Defining acceptance criteria & AQL for chassis

You build or buy a chassis because uptime matters. If the frame, rails, welds, or sheet-metal tolerances slip, fans start rattling, boards mis-seat, vibration goes up, and someone’s pager lights up at 3 a.m. Let’s pin down acceptance criteria 그리고 AQL for chassis in a way that’s simple, field-tested, and ready for production. I’ll use language from shop floors and quality rooms, not a textbook. I’ll also drop in how we handle it at IStoneCase—we design and manufacture GPU and server enclosures at scale, so this stuff isn’t theory for us.

Acceptance criteria for chassis (dimensions, function, welds, finish)

Goal: make sure every shipped chassis mounts cleanly, cools right, and survives rack life without weird squeaks or hot spots.

• Dimensions / mechanical fit: Honor drawing tolerances on all datum-driven features—rail hole pitch, motherboard standoff pattern, PCIe alignment, fan wall position, and PSU bay. For key fits (rails-to-rack, board-to-I/O), treat them as special characteristics.
• Function: Doors open/close without rub, latches lock, trays slide at spec force, fans and filters service without tools if design says so. Cable raceways must clear connectors under bend radius. No burrs in finger paths.
• Welds & joints: No cracks, no lack of fusion, no porosity clusters in load paths. Spot weld nuggets must hold per peel test; MIG/TIG seams must be continuous where the print calls it. Visible spatter removed.
• Finish: Uniform coating, no blisters, no bare edges in corrosion-risk zones. Ground points masked and conductive.
• Handling / packaging: Foam and bagging protect corners; shock indicators used on bulk shipments; label legible and scannable.

Quick rule of thumb: anything that can knock a server offline or block a card swap = critical; anything that slows technicians = major; cosmetic only = minor. Easy.

AQL for chassis (ISO 2859-1 sampling plans)

You can’t 100%-inspect every lot forever. We use AQL as the dial between risk and cost.

• Start at General Inspection Level II.
• Grade defects into Critical / Major / Minor.
• Typical targets: Critical AQL = 0, Major AQL = 2.5%, Minor AQL = 4.0%. New supplier? Tighten Major to 1.0% until their process stabilizes.
• Escalation: when you see trend drift, jump to Level III and use tightened inspection; when the process holds steady, move to reduced inspection.

Example sampling table (illustrative)

AC = accept if ≤ AC defects; RE = reject at RE or more. For Critical with AQL=0, 하나 is enough to fail the lot. Harsh? Yep. But downtime is harsher.

Welding acceptance for chassis (ISO 5817 mindset, shop-floor checks)

Chassis are not bridge girders, but weld quality still makes or breaks reliability. For racks, slide rails, and GPU bracing:

• Visual acceptance: no cracks, no undercut on edges, no crater holes at stop-starts, no burn-through in cosmetic areas. Smooth toe transition.
• Spot/Projection welds: verify nugget size via peel tests; no skipped spots on patterns that carry load (rails, handles).
• Seam welds on load paths: continuous bead where required; grind spatter; maintain bead height within drawing or work instruction.
• Rework rules: weld repairs require re-inspection; heat discoloration must be cleaned if it affects coating or conductivity.

In practice, we pair weld acceptance with AQL=0 for presence/absence type checks on safety-relevant joints. You don’t wanna gamble here, we kinda dont.

Process capability for special characteristics (Cpk/Ppk)

AQL handles attributes (good/bad). For variables (dimensions), run capability:

• Critical features (card alignment, rail engagement, board keep-out): Cpk/Ppk ≥ 1.67. If capability falls short, lock in 100% check until you lift the process.
• General features (cover flatness, slot width, cosmetic gaps): Cpk/Ppk ≥ 1.33.

Short version: hold the stack-ups where a millimeter drift can jam a sled or mis-align a PCIe riser. Don’t let tolerance creep eat your lunch.

Defect categories, actions, and typical chassis examples

Real-world scenario: rack-level fit and thermal margin

Situation: a customer racks 200 units of mixed GPU nodes and storage trays. Techs report tray binding on the upper third of racks and hot aisles getting spiky.
What we check:

1. Rail-to-rail pitch vs. rack vendor spec (process capability first, not just pass/fail).
2. Card keep-out around double-wide GPUs; riser face parallelism.
3. Fan wall to filter gap—air bypass robs CFM.
4. Coating build on grounding pads—electrical continuity.

Containment: Critical = AQL 0 on rail hole pitch and grounding points; Major = AQL 1.0 on latch force; Minor = 4.0 on paint specks.
수정: shim kit retired after die change; coating mask widened; capability raised to Cpk ≥ 1.67 on rail-pitch. Ops breathes again. Not magic, just discipline.

Supplier management playbook (tight, but human)

• Kickoff: PPAP/FAI for every new 서버 PC 케이스 또는 컴퓨터 케이스 서버 form factor. Golden sample lives in a locked cabinet (really).
• Control plan: mark special characteristics with ◆ on prints; measurement method frozen (fixture + gauge R&R done).
• Incoming: use the AQL table above; escalate on trend, relax after sustained stability.
• 변경 제어: any die, coating, or weld fixture change needs notification and delta FAI. No “we thought it’s fine.” This part are critical.
• Feedback loop: RMA tags carry exact failure mode words—“tray bind at 42U”, “riser skew”, “fan door rub”—so the factory doesn’t chase ghosts.

How IStoneCase bakes this into chassis programs

We ship to data centers, AI labs, and MSPs that can’t afford flaky hardware. So we build acceptance into the design and the line:

• Design for fit & service: tool-less trays that actually un-jam, cable channels that clear 12VHPWR bends, and fan doors that don’t snag.
• GPU airflow realities: baffle geometry modeled, then validated with smoke and temp maps, not only CFD.
• Rail & guide integrity: guide rails torqued and tested for sag; weld maps tied to holding fixtures.
• Documentation: traveler includes AQL plan, defect images, and go/no-go gauges. New operator? They still pass the same gate.

If you’re speccing a 서버 랙 PC 케이스, 서버 PC 케이스, 컴퓨터 케이스 서버또는 ATX 서버 케이스, you can loop us in early. We also customize GPU 서버 케이스, Rackmount case, 벽걸이 케이스및 NAS 장치 for oddball airflow or cable constraints. OEM/ODM is our daily bread; batch buys and wholesale are fine too.

Quick reference tables (print-friendly)

Defect grading & acceptance

Capability thresholds for special characteristics