products that reach shelves at a workable price tend to have had a manufacturing conversation early, as the account of how one household mixing device was carried from sketch to production makes clear: the geometry and the paperwork were settled in the same period, not one after the other.

Manufacturability is not a stage that follows invention. It is a constraint the invention has to be shaped around while shaping is still free.

Draft · Walls · Part count · Stack-up