Automated PMI Placement and Numbering at an Industrial Machinery Manufacturer with 100 NX Users
Automated PMI placement and numbering replaced a manual ballooning routine at an industrial machinery manufacturer with more than 9,000 employees and about 100 Siemens NX users. inteliScience SRL built an NX Open tool that places and numbers PMI elements by the X, Y and Z coordinates of a chosen coordinate system, and re-orders and renumbers them by itself when the design changes. The client reports that preparing a part so the checking software can read its PMI went from days to minutes.
This automated PMI placement and numbering case study covers the ordering half of model-based definition. Our PMI authoring automation case study covers the other half — the rule set that decides which callout belongs on a feature, at an automotive OEM with 300+ designers. Here the callouts already existed; what broke was keeping their placement and their numbers consistent through revision after revision.
An industrial machinery manufacturer with about 100 Siemens NX users now balloons and numbers PMI automatically instead of by hand, and the client reports that getting a part ready for the checking software takes minutes where it used to take days. The tool sorts PMI elements by the X, Y and Z coordinates of a coordinate system, places the balloons in that order, numbers them, and — the part that mattered most — re-runs the whole assignment when the design changes, so the numbering never has to be repaired manually. Two inteliScience engineers built it in about six months on Siemens NX 2412.
LAST UPDATED AUGUST 21, 2026 · FIGURES RELEASED BY THE CLIENT FOR ANONYMOUS PUBLICATION
- Client
- Industrial machinery manufacturer, more than 9,000 employeesNOT NAMED
- Sector
- Industrial machinery
- Users
- About 100 Siemens NX users
- Scope
- Coordinate-based PMI placement, ballooning, numbering and automatic renumbering
- Duration
- About 6 months
- Team
- 2 developers and consultants
- Environment
- Siemens NX 2412
- Our role
- NX Open development and PMI method consulting — inteliScience SRL
The situation: the checking software could not read what people had numbered by hand
The manufacturer does not use PMI as decoration. Downstream of design, a checking software reads the annotated model and works through the callouts one by one to verify the part. That handover only works if every callout carries a unique, stable identifier and sits where the inspection planner expects it. In other words, the numbering is the interface.
Producing that interface by hand is slow in a way that is easy to underestimate. A designer places each balloon, assigns each number and then walks the part again to confirm that nothing is duplicated, missing or out of sequence. For a machinery part with a serious number of dimensions and tolerances, that is a day of work at least — and the client reports it regularly ran into days before the part was fit for the checking software.
The expensive part is not the first pass, though. It is every pass after it. A design change — one added feature, one removed hole, one re-ordered face — invalidates the sequence, and someone has to walk the entire part again to repair numbers that are no longer contiguous or no longer match the inspection plan. This is not an exotic failure mode. ISO 16792:2021, the standard that governs digital product definition data, explicitly covers the preparation, revision and presentation of that data — revision is named in the standard because revision is where the practice breaks.
The industry numbers point the same way. In NIST Advanced Manufacturing Series 100-26 (2020), the cost of inadequate design and model data in US discrete manufacturing includes an estimated 8.4 billion USD a year spent on engineers answering questions and creating additional drawing documentation, plus 3.8 billion USD on machinists doing the same. The same NIST work found that only about 26.8 percent of survey respondents released 51 percent or more of their designs with PMI embedded in 3D models. Both figures are industry context, not this project’s results — the results are further down and carry their own stamps.
What changed in the working day
- manualPlace every balloon by hand, one callout at a time
- manualAssign each number and keep the sequence contiguous
- by eyeWalk the part again to find duplicates, gaps and misplacements
- repeatRepair the numbering by hand after every design change
- fragileInconsistent identifiers are of no use to the checking software
- by ruleBalloons are placed in coordinate order, X then Y then Z
- generatedNumbers follow from that order, uniquely and contiguously
- automaticA design change triggers re-ordering with no manual pass
- consistentThe same part geometry always produces the same numbering
- readableThe checking software gets identifiers it can process
What the client reports
The figures below come from the client and were released for anonymous publication. They describe what the design and inspection planning teams observe in their own work; they are not a controlled study, and the stamps say so.
Released by the client for anonymous publication.
Getting a part ready for the checking software
client-confirmed The ballooning and numbering pass regularly ran into days for a typical machinery part; with the automation the client reports minutes. No exact before and after times were released, so this is stated as the client stated it. The bar indicates the order of magnitude described, not a measured ratio.
Renumbering after a design change
client-confirmed Qualitative statement, no metric attached. Every change used to require walking the part again to repair the sequence; the algorithm now re-derives order and numbers from the geometry, so there is no repair step to schedule and no drift between revisions.
NX users working to the automated placement and numbering
client-confirmed Rolled out to the Siemens NX user base of a company with more than 9,000 employees, not to a pilot group.
How does coordinate-based PMI numbering actually work?
Coordinate-based PMI numbering works by taking the decision away from the person. Every PMI element has a position in space; a coordinate system gives that position an unambiguous X, Y and Z. Sort by those three values in a defined precedence and the result is a deterministic sequence that two engineers, two sessions and two revisions all agree on. Everything else in the tool follows from that one decision.
The coordinate system becomes the sort key
The tool reads the PMI elements of the part and orders them by the X, Y and Z coordinates of a chosen coordinate system, with a defined precedence and defined tie-breaking. This is deterministic geometry, not judgement and not a model: the same part always yields the same order. Agreeing which coordinate system is authoritative, and what happens on parts whose orientation does not match it, was the real design work.
OUTPUT — one reproducible reading order per partBalloons go where the order says and carry the number it gives them
Placement and labelling are then mechanical. Balloons are placed following the sequence, numbered uniquely and contiguously, and written back to the model so the checking software finds identifiers rather than free text. What used to be a careful manual pass becomes a command.
OUTPUT — ballooned, numbered PMI the checking software can processA design change re-runs the assignment instead of breaking it
This is the layer that decides whether such a tool survives contact with real work. When the design changes, the algorithm re-evaluates the coordinates and re-assigns order and numbering across the part. Nobody repairs a sequence by hand, and no revision inherits the previous revision’s mistakes. The documentation stays consistent because it is derived, not maintained.
OUTPUT — numbering regenerated on every change, not patchedWorth saying plainly, because the question comes up in every MBD conversation: there is no language model anywhere in this. Placement and numbering are geometry and rules, they have to be reproducible, and a reproducible rule is exactly the kind of thing that should never be handed to a probabilistic system. Where we do use AI at inteliScience, it assists people — it does not decide pass or fail on release data.
What the solution is built on
Supported Siemens extension points and the client’s own conventions. Nothing that dies with the next NX upgrade.
Related work on the same ground: MBD and PMI compliance automation for Siemens NX, NX Open custom tool development, and, for the programming side of this kind of work, NX Open programming for machine builders.
What we would do differently
We would build the renumbering diff report first, not last. The algorithm was correct early, but the inspection planning side had no way to see what had changed between two revisions — which callout kept its number, which one moved, which one was new. Without that view a correct renumbering still looks like a risk, and people compensate by checking it manually, which is exactly the work the tool was supposed to remove. A short report listing old number, new number and reason would have bought trust in week three instead of month four. We would also settle the coordinate-system convention in writing before the first line of ordering code: several early bug reports were not bugs but undocumented exceptions, parts whose functional orientation did not match the coordinate system everybody assumed was authoritative.
When does automated PMI placement and numbering pay off?
Automated PMI placement and numbering pays off when the numbers are an interface to something else, and when parts change. The case is strong when a downstream consumer — checking or inspection software, CMM programming, a first article inspection process — reads identifiers rather than a person reading a drawing; when parts carry enough callouts that a manual pass costs hours; and when revisions are frequent enough that the repair work repeats. It is a weak case when a part has a handful of callouts, when nothing downstream consumes the numbering, or when the organisation has not decided which coordinate system is authoritative — that decision is a prerequisite, not something automation can supply.
Frequently asked questions
How does the tool decide the order of PMI elements?
What happens to the numbering when the design changes?
How long does a project like this take?
When would this not be worth it?
Why is the client not named?
Your inspection data is only as good as its numbering.
Send us one annotated part and the rule your inspection team expects the numbering to follow — we will show you how much of that ballooning is deterministic, on your own model.
Talk to an MBD Automation Engineer