Jak zmniejszyć nieewidencjonowane przestoje w produkcji

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.

EU benchmark: in plants that switch from manual to automated capture, reported total downtime increases 2.5-3.5x in the first 30 days. 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 every stop from 0.4 seconds. No operator action needed.
• EU benchmark impact: visible micro-stop time goes from 0 hours/week to 8-15 hours/week on a typical packaging line. 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.
• EU benchmark: „Other" category drops from 35-45% of events to under 8%. Suddenly 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.

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.
• EU benchmark: previously invisible performance loss adds 12-18% to recorded total downtime. Most plants are stunned by the number on day 1.
• 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%.
• EU benchmark: data quality goes from „something is off" to „we trust it" in 6-8 weeks of daily reviews.
• 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 2.5x 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 2.5-3.5x 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.