SMED in a Small Plant: Changeover Reduction Step by Step

What SMED actually means

SMED stands for Single-Minute Exchange of Die. "Single-minute" means a single-digit number of minutes, so the target is a changeover under ten minutes. The Japanese engineer Shigeo Shingo developed the approach from the 1950s onwards while working with Japanese manufacturers, Toyota among them. He set it out in "A Revolution in Manufacturing: The SMED System", published in English in 1985.

Treat the ten minutes as a direction of travel. A heavy injection mould and a bar-fed lathe start from very different places. What carries over to any machine is the distinction the whole method rests on.

Internal setup is work that can only happen while the machine is stopped: removing the old fixture, mounting the new one, setting the datum. External setup is work that could happen while the machine is still running: fetching tools, checking the program, staging material and gauges.

Shingo's starting observation was that plants rarely tell the two apart. Much of the external work gets done during the stoppage anyway. Nobody decided it should be that way. The setter simply starts preparing once the previous job has finished, and the machine waits.

If you have read our piece on changeover hours that never reach the costing, the edge items it describes will look familiar. Hunting for tools, waiting for the setter, waiting for first-article release. Nearly all of that is external work that has drifted into the stoppage. SMED is a structured way of pulling it back out.

Before you start, fix the definition

You need one measure that everybody reads the same way. Changeover time starts with the last good part of the old job and ends with the first good part of the new job that goes into production without further adjustment. First-offs, program corrections and waiting all sit inside that window.

Many plants measure something narrower. They start the clock when the setter picks up a spanner, or stop it at the first part out of the machine, good or not. Both habits shrink the number, and both hide exactly the minutes where the losses are.

For a before-and-after comparison you only need a few figures:

• changeover time from good part to good part, as a median and as the range between the fastest and slowest changeover
• the number of adjustment loops before the first good part
• the number of changeovers per week on that machine
• a few weeks later, the batch sizes you are actually running

Watch the range more closely than the average. A changeover that swings wildly from one job to the next does more damage to the schedule than one that is reliably a little longer.

Step 1: Film a real changeover

Pick a machine that changes over often and whose downtime costs you delivery dates. Film one complete changeover there. A phone on a tripod is enough. Keep recording from the last good part to the first good part, including the stretches when the setter walks off to find something.

Agree this with the setter beforehand, and with the works council or employee representatives if you have them. You are studying the sequence, not grading a person. If that isn't clear, you will get a demonstration changeover instead of a real one.

Then watch the video together with the setter and write down every task with its start and end time. The list will be longer than anyone expects. Trips to the tool store, a phone call to production planning and the search for the right spanner all go on it.

Step 2: Separate internal from external setup

Go through the list line by line and ask of each task whether the machine has to be stopped for it. On a CNC lathe the result might look like this:

Shingo treated this separation as the decisive step, and it costs next to nothing. A checklist and a setup cart will do.

There is one condition. Someone has to do the external work while the machine is still running, so name that person. If one setter looks after several machines, preparation will slide back into the stoppage the first time two jobs finish together.

Step 3: Convert internal work to external

Now look at the tasks that currently need a stopped machine. Some of them can move outside if you prepare them differently:

• preset fixtures off the machine so they only need to be dropped in
• measure tools on a presetter and load the offsets in advance
• preheat injection moulds before they go into the press
• standardise clamping heights or use adapter plates so alignment disappears

This is where hardware enters, and with it cost. Check every idea against how often that machine changes over. A zero-point clamping system on a machine that is reset once a week rarely pays for itself.

Step 4: Streamline what stays internal

Whatever is still internal gets shorter. The levers are mechanical and unglamorous:

• replace bolts that need several turns with cam clamps, quick-release levers or bayonet fittings, wherever the clamping force allows
• replace adjustment with fixed stops and locating pins. Any adjustment where someone tries, measures and tries again is a candidate.
• work in parallel. Two people working both sides of the machine at once cut the stoppage, even when the total labour time stays the same.
• remove walking. Everything the changeover needs sits within arm's reach.

The order of the steps matters. Start with quick clamps and you speed up a part of the changeover that may not be the largest item at all. The video from step 1 shows where the minutes really go.

Step 5: Standardise and train

Write the new sequence down as a setup standard. One page per machine or part family is enough: the order of work, who does what, and what has to be staged before the machine stops. Photos beat paragraphs.

Then train everyone who changes over that machine, the late shift included. After a few weeks, film again and compare the two videos. Without a standard and someone who insists on it, the sequence drifts back to the old one. The piece on what survives after the consultant leaves explains why.

Where the freed-up time goes is a management decision. You can use it as capacity, or run smaller batches and bring down stock and lead time. Make that call on purpose and update the changeover times and batch sizes in the ERP. Otherwise the time disappears into daily business and nobody can say where it went.

Mistakes we see in mid-sized plants

Starting with purchasing. Quick-clamp systems get ordered before anyone has watched a changeover from start to finish.

Working from memory. A changeover reconstructed around a meeting table loses exactly the searching and waiting you are trying to find.

Involving only the setter. Many losses sit with stores, quality and production planning. If the gauge lives in the inspection room or the material isn't there, the setter alone can't fix it.

Celebrating the average. A better mean with the same spread does not make a single delivery date more reliable.

Mistaking a workshop for a programme. A SMED workshop gives you one converted machine and a list of ideas. The other machines need someone inside the business to carry the method on.

Leaving the costing alone. As long as the ERP holds the old changeover time, planning keeps scheduling the old batch sizes.

When SMED is the wrong lever

SMED pays where changeovers are frequent and every minute of downtime costs money. Plenty of machines don't fit that description. To see whether changeovers cost a machine more time than breakdowns, small stops and scrap do, calculate its OEE for one shift.

On a machine with spare capacity, a shorter changeover first produces more idle time. It only becomes interesting there if you want to turn the saved time into smaller batches. Long runs with a few changeovers a month don't offer much either.

Sometimes the problem isn't at the machine at all. If orders wait days for release, material or a drawing, SMED saves minutes on the shop floor while the days before it stay where they are. In our process audit example, the bottleneck sits in the approval step while the business is already pricing a second machine.

So before any SMED project, find out where in the order flow the time is actually lost. A process analysis answers that with timestamps from your own plant, before you spend money on clamping hardware.