A reliability test answers a question about one unit: does this design work. Quality control answers a different question: is the thousandth unit the same as the first. The second question is the one that decides whether a rollout across forty sites goes smoothly, and it cannot be answered by testing every unit harder.
This is the part of a factory visit that buyers walk past, because it looks like paperwork. Four gates and one sampling scheme are what turn a good prototype into a predictable shipment.
IQC: the three incoming items that decide the rest
Incoming quality control inspects what arrives from sub-suppliers. On a touch terminal, three categories account for nearly all of the variability that reaches the customer, and each is checked against a number rather than a look.
| Incoming item | What is measured | Typical accept limit |
|---|---|---|
| LCD panel batch | Colour difference between panels, brightness bin | ΔE ≤ 2 within a lot; brightness within ±5% |
| Lithium cell or pack | Capacity at 0.2 C, internal resistance, open circuit voltage | ≥ 95% of rated capacity; IR and OCV inside the cell maker’s window |
| Mainboard | In-circuit test plus a first-article run on a golden fixture | 100% ICT; one full functional pass per lot |
| Housing and bezel | Colour match to the master sample, sink marks, thread integrity | Visual against a limit sample; torque test on every threaded insert |
| Power adapter | No-load voltage, ripple, plug polarity | Within ±5%; polarity checked against the harness, not the label |
The panel colour bin is the one buyers notice without knowing why. Two terminals from different panel lots, side by side on the same counter, will show one slightly warmer than the other. It is not a defect and it is impossible to un-see once a customer has spotted it, which is why the bin is controlled at IQC rather than explained later.
IPQC: first article, torque, and the screw that is missing
In-process control runs on the line while it is running. The first three units off the line after any change — a new lot, a new operator, a fixture adjustment, a shift change — are inspected in full and signed off before production continues. After that, a patrol inspection every two hours pulls one unit and checks the items that drift.
What drifts in a terminal assembly is mechanical, not electrical. Screw torque on an M3 machine screw into a brass insert is typically specified at 0.5 N·m; below that the unit loosens in transit, above it the insert strips and the housing is scrap. Electric drivers with torque logging are the only way to hold it, and the log is the evidence. Screw count is checked visually against the drawing, because a missing screw is a rattle in the field, not a failure on the bench.
Two more items belong here and are usually skipped. Adhesive and gasket continuity on an IP-rated unit: a gap in a foam gasket is invisible after assembly and shows up as a failed water test or, worse, as corrosion on a board six months later. And connector seating: a display flex that is inserted but not latched passes the functional test and fails on the first lorry journey with a rough road.
FQC: AQL is a sampling scheme, not a percentage
Final quality control samples the finished lot. The scheme is defined in ISO 2859-1, published in China as GB/T 2828.1, and it is widely misquoted. An AQL of 0.65 does not mean 0.65% of units may be defective; it means that lots at that quality level will be accepted about 95% of the time under the specified sampling plan.
| Lot size | Sample (level II) | Major 0.65: AC / RE | Minor 1.5: AC / RE |
|---|---|---|---|
| 501–1,200 | 80 | 1 / 2 | 3 / 4 |
| 1,201–3,200 | 125 | 2 / 3 | 5 / 6 |
| 3,201–10,000 | 200 | 3 / 4 | 7 / 8 |
Read the middle columns as a budget. With a lot of 1,200 units sampled at 80 pieces and an AQL of 0.65, finding two defective units rejects the lot. That is a tight plan, and it is why a factory that inspects 80 units and reports zero defects on a new model is usually reporting that nobody looked.
Critical defects — anything that could injure someone or cause a fire, which on a POS means mains isolation, battery swelling and sharp edges on a handheld — are not sampled at all. Acceptance number zero, any occurrence rejects the lot.
OQC: the box is part of the product
Outgoing control checks the shipment rather than the unit: the accessory list against the order, the adapter plug type for the destination market, the manual language, the serial number and MAC address bound to the carton label, and a carton drop test on a finished box.
This is where the least glamorous and most expensive mistakes happen. A container of terminals shipped with the wrong plug type is not a quality problem, it is a customs and rework problem worth more than the goods. A carton that has not been drop tested arrives with cracked bezels, and the damage is indistinguishable from a manufacturing defect until someone opens the carton in front of the customer.
The serial-to-MAC binding matters more than it sounds. A terminal whose label says one MAC and whose radio reports another will fail activation on a managed network, and the site engineer will spend an hour on it. Doing the binding at OQC costs seconds; doing it in the field costs a truck roll.
Traceability: what one serial number should return
Traceability is the ability to take a serial number from a failed unit and walk backwards to the batch, the station and the operator. It is only real if the records were created during production rather than reconstructed afterwards.
| Record | Retrieved by | What it tells you |
|---|---|---|
| Assembly date, line, station | Serial number | Which shift built it and on which fixture |
| Panel lot and touch firmware version | Serial number | Whether a display failure is isolated or a whole lot |
| Battery cell lot | Serial number | Whether a swelling report is one pack or a thousand |
| Test log: results and timestamps | Serial number | Whether the unit was actually tested, and when |
| Firmware and application image | Serial number | What shipped, so a field fix can be reproduced |
The commercial value of this appears the first time a customer reports a fault on twenty units. With traceability, the answer is a lot number and a containment plan within a day. Without it, the answer is a replacement of everything shipped in the last quarter, because there is no way to narrow it down.
The repair loop is the quality system
Units that fail are not scrapped; they are repaired and returned to the line. The loop around that repair is what actually improves the product, provided three things are true.
- Every return gets a failure analysis, not just a repair. “Replaced board” is a repair. “Backlight connector cold joint, 3 of 47 returns this month” is an analysis, and only the second one changes anything.
- The analysis is counted, not narrated. A Pareto chart by failure mode per month, with the top three named and owned. Without the count, the loudest customer defines the quality agenda.
- The fix reaches the process, not the unit. A cold joint fixed by re-soldering one board will be back next week. The fix is a change to the reflow profile or the connector supplier, and it needs a date.
This is also why a factory that repairs quietly and reports a low defect rate is more worrying than one that reports a moderate rate and shows the Pareto. The first is managing the number. The second is managing the process.
Three numbers worth asking for
Most quality claims are adjectives. Three numbers are checkable, comparable between suppliers, and cheap to produce for anyone who actually measures.
| Metric | Definition | A reasonable mature value |
|---|---|---|
| First pass yield | Units passing all gates without rework, over units started | Above 97% on a product in volume |
| Field return rate | Returns in the first 12 months, over units shipped | Under 2%, and falling after the first year |
| Lot-to-lot variation | Spread of a key parameter, such as brightness or battery capacity, across three consecutive lots | Inside the IQC window, not drifting |
Ask for them for the specific model you are buying, not for the factory as a whole. A site average hides a new model that is running at 88% first pass yield, and the new model is the one on your purchase order.
FAQ
What is the difference between QC and reliability testing?
Quality control runs on the production lot and decides whether to ship it. Reliability testing runs on samples of a design and decides whether the design has enough margin. A factory can pass every QC gate and still fail a reliability test, because the two are looking at different populations.
Is a 100% functional test enough?
No. A functional test catches assembly errors, which is most defects but not the ones that matter most. It cannot catch a cold solder joint that opens when warm, a gasket with a gap, or a connector that is seated but unlatchable — all of which pass a five-minute test.
How do I audit a factory I cannot visit?
Ask for three things: the first-article report with the operator signature, the last month of test logs for the serial range you are buying, and the Pareto of returns. Fabricating all three consistently is more work than doing them.
Where to go next
Quality control is a sampling argument, and the argument only holds if the samples are real. For the hardware side of the same question:
DCA-155A Android terminal DCA-Q3A all-in-one with printer Windows POS systems the six test benches and their pass limits what happens at lamination and calibration fifteen questions, and the answers that should worry you · the footprint cost of an all-in-one
Specifying the hardware? The hardware side is covered in the POS system, by counter position — where each part sits and which of them you can stop buying separately.