CMM data filtering for CNC parts changes how measured points are evaluated; it does not change the part. Buyers should agree on the evaluation method before inspection, distinguish suspected bad points from real defects, and retain both original and processed results. Approve a report only when its settings, exclusions, rationale, and authority can be traced to the applicable drawing and inspection plan.
What CMM data filtering means for a buyer
A coordinate measuring machine (CMM) collects measurement data from a part. An evaluation uses those data to calculate a feature or assess a characteristic against an applicable requirement. Processing can affect the reported result, so a pass or fail is meaningful only when the evaluation method is understood.
An outlier is a point identified as unusual relative to an evaluation or the surrounding data. That label alone does not establish why the point is unusual. It might reflect an invalid measurement, or it might represent a real condition on the part. Outlier removal excludes selected points from a specified evaluation; it should leave a traceable record of what was excluded and why.
Smoothing processes data to alter the contribution of geometric variation to an evaluation. A Gaussian filter is one possible smoothing method. Its cutoff is a processing setting, not a physical improvement to the surface. A smoother plot can be easier to interpret while also making some local variation less visible.
Hexagon’s PC-DMIS 2022.2 documentation describes constructing a filter set from eligible input data and identifies outlier removal as an option applied before filtering in that workflow. This is a version- and feature-specific example, not a universal rule for every CMM program or a statement about SUUXIANG’s software. Source: Hexagon, Constructing a Filter Set
Choose the evaluation before measuring
Start with the controlled drawing, the characteristic being evaluated, and any customer inspection requirements. Then agree on the measurement strategy and how the data will be evaluated. If the drawing or inspection plan leaves processing ambiguous, resolve that ambiguity with the authorized engineering and quality stakeholders before using the result for acceptance.
Document the feature, applicable revision, acceptance requirement, evaluation method, permitted processing, and the record needed for review. Define which result controls acceptance if both original and processed evaluations will appear on the report. A supplementary plot can help explain a result, but it should not quietly replace the agreed acceptance evaluation.
Sampling, scan coverage, probe access, and alignment are separate decisions from data processing. A filter cannot establish that the correct area was measured or that the agreed datum strategy was followed. For those acquisition choices, see CMM scanning versus touch-trigger sampling. The broader inspection context is covered in CMM Inspection.
Compare the available choices
Use the following framework to discuss a feature with the supplier or inspection team:
- Original-data evaluation: Consider it when the agreed method calls for evaluation without additional exclusion or smoothing. Review the data for questionable measurements and actual local conditions rather than assuming every unusual point is invalid.
- Documented point exclusion: Consider it when there is evidence that specific points are invalid and the agreed method permits exclusion. Identify each excluded point, explain the cause, and retain the original evaluation. If the cause cannot be established, investigate or reacquire rather than treating exclusion as automatic acceptance.
- Smoothing: Consider it when the drawing and measurement strategy support a processed evaluation for the characteristic. Agree on the algorithm and settings, and examine what variation the processing could suppress. Keep the original evaluation available for review.
- Reacquisition: Choose it when the existing data cannot establish whether a questionable point is a measurement artifact or a real part condition. Define the area and purpose of the repeat measurement, and keep the original record alongside the new data.
- Hold for disposition: Use this when the required method, point history, settings, or authority is missing. A favorable number alone does not resolve missing evidence.
Separate an edge artifact from a real burr
Consider a hypothetical scan of a machined edge. An isolated point appears inconsistent with nearby data. The inspection team suspects that the point was collected at an unsuitable location near the edge, but the record does not establish the cause. Removing it immediately would turn a suspicion into an undocumented acceptance decision.
First identify the point and its location. Review the scan path and available observations, then reacquire the area using an agreed approach that can help distinguish a measurement artifact from a physical condition. Preserve the initial data and identify the repeat measurement as a separate record. If the repeat data support an invalid acquisition, document the basis for excluding the original point, provided the inspection plan permits it.
If the condition is a genuine burr, high spot, damage, or form error, do not remove its points merely to obtain a pass. Assess it against the drawing and the applicable acceptance requirements. If correction or a deviation is proposed, the authorized parties should decide how to disposition the part and what subsequent inspection is required.
Understand what smoothing can conceal
In a hypothetical dense scan of a functional surface, a processed profile may look even while the original data show a localized feature. The buyer should ask where the local variation lies, whether it affects the drawing requirement or part function, and how the agreed evaluation treats it. More collected points do not by themselves settle whether smoothing is appropriate or whether a local condition needs separate attention.
Hexagon’s PC-DMIS 2022.1 documentation describes a Gaussian option for particular constructed line features and states that cutoff wavelength controls smoothing in that context. It also describes optional outlier removal before filtering. The documentation is an example of how settings affect an evaluation, not a recommended cutoff for a buyer’s part. Source: Hexagon, Apply Gauss Filter / Cutoff Wavelength
For an approval review, ask the inspection team to show the original and processed evaluations using the same identified feature and part. Look for changes near edges, mating areas, and other locations where a localized condition could matter. If a local issue is obscured, request an appropriate additional assessment under the agreed inspection plan; do not substitute a visually smooth plot for the specified acceptance result.
Resolve supplier and receiving-report differences
A supplier may provide a filtered result while a receiving team evaluates the data without filtering. Different reported values do not, by themselves, establish that either physical part or instrument changed. The evaluations may answer different questions because their processing choices differ.
Before arguing over the numbers, establish whether both reports refer to the same part, drawing revision, characteristic, and acceptance requirement. Compare the alignment and acquisition strategy separately from the processing method. Then identify the input dataset, exclusions, algorithm, settings, and feature evaluation used for each result.
If the datasets or inspection strategies differ, a comparison of processing alone cannot resolve the disagreement. If the same original dataset is available, an authorized reviewer can compare the evaluations and identify where processing affects the result. The controlling outcome remains the one required by the agreed drawing and measurement strategy, subject to any authorized disposition.
Use a controlled report approval workflow
- Confirm scope and authority. Identify the part, lot or individual item as applicable, drawing revision, characteristic, acceptance requirement, and parties authorized to approve the inspection method and any deviation. Make sure the report applies to the item being released.
- Agree on the evaluation. Record the measurement strategy and the intended acceptance evaluation before results are known. State whether exclusions or smoothing are permitted, which algorithm and settings are intended, and how a questionable point will be investigated. Resolve drawing or plan ambiguity through the designated engineering authority.
- Preserve the original record. Retain the original dataset in a form that can be traced to the inspected part and feature. Record the inspection program or method revision and the relevant software and algorithm version. Keep original data accessible even when a processed report is the deliverable.
- Investigate questionable data. Identify the affected points and their locations. Record the reason for suspecting invalid acquisition, the evidence reviewed, and whether reacquisition is needed. Treat possible burrs, high spots, damage, and form error as part conditions to be assessed, not as automatic exclusions.
- Document processing. Record included and excluded point identities, the exclusion basis, processing order, algorithm, version, settings, and the person or role authorizing the decision. Where the applicable documented workflow combines outlier removal and filtering, retain evidence of the order used. Do not infer missing settings from the appearance of a plot.
- Compare and evaluate. Retain both original and processed evaluations, clearly labeled, with the applicable acceptance result identified. Review any material difference and assess whether local features or excluded points require further investigation. A processed result does not erase an original failed or questionable evaluation.
- Approve, hold, or disposition. The authorized reviewer checks the report against the agreed method and required evidence. Approve only when the controlling result and documentation support acceptance. Otherwise, hold the item while requesting clarification, reacquisition, correction, or an authorized deviation as applicable.
- Control subsequent changes. Retain the approved report and supporting records with the relevant revision. If a program, algorithm, setting, point selection, or requirement changes, identify the change and obtain the appropriate review before using the revised evaluation for release.
What if settings change after a failed result?
Treat the change as a controlled technical decision. Keep the original failed evaluation, the revised evaluation, the reason for the change, and the authority approving it. Determine whether the original method departed from the agreed plan or whether the plan itself requires revision. Record the decision before assigning acceptance status.
Changing a cutoff or excluding additional points simply to turn a fail into a pass is a red flag. If no defensible technical reason or approval exists, the revised result should remain blocked. Where a legitimate method correction is agreed, reviewers still need to understand its effect and retain the earlier record.
Specify report expectations in the RFQ
A useful request for quotation gives the supplier enough information to identify the intended part and discuss inspection before a report is produced. Provide the controlled drawing and revision, available CAD data, material and condition, quantity, critical characteristics, datum requirements, surface requirements, and the requested inspection documentation.
For features where CMM data processing may affect acceptance, identify the measurement strategy or ask for a proposed plan. State whether original datasets, point identities, excluded-point rationale, processing settings, and original and processed evaluations are required deliverables. Name the role that can approve a method change or disposition a discrepancy.
Approval checks and red flags
Before accepting the report, confirm:
- The part, drawing revision, characteristic, and controlling acceptance method match the order.
- Original data and both original and processed evaluations are available when required.
- Included and excluded point identities, reasons, processing order, algorithm, version, and settings are traceable.
- A suspected artifact has supporting evidence; an unresolved local condition has not been discarded as an outlier.
- Method changes and deviations have the required authority, and missing evidence leaves the decision on hold.
Red flags include an unlabeled filtered plot, exclusions with no point identities, settings supplied only after a failure, a report that omits the original evaluation, or a request to accept a result without knowing the controlling method. A polished report is not a substitute for a traceable evaluation.
Frequently asked questions
Does every CMM report need smoothing?
No. The appropriate evaluation depends on the drawing, characteristic, and agreed measurement strategy. Smoothing should be an explicit choice with documented settings and an understood purpose, rather than an assumed step for every feature.
Can an outlier be removed if it makes the part fail?
A failing result is not, by itself, a reason to remove a point. Establish whether the point represents invalid acquisition or a real part condition. Apply any permitted exclusion under the agreed method, retain the point identity and rationale, and preserve the original evaluation.
Does a smoother result mean the part is better?
No. Processing changes the evaluated data, not the physical part. Review the original data and any relevant local condition against the drawing and inspection plan before deciding whether the processed result supports acceptance.
What should a buyer request when two reports disagree?
Request identification of the part and revision, the applicable acceptance requirement, the acquisition and alignment strategies, original data, processing settings, exclusions, and both evaluations. If the records are incomplete or the controlling method is unclear, hold the decision until the authorized parties resolve it.
Can a supplier change filtering after inspection?
A change can be proposed for technical review, but it should be documented and authorized under the project’s controls. Preserve the prior result and explain the effect of the revision. A change made only to obtain a pass, without a defensible basis and approval, should not control acceptance.
Upload your drawing to SUUXIANG and include the critical characteristics and inspection documentation you need reviewed.