Publish Time: 2026-11-13 Origin: Site
If you only read one paragraph, read this one. For an automotive PCBA line in 2026, the three machines that survive a customer audit are the Koh Young Zenith for measured solder volume with an IPC paper trail, the Omron VT-S1080 for wetting-direction judgement across four oblique views, and the I.C.T AI-5146-3D for a Tier-2 plant that needs 510 × 610 mm boards, dual rail and one supplier for print, SPI and AOI on a five-figure budget. Everything below is ranked for the buying situation each machine actually answers, not for a hypothetical best machine.
One thing changed the question this year. AIAG published CQI-38, the AI Vision Inspection Application and Assessment Guideline, in June 2026 — the automotive industry's first guideline written specifically for AI-based vision inspection, and it exists to supplement IATF 16949 at every tier. It asks how you define defect criteria, how you validate the system, how you control false calls and escapes, and how you manage model changes after launch. That is a procurement document, not a technology brochure. It means the demo you should demand is no longer "watch it find a bridge"; it is "show me the measurement, the repeatability, and the record you will hand my auditor."
Table of Contents
TOP 2. Omron VT-S1080: Judging Wetting Instead of Grey Levels
Inside a Real Ramp-Up: The Inspection Stage Nobody Plans For
Seven machines, one table. Every number below is taken from the maker's own published specification or a first-party document; where a maker publishes nothing on height accuracy or gauge capability, that absence is stated in the row rather than papered over.
Machine | Standout spec | Best for | Watch out for |
|---|---|---|---|
#1 Koh Young Zenith (Alpha / UHS) | Moiré true 3D, height accuracy ±3%, multi-projection tall-part inspection to 25 mm | Plants whose customers ask for measured solder volume and a Cpk history | Highest entry cost of the seven; tall-part capability is a paid option |
#2 Omron VT-S1080 | MDMC illumination with MPS phase shift, four oblique views, 12.5 µm, 25 MP direct camera | Japanese and European Tier-1 chains that judge wetting against IPC, not grey levels | Dense parameter set and a price that lands in the flagship band |
#3 I.C.T AI-5146-3D | 15 µm at 60 × 45 mm or 10 µm at 40 × 30 mm, 510 × 610 mm on one rail, dual-rail option, AI programming | Tier-2 suppliers building a first automotive-capable line on a fixed budget | No published height accuracy or GR&R; you must generate that evidence yourself |
#4 Mirtec MV-6 OMNI | 12-projection digital blue Moiré, 25 MP, 45 mm tall parts, four 18 MP side cameras | High-mix plants that need side-view OCR on J-leads and QFNs | Solder-joint teaching is the bottleneck, and false calls bite on short runs |
#5 Viscom iS6059 | Eight angled plus one orthogonal camera, 0.5 µm Z resolution, 30 mm Z range, long-board option to 720 mm | European automotive shops with stable, high-volume product families | 508 × 508 mm standard ceiling; 610 mm boards need the long-board option |
#6 Yamaha YSi-V | 2D, 3D and four-way oblique in one pass, 12 MP, 7 or 12 µm, 80,000-part built-in library | Lines already built on Yamaha mounters that want defect data to stop the placer | Most of the value is locked to a Yamaha line; resolution trails the leaders |
#7 Saki 3Di-ZS3EX | 25 mm height range (40 mm with the Z-axis option), 686 × 1,016 mm XL envelope | Long or oversized automotive boards where height repeatability is the argument | Expect a methodical tuning phase on your first boards before it settles |
Koh Young built the first volume-production 3D AOI and has held the inspection market lead since 2006 by its own account, and the Zenith is why the phrase "true 3D" became a purchasing requirement rather than a marketing adjective. Instead of inferring height from shadows and grey levels, it projects multi-frequency Moiré fringes and reconstructs the surface, then judges the joint against quantitative limits. That distinction matters most in exactly one place: the moment a customer's quality engineer asks how you decided a fillet was acceptable.
A 2D or 2.5D system compares what it sees to what a programmer once accepted. A true 3D system reports a height and a volume, and then applies a limit. Koh Young's published Zenith specification states a height accuracy of ±3% against its own calibration target, and the maker states that its inspection criteria are based on IPC-610 acceptability requirements rather than on a golden board. For an automotive ECU or a body-control module, that is the difference between "it looked like the sample" and "the heel fillet measured 0.31 mm against a 0.25 mm floor".
The practical consequence is in the audit pack, not on the shop floor. When a Tier-1 asks for evidence that your AOI is measuring rather than pattern-matching, a system that outputs height, volume and wetting angle per joint answers in one export. Pair it with a FAT and SAT checklist that records those measurements at acceptance, and the argument is closed before it starts.
Tall parts are where moiré systems historically fell over: a 25 mm electrolytic capacitor or a relay casts a shadow that swallows the joints beside it. Koh Young answers with multi-projection imaging, and states on its own product page that the platform handles components up to 25 mm tall through a multi-projection moiré arrangement rather than a single oblique pass. The base machine is not wrong without that capability — it is simply blind in the shadow, which is fine on a board of 0402s and not fine on a power stage with a heatsink.
So the specification question is not "does it do 3D". It is: what is the tallest part on your board, is it inside the projection envelope, and is the 9-way option on the quotation. Ask for the answer in writing against your own board, with the tallest three components named. If a supplier quotes the base platform for a board with 20 mm cans and does not mention projection count, that is the first thing to challenge.
Two places. The first is budget: this is the most expensive route in the list, and the gap is not marginal once projection options, software modules and a repair station are added. The second is small-line economics. A plant running eight changeovers a day on 300 boards a week will spend more engineering time feeding a metrology platform than it recovers from the measurements. The Zenith earns its price when the board mix is stable and the customer is asking for capability data — not when the line is chasing prototype revenue.
Before choosing the Zenith, check three things: whether your customer's drawing actually calls for measured solder volume or only visual acceptance, whether the tallest part on your board sits inside the projection envelope you are paying for, and whether anyone on your team will own the KSMART data after installation. If the answer to the third is "nobody yet", budget for the training as well as the machine.
Omron's argument is narrower and, for a certain kind of customer, harder to beat. The VT-S1080 does not try to win on projection count or headline speed. It wins on the thing IPC-A-610 actually describes — the direction and extent of solder wetting — and on a Japanese supply chain that has been buying inspection as a process-control instrument for three decades. If your customer's quality manual cites wetting angle rather than "no visible defect", this machine speaks their language.
The S1080 combines MDMC (multi-direction, multi-colour) illumination with a micro phase-shift technique to reconstruct a solder surface from several lighting directions and colours at once. Omron's own documentation describes the output as solder wetting direction, which is the quantity IPC uses to separate an acceptable heel from an insufficient one. On a lead-free joint with a matte, grainy surface, that is a much more stable basis for a decision than the brightness of a specular patch.
The configuration published in Omron's VT-S10 family specification is a 25 MP direct camera at 10 µm, or 12.5 µm on some variants, with a 51.2 × 51.2 mm field of view and a 25.4 mm height range. Board capacity reaches 510 × 680 mm on a single lane, or 510 × 330 mm in dual-lane form. Those are not class-leading numbers in any single column; the claim is that the numbers hold when the board colour, the flux residue and the component finish all change at once.
A top-down camera cannot see the inboard fillet on a gull-wing lead, and it sees nothing of a J-lead. Omron puts four oblique projectors around the optical axis, which lets the system look under the lead without moving the board. On automotive connectors and QFN ground pads, that is the difference between inspecting the joint and inspecting the shadow of the joint.
The honest limit: oblique imaging is not the same as transmission imaging. A BGA ball array is still a job for X-ray, not optics, and no AOI in this list changes that. Omron's own defect list stops where the light stops.
Two things bite buyers after signature. The first is price — the S1080 sits in the flagship band, and the number in a distributor's price list is not the number on a configured quotation with dual lane and the software modules you will actually use. The second is parameter density. Omron's answer to programming effort is AI-assisted tuning, and it works, but the machine still exposes a large parameter surface, and a plant without a dedicated process engineer will use perhaps a third of it. Forum discussion of Omron AOI tends toward the same summary: excellent capability, and you need real trials before you commit.
Before choosing the S1080, do two things: run your worst board — the one with black solder mask, white silkscreen and a tall connector — through a demo, and ask who in your plant will own the tuning after the vendor's engineer leaves. If the honest answer is "the line leader", negotiate the training days into the contract.
This is the entry for a specific buyer, and it should be read that way. If you are a Tier-2 supplier or an EMS plant assembling your first automotive-capable line, the constraint is rarely "which machine measures best" — it is capital, floor space, and how many suppliers you can manage while your process is still maturing. The AI-5146-3D is built for that arithmetic: one platform covering print, SPI and AOI, one service contract, and a board envelope large enough that you do not turn work away in year two. It is not a claim that it out-measures the two machines above it.
The published envelope is 50 × 50 to 510 × 460 mm, extended to 510 × 610 mm in single-rail large-board mode, or 510 × 330 mm with the dual rail fitted. Board thickness runs 0.6 to 6.0 mm, top clearance is 25 mm, and warpage compensation is quoted at ±3.0 mm. For an automotive body module or an LED driver, a 610 mm rail length is the difference between inspecting the board and sending it out.
Optics are 15 µm at a 60 × 45 mm field of view, or 10 µm at 40 × 30 mm, from a 12 MP area camera behind a telecentric lens, with four-direction structured-light projection on top of an RGB+W source. Minimum detectable component is quoted as 01005 chip and 0.3 mm pitch IC. Those figures are in the same band as the machines above; they are not better than them.
The cost that hurts a small plant is not the machine, it is the engineering hours per new board. I.C.T quotes 0.55 s per field of view and an offline programming path where component and joint recognition is trained rather than hand-drawn, with one-click box selection and batch parameter edits. On a line running twenty part numbers a month, cutting first-article programming from a day to a couple of hours changes the payback more than any optical specification on this page.
The same applies to the inspection record. The machine reports through MES and supports barcode and OCR reads, which is the minimum an IATF-driven customer will ask for. What it does not publish is a height accuracy figure or a gauge R&R result — and that gap is the honest reason this entry sits third rather than first.
Because the maker publishes no height accuracy or GR&R number, you have to generate that evidence on your own board, and you have to do it before the machine ships, not after. Three tests close the gap: a gauge R&R on your critical characteristic, a challenge panel with intentional bridges, tombstoned parts and lifted leads run ten consecutive times with zero escapes, and a repeatability check on the same joint measured twenty times across a shift. Run those and the missing datasheet number stops being a risk. Skip them and it becomes one.
Before choosing the AI-5146-3D, price four things first: the dual-rail vs single-rail envelope against your actual panel sizes, the service response time in your country, the spare-parts lead time for camera and projector modules, and the GR&R test above. Those four answers decide whether the value case survives year three.
Mirtec's pitch is geometric: if shadowing is the enemy, project more light from more angles. The MV-6 OMNI uses twelve digital blue-light Moiré projections plus eight-phase coaxial colour lighting, which lets it reconstruct height across a 0 to 25 mm range without caring whether the object is black, white or mirror-finish. On a board with mixed component heights and a dark solder mask, that is a real advantage, and it is why the platform has a large installed base in high-mix contract manufacturing.
A projection-based system fails in one specific way: a tall part casts a shadow, and every joint inside that shadow becomes either a false call or an escape depending on how the thresholds are set. More projectors means the shadow is smaller and the reconstruction is complete from more directions. Mirtec states the 12-projection configuration covers 0 to 25 mm, and with the optional Z-axis and GPU it claims tall-part inspection to 45 mm — the highest figure in this list.
The optics are a 25 MP or 15 MP CoaXPress camera behind a 4, 7.2, 10 or 15 µm lens, with a quoted 3D accuracy of ±3 µm and a published 3D throughput of roughly 2,000 to 4,450 mm²/s depending on lens, according to Mirtec's MV-6 OMNI specification. Read the throughput figure carefully: it is area per second at a given resolution, not boards per hour, and the fastest number belongs to the coarsest lens.
Four 18 MP side cameras are the option that separates this machine from a pure top-down system. They enable side-view OCR and fillet inspection on J-leads, coils and QFNs — the packages where the joint is under the body and the marking faces sideways. On an automotive connector or an inductor, that is the difference between reading a polarity dot and guessing at it.
If your board has through-hole connectors, J-lead SOJs or QFN ground pads, this option should be treated as part of the machine, not as an upgrade to consider later. If your board is all 0402s and QFPs, it is money you can keep.
The most useful evidence on this machine is not in a brochure. In a 2025 SMTNet thread, an operator running an MV-series 3D AOI in a high-variety, low-volume contract shop described the exact squeeze this class of machine creates: several minutes per board, constant false calls, and a tuning choice between "multiple false callouts" and "it starts to miss things". A second contributor with an MV-7xi reported the same thing specifically on solder joints, and said he keeps joint inspection enabled only for high-volume assemblies.
That is not an indictment of the hardware — it is the operating cost of pattern-based joint judgement on a mix with twenty changeovers a week. Before choosing the MV-6 OMNI, measure three things in a demo: minutes per board on your real mix, false calls per thousand joints after tuning, and how long a library cleanup takes when a new package appears. If the numbers do not fit your takt, the machine will sit offline no matter how good the optics are.
Viscom solves the shadow problem with hardware rather than software: one orthogonal camera in the middle, eight megapixel cameras arranged around it. Nothing is reconstructed from a single viewing direction, so there is very little to filter out afterwards. That approach costs floor space and money, and it buys something specific — a 3D result that does not depend on how reflective the part happens to be. It is why the platform holds ground in German and Central European automotive plants where the product family changes rarely and the process window is narrow. One thing to settle before you compare quotes: the S3088 series this platform grew out of is gone from Viscom’s line-up — the current AOI range is the iS6059 and iS6052 series, and any S3088 datasheet you find describes a machine you can no longer order new.
In a pure projection system, the algorithm has to decide whether a bright patch is a solder fillet or a reflection. In an eight-camera arrangement, that question is answered by parallax before any algorithm runs. Viscom’s iS6059 specification lists three camera heads — XM8-II at 8 µm over a 40 × 40 mm field of view, and XMs-II or XMplus-II at 10 µm over 50 × 50 mm — each pairing eight angled megapixel cameras with one orthogonal head on a 0.5 µm Z stage, and the company describes the result as effectively shadow-free 3D inspection. The practical benefit shows up on shiny lead-free joints and on parts sitting next to a tall neighbour.
The trade is throughput per camera euro. More cameras means more data per board and a heavier compute load, and Viscom publishes no single area-per-second figure for the iS6059 — what it does publish is image acquisition 25 percent faster and 26 percent more pixels than the previous generation. Treat those as relative statements, then ask for a measured cycle time on your own board at your own field of view: that number, not the brochure, decides whether one machine keeps pace with the printer ahead of it.
The Z axis is quoted at 0.5 µm resolution with a measurement range up to 30 mm, which is the second-deepest range in this list after Mirtec's optional 45 mm. For an automotive board with a shielded inductor, a relay and a connector on the same panel, 30 mm of usable range means one program covers the whole board without splitting it into zones with different recipes.
Z resolution is not the same as Z accuracy, and the difference matters. A 0.5 µm step tells you how finely the system samples height; it does not tell you how close that height is to truth. Ask for the accuracy figure against a certified step gauge, not the resolution figure from the brochure. Viscom’s broader claim — 3D solder-joint measurement with integrated verification to cut pseudo-defects — is the one to test with a challenge panel rather than accept as a line item.
The iS6059 is published at 508 × 508 mm maximum board size on single-track transport, with pneumatic clamping, automatic width adjustment and 50 mm upper clearance. That is enough for most automotive ECUs and body modules, and short of a 610 mm LED driver or a long instrument cluster board — but here the size is an option rather than a wall. Viscom lists a long-board configuration that takes boards up to 720 mm in length, so a 610 mm panel does not rule the machine out; it means the long-board variant has to be written into the offer instead of assumed away.
Transport is the second constraint: the standard machine is single lane, and the dual-lane arrangement is a separate model in the same family rather than a field option. On a line running two lanes of panels, you either specify the dual-lane machine, buy two, or accept that the AOI is the takt limit. Before choosing the iS6059, do two things: measure your largest board against 508 mm including the panel frame, and model the takt with one lane unless the dual-lane variant is already priced. If either answer is tight, the machine is right and the line is wrong.
The YSi-V is the one machine in this list whose main value is not on the inspection bench at all. It is what happens after a defect is confirmed: Yamaha's M2M function can push a confirmed defect back to the mounter that placed it and stop that machine, which cuts the number of boards built with the same fault from a batch to a handful. If your line is already Yamaha, that feedback path is worth more than a micron of resolution, and it is genuinely hard to replicate with a third-party AOI.
The normal AOI workflow is reactive: inspect, flag, review, rework, and hope the engineer traces the cause back to a nozzle before the next twenty boards run. Yamaha's approach closes that loop inside the line. The YSi-V captures 2D, 3D and four-way oblique images in one pass, and the QA option feeds confirmed defect information to the mounter so it halts. On a component that is being placed wrong because of a bent nozzle or a drifting feeder, that is the difference between ten suspect boards and two.
It also changes who the machine is for. This is not primarily a metrology purchase; it is a process-interlock purchase. The customer who benefits is the one whose defect cost is dominated by repeat escape rather than by undetected single events.
Yamaha publishes a standard inspection library covering roughly 80,000 SMT component types, supplied with the machine, plus AI-assisted library matching that removes the manual step of picking a shape for every new part — both stated in Yamaha's YSi-V datasheet. For a plant introducing new assemblies weekly, library coverage is the single biggest lever on programming time — bigger than camera resolution and bigger than inspection speed.
The hardware backing it is a 12 MP camera at 7 or 12 µm, a board range from 50 × 50 to 610 × 560 mm with a 750 mm length option, and a machine footprint of roughly 1,252 × 1,498 mm. Those are mid-band numbers. The library and the M2M link are not.
Two conditions. If your mounters are not Yamaha, the M2M feedback has nothing to talk to, and the machine becomes an ordinary mid-range 3D AOI with a smaller resolution advantage than its price suggests. If your quality system needs measured solder volume with a Cpk history for an auditor, the YSi-V will give you good images and less quantitative depth than the top three.
Before choosing the YSi-V, check two things: the exact mounter models on your floor and whether they support the feedback function, and your customer's acceptance criteria — wetting judgement or measured volume. If the first answer is "Yamaha and yes", this is a strong buy. If it is "mixed brands", the integration premium is being paid for a feature you cannot use.
Saki’s current 3D AOI line is the 3Di EX series, sold in three sizes as the 3Di-MS3EX, the 3Di-LS3EX and the 3Di-ZS3EX; the BF-3Di this entry used to describe is no longer in the catalogue. The XL model is the reason the family stays in the list: it takes boards up to 686 × 1,016 mm, the largest envelope of the seven, which covers instrument clusters, LED drivers and multi-up panels that would otherwise need a specialised platform. On every other axis Saki is deliberately conservative, and the specification it does not publish is as revealing as the ones it does.
Saki’s 3Di EX specification states the height range as a function of the camera head rather than as one headline number: 25 mm of measurement range with either the 8 µm or the 15 µm camera, expandable to 40 mm when the Z-axis option is fitted. That is a different trade from Koh Young’s multi-projection or Viscom’s camera array — Saki keeps one imaging chain and buys range with a moving axis. The same page gives 4,500 mm²/s at 8 µm and 7,000 mm²/s at 15 µm on an A5-size sample, the fastest published pair in this article, and notes that the camera resolution can be switched on site in under 90 minutes.
The figure this entry used to carry — 2 µm repeatability at 3σ on a 2,500 µm step — belongs to the BF-3Di datasheet and no longer appears in the current specification. The 3Di EX page publishes resolution, range and speed; it publishes no height accuracy and no repeatability. That absence is the thing to raise in the evaluation, because a capability study runs on repeatability, not resolution. Our own comparison of Saki and Pemtron precision claims reads the same gap from the other side. Ask Saki for repeatability across your own height band, in writing, and treat the answer as a pass or fail on the audit criterion rather than a nice-to-have. The same rule applies to the envelope: measure your largest panel against 686 × 1,016 mm before you fall for the number.
Board capacity runs from 50 × 60 mm up to 330 × 330 mm on the M, 510 × 510 mm on the L and 686 × 1,016 mm on the XL, with 40 mm of upper and 60 mm of lower clearance on single-phase 200–240 V power. The XL is a large machine — 1,840 × 1,480 × 1,520 mm and around 980 kg — so check the aisle and the floor loading before you check the price. Components down to 0201 mm are inside the stated capability, and the option list covers dome lighting, side cameras for back-side electrode and connector joints, a UV head for conformal coating inspection and extra-component detection.
Saki also ships automated inspection-program tooling that builds recipes from design and centroid data, and its One Programming approach shares one program set across SPI, pre-AOI, post-AOI and — scheduled for 2026 — AXI. On an NPI-heavy line that is a real saving; on a stable high-volume line it is a convenience.
The honest weakness is time-to-stable. Saki’s workflow rewards a methodical tuning phase on the first physical boards, prioritising long-term stability over immediate setup speed — a trade that suits a plant running the same product for years and frustrates one running twenty changeovers a week. Comparable comparisons of Saki against Viscom make the same point: Viscom tends to reach a usable recipe faster on new CAD, Saki tends to hold it better once tuned.
Before choosing the 3Di-ZS3EX, schedule the tuning, not just the installation. Ask for a written plan covering the first five boards, who signs off the recipe, what the false-call target is at board five versus board fifty, and the repeatability data behind the recipe. If the vendor will not commit to that plan, the machine will still work — it will just take a quarter longer than the project plan says.
The most useful thing we can show you is not another specification table. It is what happens in the weeks around the truck, because that is where most inspection purchases are quietly lost — the machine arrives, the line runs, and then someone discovers that the AOI recipe was written against a golden board that no longer matches production.
In December 2025 our engineers commissioned a complete SMT line for an automotive PCBA manufacturer in Uzbekistan. The plant had limited floor space and a product mix that had to be inspected, not sampled. The configuration covered solder paste printing, placement, reflow, online AOI, PCB handling and the auxiliary stages, and the AOI went in as part of the line rather than as a standalone purchase.
Field | Detail |
|---|---|
Customer industry | Automotive PCBA manufacturing, Central Asia |
Configuration | Loader, automatic stencil printer, pick-and-place, reflow oven, online AOI, conveyors and supporting units, laid out for a space-constrained hall |
Product and pain points | Automotive control boards with mixed package sizes; the customer needed inspection coverage from day one rather than after a ramp period |
Production challenge | Installation, commissioning, operator training and trial production had to happen inside one visit, in a hall with no room for a spare machine position |
Solution | Layout review on arrival, machine-by-machine levelling and power-up, line-level commissioning for transfer and communication, structured operator and maintenance training, then trial production with parameter tuning on real boards |
Verified results | Line handed over after successful trial production with stable inspection results; figures are from our project team, not an independent audit |
What that project adds beyond a specification table is the sequence. The AOI recipe was built and tuned while the printer and the mounter were still being dialled in, which means the first accepted recipe was already adjusted to real paste and real placement offsets rather than to a nominal board. That is the opposite of the common sequence, where AOI is bought last, installed after the line is stable, and then programmed against a process that has already drifted.
If you are planning a similar build, two actions transfer regardless of which machine you choose. Put the AOI in the line layout from the beginning so it has a conveyor position and a takt slot, and hold the first-article inspection before handover rather than after. Our own first-article inspection and changeover guide covers the sign-off sequence in detail.
An automotive AOI is not a faster version of a consumer-electronics AOI. It is a measurement system that has to produce evidence for a third party, and the third party has just been given a new checklist. CQI-38 asks for a defined defect taxonomy, documented system validation, control of false calls and escapes, data retention and a change-control process for the model itself. Four requirements follow from that, and they are the four that separate a shortlist from a purchase order.
Automotive lines inspect every board. That sounds like a throughput problem and is really a false-call problem, because at 100 percent inspection the cost of a nuisance call multiplies by the number of boards. The figure quoted most often in automotive AOI programmes is below 5 percent at board level, and a plant running a thousand boards a shift cannot absorb a system that sits at 8 percent no matter how good its optics are.
Class 3 acceptance is written in terms of fillet height, wetting angle, side-joint length and end-joint width — quantities, not appearances. A machine that judges against a golden image is doing something different from a machine that measures those quantities and compares them to limits, and only the second one can be defended in a PPAP review when the golden board itself was made on a Monday.
Every board needs a serial number, an inspection result, images, and a link to the process data around it. Retail retention expectations for automotive records run to 15 years after end of production. That is a software requirement as much as an optical one: the machine has to export to MES or ERP in a format the quality system can actually store, which is why MES integration requirements belong in the technical evaluation and not in the IT queue after acceptance.
The number that decides whether an AOI is fit for a Class 3 line is not on any datasheet. It is the gauge R&R you run on your own board, against your own tolerance window, and the usual threshold is 10 percent of that window. IATF-driven quality systems expect a documented measurement system analysis for pass/fail inspection, and a challenge panel run ten times with zero escapes is the practical way to produce it. Budget two days and one engineer for this before you sign, not after.
Work through the five questions in order. The first two eliminate most of the market, the third and fourth set the budget, and only the fifth is a technical comparison. Answering them out of order is how plants end up comparing micron figures between machines that were never candidates for the same line.
1. What does your customer's drawing actually require? Measured solder volume with a Cpk history, or visual acceptance to IPC-A-610 Class 3. The first narrows the field to the top three; the second opens it up.
2. What is the largest board on your three-year roadmap, including the panel frame? If it exceeds 508 mm, only the machines with a long-board option stay in — Viscom’s iS6059 reaches 720 mm, and Saki’s XL model reaches 686 × 1,016 mm. If it exceeds 610 mm, most of the list is out and those two become the conversation.
3. How many changeovers per week? Above eight, programming and library maintenance dominate total cost, and AI-assisted programming moves from nice-to-have to the deciding feature.
4. Who will own the recipe after the vendor leaves? A dedicated process engineer can absorb a dense parameter set; a line leader cannot, and should buy the machine with the shortest path from CAD to a usable recipe.
5. What is your false-call budget in operator minutes per shift? Convert it into a rate before you demo. A machine that flags 3 percent of boards on a thousand-board shift costs roughly thirty reviews, and at fifteen minutes each that is most of one operator.
Your situation | Try first | Check before ordering |
|---|---|---|
OEM-nominated Tier-1, customer audits the measurement | Koh Young Zenith, Omron VT-S1080 | Projection count against your tallest part; who owns KSMART or the SPC stack |
Tier-2 building a first automotive line on a fixed budget | I.C.T AI-5146-3D, Saki 3Di-MS3EX | GR&R on your board; service response time in your country |
High-mix EMS, eight or more changeovers a week | Mirtec MV-6 OMNI, I.C.T AI-5146-3D | Minutes per board on your real mix; library cleanup effort |
Long or oversized boards, LED drivers, clusters | Saki 3Di-ZS3EX, I.C.T AI-5146-3D large-board mode | Board length including panel frame; single vs dual rail |
Line already built on Yamaha mounters | Yamaha YSi-V | Whether your mounter models support the feedback stop function |
Budget band (industry estimates) | Try first | What you give up |
|---|---|---|
Entry band | I.C.T AI-5146-3D, Saki 3Di-MS3EX entry configuration | Published height accuracy and gauge data; deepest tall-part range |
Mid band | Mirtec MV-6 OMNI, Yamaha YSi-V | Some metrology depth at the very top end; options still add up |
Flagship band | Koh Young Zenith, Omron VT-S1080, Viscom iS6059 | Nothing technically — but capital and calibration costs rise, and the machine needs an engineer to run it |
These bands are industry estimates assembled in November 2026 from distributor listings and vendor material, not quotations. Configured prices move with rail count, camera options and software licences, and the only number that matters is the one on a written offer against your board. For a longer treatment of how price interacts with payback, see our 3D AOI price guide.
This list is ordered by buying situation, most common first, and that order is an editorial judgement rather than an arithmetic result. Automotive PCBA plants are mostly Tier-2 and Tier-3 suppliers with a Tier-1 as their customer, and that buyer sits between two weightings. So we scored all seven on six dimensions out of ten and published both totals, which lets you recompute the order with your own priorities.
The dimensions are: documented 3D metrology (a published accuracy or repeatability figure, not a resolution claim); solder judgement tied to IPC-A-610 Class 3; false-call control and programming effort; traceability and data integration; board and line flexibility; and cost, delivery and service reach. Under audit-driven weights, metrology and IPC together carry 60 percent of the score. Under value-driven weights they carry 30 percent and cost carries 25.
Machine | Metrology /10 | IPC-C3 judgement /10 | False call & programming /10 | Traceability /10 | Flexibility /10 | Cost & service /10 | Audit weights | Value weights |
|---|---|---|---|---|---|---|---|---|
Koh Young Zenith | 10 | 9 | 8 | 9 | 8 | 3 | 8.85 | 7.35 |
Omron VT-S1080 | 9 | 10 | 7 | 9 | 8 | 4 | 8.65 | 7.45 |
I.C.T AI-5146-3D | 6 | 6 | 8 | 7 | 9 | 9 | 6.75 | 7.65 |
Mirtec MV-6 OMNI | 8 | 7 | 5 | 8 | 8 | 6 | 7.20 | 6.90 |
Viscom iS6059 | 9 | 8 | 7 | 8 | 6 | 5 | 7.95 | 6.95 |
Yamaha YSi-V | 7 | 7 | 8 | 7 | 7 | 7 | 7.15 | 7.15 |
Saki 3Di-ZS3EX | 6 | 8 | 6 | 7 | 8 | 6 | 6.75 | 6.75 |
Read the two right-hand columns together, because they disagree. Under audit weights the I.C.T platform scores 6.75, level with Saki at the foot of that column, because neither publishes a height accuracy or a repeatability figure. Under value weights the same machine scores 7.65, first of the seven, because a 510 × 610 mm envelope and a five-figure price beat a micron it does not need. We placed it third because the audience for this article is mostly Tier-2 buyers who need both halves of that trade, and because a plant that cannot show a GR&R will fail the audit regardless of which column it optimised.
Three things are outside the scorecard on purpose. Local service quality varies by distributor and country, so the score assumes the vendor's own published support structure and not your local agent's reputation. Software licensing and calibration contracts change the five-year cost more than the purchase price does. And none of the scores reflect a machine's behaviour two years after installation, which is the period that actually decides whether the purchase worked.
The technical figures in this article come from the makers' published specifications, product pages and brochures, read in November 2026. Where a maker publishes a number, we quoted it with its condition attached — Omron's 12.5 µm and 25 MP, Viscom's 0.5 µm Z resolution and 30 mm range, Saki’s 25 mm and 40 mm height ranges, Mirtec's ±3 µm 3D accuracy. Where a maker publishes nothing on height accuracy or gauge capability, we said so in the section rather than filling the gap. Each of those six specifications is linked to the maker's own published page at the point where it is quoted, so the condition travels with the number instead of being summarised away here.
Requirements and process claims come from a separate layer of sources, and they are not interchangeable with vendor material. The CQI-38 publication is documented on AIAG's own publications listing, which records the guideline as published in June 2026 to supplement IATF 16949. The false-call and 100 percent inspection expectations for automotive AOI are set out in ASC International's zero-defect guidance. The challenge-panel and Class 3 qualification method follows a published Class 3 line-qualification write-up. The GR&R and MSA reading of IATF is discussed, with its limits, in TechnoLynx's scoping guide.
Operator experience comes from forums, and is labelled as such. The Mirtec false-call and bottleneck observations are from a 2025 SMTNet thread in which the machine's own regional president replied. Nothing in this article is based on a lab test we ran, and no machine was benchmarked side by side on the same board for this piece.
What we did not do: we did not include 2.5D systems, offline or desktop machines, or models that are no longer in production, however good their reputation. We did not include TRI's TR7700 or Nordson's SQ-series in the seven, not because they are weak — TRI's strength is its AOI and AXI line together, and Nordson's SQ platform is an AOI, SPI and CMM hybrid that competes with a different buying decision — but because including them would have turned a seven-machine shortlist into a ten-machine catalogue. Both belong on a long list, and both should be demoed if your mix is unusual. And we did not test any of these machines on an automotive board ourselves, so treat every specification here as a vendor claim with a condition attached, and verify it with the four checks in the previous section.
You still need X-ray. AOI inspects what light can reach, which covers component presence, polarity, placement offset and the visible fillet on gull-wing and J-lead joints. It cannot see the balls under a BGA or the void inside a QFN thermal pad. Automotive guidance treats post-reflow AOI and X-ray as two layers of the same inspection strategy rather than alternatives, with 100 percent X-ray on BGAs and void measurement typically capped around 25 percent. Budget for both, and plan the AOI and AXI positions in the line layout together. Our comparison of what X-ray sees that AOI cannot covers the split in detail.
The guideline is built around lifecycle management of an AI-based inspection system rather than a one-off acceptance test. In practice that means a defined defect taxonomy agreed before the vendor is chosen, documented system validation against that taxonomy, records of false-call and escape performance in production, retention of inspection data and images, and a defined process for model changes after launch. It supplements IATF 16949 rather than replacing it, so your existing MSA and PPAP obligations still apply. Plan the acceptance evidence pack at the same time you plan the purchase order.
Below 5 percent at board level is the working target for automotive lines, and several published automotive references use that figure as the efficiency limit rather than a quality limit — above it, the review queue starts to compete with production for operator time. The number to negotiate is not a rate in the abstract but reviews per shift on your mix, because a 5 percent rate on a thousand boards is a different staffing decision from 5 percent on a hundred. Write the target into the acceptance test, along with the defect panel you will use to measure it.
Between two hours and two days, and the spread is not about the machine's speed. It is about whether the vendor's library already contains your packages, whether you have CAD or Gerber data of usable quality, and whether the person doing the work has done it before. Vendors quote figures from a clean import: Mirtec states AI-assisted matching cuts teaching time by up to 90 percent, Saki builds inspection programs from design and centroid data with its own automated tooling, and I.C.T quotes setup in minutes with offline editing. All three are plausible on a board that matches their assumptions. Ask for the first-article time on a board that does not.
Dual rail buys throughput and costs board length. On the I.C.T platform, the dual-rail envelope is 510 × 330 mm against 510 × 610 mm on a single rail, and Omron publishes a similar split at 510 × 330 versus 510 × 680 mm. If your panels are short and your volume is high, dual rail doubles the boards inspected per pass. If your boards are long, dual rail removes the capability you were buying the machine for — and you will discover that at the first large board, not at the quotation.
Yes, with two conditions that decide the outcome. The machine needs a calibration certificate and a service path for spare cameras and projectors, because a used AOI without those is a parts donor. And your customer has to accept it: an IATF-driven audit asks about the equipment's validation history, and a machine with an unknown service record is a harder conversation than a new one at the same price difference. If both conditions hold, a used flagship platform usually beats a new entry model — the optics age better than the software, and the software is what you cannot upgrade.
I.C.T has built SMT equipment in Dongguan since 2012. The figures we publish and stand behind are 1,600+ customers in 72 countries, a 12,000 m² factory, 89 staff including 20 engineers, and CE, RoHS and ISO 9001 certification. We sell printers, SPI, AOI, X-ray, reflow, handling and full-line solutions, which is why this article can describe the inspection stage in the context of the line around it rather than as an isolated box.
The conflict of interest is obvious and worth stating: our own AI-5146-3D is on this list, at number three. We published both weighted totals, including the one where it finishes last, so you can disagree with the placement in either direction. If you want the broader range rather than one model, the AOI inspection range covers 2D and 3D, inline and offline, DIP and coating inspection. And if you are still deciding whether you need 3D at all, start with 2D versus 3D AOI explained before you look at any of the seven machines above.
content is empty!