Maintenance owns the sensors. Production owns the machines. IT owns none of the equipment — and ends up owning the pipeline that connects all of it, which is exactly where most automotive PdM programs quietly break. iFactory, an on-premise AI platform for predictive maintenance and OEE analytics, treats that pipeline as plant IT's actual deliverable: the network, the sync, the data contract between CMMS and MES. Walk through your own data architecture with the team.
Maintenance Owns The Sensors. IT Owns Whether They Actually Talk To Each Other.
A PdM model can't fix a broken pipeline. This is the checklist for the team that owns the pipeline.
Three Teams, One Pipeline, Nobody Owns The Middle
Reliability picks the sensors. Production runs the MES. IT gets asked why the model is wrong — and usually nobody told IT the pipeline was theirs to keep honest.
A stamping press's ram-position sensor can be perfectly calibrated and still feed a model garbage, if the network path between that sensor and the CMMS drops packets during a shift change or the timestamp drifts by even a few seconds against production events. That's not a sensor problem. That's a plant IT problem, and it rarely shows up on anyone's checklist until the model's predictions start missing.
The ownership gap is the real issue, not any single technical failure. Reliability teams write checklists for sensors. Data science teams write checklists for models. Almost nobody writes the checklist for the layer in between — the network paths, the synchronization windows, and the shared identifiers that make the first two checklists mean anything at all.
Nine Things Plant IT Should Own Before Blaming The Model
- 01 — Network path mapped end to end. Every sensor's route to the historian, documented — not assumed from the original install diagram.
- 02 — Time sync verified across sources. Sensor clocks, PLC clocks, and MES event timestamps aligned within the tolerance the model actually needs, not just "close enough."
- 03 — Asset ID shared, not duplicated. The same equipment carries one identifier across CMMS and MES — not two numbering systems quietly drifting apart.
- 04 — Failure codes mapped, not just copied. A downtime reason in MES and a failure code in CMMS describing the same event, reconciled into one shared library.
- 05 — Sync frequency matched to decision speed. Real-time where a decision needs real-time; batch where it doesn't — not one setting applied everywhere out of convenience.
- 06 — Dropped packets and gaps monitored. A data completeness dashboard that IT watches directly, not one maintenance has to request after a model already went stale.
- 07 — Change control on data mappings. When a tag name, sensor ID, or failure code changes, the mapping updates everywhere at once — not in CMMS this week and MES next quarter.
- 08 — OT network segmented and secured. PdM data traffic isolated enough that a plant floor network issue doesn't become a corporate security incident, and vice versa.
- 09 — Audit trail on every sync. Who changed what mapping, when, and why — recoverable months later when a model's accuracy drop needs a root cause.
Which Of These Nine Is Actually Covered On Your Floor?
Walk through your current CMMS-MES setup with the team — most plants are solid on two or three of these and have never looked closely at the rest.
The Gap Nobody's Monitoring Until The Model Misses
A model doesn't announce that its input data went bad. It just gets quietly less accurate.
| Symptom on the floor | What it usually means | Who should catch it |
|---|---|---|
| MES shows a stop with no matching CMMS work order | Asset ID or failure-code mapping drifted apart | IT — mapping governance, not maintenance |
| Predictions degrade gradually, no clear trigger | Timestamp drift or a growing gap in sensor completeness | IT — monitoring the pipeline, not the model |
| Same equipment, two different asset numbers | CMMS and MES were never reconciled to one identifier | IT — shared master data, set up once |
The Model Wasn't Wrong. The Clock Was.
A body-shop welding robot's predictive model starts missing early-warning windows it used to catch reliably. Reliability engineering re-tunes the model twice. Nothing improves. IT finally traces it to a PLC clock that drifted eleven seconds out of sync with the plant time server after a firmware update three months earlier — eleven seconds was enough to misalign sensor readings against the production events the model used for context. The model was reading the right data at the wrong time, which to a prediction engine looks exactly like the wrong data.
Nobody was monitoring clock drift because nobody had been told it was theirs to monitor. That's the gap this checklist exists to close.
Four Questions To Ask Before Re-Tuning Anything
- Has the asset mapping changed recently — a line reconfiguration, a new tag naming convention, a CMMS migration?
- Is the completeness score actually being watched — or does it only get checked after someone complains?
- Are CMMS and MES agreeing on the same failure — or does a cross-check turn up events on one side with nothing on the other?
- Did a network or firmware change happen upstream — a change nobody connected to the model's declining accuracy?
Re-tuning a model before answering these is like replacing a part before checking if it was actually broken — expensive, time-consuming, and usually pointed at the wrong layer of the problem. Most accuracy complaints that reach a data science team should have stopped one layer earlier, at whoever owns the pipeline feeding it.
The Pipeline Monitored, Not Just The Model
iFactory treats the CMMS-MES-sensor pipeline as a first-class thing to monitor, not a one-time integration project that's assumed to keep working forever. Shared asset identifiers, mapped failure codes, and synchronized timestamps are maintained continuously, with drift and gaps surfaced before they reach the model.
- Asset IDs shared across CMMS and MES from one source of truth, not reconciled manually after the fact.
- Timestamp and completeness monitored as a live score, not an annual audit finding.
- Mapping changes tracked with a full audit trail, so a drifted sync has a traceable cause.
- On-premise deployment keeps the full data pipeline inside the plant's own network boundary.
None of this removes the need for good sensors or a well-tuned model. It just means the layer between them finally has an owner, a dashboard, and an alert the moment something drifts — instead of being invisible until a prediction quietly goes wrong.
What A Plant IT Lead Should Check First
A PdM data pipeline earns trust when IT can point to what's being monitored continuously, not just what was configured correctly at go-live.
What Plant IT Teams Ask About PdM Data Quality
Sensor selection and calibration are maintenance's call — but the network path, time sync, and system-to-system mapping that gets that data to the model reliably is squarely an IT deliverable. Talk through the split with the team.
Even a few seconds of drift can misalign sensor readings against the production events a model depends on for context — enough to quietly degrade predictions without an obvious cause. See your own sync tolerance assessed.
Cross-check recent downtime events in MES against work orders in CMMS for the same asset and time window — a mismatch almost always points straight at the gap. Ask support how this cross-check is automated.
Not necessarily — most plants already have the systems in place, just without continuous monitoring on the mapping and sync layer between them. Audit your current setup before assuming a rebuild is needed.
Plant IT, since it's a pipeline health metric rather than an equipment health metric — treating it as maintenance's job is usually how it goes unwatched. See how ownership is split in a live deployment.
Get Your Own Nine-Point Pipeline Review
Book a session with the iFactory team and find out which of the nine checklist items your current CMMS-MES setup actually covers.
