The paper shift reports of a machining hall show an OEE of 78 percent. After four weeks of automatic machine data, the same twelve machines show 61 percent. The first reaction in the meeting is that the new system must be wrong.
It is usually not. It is the first honest number the area has seen.
Four states are enough
For overall equipment effectiveness, a machine needs to report only a few things:
- Running: program active, parts counted.
- Stopped: no program, no setup booked. Needs a reason.
- Setup: booked by the operator. A planned loss.
- No plan: no shift planned. Not part of planned time.
Modern CNC and PLC controls deliver these states through standard interfaces, read by an edge gateway in the hall. Older machines can be read with a simple signal adapter on the stack light or the spindle current. That gives running and stopped, which is already enough for availability.
One written definition
Before the first number is shown, write the definition on one page:
- Planned time: 3 shifts × 7.5 hours × 5 days = 112.5 hours per machine and week.
- Availability: run time divided by planned time. Setups count as loss.
- Performance: ideal cycle time from the routing times parts, divided by run time.
- Quality: good parts divided by all parts.
OEE is the product of the three. Change the definition only with a recorded decision, or trends lose their meaning.
Why the gap appears
In the example, availability fell from 89 percent on paper to 74 percent in the data. Stops under ten minutes were never written down, and setups were reported with their planned instead of their actual time. Performance fell from 93 to 88 percent because slow running is invisible on paper. Quality stayed at 94 percent, because scrap had always been counted.
Say this openly in the first meeting. If the gap is treated as a scandal, people will defend the old number instead of attacking the losses.
Validate at the machine
Trust in data is built at the machine, not in a presentation. Stand at two machines for one hour each with a stopwatch, write down every state change and compare it with the data for the same hour. Fix signals or definitions until both match.
Check the data quality every week: gaps in the recording, performance values above 100 percent (a sign of wrong ideal cycle times in the routing), and the share of stop minutes without a reason.
The spread is the work list
The average says little. In the example the best machine reached 73 percent and the weakest 52 percent. The two weakest were older machines with long changeovers. They became the focus of the first improvement workshop, and the target for the area was set at 68 percent within six months.
That target was reached. Not because of the data itself, but because the meeting started acting on it every day.