This article is for plant managers, production managers, maintenance and continuous improvement leaders who have worked hard to bring MTTR down and are puzzled that output did not rise in the same proportion. It looks at what happens after the restart: the minutes when the machine runs again but below its nominal rate, the first parts that go to scrap, the adjustments that follow. And it explains how to measure that recovery loss machine by machine, so that it appears in OEE instead of being spread across categories nobody reads together.
- MTTR stops the clock when the machine is repaired, not when it is recovered.
- After a restart, reduced speed and first rejects keep costing parts.
- That loss is split across availability, performance and quality.
- Measure time to nominal rate and restart scrap for each machine.
Repaired is not recovered
Ray Harkins, who teaches through The Manufacturing Academy and is general manager of Lexington Technologies, develops this idea in Recovery Behavior: What Happens After the Stop Matters, published on Accendo Reliability and adapted from chapter 11 of his book Measuring Manufacturing Effectiveness (2026). Discussing the losses a restart creates, he writes: “Recovery behavior explains why improvements in MTTR alone may not produce proportional gains in output.”
His article draws a clear line between two things that plants often treat as one. MTTR measures the time needed to restore function. Recovery covers the whole span from abnormal operation back to stable, productive output: detecting the problem, reaching target speed again, reaching quality stability. Output can stay degraded well after the repair is closed. Harkins also notes that restart losses touch all three OEE components (availability, performance and quality), that they are split across categories and therefore underreported, and that recovery varies with operator experience, staffing, documentation and equipment design. He points out that recovery is rarely measured directly and is usually recognized through patterns: more scrap after downtime, performance that stays low after a restart, more operator intervention. This article takes the measurement side of that argument: how to turn those patterns into a number per machine.
Why a lower MTTR does not always mean more parts
MTTR stops the clock when the machine is declared available again. The production loss does not stop there. After a jam, a tool change or a breakdown, many machines go through a stabilization phase: reduced speed while the operator watches the first cycles, temperature or tension that has not settled yet, first parts checked or rejected, a second adjustment. None of this belongs to MTTR, so a project that only targets repair time can improve the indicator while leaving the real loss almost unchanged, or even moving part of it into the minutes after the restart.
Take, for example, a machine with a nominal rate of 10 parts per minute (the figures below are invented for illustration). A maintenance project brings the average repair time down from 12 to 8 minutes. To get there, restarts are made faster, with less time for setting checks, so the machine takes a little longer to settle and rejects a few more parts afterwards. In this example, the ramp-up is assumed to run at about half the nominal rate on average.
| Per stop (illustrative example) | Before the MTTR project | After the MTTR project |
|---|---|---|
| MTTR | 12 min | 8 min |
| Parts lost while stopped | 120 | 80 |
| Time to reach nominal rate after restart | 8 min | 10 min |
| Parts lost during ramp-up (about half rate) | 40 | 50 |
| Scrap parts after restart | 15 | 20 |
| Total loss per stop | 175 parts | 150 parts |
In this example, MTTR falls by a third, but the loss per stop falls by only about 14 percent. The maintenance report shows a clear success; the production report shows a modest one. Both are right, because they do not measure the same thing. Harkins makes a related point about restart choices: running below maximum rate for a while can shorten recovery and improve stability, while an aggressive restart can win throughput at the expense of quality and reliability. Without a recovery measure, that trade-off is invisible.
Recovery also changes how stop frequency should be read. Harkins observes that frequent short stops followed by slow stabilization can consume more effective production time than rare long outages. If a machine stops 30 times a shift and loses two minutes of reduced speed after each restart, the recovery loss can exceed the recorded downtime. That is the same mechanism we described for micro-stops in an OEE Pareto, seen from the other side of the stop.
Where recovery loss hides in OEE
Because the loss after a restart is spread across the three OEE components, each part of it lands in a different report, and none of them carries the label “recovery”. The table below follows the split Harkins describes and adds where each part usually ends up.
| OEE component | What happens after the restart | Where it usually shows up | What to measure |
|---|---|---|---|
| Availability | Checks, first-part approval, second adjustment, short stops right after restarting | Extra stop minutes, often coded as a new stop with a different reason | Stops occurring within the recovery window, linked to the original stop |
| Performance | Reduced speed while the process settles, cautious operation of the first cycles | Unexplained gap between theoretical and actual rate | Time to reach nominal rate and parts lost against nominal during that time |
| Quality | First parts out of tolerance, purge, material reintroduced | Shift scrap total, with no link to the event that caused it | Rejects in the recovery window, attached to the stop that preceded them |
Read separately, these three lines look like ordinary noise: a few extra stops, a slightly low performance rate, a scrap total in line with usual levels. Read together and attached to each restart, they become one loss with one cause, which can be ranked and improved like any other.
How to measure recovery loss machine by machine
- Set a nominal rate per machine and product. Recovery is measured against a reference. Use a documented rate per product reference, not the best rate ever seen.
- Write down what “recovered” means. Define a stabilization criterion with production and quality, for example a number of consecutive cycles close to the nominal rate and the first good part confirmed. Harkins lists defined stabilization criteria among the management actions that improve recovery; they are also the condition for measuring it.
- Timestamp every restart. The end of each stop is the start of the recovery window. Without automatic stop data, this moment is not recorded precisely enough to work with.
- Measure time to nominal rate. For each restart, count the time until the stabilization criterion is met. This is the recovery time, separate from MTTR.
- Convert the reduced rate into parts. Parts lost in recovery equal the nominal rate multiplied by the recovery time, minus the parts actually produced during that time. This keeps the loss in the same unit as production targets.
- Attach scrap to the restart. Rejects declared in the recovery window are counted as restart scrap and linked to the stop that preceded them.
- Report recovery next to MTTR. For each machine, show MTTR, average recovery time, recovery loss in parts and restart scrap, broken down by stop reason, shift and product. The spread between shifts or teams is often the most useful line, because it shows where a standard restart procedure would help.
With these figures, the question in the weekly review changes. It is no longer only how quickly a machine is repaired, but how much production each restart costs, and which machines, products or shifts recover worst.
Measuring recovery without a manual log
Recovery loss is made of seconds and partial rates, which is exactly what a manual log cannot capture: nobody writes down that the machine ran at reduced speed for six minutes after a restart. TEEPTRAK measures machine stops, micro-stops and cycle rate automatically, with a module wired in the electrical cabinet and a sensor at the edge of the line, without connecting to the PLC. Installation takes about an hour per machine and OEE is available within 48 hours. Because every stop and every cycle is timestamped, the time to nominal rate after each restart and the rate deficit during that window can be read for each machine, with operators qualifying the stop reasons and the rejects. TEEPTRAK is used in more than 700 plants in over 30 countries.
Further reading
- Harkins, R. (2026). Recovery Behavior: What Happens After the Stop Matters. Accendo Reliability. Adapted from chapter 11 of Measuring Manufacturing Effectiveness, Amazon KDP.
- Why micro-stops hide in frequency, not duration: what 30-second stops do to an OEE Pareto
- MTBF MTTR unplanned downtime: measuring equipment reliability
- Machine slowdowns: the hidden speed loss in OEE
FAQ
What is the difference between MTTR and recovery time? MTTR measures the time needed to restore the machine to working condition. Recovery time continues after that point, until the machine runs again at its nominal rate and produces good parts consistently. A machine can be repaired and still be far from recovered.
Should recovery loss be counted as availability, performance or quality loss? In OEE it is usually split across the three: extra stops in availability, reduced speed in performance, rejects in quality. Keep that split for the OEE calculation, but also report the total per restart, so that the loss can be ranked and attributed to its cause.
How do we define the end of the recovery window? Agree on a stabilization criterion per machine family with production and quality, for example a number of consecutive cycles close to the nominal rate plus the first good part confirmed. Apply it the same way on every shift, so recovery times can be compared.
0 Comments