Machine Monitoring Without Buying New Sensors
Machine monitoring projects die in the quotation stage. The sensor vendor counts machines, adds protocol gateways and installation labor, and the number that comes back is bigger than the number the factory manager controls. Meanwhile the machines keep running blind, not because data is impossible, but because the project was scoped as all machines, automatically, now.
There is a cheaper sequence. Start with data that already exists or costs a keystroke, and let the money follow the gaps that actually hurt.
Level 0: Operator-Reported Counts and States
A terminal at the station, an operator reporting good and scrap at the end of each run, a start and stop button. This costs a screen you already bought for the MES and zero sensors, and it produces a surprising fraction of the value: real production counts, real downtime reasons, and an OEE number computed from data instead of memory. Its known weakness is honesty and granularity: the terminal knows the run was 40 minutes, not that the machine was jammed for 6 of them. That is fine. Level 0 exists to prove the workflow and baseline the numbers.
Level 1: Mine What the Machine Already Exposes
Most machines installed in the last twenty years already speak something:
- CNCs: Focas, OPC UA, or MTConnect on many Fanuc, Haas, DMG, and Mazak controls. Cycle start, program number, spindle load, alarm history: all there.
- PLCs: if an HMI or SCADA already reads the machine, the same tags can feed a second reader. A keyed SCADA ingest that already exists is the cheapest sensor you will ever "buy".
- Older machines: a cycle-start relay and a parts-counter signal wired to a cheap I/O module gives running/stopped and counts, which is most of what a monitoring wall needs.
The discipline that keeps this cheap: read counts and states first. Process parameters (pressures, temperatures, torque) are valuable (see machine characteristics on the terminal) but they can trail the basics by a phase.
Level 2: Retrofit Sensors Only Where the Gap Costs Money
Now the quotation is different, because you know exactly which machines justify it. The bottleneck machine whose six-minute jams are invisible at Level 0 gets the vibration/current sensor. The station whose scrap nobody can explain gets the process-parameter tap. The machines that run fine with operator counts keep running on operator counts, possibly forever, and that is the correct outcome.
What the Data Feeds
Monitoring is not the deliverable. The deliverables sit one step downstream, and they are worth the sequence:
- Live andon states: red tiles with stop timers, fed by machine state where it exists and operator state where it does not (the andon board mixes both honestly).
- Downtime with reasons: machine-detected stops land in the same Pareto as operator-entered reasons, so the biggest loss is a number, not an argument.
- OEE per order and shift: availability from states, performance from counts versus standard, quality from the inspection record. All three from the same keyed ingest.
Frequently Asked Questions
Isn't operator-reported data too inaccurate?
It is less granular and occasionally optimistic, and it is still the data most factories run on today, except theirs lives on paper. Level 0 replaces paper with a record you can query. When the numbers start driving decisions, upgrade the machines where the error actually costs.
Will connecting to our PLCs void warranties or break anything?
Reading is non-invasive: a second connection that observes tags the HMI already uses. Write nothing. Agree the tag list with your controls tech, use read-only access, and the machine cannot tell you are there.
How many machines before this stops working?
The scaling limit is keys and stations, not sensors: a keyed ingest per data source, a terminal per station. Factories run dozens of mixed sources (some automatic, some manual) on one board; the mix is a feature, because each machine gets the cheapest level that produces trustworthy data.
Start With the Terminal, Upgrade Later
Voltrus MES ingests machine states through keyed SCADA connections and takes operator-reported data everywhere else: the andon board and OEE mix both. One line to start, sensors only where they pay.
See Voltrus MES