Sampling Inspection and Acceptance Methods for 3D-Printed Sample Validation: From Sample Size and AQL to Engineering Rules for Release and Rejection

Introduction: Why “Inspecting Everything” Is Not the Answer to Sample Validation

In low-volume 3D printing deliveries, the most common approach to sample validation is to measure the first article and every selected sample part, then release the lot if the dimensions pass. This approach is barely workable for one-off customized parts, but once production reaches lots of 10 or 50 parts, two problems become apparent. First, the labor and equipment time required for 100% inspection are severely underestimated: measuring one complex part on a CMM often takes 30–60 minutes, meaning 50 parts can consume an entire day. Second, “100% inspection” itself does not define acceptance rules—what happens if one part is out of tolerance, or two parts are out of tolerance? These decisions are often made ad hoc by on-site engineers, making conclusions neither reproducible nor traceable.

A truly executable sample-validation process must define rules for “how many to sample, how to decide, and what to do after the decision.” This article presents a practical engineering method covering sampling plans, sample sizes, acceptance thresholds, and release or rejection decisions.

1. Start with Tiers: Three Inspection Levels for Sample Validation

Sample validation for 3D printing should not be treated as a generic “inspection” activity. It should be divided into three levels according to the validation objective. The first level is first article inspection (FAI), applied to the first complete part and requiring full dimensional coverage. Its purpose is to confirm that the selected process-parameter combination can consistently reproduce the design intent. The second level is in-process sampling inspection, applied to sampled parts within a lot and focused only on critical characteristics, with the purpose of monitoring process drift. The third level is final release inspection, applied to the delivery lot and used to accept or reject the entire lot according to a sampling plan.

The sample sizes and decision logic for these three levels are fundamentally different: FAI is a full-dimensional measurement of n=1; in-process sampling typically inspects one part for every 5–10 parts and measures 3–8 critical characteristics; final release inspection determines sample size from a table based on lot size. Confusing these three levels is the starting point for most failed validation processes.

2. How to Determine Sample Size: From Lot Size to Sampling Plans

Typical lot sizes for low-volume 3D printing range from 10 to 50 parts. Following the principles of GB/T 2828.1, lot size N and sample size n have defined relationships: for N of 2–8, perform 100% inspection; for N of 9–15, sample n=2 or 3; for N of 16–25, sample n=5; for N of 26–50, sample n=8; and for N of 51–90, sample n=13. This mapping is not arbitrary; it is based on a statistical balance between the acceptable quality limit and producer risk at inspection level II.

It should be noted that defect modes in 3D printing differ from those in conventional machining. Defects are often systematic—such as powder-lot variation, recoater-blade wear, or laser-power degradation—rather than randomly independent. Therefore, sampling plans should prioritize distributed sampling: samples should cover different powder-bed areas, different build heights, and parts removed from the machine at different times, rather than consecutively selecting parts from the same tray. If a lot is produced in three separate print jobs, samples should be distributed among the jobs, with at least one part sampled from each job.

3. Acceptance Thresholds: Characteristic Classification and AQL Selection

Which dimensions are measured is more important than how many parts are sampled. Measured characteristics should be divided into three categories: critical-to-quality (CTQ) characteristics are features affecting assembly fit, sealing, or load-bearing performance, such as mating hole locations, sealing-surface flatness, and thread go/no-go requirements; major characteristics affect appearance or secondary functions; and minor characteristics are non-mating free dimensions.

Different acceptable quality limits (AQLs) should be assigned accordingly: an AQL of 0.65, or even 0.4, is recommended for critical characteristics; 1.5 for major characteristics; and 2.5 for minor characteristics. AQL represents the “maximum tolerable lot nonconforming rate.” The stricter the value, the lower the rejection threshold for a given sample size. For low-volume production, a more practical approach is to state the acceptance/rejection numbers (Ac, Re) directly. For example, for n=8 and AQL=0.65, the decision array is (0,1), meaning no nonconforming sample is allowed; one nonconforming part results in rejection of the entire lot. At AQL=1.5, the array may be (0,1), or relaxed to (1,2) under switching rules. Recording the decision array in the inspection form eliminates the need for on-the-spot judgment.

4. After an Out-of-Tolerance Result: Paths for Reinspection, Concession, and Rejection

Decision rules must include actions to be taken after nonconformance; otherwise, the process will stall in disputes. Three paths are recommended for standardization.

Path one is reinspection and expanded sampling. When the number of nonconforming parts in the sample has not reached Re, expanded sampling at twice the original sample size may be permitted for the same lot, retesting only the nonconforming characteristic. The second decision is based on the combined samples. Reinspection is permitted only when the cause of nonconformance has been identified and is incidental, such as localized deformation caused by support residue on a single part. Reinspection must not be used when systematic deviation is present.

Path two is acceptance by concession. This applies only to minor characteristics and selected major characteristics, and only when all three conditions are met: assembly and function are unaffected, the customer provides written approval, and the lot is marked, segregated, and traceable. Critical characteristics must not be accepted by concession; this is a hard boundary of the concession rule.

Path three is rejection of the entire lot. Rejection is triggered when the number of nonconforming samples reaches Re, when any critical characteristic is out of tolerance, or when systematic deviation is identified. After rejection, a nonconforming-product control process should be initiated: the entire lot should be segregated, identified, and reviewed. It is not sufficient to remove defective parts and release the remainder, because sorting cannot ensure the conformity probability of the remaining parts when defects are systematic.

5. Making Decisions Reproducible: The Minimum Record Set

For sampling validation to be reviewable and accepted by customers, records should include at least: the relationship between lot size N and sample size n; sample selection locations, including print-job number, tray coordinates, and build height; measured values and tolerance zones for each characteristic; measuring-equipment identification numbers and calibration validity periods; the decision array (Ac, Re) and final conclusion; and the disposition path and approver when nonconformance occurs. These six items form the minimum closed loop. Without any one of them, the validation conclusion cannot be substantiated in a dispute.

Particular attention should be paid to measuring equipment: the calibration status of CMMs, calipers, and surface-roughness testers directly affects the validity of acceptance decisions. Inspection records should consistently document the equipment ID and calibration expiry date. Data from equipment with expired calibration must not be used as the basis for release.

Conclusion

The transition in 3D-printed sample validation from “inspect everything” to “sampling-based acceptance” is fundamentally a shift from experience-based judgment to rule-based judgment. Three elements are central: determine sample size by lot size and use distributed sampling to cover variation in print location; establish AQLs and acceptance/rejection numbers (Ac, Re) by characteristic class; and define three predetermined paths for reinspection, concession, and rejection, while making clear that critical characteristics cannot be accepted by concession. Once these rules are incorporated into inspection specifications, quality conclusions for low-volume deliveries can achieve consistency and traceability—and withstand both customer audits and internal reviews.