Key takeaway
Connect a few machines, agree the numbers with the people who run them, then switch the paper off. Do not skip the agreeing.
01Week 1: connect the machines that matter
Start with one line, ideally the one everyone argues about. Connect the controls that expose data directly. Add sensors to the machines that do not. Do not wait for the perfect integration to every machine; six connected machines that the plant trusts beat twenty-two that it does not.
- Controls with OPC UA or MTConnect connect in hours
- Legacy machines get a current clamp or IO signal
- Operators get a terminal, not a training course
02Week 2: validate against the floor
Run the new data next to the old shift sheet for a week. Every day, compare. Where they differ, walk to the machine and find out why. Usually the system is right and the sheet was optimistic. Sometimes a signal is wired wrong. Either way, by the end of the week the numbers are agreed, and that agreement is what the whole rollout rests on.
03Week 3: switch the paper off
Stop the manual report. Not 'in parallel', not 'as a backup'. If the sheet still exists, it will still be the source of truth in the daily meeting. Put the live screen where the sheet used to be pinned. Start the shift handover from it.
Expect discomfort for a few days. The live data shows stops that never used to be written down. That is the point.
04Week 4: pick one loss and remove it
Look at the Pareto for the first three weeks. Pick the largest bar that the line can influence without capital. Material wait is the usual candidate. Assign an owner, set a target in minutes per shift, and review it every day for two weeks. The first visible win is what turns a monitoring system into a way of running the plant.
05Then: the next line
Once one line runs on live data, the second line asks for it. Rollouts that start with the whole plant tend to stall in integration. Rollouts that start with one line and one win tend to finish.