Micro-stops and Their Impact on Performance
Estimate time lost to minor stops, understand duration measurement and relate the findings to equipment performance.
Updated:
Define what counts as a micro-stop
“Micro-stop” needs an operational definition. The threshold separating a small stop from a logged stoppage depends on the measurement system; there is no universal one-second-to-five-minute rule. An operator beside a stopped machine does not establish a micro-stop: setup, lack of demand, or intervention may explain the state.
A slow cycle still produces at a lower rate. A snapshot does not measure speed on its own. Check classification against cycles, counts, or machine records without automatically attributing slower work to motivation or fatigue.
What sampling estimates
Random state observations can estimate the time share classified as small stops in the observed population. They do not directly provide event count or mean duration per event.
Long events are more likely to overlap a reading. Timing only stops encountered at random selects by duration and may capture only their remaining time. This is not an unbiased event-duration sample. Construct duration histograms from complete starts and finishes under a defined capture rule.
Numerical planning
These are mathematical examples, not sector recommendations. With p₀=0.08, n=ceiling(z²p₀(1−p₀)/e²) gives:
| Confidence | z | Absolute margin | n |
|---|---|---|---|
| 95% | 1.96 | 2 points | 707 |
| 95% | 1.96 | 1.5 points | 1,257 |
| 99% | 2.576 | 2 points | 1,221 |
| 90% | 1.645 | 2 points | 498 |
Study days depend on routes, resources, and coverage. They cannot be inferred from n, and there is no universal five-day minimum. Nor does n>30 guarantee an adequate normal approximation for small proportions. NIST describes the proportion model.
Simulated example: 64 out of 800 readings
With 64 micro-stop states among 800 readings, p̂=8%. The approximate Wald interval is 6.12–9.88%; the 95% Wilson interval under the same binomial model is approximately 6.31–10.09%. See both methods in NIST.
If the other categories were 680 productive, 32 major stop, and 24 inactive, they would total 800. The ratio 680/(800−32)=88.5% describes classified activity within the remaining denominator. It is not automatically OEE performance: compatible ideal-cycle, output-count, and run-time data are needed.
OEE without double counting
Under an OEE convention, small stops retained in run time reduce performance; stops removed from run time reduce availability. Define the boundary and do not subtract the same event twice. Quality is good count/total count, not the proportion of inspection observations. OEE formulation.
Estimating recoverable units from a share of observed states requires data and assumptions about capacity, demand, blocking and speed. Check the estimate through a trial of the change before treating it as recovered output.
Investigate and improve
Separate state, event duration, and confirmed cause. A frequent category indicates where to investigate, not proof of a feeding or ergonomic problem. Trial a change and check good output and safety under comparable conditions.
Do not discard early days because workers supposedly return to normal behavior. Predefine any training phase and exclusions. Randomness neither eliminates reactivity nor justifies covert observation.