Turning Changeover Hours Into Production Hours

Changeover hours disappear when you stop making expensive transitions, not when you make each transition faster. On a machine where light-to-dark costs 60 minutes and dark-to-light costs 240, the order of the jobs is worth more than any procedure improvement. EDGEBIC by User Solutions models changeover as a per-machine, product-pair lookup, so the schedule charges the real cost of each transition and can then be sequenced to avoid the expensive ones.

This post shows the mechanism, walks the documented arithmetic in full, and is clear about what sequencing cannot fix. For the general treatment, see changeover time reduction and the SMED quick changeover approach, which attacks the same cost from the opposite direction. This post sits under the EDGEBIC results guide.

The Problem With One Setup Number

Most scheduling systems store a single setup time per routing step. Paint booth: 30 minutes. Every job, every sequence, every time.

The floor does not work that way. Going from white to cream is a rinse. Going from black to white is a full solvent purge. A furnace switching alloys is 90 minutes; running the same alloy again is zero. One number cannot represent both, so it represents neither, and the schedule is wrong in a direction that always favors optimism.

The documented paint booth case shows how wrong. Three jobs on one booth: MO-2001 white, MO-2002 black, MO-2003 white. With a flat 30-minute setup on each, the schedule claims the day ends at 12:45. The actual changeovers the floor would live through are 30 minutes cold start, then 240 minutes white to black, then 240 minutes black to white, and the real day runs about 11 hours. The plan overpromised by roughly half a day, on three jobs, on one machine.

Every downstream decision inherited that error: the promise sales made, the overtime nobody pre-authorized, the truck booked for an afternoon pickup.

Mechanism: Charge the Real Transition

EDGEBIC's setup matrix replaces the flat number with a lookup keyed on three things: which machine, what it last ran, and what it is about to run.

The data model has three tiers, which is what keeps it maintainable:

Eight families produce 64 cells per machine instead of thousands of product pairs. When one specific pair genuinely differs (red to black needing an extra solvent flush, 90 minutes rather than the family's 60) a product-level override handles it without disturbing the rest of the table. For the definition of the grouping concept, see what is a setup family.

At schedule time a five-level chain resolves each step, first hit wins: cold start uses the routing's own setup, the same product back-to-back charges zero, then the product override, then the family cell, then the routing or work center default if no cell exists. Every resolved value is stored with its source, so the Job View can tell you exactly why a step was charged four hours instead of thirty minutes.

The Arithmetic, Three Ways

Same three jobs, same booth, three configurations. The families are Light (white, cream) and Dark (black, dark grey), and the matrix for this booth reads: Light to Light 0, Light to Dark 60, Dark to Light 240, Dark to Dark 0.

Scenario A, flat setup (the dishonest plan). 30 minutes charged per job, 90 minutes total, day ends 12:45 on paper. Floor reality is about 11 hours.

Scenario B, matrix loaded, same due-date order (white, black, white). Now the resolution is honest:

Total setup: 330 minutes. The day runs 08:00 to 16:45, an 8.75-hour day that visibly overflows the shift. This plan looks worse and is worth money immediately, because the planner can now see the overflow before it happens instead of discovering it at 16:00.

Scenario C, matrix loaded, resequenced light before dark (white, white, black).

Total setup: 90 minutes. The day runs 08:00 to 12:45 and all three jobs fit in one shift with more than three hours to spare. The Source value on each row is also what makes the saving auditable, because a matrix can be fully configured and quietly not firing.

From 330 minutes to 90 is a 73% reduction, from sequencing alone. No new equipment, no faster purge, no overtime. The 4.75-hour day that the dishonest plan promised turns out to be genuinely achievable, but only in the right order.

Notice the two-step structure, because it is the shape of the whole outcome. The matrix turns a deceptive plan into an honest one. Sequencing turns an honest plan into an efficient one. You cannot do the second without the first, which is why loading the real numbers comes before any optimization.

Scaling It Past What You Can Sequence by Eye

Three jobs on one booth is a puzzle you can solve on paper. Forty jobs across twelve work centers, where the best paint order fights the best mill order and both fight the due dates, is not.

EDGEBIC's optimizer carries a least-setup goal alongside on-time-first, minimum makespan, and balanced presets. Two properties make it usable rather than alarming: the multi-run layer is clamped never worse than the schedule you already have, and nothing is written until a planner reviews the comparison and accepts it. When you choose least-setup, you see what it costs in tardiness before you commit, and when you choose on-time-first, you see what it costs in setup hours.

One more place the same mechanism pays: work center groups. Each pool member can carry its own setup override, because an older machine with simpler fixturing genuinely changes over faster. In the documented three-mill example, the 1998 Bridgeport is 50% slower per unit and carries a 0.25-hour setup against the group's 0.5, which is exactly the kind of trade-off a scheduler should evaluate rather than a planner remember.

What the Software Cannot Do Alone

It cannot know your changeover times. Every number above came from a table a human filled in. The 73% exists because the matrix held real minutes. Guessed values produce a confidently wrong plan, which is worse than an obviously wrong one.

It cannot make a changeover faster. Sequencing avoids transitions; it does not shorten them. If your dark-to-light purge takes four hours, it still takes four hours whenever you have to do it. SMED work, quick-change tooling, and standardized procedures attack that number, and they compound with sequencing rather than competing with it.

It cannot override physics or quality rules. Some transitions are mandatory regardless of cost (an allergen washdown, a validated cleaning step). Model them as high-cost cells and let the schedule route around them where it can, but never let a sequence decision quietly skip a required step.

It cannot resolve the due-date conflict for you. Grouping like with like saves hours and can delay a specific customer. The optimizer will show you both numbers. Choosing between them is a business decision.

It cannot maintain the table. New products need a family assignment. A changed process changes a cell. A matrix that was true two years ago and has drifted since produces schedules that are precisely wrong.

Want to know your own 330-to-90? Bring your worst machine's changeover times and a week of orders to a demo, and we will sequence them both ways.