An inspection report for a machined part records what was measured, with what instrument, and what decision followed. It is evidence for one drawing revision and one batch, and it is only as strong as the uncertainty behind the numbers it contains.
What belongs in an inspection report for machined parts?
A report is written against a drawing, so the first block identifies it: part number, drawing revision, batch or serial numbers, quantity, date and the location where the parts were measured.
The characteristic list follows. For each feature the drawing tolerances:
- the nominal size and the tolerance, with the measured value, in the drawing's units
- the instrument and method, with its calibration status
- the measurement uncertainty, or how it was derived
- the decision, and the rule that produced it
- the material certificate number, where the customer requires one
- the person or station that measured, and the software version for a CMM program
A report that prints "OK" per line and nothing else records a decision without the data behind it, and nobody can check it against a re-measurement later.
What does a CMM measurement actually state?
A CMM does not measure a diameter. It measures the coordinates of points on a surface, and software fits a feature to those points. The result depends on how many points were taken, where they were placed, the stylus tip diameter, the fitting method, the datum reference frame, and the temperatures of part and machine.
ISO 10360-2:2009, adopted in Germany as DIN EN ISO 10360-2:2010-06, defines the acceptance and reverification tests for Cartesian CMMs with contacting probing systems. The machine is tested with calibrated test lengths, and the manufacturer declares a maximum permissible error for length measurement. A machine that passes is proven for length measurement under test conditions. It is not thereby proven for a 0.8 mm wall, an interrupted bore, or a position tolerance evaluated from twelve points.
A report that names the CMM model but not the probing strategy hides the part of the uncertainty that comes from the measurement plan rather than the machine.
How does measurement uncertainty change a pass/fail decision?
Uncertainty turns the comparison of a measured value with a limit into a statistical question. ISO 14253-1:2017, third edition, published October 2017, sets the rules. The measured value is compared with the specification limit, and the region around each limit that is narrower than the uncertainty is neither conforming nor nonconforming. In the 2017 edition the default coverage factor k = 2 was replaced by a default conformance probability of 95%, so the risk stays constant whether the tolerance is wide or narrow against the uncertainty. The standard expects supplier and customer to agree in advance how values in that zone are handled.
The uncertainty itself follows the GUM, JCGM 100:2008. The combined standard uncertainty uc is multiplied by a coverage factor k, and for an approximately normal distribution k = 2 gives an interval of about 95% confidence, k = 3 about 99%.
Take an upper limit of 25.050 mm and a standard uncertainty of 0.002 mm. A measured 25.048 mm lies inside the limit by one standard uncertainty (0.002 mm), which is not enough to demonstrate conformity at the 95% level. The value falls in the uncertainty zone. A measured 25.044 mm is clear of the zone and conforms; 25.056 mm is clear on the other side and does not.
There are three honest ways to handle the zone: agree a guard band and inspect against a tightened limit, re-measure with a method of lower uncertainty, or accept the shared risk in the quality agreement. What does not work is comparing a rounded value with the limit and ignoring the uncertainty behind it.
What are Cp and Cpk, and what do they not tell you?
Capability indices put the process spread against the tolerance. The definitions below are the ones in the NIST/SEMATECH e-Handbook; they assume a normal distribution and an in-control process.
Table 1. Capability and performance indices, definitions
| Index | Formula | What it compares | Centring |
|---|---|---|---|
| Cp | (USL − LSL) / 6σ | tolerance width to process width | ignored |
| Cpk | min[(USL − μ)/3σ, (μ − LSL)/3σ] | nearer limit to the mean, in 3σ | included |
| Cpu | (USL − μ) / 3σ | upper limit only | one-sided |
| Cpl | (μ − LSL) / 3σ | lower limit only | one-sided |
| Cpm | (USL − LSL) / 6·√(σ² + (μ − T)²) | tolerance width to variation about the target T | included, with target |
Source: NIST/SEMATECH e-Handbook of Statistical Methods, section 6.1.6. USL and LSL are the upper and lower specification limits, μ the process mean, σ the process standard deviation, T the target value.
A worked example. A bore is drawn 25.00 ±0.05 mm, so USL is 25.050 mm, LSL is 24.950 mm and the tolerance width is 0.100 mm. Thirty parts give a mean of 25.010 mm and a standard deviation of 0.012 mm.
Cp = 0.100 / (6 × 0.012) = 1.39.
Cpk = min[(25.050 − 25.010) / 0.036, (25.010 − 24.950) / 0.036] = min[1.11, 1.67] = 1.11.
The spread is adequate but the mean sits 0.010 mm high. Moving the mean to 25.000 mm raises Cpk to 1.39 without touching the machine. Cp says whether the process is wide enough; Cpk says whether it is in the right place.
Neither number says whether the parts in the box conform. Capability describes a distribution. A process at Cpk 1.4 still produces an occasional part outside the limit, and the index does not identify which one.
How many parts does a Cpk value need?
The NIST handbook gives the working numbers: at least 25 values before the Cpk approximation is valid, about 50 independent values generally considered large enough, and n ≥ 100 for a capability study. A Cpk printed from five parts describes five parts.
Independence matters as much as count. Thirty parts turned from one bar in sequence share a material lot, a setup and a temperature history. They do not sample the process the way thirty parts spread over three days do, and the standard deviation from the first group is usually too small.
The normal assumption is the other trap. A characteristic with two peaks from a two-cavity tool, or a distribution cut off by a pre-machining operation, breaks the formula. NIST recommends a transformation or a nonparametric method in that case. ISO 22514-2:2017 sets out calculation models for processes whose distribution is not normal and changes over time.
What Cpk should a drawing require?
There is no general value. Neither ISO 22514-2 nor ISO 14253-1 sets an acceptance limit for Cpk. The requirement belongs in the drawing note, the quality agreement or the customer's supplier manual, written for a named characteristic.
The number only means something together with the tolerance and the measurement uncertainty. A Cpk requirement of 1.67 on a 25.00 ±0.05 mm bore needs σ ≤ 0.010 mm, since the tolerance width divided by six times 1.67 gives 0.100 / 10.0. If the measurement uncertainty at that size is 0.002 mm, the measurement alone consumes a fifth of the permitted standard deviation, and the process has to be tighter than the index suggests. On a ±0.5 mm dimension the same Cpk permits σ = 0.10 mm, which most turning centres hold without effort.
So ask for Cpk on the characteristics that carry function, and ask for the sample size, the date range and the measurement method with it.
What should come with the parts?
- the dimensional report against the drawing revision
- the material certificate: EN 10204:2004 type 2.1 is a declaration of compliance with the order without test results, while type 3.1 rests on specific inspection, carried out on the products to be supplied
- the first article report, where the customer requires one
- a non-conformance report for any characteristic outside the limit, with the disposition
- the CMM program and point lists, if the customer audits the measurement
EN 10204 separates non-specific inspection, where the products inspected are not necessarily the products actually supplied, from specific inspection on the products to be supplied. Neither document type is a dimensional report: both describe the material against its product specification.
Our audited network in Bursa covers milling, turning, sheet metal, welding, casting with machining and assembly, and works to ISO 2768-m as standard, tighter where the drawing specifies. The quality documents guide lists the rest of the document set.
How do you check a report before you accept it?
Read it against the drawing, not against the previous report.
- The drawing revision matches the order.
- Every toleranced characteristic appears, not only the ones that passed.
- Units and decimal places match the tolerance. A ±0.005 mm tolerance reported to two decimals cannot be judged.
- The instrument is named, and its calibration is current.
- Uncertainty and the decision rule are stated.
- A quoted Cpk carries n, the date range and the machine.
- The values are measured values, not nominal values copied from the drawing.
A report does its job when a second person can repeat the measurement and reach the same decision. That is the standard to hold it to, whether the part is a bracket at ±0.2 mm or a bearing seat at ±0.005 mm.
Frequently asked questions
- What is the difference between Cp and Cpk?
- Cp compares the tolerance width with six process standard deviations and ignores where the mean sits. Cpk measures the distance from the mean to the nearer specification limit in units of 3σ, so it falls as the process moves off centre. When the mean is exactly at the middle of the tolerance, Cp and Cpk are equal.
- Does a Cpk of 1.33 mean the parts are good?
- No. Cpk describes a distribution, not the individual parts in the box. A process at 1.33 still produces parts outside the limit. The index also assumes a normal, stable process and a sample large enough to estimate the standard deviation, so a value quoted without sample size and date cannot be judged.
- How does measurement uncertainty affect a pass/fail decision?
- ISO 14253-1:2017 defines an uncertainty zone around each specification limit. A measured value inside that zone cannot be declared conforming or nonconforming under the default rule, which works at a 95% conformance probability. The supplier and customer have to agree in advance how such values are treated.
- How many parts are needed for a Cpk value?
- The NIST/SEMATECH e-Handbook gives at least 25 values before the Cpk approximation is valid, about 50 independent values as generally large enough, and n ≥ 100 for a capability study. Parts must also be independent: thirty parts from one bar sample the setup, not the process.
- Is a 3.1 certificate an inspection report?
- No. A 3.1 certificate to EN 10204:2004 is a material document based on specific inspection of the products supplied; it reports results against the product specification. It says nothing about the dimensions on your drawing. The dimensional report is a separate document.
- What should a CMM inspection report include?
- The drawing revision, every toleranced characteristic with nominal, tolerance and measured value, the CMM model and probing strategy, the evaluation method, the datum reference frame, the measurement uncertainty and the decision rule. A pass/fail column without measured values cannot be checked later.
Sources
- ISO 14253-1:2017, Decision rules for verifying conformity or nonconformity with specifications (preview PDF of the English text, STN EN ISO 14253-1:2018)
- DIN EN ISO 14253-1:2018-07, German adoption of ISO 14253-1:2017 (DIN Media)
- DIN EN ISO 10360-2:2010-06, Acceptance and reverification tests for coordinate measuring machines, Part 2 (DIN Media)
- NIST/SEMATECH e-Handbook of Statistical Methods, 6.1.6 Process Capability
- JCGM 100:2008, Evaluation of measurement data — Guide to the expression of uncertainty in measurement (BIPM)
- DIN EN 10204:2005-01, Metallic products — Types of inspection documents (DIN Media)
- EN 10204:2004, Types of inspection documents, preview PDF with terms and definitions (ILNAS)
- DIN ISO 22514-2:2019-07, Statistical methods in process management — Capability and performance, Part 2 (DIN Media)
Questions on a drawing? sales@ertas.engineering