Unknown downtime is any stop logged as other, unknown or with no reason, so it tells you the logging failed and tells you nothing about the machine. Shrink it by sorting those stops by length: capture short stops from a signal, improve the pick list for medium ones, and ask about long ones during the same shift.
Why this bucket grows
Nobody picks "other" to be difficult. They pick it because the right code is missing, the list is too long, or the line is already running again and they have parts to deal with.
The bucket is also not rare in published data. A study of 884 machines in 23 Swedish companies found that "unclassified losses" were about 19% of scheduled production time (Hedman et al., 2016), which was more than half of the recorded loss time, and that was with automatic measurement in place.
So do not expect signals alone to solve it. A machine can tell you when it stopped, while a person or a camera still has to say why.
Measure the share first
Minutes and counts give different pictures. In this example unknown stops are 30% of the minutes yet 46% of the events, which already suggests that many of them are short.
Any export with a reason and a duration column is enough for this. Count blanks, "other", "unknown", "misc" and any other catch-all together, because crews tend to invent new names for the same bucket.
In a spreadsheet, flag those rows in a helper column, then use SUMIFS on the flag and the duration to get each length band. Ten minutes of setup gives you a number you can refresh every week.
Sort the bucket by stop length
This is the step that makes the problem workable. Stops of different lengths go unexplained for different reasons, so a single fix will not reach them all.
Under 2 minutes: too short to log by hand
Operators clear these and keep running, and asking them to type a reason for each one will fail. Treat them as a group, and read what micro-stops are if the term is new.
Group them by machine and alarm where you can, and decide where the line falls with a stop threshold. A camera on the worst station can show the cause, and computer vision for OEE explains how.
With no signal or camera, sample them. Stand at the worst machine for two hours, tally each short stop by what you see, and treat the result as a rough split and not a measurement.
2-10 minutes: the pick list failed
Someone saw the cause, but the code was hard to find or did not exist. Read the notes attached to these stops, then count what people wrote.
Three repeat phrases usually become three new codes. Remove codes that nobody used in the past three months, so that the list stays short.
Over 10 minutes: someone knows
A stop this long had people around it. Set one rule: any stop over 10 minutes with no reason gets a question from the team lead before the shift ends.
Next-day questions get worse answers, since people have moved on and memories have blurred. This group is small in minutes, and it should reach close to nothing within weeks.
Read the pattern before you fix it
| What you see | What it usually means |
|---|---|
| One shift has far more unknown | Training or habit on that shift. Ask them first, as they may have a good reason. |
| One machine has far more | Codes do not fit that machine. Add a second level for it. |
| Many entries at the end of the shift | People are catching up from memory, so log nearer to the machine. |
| The first code on the list is suspiciously popular | Fast tapping, with no real choice. The list needs better order. |
| Unknown falls, then returns | Nobody acted on the data, so people stopped seeing a reason to log. |
Rules that keep the bucket small
- Split "other" from "not yet known". Other means no code fits. Not yet known means someone is still looking.
- Require a short note on "other". Three words are enough, and they feed your next code review.
- Show the share on the line's board. A visible number gets attention at no cost.
- Act on what people log. Tell the crew what got fixed because of their entries. This does more than any reminder.
- Review the list monthly. Add from the notes and retire what goes unused.
The two pages that own the detail are how to build reason codes and operator downtime logging, so start there if your list is new.
A four-week plan
Export the stops, split the bucket by length, and put the share on the board.
Start the same-shift question rule for stops over 10 minutes.
Read the notes, then add the top three codes and remove the unused ones.
Pick the worst machine, then watch it or film it for one shift and group what you see.
What target to set
We found no neutral published figure for a good unknown share, so set your own. A fair first goal is half of today's share within a quarter, with no stop over 10 minutes left without a reason.
Beware a share that falls overnight. It usually means someone picked a new default code, and your data just got worse while looking better, so spot-check the most used code against what you see at the line.
Fixes that make the data worse
- Deleting the "other" code. People then pick a wrong code, and a false reason is harder to spot than an honest unknown.
- Adding 40 codes at once. Long lists slow people down, and that sends them back to "other".
- Ignoring unknown in the Pareto. Show it as a bar. If it is the tallest one, that is your first project.
- Having a supervisor code everything at shift end. The bucket shrinks and the answers become guesses.
An unknown stop is hard to explain after the fact. Where cameras are fitted, Fabrico shows video of each stop, streamed on request next to its machine data, and operators tag the reason at the line. See how Fabrico uses cameras for OEE.
What people ask next
What is a good percentage for unknown downtime?
We found no neutral published benchmark, so measure your own share and aim to halve it in a quarter. Keep watching for false falls caused by a new default code. Downtime targets covers how to set goals.
Should I remove other from my reason codes?
No. Removing it pushes people to pick a wrong reason, which is harder to find than an honest unknown. Keep it, require a short note, and review the notes each month.
Why does automatic data still leave so much unclassified?
A signal shows that a machine stopped, but it does not show why. Short stops are also hard to tag by hand. OEE data collection methods compares what each method can capture.
How do I get operators to log the reason?
Keep it to one or two taps at the line, only ask about stops above your threshold, and show them what got fixed as a result. Operator downtime logging has the full rules.
Does unknown downtime affect the OEE number?
No. The minutes are lost whatever you call them, so OEE stays the same. What suffers is your ability to decide what to fix next.
Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.
Related guides: OEE for One Machine: A Simple Way to Start · Shop Floor Visibility: Find What You Cannot See · Machine Downtime Reasons: 12 Causes to Check