Check downtime data against a second source. Do run time and stops add up to scheduled time, do counts agree with run time, and does an hour of watching the line match the log? Then test the reason codes and the rounding of the times, because a log can look fine and still blame the wrong loss.
Every decision about stops rests on that log: which machine gets the budget, which cause gets the project, and which crew gets the awkward question.
Yet it often goes untested. Here is an audit you can run on one line in an afternoon, using a counter, a clock and a pencil.
Why bad data can look fine
Here is the trap, in numbers. Example: a line with an ideal rate of 60 units a minute and a 480-minute shift makes 21,900 units, and 21,600 are good.
The log shows 52 minutes of stops, while an hour meter on the machine shows only 398 running minutes. So real stops are 480 - 398 = 82 minutes, and 30 minutes never reached the log.
| From the log | From the hour meter | |
|---|---|---|
| Stops | 52 min | 82 min |
| Run time | 428 min | 398 min |
| Availability | 428 / 480 = 89.2% | 398 / 480 = 82.9% |
| Performance | 365 / 428 = 85.3% | 365 / 398 = 91.7% |
| Quality | 21,600 / 21,900 = 98.6% | 98.6% |
| OEE | 75.0% | 75.0% |
That is why OEE cannot warn you. A bad log does not change the score, but it does send your team to fix speed when the real problem is stops.
Two habits keep any check fair: use the same data rules for all crews, and test reported times against a clock, a meter or a count.
The 7-step audit
Run time plus stops must equal scheduled time, on every shift.
Units / ideal rate must not exceed run time.
Hour meter, PLC run signal or an added sensor.
One per shift. Write every stop, then compare with the log.
Take 20 stops you saw and compare the codes.
Share of times ending in 0 or 5.
Share of minutes under unknown or other.
Step 1: add up the time
Pull a week of shifts and check the sum. Look for stops that overlap, that run past the end of the shift, or shifts where the numbers leave minutes unaccounted for.
This test takes minutes and needs nothing but a spreadsheet. If it fails, nothing further is worth much until it passes.
Step 2: compare with counts
Divide total units by the ideal rate. That gives you the minimum time the line must have run.
If the result is more than the logged run time, then either a stop is overstated, the ideal rate is too low, or the count is wrong. In that case performance reads over 100%, as performance over 100% explains.
If performance is low and steady, that is a clue too. It may be slow running, or it may be stops nobody logged.
Step 3: compare with a machine source
Use anything that runs only while the machine runs: an hour meter, a PLC signal or a retrofit sensor. Then compare its total with the log, as in the example above.
Machine data needs an audit as well. A run signal can stay on while the machine idles, or when it is blocked, so check what it means, using machine state detection and PLC signals for OEE.
Automatic systems also have limits of their own. In a study of 884 machines in 23 Swedish companies, unclassified losses were about 19% of scheduled production time, and that was more than half of all recorded loss time.
Step 4: watch the line
Nothing replaces this. Stand at the line for one hour and write down every stop, with a clock time and a length, then pull the log for that hour.
Example: you saw 9 stops totalling 11 minutes, and the log has 3 stops totalling 7 minutes. It caught 33% of the stops but 64% of the minutes.
Expect that pattern. Long stops tend to reach the log and short ones do not, which is why your stop threshold and your actual logging habit may be two different things.
Step 5: test the reason codes
Take 20 stops you saw yourself, and compare the code in the log with what really happened. Do it for different crews if you can.
Look for two things: wrong codes, and the same event coded differently by different crews. If a list has too many codes, or vague ones, fix it with downtime reason codes.
Step 6: test the rounding
This is a neat one, and it needs no walking at all. Take the minutes of the stop start times, and look at the last digit.
Clock times read honestly should end in 0 or 5 about 20% of the time, because that is 2 digits out of 10. If 70-80% do, those times were probably written from memory.
The same check works on stop lengths. A log full of 5, 10, 15 and 30 minutes is an estimate, and that is worth knowing even if the total looks right.
Step 7: measure the unknowns
Add up the minutes coded as unknown, other or blank, then divide by all stop minutes. High numbers mean your Pareto chart is drawn from what is left over.
See unknown downtime for how to shrink this bucket. Do not delete the code, because people will simply pick another wrong one.
Score your line
Put the results on one page. Here is an example scorecard, and the pass marks are our suggested starting points, not a standard.
| Test | Result | Suggested pass | Verdict |
|---|---|---|---|
| Shifts where time adds up | 100% | 100% | Pass |
| Stop minutes missing, against meter | 37% | Under 10% | Fail |
| Minutes caught in the line watch | 64% | Over 90% | Fail |
| Codes that match what you saw | 14 of 20 = 70% | Over 90% | Fail |
| Times ending in 0 or 5 | 78% | Near 20% | Fail |
| Minutes coded unknown or other | 18% | Under 10% | Fail |
What to fix first
- Time does not add up: fix the form or the export first.
- Minutes missing: make logging easier at the line, using operator downtime logging, or add a machine signal.
- Codes wrong: cut the list, and show crews how the codes get used.
- Rounded times: use a shared clock, or a timestamp at the moment of entry.
- Short stops missing: a paper log will not catch them. Compare the options in OEE data collection methods.
Never use an audit to punish a crew. If people expect blame, the next log will look better and be less true.
Repeat the audit monthly until it passes, then each quarter. For errors in the formula rather than the data, read OEE calculation mistakes.
A signal from the machine takes memory out of the stop times this audit tests. Fabrico takes stops, speed and counts live from PLCs or IoT sensors, including external sensors on older machines, and operators tag the cause at the line. See how Fabrico connects to machines.
What people ask next
How accurate is manual downtime data?
We found no general figure we would stand behind, and it varies by plant. Run steps 3 and 4 on your own line, since your own gap is what you need. OEE data collection methods compares manual and machine sources.
Is automatic downtime data always right?
No. It can get the time right and the reason wrong or blank, and a signal may read as running while the machine waits. Check it by watching the line, and see machine state detection.
How often should we audit downtime data?
Monthly until the line passes, then about each quarter. Also audit after any change to codes, crews or systems. One line per week across a plant keeps it light.
What if I have no hour meter or PLC access?
A low-cost external sensor on a motor or at the line exit can give you run time and counts. Even one shift of data makes the comparison possible. See retrofit sensors for OEE.
Why does the audit not change our OEE?
Because OEE is good units against scheduled time, wherever the stops were logged. A log fix moves the loss between availability and performance. See the worked table at the top of this page, and how to calculate OEE quality for counts.
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.