When Should DFM Begin?

Most teams start too late. They treat DFM as a gate in front of steel, then discover that the enclosure split, the boss pattern and the cosmetic faces are already promises to industrial design or to a customer. A late review can still catch a wall that will not fill. It cannot cheaply rearrange the product. The sums below are hypothetical teaching maths, not BrahmWorks prices, yields, timelines or test results. What a build must contain once that architecture is frozen is set out in How to Take a Hardware Prototype to Production.

At architecture you decide how the product comes apart, and by which processes. One moulding or two. A sheet-metal chassis or a moulded frame. A cable that exits in the draw direction, or a port that will demand a slide for the life of the tool. A battery the user opens, which forces a joint you can reclose, or a battery the factory closes once.

These choices set cost and risk more than any later fillet. A slide is not a detail to be "DFM'd" after the industrial designer has signed a silhouette with a side hole. The side hole is the slide. Moving it to the parting line is an architecture change wearing a detail's clothes. Do it while the other parts, the artwork and the board outline can still move.

You do not need a moulder in every architecture meeting. You do need someone who will say which processes are in play and which features they punish. A zero-draft wall is a decision. Name it as one.

Detail DFM is draft, ribs, bosses, datums, gate and ejector faces, tolerances on the features that function, and the assembly order those features assume. This is real work, and it is the wrong time to discover that the product should have been three parts rather than seven. If the part split is still moving, you are not detailing. You are still in architecture, whether the Gantt chart agrees or not.

Supplier DFM is the layer you cannot finish alone: shrink for the grade they will mould, a texture depth they will actually apply, a press or a brake they own, a panel size their line can handle. Invite that feedback when the architecture is stable enough that comments are about a real part, and early enough that a boss can still move. Feedback that arrives after steel is cut is a change order. Feedback that arrives before the requirements are written is guesswork you will be asked to obey.

Copying a generic draft angle into a concept model, before the resin and the texture exist, makes the CAD look finished. It is not finished. It can also lock fits against a taper nobody has accepted. The same is true of a resin chosen because a previous product used it, before chemical, heat and colour requirements exist. Early DFM sets the questions. It does not pretend to close them with default numbers from a poster.

A second early mistake is refusing to prototype until the production process is fully designed. Printed and machined units may answer functional questions. They are not the production process in disguise. Use them, and label what they cannot teach. Waiting for a perfect package only delays the news that the function itself is wrong.

The teaching product is a countertop appliance. It is not a BrahmWorks product. An industrial design was signed with a flush side button, a cable exiting the rear wall, and a battery door on the same rear face. No process had been named. The mechanical team then assumed injection moulding.

The side button, square to the draw, is a slide or a redesign. The rear cable and the rear door want the parting line in a place the flush front also wants. The battery door wants a seal, and the seal wants a land the slide will interrupt. None of this is a fillet problem. A teaching recovery, not a project history: move the cable exit to the split line, accept a small witness at the button by making the button a separate moulding from the top, and keep the door on the rear with a shut-off instead of a second slide. That recovery is still possible in CAD. It would have been cheaper before the industrial design was shown to a customer as frozen.

If the customer promise cannot move, the architecture pays: a slide, a witness you explain, or factory-sealed cells instead of a door. Those are on time only before the tool. After it, they are scrap and a recovery plan.

Hypothetical teaching maths. Not a quotation and not a BrahmWorks timeline or price.

Assume one side action is the consequence of leaving the button where the sketch put it. Assume three moments when you could still move the button into the draw.

• In architecture, the change is CAD only. Assume 16 hours of design time at an invented INR 2,500 an hour: 16 × 2,500 = INR 40,000. No steel exists.
• After a soft tool is cut, assume a local rework of INR 90,000, and assume 200 parts already moulded are scrap at INR 80 each: 200 × 80 = INR 16,000. Moment cost = 90,000 + 16,000 = INR 1,06,000. The hours of design do not vanish; they sit inside the rework.
• After a hard tool and a released cosmetic, assume the insert is INR 1,50,000 and a customer approval cycle you do not pretend to price here. The steel line alone is already 1,50,000 / 40,000 = 3.75 times the architecture edit.

These assumptions are not a forecast that rework will be needed. They are the shape of the decision. Moving a feature is cheapest while only files exist. A DFM review scheduled for "before tool kick" is on time only if the part split and the draw are still allowed to change at that meeting. If they are not, the review is an autopsy.

• Process class and part split are chosen, and written, before cosmetic sign-off.
• Every side hole, hook and port is marked as in-draw, a slide, or a separate part.
• User-opened joints are known, because they force a reclosable design.
• Detail draft and ribs are not being used to hide an architecture problem.
• Supplier comments have a model stable enough to comment on, and steel that is not yet cut.
• Default angles and a favourite resin are labelled as unset, not as decisions.
• Prototypes in other processes are running, and their limits are labelled.
• A customer-facing picture is not treated as a tool release.
• The person who can change the silhouette is in the room when the draw direction is chosen.

Not if it is an architecture review. You do not need a polished prototype to know that a side port is an undercut. You do need the prototype later, to learn whether the function is worth the process you just chose. Skipping the early pass because "the prototype will teach us" usually means the prototype teaches you a lesson you then cannot afford to apply.

When you are choosing moulding, a short review with someone who cuts tools is worth more than a long internal debate about draft tables. They should see the part split, the cosmetic faces and the quantity. They should not be asked to invent the product. Bring them a proposal and a list of features you suspect are expensive. An open-ended "what do you think" invites a redesign of the brief.

Then DFM begins as a comparison, not as a single checklist. Sketch the same function as a moulding, as sheet metal, and as a machined part, and write what each version punishes. Choosing later, after one version has been rendered and promised, is how the comparison quietly dies. You can leave the resin unset. You should not leave the process class unexamined while the shape hardens.

The architecture pass finishes when the split and the process class are stable. Detail DFM finishes when the maker accepts the model for the tool or fixture you are actually buying. It reopens when a duty test, a resin change or a supplier change moves a feature. A stamped "DFM complete" on a part whose process later changes is a historical note, not a permission to skip the new process.

Bring the architecture, the cosmetic constraints you have already promised, and the processes you think you are choosing between. BrahmWorks can mark which promises are already tool decisions, while the files are still the only thing that would have to move.

Review the next stage with BrahmWorks