Kako da smanjite neevidentirane zastoje u proizvodnji

Step 3: Treat slow-running as downtime, not as „running." A line at 65% of nominal speed for 4 hours has produced the equivalent of 1.4 hours of zero output. That is downtime. It should appear on the dashboard.

• How: track actual cycle time per minute against ideal cycle time. Any 1-minute window below 75% nominal speed becomes a „performance loss" event with the same weight as a stop.
• Expect the recorded total to rise once this loss is counted. It was always there, it just was not on the dashboard.
• This is the difference between availability and OEE. OEE calculation covers the full math.

Step 4: Cross-check operator logs against sensor data daily, not weekly. When operator says „cleaning" and sensors show the line was producing for 8 of those 12 minutes, you have a classification error. Daily cadence catches it before the pattern becomes habit.

• How: automated report comparing logged events to sensor + CV data. Highlight mismatches over 25%.
• The goal of the daily review: the team moves from „something is off" to „we trust it".
• The reviews are 5 minutes at the end of each shift. Not a project, a habit.

See how unrecorded downtime feeds the MDT latency tax.

The hardest part of closing the recording gap is the conversation in week 2 when reported downtime is well above what management expected. That is not the line failing. That is reality finally showing up.

Frame the conversation before you start. Tell the CFO and plant director on day 1: „We are turning on automated capture.

Reported downtime will go up in the first month. The line did not get worse.

We finally see what was always there. By month 4, the new visible total will drop 30-40% as we attack what we now see."

Plants that skip this conversation get pulled off the project in week 3 when the number spikes. Plants that have the conversation upfront finish the project.

A modern OEE solution with native CMMS handles all 4 steps automatically and produces the daily classification review without operator labor. That is the difference between Fabrico and a clipboard system pretending to be a dashboard.

Quick answer: Unrecorded downtime is the gap between actual line stoppage and what the OEE dashboard reports, typically 30-50% of all losses in plants relying on manual operator logs. The fix is automated capture (PLC tap or Computer Vision) that sees stoppages the moment they start, so the dashboard reflects ground truth instead of end-of-shift memory.

Related deep-dives: OEE data collection methods · Computer Vision OEE · visual downtime verification · closing the OEE-CMMS loop.

Ask any plant manager and they will tell you operators are diligent at logging downtime. Ask the operators and they will agree. Look at the data alongside automated capture and the numbers tell a different story.

Typical European packaging plant comparison, same line, same week:

• Operator-logged downtime: 18 hours
• Computer Vision-captured downtime: 47 hours
• Ratio: operators see 38% of reality

This is not about operators being lazy. Three structural reasons they miss the rest:

• Micro-stops too short to walk to terminal. A 45-second jam gets reset and forgotten. By shift end nobody remembers the 12 jams that day.
• Wrong category is easier than right category. The terminal has 30 reason codes. Picking the right one takes thought. „Other" is one click and lets the operator return to the line.
• Slow-running is not stopped. The line is moving, so it does not feel like downtime. But running at 65% speed for 4 hours is 1.4 hours of effective downtime that never gets logged.

When a plant switches from manual to automated capture, expect reported total downtime to increase. That is the gap becoming visible. See data collection methods.

Step 1: Capture every stop under 5 minutes. This is where most of the hidden downtime lives. Operators do not log them. PLCs often do not signal them. They look like „normal operation" in the data.

• How: Computer Vision on the line output catches short stops. No operator action needed.
• Micro-stop time that used to show as zero starts to appear on the dashboard. That is not new downtime, it always existed.
• Expect operator pushback: they will think the system is wrong. Show them the timestamped video clips. Pushback ends in week 2.

Step 2: Make classification automatic, not a 30-code menu. The terminal with 30 dropdown options is a guarantee that „Other" wins.

• How: limit operator-facing categories to 4-6 visible buckets (mechanical / quality / changeover / external / micro-stop / other). Auto-classify deeper levels in the background using event patterns and machine state.
• Watch the share of events coded „Other". As it falls, you can see what is actually happening.
• Bonus: operator engagement goes up because the system feels respectful of their time.

See the 6 OEE losses framework for the right classification structure.