Taking a mold changeover off the stopped machine

This entry describes a five-day kaizen run in the spring of 2019 at a precast concrete manufacturer in the western United States, one of a series of kaizens the practice carried out at the plant. The work was on the changeover of a jacket-and-core casting machine, which casts a family of concrete vault and manhole components in a mold set into a casting pit and filled from a hopper overhead. The brief was to shorten the change from one mold to the next. Its stakes were set by chemistry: concrete left waiting in the hopper hardens, so a slow changeover is not only lost production but a standing risk to the machine. The client’s name and certain operational details have been altered or omitted at the client’s request.

The plant cast its concrete vault and manhole components on the casting machine, and to move from one product to the next the crew had to strip the old mold and set the new one. The jacket’s hydraulic lines came off; the jacket was unclamped and lifted away on the yard crane; the core was pulled and landed on a pallet; the next form was brought in, seated in the pit, and reconnected. The changeover was slow, and — more to the point — nobody could say how slow, because it varied from one day to the next and had never been written down. An early attempt to simply time it was given up past forty-five minutes, the stopwatch still running and more than forty steps recorded before the count was abandoned. As it ran, the changeover pulled operators in from around the yard to rush it and put a forklift to work in the pit, and still went long enough, often enough, to leave concrete setting in the hopper. Long, variable, and undocumented, it forced the yard into long production runs: to avoid changing over, the crew made more of a given product than the schedule needed and stacked the surplus outside. The same changeover that filled the yard with surplus of one product was also, in the same motion, the reason the yard went short on another. It had missed customer orders that winter for want of product it had not had time to switch the machine over to make.

The charter’s brief was to make the steps faster and get the total under twenty minutes. But watching the crew work, the more useful question was not how fast each step ran; it was whether the machine had to be stopped for the step at all. A changeover holds two kinds of work. Some of it can only happen with the machine down and the mold open — internal work. The rest could be done while the previous run is still casting — external work — if only it were staged to happen there. And most of the time on the machine was going on work that did not need it stopped: the next form was fetched and rigged after the old one came off, when it could have been standing ready; the hydraulic connections were hand-threaded, kneeling, a turn at a time, with the pit idle and the concrete in the hopper going off; the forklift and the crowd of extra hands were there to force through a sequence that, sorted properly, needed neither. The clock was not set by how fast the steps were. It was set by how many of them sat on the stopped machine.

A mold on a scissor-lift cart in the changeover bay. — **A mold staged on a scissor-lift cart, off the machine.**·Photograph: house archive·Spring 2019

From there the team walked the changeover back and sorted each step — internal or external — then set about moving work out of the first column into the second and shortening what was left. The next mold would be staged and rigged before the changeover began, not after. A second set of hydraulic clamps, put on order during the week, meant the incoming mold’s clamps could be prepared off-line while the machine was still running, turning the clamp-install steps external. Quick-disconnects would replace the hand-threaded hydraulic connections. The steps that genuinely had to stay internal were shortened where they could be: a sliding setup for the jacket clamps, a platform so the crane could be hooked faster, the spreader left on the crane with a longer chain so the crew stopped re-rigging it every cycle, stop blocks in the pit so the crane came free sooner. And the forklift came out of the changeover altogether — where the old sequence had called in a crowd to rush it, the sorted one was done with two operators, and a third for the press head. Sorted this way, the changeover that had run past forty-five minutes and climbing, at forty or fifty steps, collapsed to a sequence of nineteen — and the first clean run the team could put a stopwatch to came in at sixteen minutes forty-five seconds.

Three things were put in place so the gain would outlast the week. The first was standard work: the nineteen-step timed chart itself. The simplification to nineteen steps was what made documentation possible at all — before the event the changeover had been too variable to write down, and once it was sequenced and stable it could be. The same chart became the training document for new crew, where before the changeover had lived only in the heads of the operators who had always run it. The second was a measure the area could keep for itself: log the first changeover of each week and plot the single point against a line drawn at 600 seconds — ten minutes — on a chart on the wall. The third was an improvement list of twenty-one items, each with an owner and a date. The area manager held the hardware items — the second set of clamps, the vibration-resistant lock nuts. The maintenance lead carried most of the rest: the quick-disconnects, tool boards hung in the pit, stirrups for the clamping blocks, index marks so the clamps landed in the same place every time. A maintenance technician laid a self-leveling epoxy floor and built the product gages. Most of the twenty-one were closed within about six weeks of the event.

That sixteen minutes forty-five seconds was where the documented process started, not where it would stop. Ahead of it lay a costed and sequenced path: twelve minutes thirty once the ordered clamps arrived, eight minutes thirty-five as three further changes landed — under the ten-minute goal the KPI had been drawn for, and well under the twenty-minute target the charter had set.

Results, as the team measured them on the closing day:

Baseline, before the event: Undocumented; when first timed it passed 45 minutes and was still climbing, at 40 to 50 steps, before the count was abandoned — run by a crowd of operators with a forklift working the pit.
First documented run: 16 minutes 45 seconds, in 19 sorted steps, with two operators and no forklift.
Engineered path: 16:45 → 12:30 with the second set of clamps → 8:35 with three further changes. Charter target: under 20 minutes. KPI goal: 10 minutes.
Changeover steps: 40 to 50-plus (uncounted) → 19, documented for the first time.
Crew and equipment: A crowd called in to force the changeover, plus a forklift → two operators (a third for the press head), no forklift.
Standard work: Written, posted, and doubling as the new-hire training document.
KPI: First changeover of each week logged against a 600-second goal line, kept by the area.
Improvement list: 21 items, each owned and dated; most closed within about six weeks.

The changeover had looked like a speed problem, and speed problems invite the obvious answer — do each step faster, buy faster hardware. What the week found instead was that a changeover’s duration is barely about the pace of the work; it is about how much of the work sits on the stopped machine, and almost any of it can be moved off, staged to happen while the previous run is still casting. That distinction — internal work that needs the machine stopped, external work that does not — is not new. It is the heart of a method the Toyota Production System calls SMED: single-minute exchange of die, meaning a die changed in single-digit minutes, under ten. The term is Shigeo Shingo’s, and it comes from the press shop — the replacement of a steel stamping die set in less than ten minutes, achieved by exactly the sorting of internal from external work that the crew did that Monday. The changeover this kaizen was driving toward was one of that kind, arrived at by that method.

The reason a single-digit-minute changeover is worth the trouble sits a level above the machine. The changeover time is what sets the smallest batch a plant can afford to run — the economic order quantity, the run just long enough to earn its setup back. The longer the changeover, the larger that batch has to be; and the larger the batch, the more a plant makes in excess of what any customer has actually ordered. That excess — product made beyond the order — is overproduction, the waste the Lean canon names first and counts worst, because it is the parent of the others: it settles into inventory, ties up capital in goods no one has asked for, and hides every other problem behind the stock. Once the changeover is negligible there is no longer much value in continuing to run, so the batch shrinks and the surplus with it. Sixteen minutes at the machine was never a sixteen-minute problem. It was the governor on how the whole yard scheduled and on how much concrete stood outside waiting to be needed. Bring it under ten minutes and the yard can make what its customers are actually ordering, in the order they order it. That a change this modest sits so near the center of a Lean transformation, rather than at its edge, is what made this a classic instance of the kind — which was the point of the exercise, and never really the stopwatch.

· · ·