Say you slap an OEE dashboard up on Line 3 on a Monday. Big TV, nice colors, gauges galore. By Wednesday the line lead swears the number is wrong. By Friday...nobody's looking at it. Poor little dashboard.
Happens all the time - and the screen is almost never the culprit. The trouble starts the first time somebody asks what the number actually means.
Quick refresher, since everything hangs on it. OEE is availability x performance x quality. How much of the planned time the machine actually ran, how fast it ran vs. how fast it should, and how much of what it made was good. Six of eight planned hours, at 90% of ideal speed, with 98% good parts - that's about 66%.
Now, back to that Wednesday squabble. The line lead says the 40-minute changeover shouldn't count against them, because changeovers are "planned." Fair enough...maybe. But call changeovers planned and they vanish from your numbers entirely, and on plenty of lines they're the biggest chunk of lost time you've got. I'd rather count 'em and stare at 'em than sweep 'em under the rug. Whatever you decide, write it down. That same argument comes back every time someone new walks up to the screen.
Next squabble: ideal cycle time. Compare against a speed that's too slow - your slowest good day, or some number somebody guessed years back - and performance looks fantastic forever. Fantastic and useless. Use what the machine can really do.
Then downtime reasons. Hand an operator a tablet with 40 reasons to scroll through and guess what they'll pick? "Other." Every time. Your Pareto chart turns into one giant "Other" bar. Keep it to eight or ten reasons, two taps, done. You can always split one up later once you see where the time is going.
On the data side, let the machine tattle on itself as much as possible: running, faulted, good count, reject count. If your PLCs are on Ethernet, those come straight off tags, usually through a little edge gateway or something like Ignition into a database. Older machines with no network port can still join the party with a couple of sensors and a small I/O block. The only thing people should have to type in is the one thing the machine can't know...why it stopped.
Automatic capture also catches the sneaky stuff - the short stops. A machine that hiccups for 20 seconds forty times a shift loses more time than one 10-minute jam, and nobody ever writes the hiccups down.
The screen itself? Keep it boring. What each machine is doing right now, today's OEE with the three parts broken out, and the top downtime reasons. Shift comparison if folks want it. Forty gauges is a Christmas tree, not a dashboard.
And don't sweat that "85% is world class" number floating around every slide deck on the planet. Track your own line. Line 3 going from 55% to 65% is real money, whatever the brochure says.
If I were starting from scratch, I'd pick the bottleneck line, collect data for two or three weeks before showing anybody a chart, and hash out the definitions as they come up. Then on to the next line.