CAD Data Quality Metrics: The Short Version
CAD data quality metrics are the measurable indicators that describe how well NX parts, assemblies and drawings comply with defined rules: first-pass yield, findings per part, critical findings share, auto-healed share and downstream rejection rate. Automated checks produce them on every release. inteliScience, Siemens NX specialists and official DOCUFY partner, baselines them and wires them into Teamcenter gates.
The Problem: Quality Is Discussed, Never Measured
Ask an NX methods team how good their CAD data is and you get an opinion. Ask for a number and the room goes quiet. Most organizations we meet run automated checks, yet the results stay in the checker: a list of findings per part, read once by the designer, then gone. Nobody aggregates them, so nobody can say whether data quality improved after the new template.
The cost shows up elsewhere: the ERP import that rejects a BOM, the supplier who asks which of two material attributes is correct, the CAM programmer who rebuilds a face instead of offsetting it. Each is an escaped defect, invisible in engineering’s numbers.
Measurement changes the conversation. A rule that fails on 38 percent of released parts needs a template fix, not a stricter reviewer. A gate that would have blocked 400 parts last month is a gate you do not switch on yet. What CAD quality covers and why manual review does not scale is the subject of our overview of automated model validation in Siemens NX; this post is about what to do with the results.
The Six Dimensions of CAD Data Quality
CAD data quality has six dimensions, and every metric belongs to one of them: geometry validity, model structure, attribute and metadata completeness, standards compliance, PMI and drawing completeness, and downstream fitness. Five are measured by checks inside NX; the sixth is measured by the systems that consume the data, and it validates the other five.
Two standards give the dimensions a vocabulary. ISO 8000-8:2015, the measurement part of the ISO 8000 data quality series, distinguishes syntactic quality (data conforms to its format), semantic quality (data encodes the intended meaning) and pragmatic quality (data is fit for the user’s purpose) [1]. An empty attribute fails syntactically; a valid material string naming the wrong alloy fails semantically; a part that passes every check and is still rejected by the ERP fails pragmatically. For geometry, the SASIG product data quality guideline, published as ISO/PAS 26183:2006 [2], and its German counterpart VDA 4955 “Umfang und Qualität von CAD/CAM-Daten” [3] define the gaps, micro edges and degenerate faces that make a model unfit for exchange.
| Dimension | Example checks | Metric it feeds |
|---|---|---|
| Geometry validity | Closed solids, sliver faces, PDQ criteria per VDA 4955 | Findings per part, migration acceptance rate |
| Model structure | Suppressed features in released parts, broken WAVE links, unused sketches | Findings per part, mean time to fix |
| Attributes and metadata | Required attributes present, values from lists, part and drawing consistent | Critical findings share, auto-healed share |
| Standards compliance | Layers, naming patterns, standard parts from the reuse library | Repeat-offender rules |
| PMI and drawings | Datum references, PMI view count, view alignment, sheet template | First-pass yield per release |
| Downstream fitness | ERP import success, supplier queries, CAM rework | Downstream rejection rate |
Our clear recommendation: report every dimension separately. A single quality score hides the fact that attributes are fine and PMI is not, and nobody can act on it.
Eight CAD Quality KPIs and How Automated Checks Produce Them
Eight CAD data quality metrics cover what an engineering organization needs to steer with. All are computed from the same raw material, one row per check execution (part, release, run number, check identifier, severity, result, healing status), plus two feeds from outside NX. If the checker logs those fields, the KPIs are arithmetic.
| KPI | Definition | Computed from check results | Where the data comes from |
|---|---|---|---|
| First-pass yield per release | Share of parts with zero error findings in the first check run of a release | Parts without errors in run 1 / parts checked in run 1 | Check log, Teamcenter Item Revision properties |
| Findings per part | Average findings per checked part, by severity | All findings / parts checked | Check log |
| Critical findings share | Share of findings classified as error (blocking) | Error findings / all findings | Check log plus severity from the rule catalogue |
| Mean time to fix | Average time from first failed run to first passed run | Timestamp of first pass minus first fail | Check log timestamps or workflow history |
| Auto-healed share | Share of findings corrected without designer action | Healed findings / all findings | Healing log of the checker |
| Repeat-offender rules | Rules in the top five failing checks for three consecutive months | Monthly ranking of checks by parts affected | Monthly aggregation of the check log |
| Downstream rejection rate | Released parts rejected by ERP or BOM import, or queried by a supplier, per 100 releases | Rejections / releases, times 100 | ERP import log, supplier ticket system |
| Migration acceptance rate | Share of migrated parts passing the acceptance profile without manual rework | Passed parts / migrated parts | Batch run over the migrated dataset |
First-pass yield is the number for management: how much of what we release is right the first time. Findings per part is the number for methods: where the template fails the designer. Downstream rejection rate proves the other two matter; if the checks report a 95 percent first-pass yield and the ERP still rejects one BOM in ten, the catalogue is checking the wrong things. Those rejections are the subject of our guide to automating BOM and ERP sync from CAD models; the migration acceptance rate is described in our article on what to validate after a migration to Siemens NX.
Baselining: One Batch Run Over the Vault
A baseline is a single batch run of the full check profile over the existing vault before any rule is enforced, and it is the step organizations skip most often. Without it, every threshold is a guess.
Both tool families support this. Siemens NX Check-Mate, the native validation framework, ships hundreds of standard checks, runs in batch mode and can save its results to Teamcenter [4][5]. DOCUFY CAx Quality Manager (DQM), fully integrated into Siemens NX with more than 450 standardized and configurable best-practice checks for 2D drawings and 3D parts, provides a batch mode for mass data analysis of large data volumes; it is the product inteliScience runs baselines with as an official DOCUFY partner [6].
What the baseline run should produce, in this order:
- Findings per part and critical findings share for the whole vault, per dimension.
- The top ten failing checks with the share of parts they hit: the rule review agenda.
- The share of findings healed automatically. Intelligent healing functions in DQM correct errors for supported checks and give guided correction instructions for the rest, so the healed share shows how much of the problem disappears without training [6].
- A simulated gate: how many parts released last quarter an error-severity rule would have blocked.
We’d advise against publishing the baseline as a league table of designers. The point is to find bad rules and bad templates; a baseline that names people produces the wrong kind of quiet.
The checks run. What’s missing for production: a rule catalogue with owners, a baseline over your vault, KPI thresholds and a Teamcenter gate that reads the results. inteliScience closes the gap from “the checker reports findings” to “the organization steers on data quality metrics”, with NX specialists delivering nearshore from Romania (EET, one hour ahead of CET), starting with a 90-day evaluation on your own NX data.
Where Results Live: Teamcenter as the Source of Truth
A quality gate can only read what the checker writes to a place the release workflow can see, and for NX organizations that place is Teamcenter. A PDF report attached to an email is not a metric; a property on the Item Revision is.
The integrations have matured. The native framework saves results alongside the part [5]. DQM version 2026.1, published on 30 March 2026, supports NX up to and including NX 2512 and writes check results as properties on Item Revisions, transfers them to datasets and reads and writes the Validation Master Form [7]. The same release added seven base check routines, from a PMI view check and validation of composite datum references to alignment of drawing views and a clash analysis check; each is a new source column in the KPI table above.
Once the result is a property, the workflow handler needs to know nothing about CAD. It reads “errors = 0” or it does not, and the monthly report becomes a query over Item Revisions. A property per rule family lets the gate block on attributes today and merely warn on PMI until that family is ready.
Computing the KPIs From Check Results
The arithmetic fits in forty lines of standard-library Python. The script reads a CSV with one row per check execution, computes first-pass yield, findings per part, critical and healed share, and lists the top failing checks by the share of parts they affect. It runs on the export of any checker that logs these seven fields.
import csv
from collections import Counter, defaultdict
# One row per check execution: part_id, release_id, run_no, check_id, severity, result, healed
# result is "pass" or "fail"; run_no 1 is the first check run of that release cycle.
RESULTS = "check_results.csv"
rows = list(csv.DictReader(open(RESULTS, newline="", encoding="utf-8")))
# First-pass yield: parts with zero error-severity failures in run 1 of their release
first_run = [r for r in rows if r["run_no"] == "1"]
parts_in_first_run = {(r["part_id"], r["release_id"]) for r in first_run}
parts_with_error = {
(r["part_id"], r["release_id"])
for r in first_run
if r["result"] == "fail" and r["severity"] == "error"
}
first_pass_yield = 1 - len(parts_with_error) / len(parts_in_first_run)
# Findings per part, all severities, all runs
findings = [r for r in rows if r["result"] == "fail"]
parts_checked = {(r["part_id"], r["release_id"]) for r in rows}
findings_per_part = len(findings) / len(parts_checked)
# Critical findings share and auto-healed share
critical_share = sum(r["severity"] == "error" for r in findings) / len(findings)
healed_share = sum(r["healed"] == "yes" for r in findings) / len(findings)
# Top failing checks, with the share of parts they hit
parts_per_check = defaultdict(set)
for r in findings:
parts_per_check[r["check_id"]].add((r["part_id"], r["release_id"]))
top_checks = Counter({c: len(p) for c, p in parts_per_check.items()}).most_common(5)
print(f"Parts checked: {len(parts_checked)}")
print(f"First-pass yield: {first_pass_yield:.1%}")
print(f"Findings per part: {findings_per_part:.2f}")
print(f"Critical share: {critical_share:.1%}")
print(f"Auto-healed share: {healed_share:.1%}")
print("Top failing checks:")
for check_id, n in top_checks:
print(f" {check_id:<32} {n / len(parts_checked):.1%} of parts")Two details matter. First-pass yield counts only run 1 of each release; count every run and a part that failed four times before passing looks like a success. And the top failing checks are ranked by parts affected, not by finding count, because one part with 200 unused sketches should not outrank a rule that fails on half the vault.
Rollout: Monthly Report, Thresholds, Gate
Rolling out CAD data quality metrics takes three months of reporting before the first gate blocks anyone, and we hold to that sequence even when a client wants the gate on day one. Numbers nobody has seen cannot be defended in the release meeting.
- Month 1: report only. Publish the baseline and the first monthly KPI report per dimension. Fix the three worst rules or templates, not the designers.
- Month 2: thresholds. Agree a warning threshold per rule family, for example first-pass yield on attributes above 90 percent, and let the checker warn at save without blocking.
- Month 3: gate on one family. Make error findings in the attribute family a release condition in Teamcenter. Attributes heal best and break ERP first, so this is where a gate earns trust.
- Quarterly: extend and re-baseline. Add the next rule family and rerun the batch after every NX upgrade. Check behaviour shifts between versions: Siemens’ June 2026 release renamed NX to Designcenter [8], and the Early Access Program for Designcenter 2612 ran its on-site weeks from 14 to 25 September 2026 in Mason, Ohio [9], so re-validate the check profile on each release before the gate meets it.
The monthly report is one page: the eight KPIs with trend, the top five failing checks, the repeat offenders, and one decision per dimension. The rules behind it come from the catalogue work in our guide to turning company design rules into executable checks; metrics are worthless if nobody owns the rules.
Frequently Asked Questions
What are the most important CAD data quality metrics for an NX organization?
First-pass yield per release, findings per part by severity, auto-healed share and downstream rejection rate. The first three come directly from the automated checks; the fourth comes from the ERP import log and supplier queries and confirms that the checks measure what matters.
How do you baseline CAD data quality before introducing a gate?
Run the complete check profile in batch mode over the existing vault, aggregate findings per dimension and rule, list the top ten failing checks with the share of parts they hit, and simulate the gate against last quarter’s releases. Publish the result by rule, not by designer.
First-pass yield vs findings per part: which KPI should we steer with?
Both, for different audiences. First-pass yield is the management number: how much released data was right the first time. Findings per part is the methods number: which rules, templates and trainings need work. Steering with one of them hides either the causes or the outcome.
How much does it cost to set up CAD data quality measurement in Siemens NX?
The tooling is the smaller part. DOCUFY offers a 90-day DQM evaluation package with company-wide unlimited use at a fixed price of EUR 2,950 plus VAT for Germany, Austria and Switzerland, including installation, basic customizing and phone support. The larger part is the methods effort for the rule catalogue, the baseline and the Teamcenter workflow change.
Can a Teamcenter release workflow block on CAD quality metrics?
Yes, provided the checker writes its results where the workflow can read them: as properties on the Item Revision or in the Validation Master Form. A workflow handler then checks the error count per rule family and stops the release. Reports that live only in the checker or in email cannot serve as gate conditions.
Next Step
Related Topics
- DOCUFY CAx Quality Manager for Siemens NX
- Configured checks vs custom checkers: which quality solution fits your process
- The Siemens NX CAD Quality Playbook (whitepaper)
References
- ISO, “ISO 8000-8:2015 Data quality, Part 8: Information and data quality: Concepts and measuring,” https://www.iso.org/standard/60805.html, retrieved 24 September 2026.
- ISO, “ISO/PAS 26183:2006 SASIG Product data quality guidelines for the global automotive industry,” https://www.iso.org/standard/43436.html, retrieved 24 September 2026.
- VDA, “VDA 4955 Umfang und Qualität von CAD/CAM-Daten (12/2006),” https://webshop.vda.de/VDA/en/vda-4955-122006, retrieved 24 September 2026.
- Siemens, “CAD model design validation,” Designcenter NX product page, https://www.siemens.com/en-us/products/designcenter/nx-cad-software/mcad/design-validation/, retrieved 24 September 2026.
- Siemens, “Ensure CAD quality with NX Validation tools: What’s new,” NX Design blog, 25 August 2021, https://blogs.sw.siemens.com/nx-design/ensure-cad-quality-with-nx-validation-tools-whats-new/, retrieved 24 September 2026.
- DOCUFY GmbH, “DOCUFY CAx Quality Manager,” product page, https://docufy.de/en/products/for-your-product-lifecycle-managementfuer-ihr-product-lifecycle-management/docufy-cax-quality-manager/, retrieved 24 September 2026.
- DOCUFY GmbH, “NX-Datenqualität sichern: Neue Version des DOCUFY CAx Quality Managers” (DQM 2026.1), 30 March 2026, https://docufy.de/nx-datenqualitaet-sichern-neue-version-des-docufy-cax-quality-managers/, retrieved 24 September 2026.
- Siemens, “Designcenter is for everyone: June 2026 release,” Designcenter blog, 23 June 2026, https://blogs.sw.siemens.com/designcenter/designcenter-is-for-everyone-june-2026-release/, retrieved 24 September 2026.
- Siemens, “Welcome to the September 2026 Early Access Program for Designcenter,” https://resources.sw.siemens.com/en-US/presentation-welcome-to-designcenter-early-access-program/, retrieved 24 September 2026.