Changeovers & lot sizing

A changeover is the time a line stands idle while it switches from one product to the next: cleaning tanks, swapping moulds, changing labels, running test samples. If a changeover takes 90 minutes, nobody wants to do it often, so the plant runs long batches. Long batches mean big piles of stock and a slow response when demand shifts.

That’s the classic trade-off in lot sizing: fewer, longer runs save changeover time but build inventory; shorter runs keep stock low but lose capacity to changeovers. It is the production version of EOQ, with the changeover playing the role of the ordering cost.

The way out is to make changeovers shorter. SMED (single-minute exchange of die), developed by Shigeo Shingo, works in three steps:

1. Separate internal work, which can only be done with the line stopped, from external work, which can be done while it runs.
2. Move as much as possible to external: pre-stage the next labels, pre-mix the next batch, pre-heat the moulds.
3. Streamline what remains with quick-release fittings, standard settings and practised teams.

If a 90-minute changeover drops to 30, the plant can switch three times as often for the same lost time. Batches shrink, inventory falls and the line can follow demand more closely.

Many process plants also use a product wheel: a fixed, repeating sequence of products, ordered to make each changeover as easy as possible.

Changeovers eat capacity

On a line making several products, time is spent either running or changing over. If running the required volumes uses a fraction $\rho$ of the line’s time, only $1 - \rho$ is left for changeovers. A product wheel that runs each product once per cycle of length $T$ needs:

where $s_i$ are the changeover times in the cycle. Each product’s batch must cover demand for one whole cycle, so batch sizes and cycle stock grow in proportion to $T$.

In practice, plan the wheel somewhat longer than the minimum to leave room for breakdowns and demand swings.

SMED in practice

The order matters: first move work outside the stop (often half the gain at almost no cost), then simplify what must stay inside it.

Sequence matters

Many changeovers depend on what ran before. In shampoo or paint, light to dark colours is quick and dark to light needs a full clean; in food, allergen-free products run before those with allergens; in plastics, low-melt materials before high. Finding the best order is like the travelling salesman problem (visit every product once at the lowest total changeover). Product wheels fix a good sequence and repeat it, which also makes the plant predictable for planners and suppliers.

Where changeovers really cost money

• At the bottleneck: a changeover hour is a lost hour of plant output, worth the bottleneck’s throughput per hour.
• Away from the bottleneck: a changeover uses spare time and costs little beyond labour and materials. Running smaller batches there is almost free.

So SMED effort should go first to the bottleneck, and batch-size rules should differ between bottleneck and non-bottleneck resources.

Lot sizing under capacity limits

With several products sharing a line and limited capacity, choosing batch sizes becomes a scheduling problem in its own right. In the research literature it appears as the economic lot scheduling problem, often solved with a common-cycle rule (every product once per cycle) or with cycles that are multiples of a base period, so fast movers run every cycle and slow movers every second or fourth. In process industries, campaigns group several batches of similar products to reduce cleaning and validation effort.

Measuring changeovers honestly

Measure from the last good unit of the old product to the first good unit of the new product at full speed. Ramp-up time at reduced speed and the scrap during start-up are part of the changeover, and leaving them out hides much of the loss. Track the changeover time per changeover type, and investigate the long tail rather than the average.