Prototype to Production LiDAR: How to Choose a Machining Partner
A LiDAR build moves from a handful of housings and optical mounts to a repeatable production line. This guide is for engineers and buyers who must pick a partner that can hold the transition. Read it and you can judge a supplier on tolerance, lead time, MOQ, traceability and DFM feedback before you send a drawing.

Key takeaways
Which process fits which LiDAR part
Match the process to the feature, not to the part name.
| Part or feature | Process | Typical tolerance | When it is wrong |
|---|---|---|---|
| Optical mount, mirror seat | 5-axis CNC | ±0.005 mm | Casting if bores need reaming |
| Sensor housing, base plate | 3-axis mill + turn | ±0.02 mm | 5-axis if no angled face |
| Lens barrel, bore | Mill-turn | ±0.01 mm | Drilling alone, roundness drifts |
| Prototype shell, 20 pcs | Vacuum casting | ±0.1 mm | For optical datum faces |
| Mounting bracket, 5,000 pcs | Die casting + CNC | ±0.05 mm | For sealed optical cavities |
| Waveguide plate, thin wall | CNC + stress relief | ±0.01 mm | Rush cutting without relief |
| Rotating spindle cap | 5-axis turn | ±0.008 mm | Manual lathe, runout too high |
Pick the partner who can hold the datum, not the one with the lowest piece price
If the optical faces are machined on the same platform as the prototype and inspected to a written plan, the transition is boring. Boring is what you want. If the quote is cheap because the housing was re-designed for casting, the beam alignment will move and the saving will disappear in rework.
What makes prototype to production LiDAR hard
A LiDAR assembly is not a box with a board inside. Light leaves a source, bounces off a spinning or solid-state mirror, crosses a window and returns to a detector. Every metal face in that path is a possible source of angular error. A 0.02 mm step on a mount can move a return spot more than the detector pixel pitch. That is why the first question is never price. It is which faces carry the optical datum.
Most programs break at the handoff from prototype shop to production shop. The prototype was machined on a 5-axis center with one setup and checked on a CMM. The production quote comes from a shop that will cast the housing and ream two bores. The casting has 0.3 mm of draft and a parting line in the wrong place. The datum moves. The beams no longer overlap at range.
Volume changes the material story too. A 6061-T6 prototype housing holds its shape. The same geometry in ADC12 die casting needs wall thickness above roughly 1.5 mm and a draft of 1.5° to 2°. If your optical cavity wall is 1.0 mm, the casting supplier will either thicken it, which changes your optical path, or reject it. Better to know that before tooling is cut.
Then there is the question of who owns the metrology. If the prototype shop measures roundness and the production shop only checks a diameter with calipers, you have two different definitions of a good part. Put the inspection method on the drawing. Say which features go to the CMM and which get a gauge. Write down the sampling rate before the first production lot ships.
- 1Count the setupsEach new setup adds roughly 0.005–0.01 mm of position error.
- 2Flag thin walls earlyBelow 1.5 mm, die casting and vacuum casting both get difficult.
- 3Keep one datum schemePrototype and production must use the same A-B-C datum set.
- 4Freeze the finishHardcoat anodizing adds 20–40 μm per surface and can close a bore.
Criteria that separate real prototype to production LiDAR partners
Start with the machine list, but read it as capability, not count. Sixteen simultaneous 5-axis centers and a Ø400 mm rotary table mean angled optical faces and undercuts can be cut in one setup. A 4,000 mm travel machine matters for a long sensor bar or a gantry frame. If a shop has no mill-turn center, a lens barrel with a bore and an external thread becomes two operations and two chances to lose concentricity.
Lead time is the second filter. A quotation and a DFM analysis inside 12 hours tells you the shop read the file. Production that can start within 24 hours tells you the material is on the floor or in a nearby rack. Parts that ship in 3–5 days is normal for machined prototypes. Ask what the shipping interval is for the tenth lot, not the first.
MOQ is the third filter and it is easy to test. A supplier that accepts one prototype and a 10,000 part run does not need to change its process between them. Same fixture. Same cutting data. Same inspection report. A supplier with a 500 piece minimum will either refuse the prototype or quote it at a rate that hides the setup cost.
Certifications tell you which markets the shop already serves. ISO 9001:2015 covers the general quality system. IATF 16949:2016 is required for automotive LiDAR programs. ISO 13485:2016 applies if the scanner feeds a medical device. ISO 27001:2022 matters when your CAD and point-cloud data leave your site. Ask for the certificate scope, not just the logo.
- 1Process capability dataAsk for Cpk on two critical features, not a sample part.
- 2Material certs6061-T6 and 7075 should arrive with a mill certificate per lot.
- 3Fixture ownershipConfirm who stores and maintains the production fixture.
- 4Change controlAny drawing revision needs a written re-quote and a new first article.
Where LiDAR production cost actually comes from
Machining cost is driven by removal volume, not by part size. A small bracket with a deep pocket can cost more than a large flat plate. On a LiDAR housing, the expensive features are usually the optical bores, the window seat and any face that must be flat to better than 0.02 mm. Those features need a finishing pass at low feed and a separate inspection step.
Surface finish is a cost lever you can control. A cosmetic exterior at Ra 0.8–1.6 μm is easy. An optical seat at Ra 0.2–0.8 μm needs a slower finishing pass, a sharper tool and often a second setup on a different machine. If the drawing calls out Ra 0.4 μm on a face that only touches a gasket, you are paying for nothing.
Anodizing and plating change dimensions. Type III hardcoat adds 20–40 μm per surface. Electroless nickel adds a more uniform 10–25 μm, which helps on a bore. If a mating bore is tolerance-critical, mask it or finish it after coating. Tell the shop which surfaces are cosmetic and which are functional before the finish is applied.
Volume is the last lever. Die casting amortizes tooling over thousands of parts and then needs CNC on the datum faces. Machining from billet has no tooling cost and no minimum, which wins up to roughly a few hundred units. The crossover point depends on geometry. A simple housing crosses over early. A housing with five angled optical faces may never justify a casting.
- 1Split the calloutsFunctional and cosmetic surfaces rarely need the same finish.
- 2Mask critical boresCoating thickness can eat a 20 μm fit clearance.
- 3Stress relief7075 and large 6061 plates should be relieved before finishing.
- 4One finish houseSplitting machining and coating adds a week of logistics risk.
How to run the prototype to production LiDAR handoff
Eight steps, in order. Skipping step 2 or 5 is the usual cause of a failed transfer.
- 11. Define the optical datum setMark A, B and C on the drawing and tie every optical face to them. Use a 3-2-1 scheme on machined surfaces, not on a casting skin. This is the single decision that controls beam alignment.
- 22. Split tolerance by functionOptical seats and bores: ±0.005 mm. Mounting faces: ±0.02 mm. Cosmetic shell and clearance holes: ±0.05 mm to ±0.1 mm. A drawing where everything reads ±0.005 mm gets quoted high or ignored.
- 33. Send STEP plus a 2D PDFThe 3D model drives toolpath. The 2D sheet carries datums, finish callouts and inspection notes. Ask for a DFM report with the quote and read it before you compare prices.
- 44. Prototype on the production processMachine the first articles on the same 5-axis or mill-turn platform that will run production. Vacuum casting is fine for a shell, but not for a face that sets the optical axis.
- 55. Inspect with the production methodCMM for position and roundness, gauge for bores, surface roughness tester for optical seats. Record the results. A first article without numbers is a sample, not a qualification.
- 66. Freeze the finish before toolingChoose clear or hardcoat anodizing, electroless nickel or powder coat, and state masking zones. Finish changes of 20–40 μm on a bore can scrap a lot.
- 77. Pilot 50 to 200 unitsRun a pilot lot on the production fixture and track Cpk on the two most critical features. Watch for tool wear drift after roughly 100 parts and adjust offsets before the main lot.
- 88. Lock the change processAny revision gets a written re-quote, a new first article and a fixture review. Keep the NDA and the inspection reports in the same folder as the released drawing.
Questions engineers ask before releasing a supplier
What tolerance should I put on a LiDAR optical mount?
Start at ±0.005 mm on the bores and faces that set the beam path, and loosen everything else. A mount that positions a lens or mirror needs both tight size and tight position, so call out GD&T position relative to the datum set, not just a diameter.
If the mount only holds a PCB or a cover, ±0.05 mm is usually enough. Over-tolerancing the whole part raises the price without improving alignment.
Can a prototype shop also run production?
Yes, if it machines the prototype on the same platform it will use for production and writes the inspection plan at the same time. The risk appears when the prototype is made one way and production is subcontracted to a casting house with a different datum scheme.
Ask the shop to name the machine, the fixture and the inspection method for the tenth lot before you approve the first one.
How do I keep the design confidential?
Sign an NDA before you release CAD, and confirm how the shop stores and transmits files. A supplier with ISO 27001:2022 has a documented information security system, which is a reasonable filter for point-cloud and optical design data.
Keep the optical prescription and the mechanical model in separate releases where possible, so a shop that only machines housings never holds the full optical stack.
When should I switch from machining to die casting?
Switch when the annual volume justifies tooling and the geometry allows 1.5° to 2° of draft with walls above about 1.5 mm. Simple housings and base plates cross over at a few hundred to a few thousand units per year.
Do not switch an optical cavity. Keep the optical faces as machined inserts or machine them after casting, because a parting line and draft will not hold ±0.005 mm.
What surface finish do LiDAR parts need?
Interior optical seats benefit from Ra 0.2–0.8 μm to reduce scatter and to seat a lens or mirror cleanly. Exterior cosmetic surfaces are fine at Ra 0.8–1.6 μm, and non-critical internal faces at Ra 1.6–3.2 μm.
Every tighter finish callout adds a finishing pass and inspection time. Mark only the faces where light actually travels or where a seal must sit.
How do I compare two quotes fairly?
Compare the DFM notes first, then the process, then the price. A quote that lists the machine, the fixture, the inspection method and the finish supplier is comparable. A quote that is a single number is not.
Ask both shops to quote the same revision, the same lot size and the same finish. If one shop quietly loosened a tolerance to hit the price, the DFM report will show it.
Send a drawing and get a DFM report back
We quote from your STEP and 2D sheet, return a quotation with free DFM analysis within 12 hours, and run one prototype or 10,000 parts on the same process. Uploads are secure and confidential.
12-hour quote and DFMNo minimum order quantity100% inspection before shipmentNDA available on request