How should sampling frequency be set for batch inspections?

Sampling frequency for batch inspections should be set in two linked plans: a lot-acceptance sample that decides whether to accept the batch, and in-process checks that detect change during production. Set both from feature severity, lot definition, process stability, failure opportunities, measurement-system reliability, and the buyer’s risk tolerance. A stable low-risk feature may use routine sampling; a new setup, critical characteristic, tool failure, mixed lot, or recent nonconformity may require event-based checks, tightened sampling, or 100% inspection. The RFQ must define CTQs, defect classes, lot boundaries, the sampling standard, switching rules, and containment actions before production.

Separate Lot Acceptance from Process Monitoring

Acceptance sampling uses a defined sample and acceptance number to make a decision about a completed lot. In-process inspection checks parts at planned intervals or after process events to find drift before the lot is complete. Increasing one does not automatically replace the other. A lot can pass its acceptance sample even when an unsampled part is nonconforming, while frequent process checks can miss a random defect that occurs between checks.

Use AQL as a Scheme Index, Not a Defect Promise

The acceptable quality limit (AQL) is an index used with an acceptance-sampling scheme. It is not a statement that every accepted lot contains no more than that percentage of defects, and it does not authorize shipping known nonconforming product. The selected plan has operating-characteristic risks for both buyer and producer. Defect classification matters because a cosmetic mark, missing thread, sealing-height error, and safety-related characteristic do not have equal consequences. The quality agreement should state the AQL or other scheme, inspection level, single/double/multiple plan, acceptance and rejection numbers, and action after rejection.

Select Sample Size from the Agreed Standard

ISO 2859-1 supports attribute inspection by lot size, inspection level, AQL, and the corresponding sampling-plan tables. General Level II is a common contractual starting point, not a universal requirement. Special levels may reduce sample size when testing is expensive or destructive, but that choice changes discrimination and risk. For Medical Device or Aerospace and Aviation parts, customer, regulatory, drawing, or product-specific rules can override a generic plan. Variable measurements may need a customer-approved variables plan or process-control method rather than being converted casually into pass/fail counts.

Set In-Process Frequency from Failure Opportunity

In-process frequency should be high enough to limit the number of parts exposed between the creation and detection of a defect. Time alone is a weak basis. Use pieces produced, tool life, setup changes, material transitions, fixture interventions, finishing batches, and the maximum acceptable containment window. First-off and last-off checks help bound the interval, while periodic checks show whether the process stayed stable between those points.

Use Capability Only After Stability and MSA Are Proven

Process capability can support reduced inspection only when the process is statistically stable, the measurement system is suitable, and the capability study represents the current machine, tool, fixture, material, program, and feature. A CNC Turning Service may show predictable diameter wear across a tool life, allowing checks at defined part counts. Titanium CNC Machining or Inconel 718 machining can introduce different tool-wear and heat risks. A historical Cpk value does not justify lower frequency after a new insert, fixture repair, program revision, or material change.

Match Detection Effort to Failure Consequence

Feature criticality determines how much undetected exposure the plan can tolerate. A practical classification is:

• Critical: A failure can affect safety, regulation, sealing, motion, or system function. The feature may require 100% inspection, automated monitoring, or a customer-mandated plan for Automotive, aerospace, or medical applications. One hundred percent inspection still needs a capable measurement method and a defined response to gage failure.

• Major: A failure can prevent assembly or intended use and may create rework or field risk. Use smaller containment intervals, additional process checks, or a stricter lot plan when capability or history is incomplete.

• Minor: A failure has limited functional impact. Routine sampling may be suitable when cosmetic criteria are objective, the lot is homogeneous, and the customer accepts the selected risk.

Do Not Sample a Mixed Batch as One Homogeneous Lot

A sample represents its lot only when the lot definition is credible. Different material heats, machines, fixtures, tool-life stages, operators, subcontract finish batches, or rework routes can create separate risk populations. These sources should be segregated or deliberately represented in the sample. A supplier providing One-Stop Service coordination may improve traceability, but it does not make machining and finishing outputs statistically identical. Reduced sampling needs source-level defect history, stable controls, and measurement evidence rather than a general supplier rating.

Define Escalation, Containment, and Exit Rules

A dynamic sampling plan must state its normal condition, the evidence that triggers more inspection, the containment boundary, and the evidence required to return. A failed CTQ check should identify the last known conforming point, stop or segregate affected production, verify the measurement system, correct the cause, and remeasure the exposed interval. Frequency should decrease only after the agreed number of acceptable lots or stable process intervals.

Apply Standard Switching Rules as a Complete System

ISO 2859-1 provides normal, tightened, reduced, and discontinuation rules for its lot-by-lot attribute scheme. Use the actual switching sequence in the contract rather than inventing local labels. Reduced inspection is not simply a smaller convenient sample, and tightened inspection is not any extra check after a tool change. Internal process-control frequency can still increase after tool breakage, fixture repair, material change, or unexplained drift, but the procedure should distinguish those event checks from the standard’s formal lot-switching status.

Use FAI as a Baseline, Not Permission to Stop Sampling

First article inspection confirms that the defined drawing, program, setup, tooling, material, and inspection route produced an acceptable initial result. It does not establish the defect distribution of later production. A change to design, material source, program, machine, fixture, tooling, subcontract process, or location may require renewed validation. The Prototyping Service or first-off record should hand specific CTQs and process risks into the Mass Production Service control plan. High-risk features can remain at 100% inspection even after FAI.

For the RFQ, provide lot size, source segregation, CTQ list, defect classes, process stages, expected tool life, inspection method, destructive-test constraints, and required records. Ask the supplier to separate the ISO 2859-1 lot plan from first-off, periodic, event-triggered, and last-off checks. Require explicit acceptance/rejection numbers, escalation triggers, last-known-good containment, reduced-sampling evidence, and revalidation rules. That structure controls buyer risk without spending equal inspection effort on every feature and every stable production interval.