Commercial Display Acceptance Criteria: Pixel Defect Classes and AQL Sampling

1. AQL 2.5 on a purchase order is not an acceptance criterion

"AQL 2.5" is the most common quality term in display purchase orders and the one most likely to collapse when a dispute starts. AQL 2.5 sets how many units in a random sample may be nonconforming before the lot is rejected. It says nothing about what makes one display nonconforming. If the order does not define that, the inspector on the day decides it, and the inspector is usually working to the factory's internal standard rather than yours.

Two separate systems govern display acceptance. One counts pixel defects inside a single panel and comes from the flat panel ergonomics standards. The other counts defective units inside a shipment and comes from ISO 2859-1, the attributes sampling standard also published as ANSI/ASQ Z1.4 in the United States and NF X06-022 in France. They answer different questions, they use different units, and neither one covers the defects behind most commercial display complaints: uneven backlight, colour mismatch between two screens mounted side by side, and image retention after a fixed layout has run for weeks.

This article sets out both systems with the actual numbers, converts the per million pixel allowances into counts you can check on a 4K screen, works a 600 unit order end to end, and gives the clause to put in the RFQ. The two reference product families are the wall-mount signage line for single-panel lots and the LCD video wall line for multi-panel ones, because those two cases need different acceptance logic.

2. The three types of pixel defect

ISO 13406-2:2001 defined three defect types, and this vocabulary is still what panel makers and factories quote, even though the standard itself has been withdrawn.

• Type 1: a hot pixel, permanently lit and appearing white on a black field.
• Type 2: a dead pixel, permanently dark and appearing black on a white field.
• Type 3: a stuck sub-pixel, where one or more of the red, green or blue sub-pixels is permanently on or permanently off.

The difference between a pixel and a sub-pixel is where most arguments start. A single red sub-pixel stuck on is a type 3 defect. Buyers routinely call it a bright dot and expect it to be counted as type 1. Under the standard it is not, and the allowance for type 3 is larger than for type 1 in every class. A screen with three stuck sub-pixels scattered across the panel conforms under class II, and if the order did not name the types, arguing about it later is expensive.

Clusters matter more than raw counts. ISO 13406-2 permits zero clusters of more than one type 1 or type 2 fault in class II, so two adjacent bright dots fail a panel that would pass with 16 scattered ones. A cluster reads to the eye as one obvious blob, while the same count spread across the panel is close to invisible at normal viewing distance. Write the cluster rule into the order as its own line, with a defined distance between defect centres, commonly 5 mm.

3. What the fault classes allow on a real 4K panel

The class limits are published per million pixels, which makes them hard to picture. A 3840 by 2160 panel has 8,294,400 pixels, so 8.29 million. Multiply the per million limits by 8.29 and the class stops being abstract.

• Class I: zero of every type and no clusters. This is the zero tolerance class.
• Class II: about 16 type 1, about 16 type 2 and about 41 type 3 on a 3840 by 2160 panel, with no cluster of more than one type 1 or type 2 fault.
• Class III: about 41 type 1, about 124 type 2 and about 415 type 3.
• Class IV: roughly 415, 1240 and 4150.

Most panel makers specify class II. That sentence is the one to read slowly: on a 55 inch 4K screen, class II permits roughly 16 permanently lit pixels and roughly 16 permanently dark ones, and the display conforms. Buyers who discover this after installation usually assumed the number was one or two.

The same class on a 1920 by 1080 panel, which has 2.07 million pixels, gives about 4 type 1 and about 4 type 2. The class describes a defect density, not a defect count, so the count scales with resolution rather than with screen size. A 43 inch 4K screen and an 86 inch 4K screen carry the same numerical allowance, because they carry the same number of pixels.

ISO 13406-2:2001 was withdrawn and revised by the ISO 9241-302, 303, 305 and 307:2008 series, and ISO 9241-307 is the part covering pixel defects and conformance test methods. The class structure was revised in the successor series, so a policy that just says "class II" is ambiguous. Name the standard and the edition in the order, and ask the factory which edition its own line test follows before you agree to it.

4. What AQL actually controls

ISO 2859-1 indexes acceptance sampling plans by an acceptance quality limit. AQL 2.5 does not mean 2.5 percent of the sample may be defective. It means the plan is built so that a lot running at a 2.5 percent process average is accepted about 95 percent of the time. The decision on the day is made by comparing the count of nonconforming units in the sample with the plan's acceptance number (Ac) and rejection number (Re).

Sample size comes from the lot size and the inspection level through a code letter. At General Inspection Level II, the default in most pre-shipment inspection protocols:

• Lot of 501 to 1200 units: code letter J, sample of 80, Ac 5 and Re 6 at AQL 2.5.
• Lot of 1201 to 3200 units: code letter K, sample of 125, Ac 7 and Re 8 at AQL 2.5.
• Lot of 3201 to 10000 units: code letter L, sample of 200, Ac 10 and Re 11 at AQL 2.5.

The common commercial split is AQL 0 for critical defects, 2.5 for major and 4.0 for minor. Note what AQL does not do: it does not classify defects. Critical, major and minor are buyer-defined categories that have to be listed in the order, defect by defect. A scratched bezel and a screen that will not power on are both "defects" until somebody writes down which is which. Inspection levels also matter: Level I halves the sample for trusted suppliers, Level III enlarges it after a failed lot, and the special levels S-1 to S-4 exist for destructive or expensive tests.

A passing result is a statement about the lot, not about the units that were never opened. If 200 units are sampled from a lot of 5000 and 10 fail, the lot is accepted and the remaining 4800 ship on the strength of that sample.

5. A worked example: 600 units, class II, AQL 2.5

Take an order of 600 wall-mount units, 55 inch, 3840 by 2160, quoted to ISO 13406-2 class II with AQL 2.5 at General Inspection Level II.

On pixel defects, each panel may carry about 16 permanently lit pixels, about 16 permanently dark ones and about 41 stuck sub-pixels, provided no two type 1 or type 2 defects sit within the cluster distance. All 600 screens can arrive at that limit and all 600 conform.

On the lot, 600 units falls in the 501 to 1200 band, so the code letter is J, the sample is 80 units, and Ac 5 / Re 6 applies at AQL 2.5. Five or fewer nonconforming units and the lot ships. Six or more and the lot is rejected.

Now the failure mode that catches buyers out. One panel arrives with 20 bright dots, which is worse than class II permits, so the unit is nonconforming, and it counts as one unit in the sample. Four more units in the same condition and the lot still passes at exactly five. The buyer receives 600 screens, 80 were inspected, and there is no claim on the 520 that were never opened.

The correction is not a tighter AQL. Tightening from 2.5 to 1.0 changes the sample size and the chance that an acceptable lot gets rejected, and the supplier prices that risk. The correction is a definition of nonconforming that the buyer controls, a remedy that survives a passing sample, and a retention sample held back from every lot.

6. The defects neither system catches

Pixel classes and AQL plans are both attribute checks: a unit passes or fails a named condition. The conditions that generate the most service calls on commercial displays are continuous variables, and they are missing from most purchase orders.

• Backlight uniformity and mura. SID publishes measurement methods in the International Display Measurement Standard (IDMS), but there is no industry-wide pass threshold. Unless the order names a number, a grid and a measurement geometry, uniformity is whatever the factory ships.
• Panel-to-panel colour and brightness difference. Two screens from the same production run can sit a few percent apart in white luminance and both be inside panel tolerance. Mounted side by side on one wall, that difference is visible.
• Viewing-angle colour shift. A specification quoting 178 degrees horizontal and vertical describes the angle at which contrast falls to a defined fraction, not the angle at which colour stays acceptable.
• Image retention. Burn-in after a fixed layout has run for weeks is a warranty question rather than an incoming inspection one, and it will not appear in a 30 minute pre-shipment test.
• Touch dead zones. On a touch model the pixel test passes and the touch layer can still have unresponsive bands, which is why interactive flat panels need a separate touch acceptance line.

Each of these needs its own clause with a measurement method and a threshold. Where no published industry threshold exists, the honest move is to ask the factory what it tests to and write that number into the order, rather than inventing one that nobody can measure repeatably.

7. Multi-panel orders: judge the wall, not the panel

A video wall is one display built from many panels. AQL treats each panel as a unit, while the buyer experiences the assembled wall. A three by three wall of nine 55 inch panels needs all nine to match, and a single panel a few percent brighter than its neighbours is visible across the whole image.

For multi-panel lots, add a matching clause on top of the pixel and AQL clauses: panels for one wall come from the same production batch and the same panel bin, and the buyer approves the wall assembled and powered at the factory before shipment. The LCD video wall panels are supplied at 46 and 55 inch with a 3.5 mm ultra-narrow bezel, and bezel consistency belongs in the same conversation, because a fraction of a millimetre of variation across a seam reads as a step in the image.

Local dimming adds another variable. On a Mini LED panel the backlight is divided into zones, and the zone count drives both contrast and the blooming visible around bright objects. A Mini LED display quoting 1,000,000:1 is quoting a figure measured with local dimming active. The same panel measured with dimming disabled reports a number in the same order as a conventional edge-lit screen. State which figure the acceptance test uses, or the two sides will test the same panel and get different answers.

8. The acceptance clause to write into the order

Twelve lines, in this order. Each one is cheap at RFQ stage and expensive after the container ships.

1. Name the pixel standard and edition: ISO 13406-2 class II, or ISO 9241-307 with the class stated.
2. State the cluster rule separately: no two type 1 or type 2 defects within a defined distance, commonly 5 mm between defect centres.
3. Define critical, major and minor by listing actual defects, not by category name alone.
4. Set the AQL for each class and name the sampling standard, edition, inspection level and severity: ISO 2859-1, General Inspection Level II, normal inspection.
5. Name the inspection point: at the factory, before packing, with the buyer or a nominated third party present.
6. Require a burn-in period before inspection, typically 24 to 72 hours of continuous operation, and say who pays for it.
7. Set a uniformity threshold together with the measurement method (IDMS or an agreed grid) and the number of measurement points.
8. For multi-panel orders, require single-batch supply and a factory acceptance test of the assembled wall.
9. Set touch layer acceptance where applicable: dead zone size, edge exclusion and the number of touch points tested.
10. Define what happens on rejection: rework, sort, replace or cancel, and who pays for re-inspection.
11. Require retention samples: two units held from each lot for the warranty period.
12. State the test environment: ambient light level, input signal and warm-up time, because a pixel test run in a bright room misses type 2 defects.

9. What tighter criteria cost, and what loose ones cost

Tighter acceptance criteria are a commercial term, not a technical one. Class I panels exist, but sorting yield is lower, so the unit price is higher and the lead time longer. The same applies to a tightened AQL: moving from 2.5 to 1.0 enlarges the sample and raises the chance that an acceptable lot is rejected, and the supplier prices that risk into the quote. The exact premium and the lead time impact are confirmed per project, so ask for both numbers rather than assuming them.

The cost of not specifying is higher. A 4K screen with 16 permanently lit pixels conforms under class II, and no claim against it will succeed. If the installation is a flagship retail window or an airport information wall where one bright dot is visible for eight hours a day, pay for class I or for a sorted batch, and say so at quotation stage. Retrofitting an acceptance standard after the lot is built is not possible.

One last point for comparing quotes. Two suppliers quoting the same panel can be quoting different fault classes, and the price difference between class II and class I supply is real. A quote that does not state the class, the sampling standard and the inspection point is not comparable to one that does, no matter how close the headline unit prices are.

Acceptance criteria matrix for a commercial display order

Item	What to specify	How it is measured	Typical clause
Pixel defects	Fault class, standard edition, limit per type and cluster rule	Full-screen white, black, red, green and blue fields at a stated distance	ISO 13406-2 class II, or ISO 9241-307 class II
Defect density on 4K	Allowance scales with resolution, not with screen size	3840 x 2160 = 8,294,400 pixels, so 8.29 million	Class II: about 16 type 1, 16 type 2, 41 type 3
Lot sampling	Standard, edition, inspection level, AQL per defect class	ISO 2859-1 General Inspection Level II, normal inspection	AQL 0 critical, 2.5 major, 4.0 minor
Sample size	Derived from lot size through a code letter	3201 to 10000 units: code letter L, n = 200	Ac 10, Re 11 at AQL 2.5
Backlight uniformity	Threshold plus grid size and geometry	IDMS multi-point luminance method	No industry threshold; state the number
Panel-to-panel match	Brightness and colour delta across one wall	Assembled wall, same content, same input signal	Single production batch, factory acceptance test
Burn-in	Duration before inspection and who pays	Continuous operation at rated brightness	24 to 72 hours
Rejection remedy	Rework, sort, replace or cancel	Agreed in the order before deposit	Supplier funds rework; buyer funds re-inspection

Frequently Asked Questions