Good measurement and inspection starts from the drawing and the control plan: decide which characteristics matter, then where in the flow to check them, from receiving through first piece, in-process and final inspection. For each one, pick a gauge with resolution about a tenth of the tolerance, keep it calibrated with traceability, and prove it with a gauge study. Set a sampling plan, control temperature and method, and write down how close to a limit a reading can be and still pass. Record the results, and react to any reject with containment.
Inspection is cheapest where the defect is created and most expensive where the customer finds it. Build an inspection plan that catches each characteristic as early as possible, then keep a final check for what can only be seen at the end.
A gauge has to see far smaller steps than the tolerance it judges. The common rule of thumb, often called the rule of ten, is that resolution should be about one tenth of the tolerance. A ±0.05 mm tolerance is 0.10 mm wide, so the gauge should read to 0.01 mm or finer. Resolution is only the starting point. Accuracy, repeatability, the fixture, the operator and the environment all add to the error.
Measure the feature the way the drawing defines it. A diameter controlled by a cylindricity tolerance is not proven by two micrometer readings, and a position tolerance relative to datums needs a setup that establishes those datums. If you need a refresher on reading the drawing, the GD&T basics guide covers datums and the common symbols.
For go/no-go checks, follow Taylor’s principle: the GO gauge should check the full form of the feature at its maximum material limit, and the NO-GO gauge should check individual dimensions at the least material limit. Attribute gauges are quick and robust but give no data on how close a part is to the limit, so they suit stable processes better than new ones.
Temperature matters more than many shops expect. The reference temperature for dimensional measurement is 20 °C, per ISO 1. Parts and gauges made of different materials grow at different rates, so a part measured warm off the machine can read differently once it cools. For hardness, use the method and scale the drawing calls for. Converted values are approximations, and the hardness conversion chart is a reference, not a substitute for testing on the specified scale.
ISO 9001 clause 7.1.5, in paraphrase, asks that measuring equipment be suitable for its use and maintained, and, where traceability is required, calibrated or verified at set intervals against standards traceable to international or national standards. Each instrument must be identified so its status is clear, protected from adjustments and damage that would invalidate it, and when it is found out of tolerance you must assess whether earlier results are still valid.
That last part is the one shops skip. When a gauge fails calibration, list the parts measured with it since the last good calibration and decide whether any could be wrong. The calibration management guide covers intervals, recall and out-of-tolerance reviews, and the calibration log template holds the records.
Use calibration suppliers accredited to ISO/IEC 17025 for the measurement in question. Their certificates state the measurement uncertainty, which you need for the decision rules below.
A calibrated gauge can still be the wrong gauge. Calibration says the instrument reads correctly against a reference in a lab. It does not tell you how much the reading varies when different operators measure real parts in your fixture on the shop floor. That is the job of measurement systems analysis.
For variable gauges, run a gauge repeatability and reproducibility study on the characteristics that matter most. The steps are in how to run a gauge R&R study, and the quality core tools guide explains where MSA sits in a product launch. Run the study before you use the gauge for capability data or for accept and reject decisions on critical features.
You can inspect everything, inspect at a frequency, or sample lots. 100% inspection suits critical features and automatic gauging, though manual 100% visual inspection still misses defects because people tire. Frequency checks, such as five parts per hour, suit in-process control and feed SPC charts. Lot sampling suits receiving and final inspection of stable product.
For lot sampling, use a recognized plan such as ANSI/ASQ Z1.4 for attributes. You choose the inspection level and the AQL, and the lot size gives a code letter, a sample size and accept and reject numbers. The AQL sampling calculator does the lookup. Remember what AQL means: the worst average quality the plan treats as acceptable over many lots. It does not promise that any single accepted lot is that good, and many customers require zero-acceptance plans for critical characteristics.
Every measurement has uncertainty, and near a limit it decides whether a part is good. ISO 14253-1 sets out decision rules for this. In paraphrase, its default rule says conformity is proven only when the measured value falls inside the specification reduced by the measurement uncertainty at each end, and nonconformity is proven only when it falls outside the specification widened by the uncertainty. Readings in between are neither proven good nor proven bad.
Most shops handle this with a guard band: a reduced acceptance zone inside the tolerance, sized from the uncertainty of the gauge. Agree the rule with the customer if it matters, and write it in the inspection instruction so every inspector treats a borderline reading the same way. The JCGM Guide to the Expression of Uncertainty in Measurement (GUM) is the reference for estimating the uncertainty itself.
Record measured values, not just ticks. Variable data shows drift long before parts fail, it supports capability studies, and it proves to an auditor that the inspection happened. Include the gauge ID on the record, so a failed calibration can be traced to the parts it measured.
Every inspection point needs a reaction plan. When a part fails, stop the process, contain product back to the last good check, record an NCR, and decide whether the cause needs a corrective action. The root cause analysis and CAPA guide covers what happens next. A reaction plan that only says "notify supervisor" is not a plan.
Many characteristics cannot be measured with a gauge: scratches, burrs, weld appearance, color, missing labels. Visual inspection is still inspection, and it needs the same discipline. Define what is acceptable with boundary samples or photographs that show the limit, not just a perfect part. State the lighting, the viewing distance, the time per part and which surfaces matter.
Attribute checks are prone to inspector disagreement, so test them. An attribute agreement study, described in the AIAG MSA manual, has several inspectors judge the same set of known good, bad and borderline parts more than once, then compares their answers with each other and with the reference. Where agreement is weak, improve the boundary samples and the instruction before blaming the inspectors.
A material certificate is evidence, not proof. Check that it matches the purchase order and the drawing: material grade, condition, heat or lot number, the standard it was made to and the required test results. Then check that the heat number on the certificate matches the marking on the material in front of you. Most receiving escapes are not bad material but the right certificate for the wrong lot.
Decide how much to verify yourself from the supplier’s history and the risk of the part. A new supplier, a safety-critical part or a recent escape justifies independent checks such as dimensional sampling or hardness testing. A long record of good lots may justify certificate review only. Write the rule down so it is applied consistently, and adjust it when supplier performance changes.
Two inspectors with the same part, gauge and instruction should get the same answer. That only happens if the instruction says exactly how to do the check: which gauge and fixture, how to locate the part, where on the feature to measure, how many readings to take and how to combine them, the temperature conditions and the decision rule. Pictures of the setup do more than paragraphs.
Treat inspection instructions as controlled documents, revised when the drawing, gauge or method changes. Put the inspection steps in the same format as the production work instructions, so operators who do in-process checks see them as part of the job rather than an extra form.
Example: Picking a gauge with the rule of ten (fictional pin). Diameter 8.00 ± 0.015 mm, so the tolerance is 0.030 mm wide.
Result: Use the 0.001 mm micrometer, or a bench gauge, and prove it with a study. The caliper stays for rough checks only.
Example: Thermal growth of a warm aluminum part (fictional). Length 200.000 mm at 20 °C, measured straight off the machine at 26 °C. Aluminum expands about 23 × 10⁻⁶ per °C; the steel gauge is assumed to stay at 20 °C.
Result: On a ±0.02 mm tolerance, the warm reading would reject a good part. Let parts soak to room temperature or compensate for temperature, and write which into the inspection instruction.
Example: Guard band from uncertainty (fictional bore). Specification 10.000 ± 0.020 mm. The CMM’s expanded uncertainty for this feature is 0.003 mm.
Result: The written rule removes the argument at the bench. Agree it with the customer for critical features.
Example: Lot sampling with ANSI/ASQ Z1.4 (fictional receiving lot). Lot of 1,000 brackets, general inspection level II, normal inspection, AQL 1.0.
Result: Check the plan against the current edition of the standard, or look it up in the AQL sampling calculator. For critical characteristics, many customers require accept on 0.
Measurement produces a value. Inspection compares a characteristic with its requirement and decides whether it conforms, using a measurement, a gauge or a visual check.
A rule of thumb that the gauge resolution should be about one tenth of the tolerance being checked. It is a starting point; a gauge R&R study gives the real answer.
At intervals you set and justify from use, stability, the manufacturer’s advice and calibration history. Shorten the interval when a gauge is found out of tolerance and lengthen it when it is consistently stable.
For critical or safety characteristics, for new or unstable processes, and wherever automatic gauging makes it cheap. Sampling suits stable processes and lot acceptance.
A reduced acceptance zone inside the tolerance, sized from the measurement uncertainty, so that accepted parts are proven to be within specification.
No. A first article inspection is a full, documented check of a new or changed part, often to AS9102. A first piece inspection confirms each setup before a production run.
COPLAIN AI PLATFORM
Upload your existing procedures and Coplain AI drafts audit-ready work instructions, job aids and forms from them.
Generate yours automatically with Coplain →