The most expensive failure we see on a newly commissioned IR optics line isn't a broken machine. Every station passes its own factory acceptance test, the operators are trained, and the first finished germanium lens still misses drawing by 2–3 fringes. Nobody is wrong, and nothing is fixable — because five suppliers each hit their own datasheet, and no one owned the number that mattered.

That gap is the real subject of any conversation about an infrared optics production line. Cutting, centering, grinding, polishing and AR coating are each well-understood processes. What isn't well-understood, at procurement time, is how the output tolerance of one station becomes the input assumption of the next — and what happens when that assumption is never written down.

Why the stack-up breaks between stations, not inside them

Machine builders specify against what they control. A wire saw vendor quotes kerf, cut tolerance and surface roughness. A generator vendor quotes form error against nominal radius. A polisher vendor quotes Ra and form at 633 nm. All true, all measured, all useless as a system spec.

The finished lens spec is a different object. It's an ISO 10110 drawing with form, centration, surface quality and thickness callouts that only exist after the last process step. Getting there requires that each station leave the part in a condition the next station was actually designed to accept. That's a line-level design decision, and in multi-vendor procurement it usually belongs to nobody.

Three examples of the handoff that gets skipped:

Handoff What the upstream vendor quotes What the downstream station actually needs
Cut → grind Kerf width, cut tolerance Thickness variation (TTV) across the blank, plus predictable stock allowance
Generate → fine grind Form error vs nominal R Subsurface damage depth, so the grind sequence can remove it
Polish → coat Ra, form at 633 nm Surface chemistry and cleanliness at the deposition temperature

None of those right-hand columns appear on a standard machine datasheet.

Stage 1: the saw sets the budget for everything downstream

Blank cutting is where most of the line's material economics and a surprising amount of its tolerance budget get decided.

On closed-loop (endless) diamond wire, our production configuration is 0.35 mm wire running at roughly 0.4 mm kerf. The loop runs one direction continuously rather than reciprocating, which removes the reversal marks that show up on brittle IR materials when the wire changes direction under tension. As-cut roughness typically lands in the Ra 0.6–1.2 µm range on germanium — rough, but that's fine, because the grind sequence was going to remove that stock anyway.

The number people underweight is thickness variation, not kerf. A blank with 8–15 µm TTV across Ø50 mm and a blank with 40 µm TTV both look like acceptable parts on a caliper. The second one forces the generator to take extra stock on one side, which changes the subsurface damage depth asymmetrically, which shows up as astigmatism three stations later. We've traced more than one "polisher problem" back to the saw.

Pitfall: buying the cutting station on kerf alone. Kerf is a material-cost line item. TTV and subsurface damage are a yield line item, and yield is bigger.

Stage 2: centering fails on the datum, not the machine

Mechanical centering to ≤ 20 arc-seconds of decenter is not hard with a decent bell chuck and centroscope verification. Specifying what it's referenced to is where projects go sideways.

If centration is set against the blank's as-cut edge, you inherit every bit of the saw's edge geometry into the optical axis. If it's set against the first ground surface, you inherit the generator's fixturing repeatability instead. Those are different error budgets with different magnitudes, and the correct choice depends on your aperture range and whether you're running spherical or aspheric.

Fair warning: this is the single most common question we get asked after installation rather than before, and by then the fixturing is already built.

Stage 3: you cannot polish out what generation put in

Curve generation on germanium leaves a subsurface damage layer typically 10–30 µm deep, depending on tool grit and infeed rate. Fine grinding and lapping exist to remove it. Polishing does not — polishing removes a couple of microns and improves finish, and if SSD is still present underneath, it opens up as scatter or as a figure that won't stabilize.

So the grind sequence has to be specified against the generator's actual damage depth, not a generic recipe. A 3-step lap that works behind a 40 µm diamond tool will not work behind a 20 µm tool run aggressively to save cycle time, and vice versa — you'll either leave damage or waste twenty minutes per part removing stock that isn't there.

The practical consequence for procurement: the generator's infeed parameters and the lapping grit sequence are one specification, not two. If they're bought from two vendors, someone on your side has to own the interface, and that person needs to be named in the project plan.

The handoffs nobody writes into the PO

Beyond the process chain itself, four line-level items routinely go unspecified:

Stock allowance flow. How much thickness is left at each stage, and who set it. Without a single owner, each station adds its own safety margin and the blank comes out too thin by stage four.
Fixturing and bonding. Blocking adhesive, marble or glass substrate, magnetic clamping — all of it has to survive the coolant and the temperature of the next operation. Germanium's density (5.33 g/cm³) makes this less forgiving than glass.
Metrology wavelength. Measuring form at 633 nm and shipping an optic that has to perform at 10.6 µm is standard practice, but the conversion has to be agreed and documented. If the coating cell measures differently from the polish cell, you get arguments instead of acceptance.
Who signs the FAT. A station-by-station factory acceptance test proves five machines work. A line-level FAT — one representative lens, cut to coated, against one ISO 10110 drawing, before crates ship — proves the line works. Those are not the same test, and only one of them protects your schedule.

This is the case for treating the whole thing as a single procurement. When one supplier owns the drawing flow from ingot to coated lens, the interface specs above stop being optional. We've laid out the station-by-station configuration, site readiness requirements and commissioning sequence for a five-station infrared optics production line — including the parts that usually get discovered on site rather than in the proposal.

What single-source procurement does not fix

It isn't the right answer everywhere, and it's worth being direct about the limits.

If you already run an integrated IR optics line with centering and grinding equipment under warranty, replacing the whole thing to solve a cutting-stage yield problem is bad economics. Replace the lagging station. Turnkey procurement earns its premium when integration burden is high — greenfield capability, a full generational replacement, or a program that demands a single technical owner for the audit trail. It earns nothing when your line is already integrated and working.

It also doesn't remove aperture-range constraints. A line configured for Ø 10–250 mm spherical and aspheric production is not going to quietly handle a 400 mm dome because the PO said "turnkey." Edge cases still need to be specified as edge cases.

And single-sourcing concentrates supplier risk. That's a real trade, not a rhetorical one. The mitigation is a documented FAT procedure and a parts/engineering warranty that survives the commissioning window — not optimism.

Where to start

If you're scoping a new IR optics line or a major expansion, three things are worth settling before any RFQ goes out:

Fix the product mix first. Material, aperture range, spherical vs aspheric ratio, and monthly volume. Every station downstream is sized off those four numbers, and changing them mid-project is what turns a 16-week build into a 12-month one.
Write the tolerance flow before the equipment list. One ISO 10110 drawing, annotated with what each station has to deliver to the next. If a vendor can't quote against it, that tells you something.
Send material for a sample cut. On germanium, ZnSe, ZnS, silicon or sapphire, measured behaviour on your actual material beats any datasheet comparison. Standard turnaround is two to three weeks.

For the full equipment matrix, footprint and utility requirements, and the commissioning sequence week by week, see the Vimfun turnkey IR optics line overview. Technical questions can go directly to our engineering team at [email protected].