This article is for plant managers, production managers, maintenance leaders and continuous improvement engineers who look at a weekly downtime report and have to decide, often in a Monday meeting, whether a bad week deserves an investigation or just a note in the minutes. It shows how to apply control chart rules to two numbers most plants already collect every day, the count of machine stops and the count of scrap parts, so that the team reacts to real signals and stops chasing noise.
- A single bad day is not always a real change in the process.
- Daily stop and scrap counts fit simple attribute control charts.
- A short set of agreed rules separates signals from noise.
- Reliable, automatic stop data makes the chart trustworthy.
The Monday meeting problem: every bad week looks like a crisis
Picture a familiar situation. For example, a line that usually logs somewhere between 15 and 25 stops per day shows 31 stops on Wednesday. Someone asks what happened, a maintenance technician is pulled off planned work, an action is opened, and by Friday the count is back to its usual level. Was Wednesday a real change in the process, or one of the high days that a stable process produces from time to time? Without an agreed rule, the answer depends on who speaks loudest in the meeting.
The opposite mistake is just as costly. A slow drift, for example two or three extra stops per day over three weeks, never looks dramatic on any single day, so nobody reacts until the monthly report shows availability going down. By then, the cause (a worn component, a new material batch, a changed setting) has had weeks to do damage.
Both errors come from the same habit: comparing today’s number with yesterday’s number, or with a target, instead of comparing it with the process’s own natural variation. That is precisely the job of a control chart.
Jay Arthur’s point: SPC is not as hard as people think
Jay Arthur, of QI Macros (an SPC add-in for Excel), tackles this habit head-on in Many People Fear Statistical Process Control (SPC). His diagnosis is blunt: “Most people fear math and statistics.” He argues that this fear, more than any technical barrier, keeps teams on the line, bar and pie charts they already know, and that with fill-in templates anyone who can enter data can build a control chart without statistical training. He even suggests reading SPC as “Smart Performance Charts” to take the intimidation out of the name.
That message lands squarely on the shop floor. Downtime and scrap are exactly the kind of data that teams plot as plain line charts, and plain line charts invite overreaction: every peak looks like an event. Jay Arthur’s article makes the case for using control charts at all; this one looks at a single, very concrete application of them: deciding whether this week’s downtime spike is a signal or noise.
Which chart for daily stop counts and scrap?
Stop counts and scrap counts are attribute data: you count events or defective parts. The right chart depends mostly on whether the opportunity for the event is roughly the same from one day to the next.
| Daily data | Typical situation | Suggested chart | What the limits rely on |
|---|---|---|---|
| Number of stops per day | Similar planned production time every day | c chart, or XmR chart | The average daily count |
| Number of stops per day | Planned time varies (short days, overtime, partial shifts) | u chart: stops per hour of planned time | The average rate; limits widen on short days |
| Scrap parts per day | Output varies from day to day | p chart: share of scrap in parts produced | The average proportion and each day’s output |
| Minutes of downtime per day | Continuous measure rather than a count | XmR chart (individuals and moving range) | The average moving range between consecutive days |
When in doubt, an individuals and moving range chart is a robust default for any daily metric. What matters more than the perfect chart type is a stable definition of the data: the same threshold for what counts as a stop, the same definition of scrap, the same planned time every day. If the counting rule changes, the chart will show a “signal” that is really a change in how you measure.
Five rules to read the chart before anyone reacts
Once a few weeks of data (20 to 25 points is a common starting point) are plotted with a center line and control limits calculated from the process itself, not from a target, use a short set of detection rules. The ones below are widely used in SPC practice. Agree on them as a team before the meeting, not after the spike.
- One point beyond a control limit. A day above the upper limit is a signal worth investigating that same day, while operators’ memories and machine logs are still fresh.
- A run of points on one side of the center line. Seven or eight consecutive days above the average stop count usually means the process has shifted, even if no single day looks alarming. This is the rule that catches slow drifts.
- A steady trend. Six or so days in a row moving in the same direction point to wear, a gradual setting change or a progressive supply issue.
- Two out of three points close to a limit. Points in the outer zone of the chart, on the same side, are an early warning before a point actually crosses the limit.
- Everything else is noise. A day that is high but inside the limits, with no run and no trend, is part of the normal variation of the current process. Investigating it on its own rarely finds a cause, because there is usually no single cause to find.
Rule 5 is the hardest to accept in a meeting, and the most useful. It does not mean the current stop level is acceptable. It means that lowering it requires changing the system (a Pareto of stop reasons, a changeover reduction project, a revised maintenance plan) rather than reacting to individual days. Control limits describe what the process does today; targets describe what you want it to do. Mixing the two is how teams end up adjusting a process that was not actually changing.
Reading stops and scrap side by side
The most informative signals often appear when both charts are read together, for the same machine and the same period.
| Stop chart | Scrap chart | Where to look first |
|---|---|---|
| Signal | Signal on the same shift | Restarts after stops, unstable feeding, first parts after an intervention |
| Signal | In control | Breakdowns or jams without quality impact: maintenance, upstream or downstream flow |
| In control | Signal | Material batch, tooling wear, a changed inspection or measurement method |
| High but in control | High but in control | A system issue: reason Pareto, standard work, structured improvement project |
For example, a day with a stop count above the upper limit followed by a scrap count above its own limit on the same shift often points to restarts: parts made while the machine stabilizes after repeated interruptions. A run of high scrap with a normal stop count points elsewhere.
One practical tip: chart stops per machine or per line, not per workshop or per plant. Aggregating several machines averages out the very signals you are looking for, in the same way that a weekly total hides a single bad shift.
Where the data comes from changes what the chart can tell you
A control chart is only as honest as the data behind it. Manually logged stops carry known biases: short stops are under-reported, end-of-shift entries are rounded, and the definition of a stop shifts from one operator to the next. A chart built on that data can show signals that are really changes in logging discipline, and miss real ones.
TEEPTRAK measures machine stops and micro-stops automatically, using a module wired in the electrical cabinet and a sensor at the edge of the line, with no connection to the PLC. Installation takes about an hour per machine and OEE is available within 48 hours, which gives a consistent, timestamped daily stop count to chart from the first weeks. TEEPTRAK is used in more than 700 plants in over 30 countries, where users record on average 30% fewer unplanned stops. The control chart then does the rest: it tells the team which days deserve a root cause analysis and which ones are simply the current process speaking.
Further reading
- Jay Arthur, QI Macros: Many People Fear Statistical Process Control (SPC)
- QI Macros: SPC, Smart Performance Charts webinar (September 17, 2026)
- Downtime reason codes: the standard list and how to use them
- Process stability before digitalization: what downtime measurement changes
FAQ
Can I use a control chart on daily stop counts if my planned production time changes every day? Yes, but chart a rate rather than a raw count: stops per hour of planned time on a u chart. Otherwise a long day will look like a signal simply because the machine had more time to stop.
How many days of data do I need before trusting the control limits? A common starting point is 20 to 25 points. You can draw provisional limits earlier, then confirm them once you have enough data, and recalculate them only after a deliberate, confirmed change to the process.
Should we stop investigating high days that stay inside the control limits? Stop investigating them one by one. If the overall level is too high, work on the system: analyze the stop reasons over the whole period, rank them in a Pareto and launch an improvement project on the largest ones.
0 Comments