Measure changeover time before you try to cut it

You measure changeovers fairly by timing each one from the last good part of the old run to the first good part of the new one, logging every element as internal or external, and comparing only changeovers of the same type by their median. Those three rules turn a stopwatch reading into a baseline you can compare across shifts, machines and months before you change anything.

This guide is for plant managers and process engineers starting a changeover reduction project, or running one and unsure whether it worked. It times one mold change element by element, then builds the comparison and the weekly loss from it. To see where changeover ranks against your other losses first, build a downtime Pareto from real minutes.

What counts as changeover time?#

The wording follows the Lean Enterprise Institute's lexicon, which treats changeover and setup as one thing: switching a machine or a line from one part number to another by changing parts, dies, molds or fixtures.^[1]

The definition matters for what it includes. The clock doesn't stop when the machine restarts. It stops when a part passes, so purge shots, first-article inspection and the adjustments between them all sit inside it.

OEE counts those two halves in different places. Setup and adjustment time is an availability loss, and defective parts made between startup and stable production are a quality loss, most often tracked after changeovers.^[3] The OEE calculation guide works through both factors on one shift.

In this example the machine stands still for 72 minutes, then runs for 14 more before a part passes. A machine monitor records the 72 minutes as downtime. The changeover took 86.

Choose a start and a stop rule#

Write down where the clock starts and stops before the first measurement, and use the same rule on every machine of a type. Four rules are in common use, and each one hides something.

Use last good part to first good part as your changeover time. Record the restart as well, so every changeover splits into stopped time and run-up. The split tells you where to work: a long stopped time is a swap problem, and a long run-up is a settings or first-article problem.

The stop and the restart come from machine data or a stopwatch. The first good part comes from whoever approves it: the first-article sign-off, the quality technician's release, or the first part logged as good. Name that event for each machine type, since a molder's first good shot and a CNC's approved first article are signed off by different people.

Log each element as internal or external#

Separating the two, then moving internal work outside the stop, is the first move of SMED (single-minute exchange of die). Its target is a changeover in single-digit minutes, under 10.^[2] You can't separate work you haven't logged, so the timing sheet comes first.

Observe a typical changeover, film it if the crew agrees, and finish with a table of the steps, their sequence and their duration.^[4] An element is one piece of work with a clear start and end, such as "disconnect water lines". With more than one person on the changeover, add a column for who does each element, so two people working side by side aren't counted as twice the machine time.

The sheet shows two elements that don't need a stopped machine. Stage the next mold beside the press and preheat it before the stop, and both leave the changeover. Internal time falls to 6 + 9 + 12 + 7 + 14 = 48 minutes.

That 44.2% cut is invisible in a single stopwatch total, because the crew's work stayed at 96 minutes. Only the element log shows which minutes moved. The counting side of a mold change, re-picking the operation so shots land on the new part, is covered in monitoring injection molding machines.

Compare like with like#

A changeover number compares only with changeovers of the same type. A 15-minute color change and a 90-minute mold change averaged together describe neither. Put every changeover under a type before you compare anything.

A shift changeover changes the crew, not the part. Track it as its own type, because its fix is the handover, not the tooling. Give each type its own reason code so the minutes never mix, using a short list of downtime reason codes.

Inside a type, direction matters. A molder going from black resin to white purges longer than one going from white to black, so log the from and the to as well as the type.

Then use the median, not the mean, and never the best run. One changeover that ran into a breakdown drags the mean for weeks. Split that stop at the moment the breakdown began, so the changeover keeps its minutes and the breakdown keeps its own, as in splitting a downtime.

Collect at least 10 changeovers of a type before you call it a baseline, and report the range beside the median. Ten die changes from 42 to 58 minutes say the method is steady. Ten from 20 to 90 say there's no standard method yet, and that's the first thing to fix.

Turn changeovers into a weekly loss#

The weekly loss adds up every changeover minute, overruns included, so it uses the mean. The median judges the method; the mean prices it.

State the loss as a share of scheduled hours whenever you compare weeks. A four-day week has fewer changeovers and fewer hours, and only the share tells you whether anything improved.

Marking changeovers as planned downtime labels them; it doesn’t take them out of this loss. Under the common OEE convention, which planned vs unplanned downtime sets out, a scheduled changeover still counts against availability, and it still takes 10 hours the press could have run.

Prove a fix with before and after#

Compare the same type, the same direction and the same start and stop rule, with at least 10 changeovers on each side. Then report three numbers, not one: the median per changeover, the number of changeovers and the weekly total.

The three can move in different directions. Shorter changeovers make smaller lots affordable, and smaller lots mean more changeovers.

Neither number is wrong. The shorter changeover bought smaller lots and less inventory, and the plant chose to spend it that way. Judge the method by the median per changeover, and judge capacity by the weekly total and its share of scheduled hours.

Keep measuring after the fix. A changeover that falls from 86 to 48 minutes in week one and drifts back to 70 by week eight has lost its new standard, not its method. Watch the weekly median for at least eight weeks after any change.

Measure changeovers automatically#

A sensor on each machine records the stopped part of every changeover without a stopwatch. IoTFlows downtime tracking detects each stop from that sensor, with no PLC connection, and the operator tags it with a changeover reason in a tap.

Turn on the shift-edge rules, and stops at the start and end of a shift classify themselves. Assign every part a machine runs, so counts move to the new part at the changeover.

The Downtimes report trends each reason by week. Its CSV export carries one row per stop, so you compute the median and range per type in a spreadsheet.

Datwyler's injection molding plant classifies stops this way and reviews the report weekly and monthly. It cut shift-changeover downtime by 50% and material-changeover downtime by 30% within six months, as Datwyler's story describes. Criollos Foods cut average product setup and changeover time by 20 minutes.

Set up your first baseline#

1. Write the rules down

   Use last good part to first good part, name who signs off the first good part on each machine type, and post both at the machines.

2. Create one reason per type

   Add a reason for each changeover type in the comparison table, and nothing broader than that, so no two types share minutes.

3. Time two changeovers by element

   Fill in the timing sheet on your most frequent type, filmed if the crew agrees, and mark each element internal or external.

4. Log every changeover

   For each one, record the stop, the restart, the first good part, the type, the from and the to, and anything else that went wrong inside the stop.

5. Split and compute

   Split any changeover that ran into a breakdown. Once a type has 10 changeovers, compute its median, range and weekly total.

6. Raise alert thresholds above the median

   Set each machine's downtime alert above the median of the changeover type it runs most, so a normal changeover pages nobody. See setting the alert threshold.

7. Pick the first type to cut

   Choose the type with the largest weekly total, not the longest single changeover, and start moving its internal elements outside the stop.