Give the stop to the source machine: the first one on the line to enter its own fault while it still had product coming in and room to send it out. Machines that stopped because of it are recorded as blocked or starved, and the line loss is counted once, at the line output.
Why the argument starts
Every manager has sat through it. The packer's shift report says 40 minutes down, the filler's says the same 40, and everyone is sure the fault lay somewhere else.
A travelling stop has one beginning and several echoes. Your data needs a rule that separates them, and the rule needs to be written down before the next argument.
See a stop travel
Example: a labeller jams at minute 2 and runs again at minute 7. The conveyors absorb the first moments, and then the wave moves both ways.
Added up, the four machines show 5 + 4 + 5 + 4 = 18 stopped minutes. The line output stopped for just 4, the palletizer's bar, so the source's 5 minutes of fault is the number to chase.
The rule in four steps
Put each machine's stop start in time order, using one shared clock.
Look for the first machine with an alarm or a stop that has no blocked or starved signal.
That machine had product at its infeed and room at its outfeed. If it did not, keep walking along the line.
Every later machine gets blocked or starved, with the source machine named on the record.
The states themselves are defined in blocked and starved, so this page only deals with who owns the minutes.
Record where and why as two fields
The source machine is where the stop began. That can differ from why it happened, and this is where well-meant rules go wrong.
Example: a labeller jams on crushed bottles. The source machine is the labeller, because that is where the stop began, yet the cause may be a depalletizer damaging bottles three machines earlier.
- Source machine comes from the timeline, so it can be filled in from data.
- Cause comes from a person who looked, so it uses your reason codes.
If the cause keeps pointing away from the source machine, run a proper root cause analysis. That step, more than any blame rule, is what removes the repeat stops.
Stop counting the same minutes twice
This step can make reported downtime fall sharply. The line runs just as it did before, and you have simply stopped adding the echoes together.
Build your Pareto chart from the 125 source minutes only. The 185 are real machine states, but they do not point to a problem on those machines.
Line-level OEE then takes its loss once, from the output or the constraint. Line OEE vs machine OEE sets out both views.
When a buffer absorbs the stop
Some source faults never reach the line output. A 30-second stop can vanish into a conveyor with two minutes of buffer, and the line output never dips.
Keep those events on the source machine, because the fault is real and a longer one will get through. But rank them below the faults that cost output, since fixing them returns less.
The same logic decides where improvement starts: usually on the bottleneck, because a minute gained elsewhere often adds no output. Bottleneck analysis shows how to confirm which machine that is, and six big losses shows where each stop type sits in OEE.
Grouping stops into one event by hand
A machine export gives you a flat list of stops, so you need a way to join the ones that belong together. A spreadsheet can do this once the clocks agree.
- Sort all machine stops by start time, across the whole line.
- Choose a window equal to your longest buffer time, for example 3 minutes.
- Give a stop the same event number as the one before it if it starts inside that window and carries a blocked or starved state.
- Start a new event at any stop in its own fault, then read the source machine from the first row of each event.
Check 20 events against what the crew remembers before you trust the result. The window will sometimes join two separate stops, so tune it to your line.
Hard cases and how to call them
| Case | How to call it |
|---|---|
| Two faults overlap on two machines | Two source events. Share the line minutes by time, with the overlap going to the fault that started first. |
| A second machine trips while the line restarts | New event, new source machine. If it trips each restart, flag the restart itself. |
| Air or power drops for the whole line | Source is the utility, not a machine. Give it a code of its own. |
| A bad batch stops the line | Source is the machine where it stopped. Cause is the material. |
| The first machine runs out of material | Nothing is in fault. Source is the supply, so log it as a material wait at that machine. |
| One machine runs slowly with no alarm | Blocked before it, starved after it, and no fault anywhere. Log it as speed loss on the slow machine. |
What one stop record should hold
- Line event number, shared by every machine the stop touched.
- Source machine, with its start and end time.
- Line minutes lost, taken at the output.
- Cause, and who tagged it.
- Echoes, listing each machine that was blocked or starved, and for how long.
Keep the blame out of it
The word "blame" leads people to protect themselves, and then they guess reasons or stop logging. Use "source machine" on your forms, since it describes the line and says nothing about a person.
Review sources at the daily production meeting with a simple question: what would stop this returning? The review works best when the operator from the source machine helps answer.
What makes the source come out wrong
- Clocks that disagree. Two PLCs a minute apart will name the wrong source, so sync them first.
- Picking the machine that shouts. The loudest alarm may sit on the machine that stopped last.
- Blaming only the bottleneck. It often gets the minutes because it runs closest to the limit, even when others started the stop.
- Dropping source machines with short stops. Their frequency matters, and mean time between stops shows it.
Assigning a travelling stop needs each machine's stop on one clock. Fabrico takes stops live 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, so OEE per line and shift stands on the same record. See how Fabrico tracks stops.
What people ask next
How do you assign downtime when one machine stops the whole line?
Give it to the first machine that entered its own fault while it had product and room. Later machines are labelled blocked or starved. Count the line's loss once, at the output.
Should blocked and starved time count against each machine?
It lowers that machine's own availability, so keep it visible with its own label. Do not treat it as that machine's fault, and leave it out of your fault Pareto. Blocked and starved covers the states.
What if the source machine stopped because of something upstream?
Keep the source machine as the place the stop began, and record the upstream problem as the cause. Two fields stop the argument. A root cause analysis can then follow the cause back.
How do you do this without PLC data?
Start with a shared clock and one person watching for a few shifts, writing the first machine down for each stop. The result is rough, yet good enough to show the top sources. OEE data collection methods compares the next steps.
Does the line OEE change when I use this rule?
No, not if you already measure it at the line output. The rule changes the machine reports and the Pareto, so effort moves to the source of each stop.
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 Parallel Machines on One Process Step · Why OEE Drops After You Add a New Machine · Conveyor Short Stops: Jams, Tipping and Gaps · Machine Downtime Reasons: 12 Causes to Check