Changeover Tracking: Measure SMED Before You Run It
Ask a production head how long changeovers take and you get a number from a good day. Ask the shift report and you get "setup" as one lumped line that also hides waiting for materials, breaks, and the forty minutes somebody spent finding a jig. SMED workshops then optimize a number nobody measured. Changeover tracking fixes the measurement first — and in a MES it falls out of data you are already collecting.
What Counts as Changeover Time
The SMED definition is crisp: the time between the last good piece of product A and the first good piece of product B. Not "when setup started" — when the last good piece came off. Everything inside that window counts: the last cycles of A finishing quality, teardown, change, first-article inspection, and the adjustment cycles before B runs to spec.
That definition is why spreadsheets undercount. A shift report records what people remember; the window between two good pieces lives in the run record — the exact events a MES writes anyway.
How a MES Captures It
Three events define the window, and all three already exist at the terminal:
- Last good report of run A (or the complete event of A's last operation at the station).
- Start of run B — including the internal steps, because the operator opens the next job on the terminal as they begin setup.
- First good report of run B — the moment B actually produces to spec.
The changeover duration is the span; the parts in between (teardown, first-article, adjustment) are its anatomy when you record pause reasons during the window. A factory that tracks downtime reasons already owns the vocabulary: "first-article approval waiting" is a reason code, and its total per month is an argument for a faster inspection path.
Why the Record Beats the Stopwatch
A SMED workshop typically times five changeovers with a stopwatch. That is a good workshop and a bad baseline: five observations, one product family, one shift, usually the operator who is best at it. A tracked record gives you every changeover since go-live, sliced by product family, machine, and shift — which is where the real findings live:
- The spread, not the average. Average 42 minutes means nothing; "A→B is 20 minutes, C→A is 70, and C→A happens weekly" is a scheduling lever. The scheduler sees the matrix and stops sequencing C before A on the tight day.
- Who did it matters less than what differed. The record shows the fast changeovers ran first-article on pre-staged gauges. That is a SMED idea arriving with its own evidence.
- External versus internal time. The window's anatomy shows how much of the "setup" was waiting for a forklift or a crane — external work that should never have been inside the window at all.
Making the Number Move
Once changeovers are tracked per family and machine, the loop is standard SMED with better inputs: separate internal from external steps, convert internal to external (pre-stage tools and materials while the machine still runs A), streamline what remains (quick clamps, shadow boards, standard settings sheets attached to the work instruction). Then let the chart prove the workshop worked — per family, not as a vague average.
Scheduling joins the win: shorter, more predictable changeovers make smaller lots economic, which is how the schedule starts shipping smaller batches without penalty.
Frequently Asked Questions
Do operators need to press a "changeover started" button?
No — and that is the point. The window is derived from run events the operator already produces: complete A, start B, first good report of B. Derived windows cannot be forgotten, and they match the SMED definition instead of someone's stopwatch opinion.
Our machines are not connected — can we still measure?
Yes. The three events are terminal events, not machine signals. Operator-reported runs give you the full changeover record at Level 0; automatic cycle detection refines it later.
What is a good changeover time?
The one that moves. Benchmarks vary wildly by process — a press changeover and a packaging line live in different worlds. Your baseline is your own record; the target is the best family-pair you have already achieved, applied to the rest of the matrix.
Every Changeover, Measured From Go-Live
Voltrus MES derives changeover windows from the run record — per machine, family, and shift — with pause reasons for the anatomy inside. One line to start, every changeover measured from day one.
See Voltrus MES