Changeover loss has two parts: the time the machine is stopped switching products, and the scrap it makes before the first good part. They are two different losses with two different fixes, and attacking them as one number hides where the money is.
Most changeover projects chase a single stopwatch figure and stop when the machine restarts. That misses half the loss. Every switch has a time cost, the minutes the line is down while you swap and adjust, and a quality cost, the trial parts and reject material you burn getting the process back to spec. On a line that changes over ten times a day, the scrap side can quietly outweigh the downtime side, and no amount of faster wrenching will touch it. This post separates the two losses, gives you a tool for each, and shows how to know which one to attack first.
If you are just getting started on the mechanics of the switch itself, read SMED quick changeover first, then come back here to split the loss apart. And before you can reduce anything, you have to measure it the same way every shift, which is the job of changeover time measurement.
What are the two kinds of changeover loss?
Changeover loss splits into time loss and scrap loss. Time loss is the stopped-machine window, from the last good part of the old product to the first good part of the new one. Scrap loss is the defective and trial material made during ramp-down and ramp-up, before the process is verified stable at rate. The two are governed by different physics and respond to different countermeasures, which is why lumping them into one "changeover cost" number usually points the team at the wrong fix.
Time loss lives in the availability term of OEE, where it shows up as setup and adjustment, one of the classic six big losses. Scrap loss lives in the quality term, as startup rejects, a separate one of the six. Two losses, two OEE terms, two toolkits. A team that only films the swap and never counts the trial parts is optimizing one term while the other bleeds.
How do you reduce changeover time loss?
You reduce time loss by moving work out of the stopped-machine window and simplifying what is left. That is the whole of SMED: separate internal setup, which can only happen while the machine is stopped, from external setup, which can happen while it still runs, then convert as much internal to external as you can, then streamline the remainder with quick clamps, locating pins, and standardized fasteners. The first pass, just sorting tasks and staging them on a cart, usually removes a third or more of the stop before any hardware is bought.
Time loss is a logistics and mechanics problem. The questions are physical: what is the crew walking for, what are they hunting for, what are they adjusting by trial. Every walk to the tool crib during the stop is external work trapped inside the internal window. The countermeasures are staged kits, shadow boards, duplicate fixtures dialed in offline, and torque tools preset to the right value. None of these touch the recipe or the material; they only touch the clock. Track the stop as machine downtime by reason code and changeover will usually sit near the top of the Pareto.
How do you reduce changeover scrap loss?
You reduce scrap loss by making the process land on-spec the first time instead of dialing it in by trial. Startup scrap comes from the ramp: the machine is running but the settings, temperatures, tensions, or registrations are not yet where they belong, so the parts are junk until someone tweaks them true. The fix is to remove the tweaking. Record the exact parameters from the last good run of each product, store them as a recipe, and recall them at setup so the machine starts where it ended last time rather than where the last operator left the dials.
The other half is catching the first bad part before you make a thousand more. A first-piece verification gate, a real check that the first part off the new setup meets spec, is what separates a controlled ramp from a hopeful one. Combine it with mistake-proofing so the wrong die, wrong material, or wrong orientation physically cannot be loaded. These are poka-yoke and preset-parameter tools, not SMED tools, and they move the quality term, not the availability term. A line can have a beautiful eight-minute changeover and still throw away forty minutes of product on the ramp because nobody attacked the scrap.
Which loss should you attack first?
Attack the bigger one, and the only way to know which is bigger is to measure both. For one representative changeover, log the stopped-machine minutes and count the reject and trial parts on both the ramp-down and the ramp-up. Convert the scrap to minutes of lost production at rate so the two losses share a unit. Now you have a fair comparison, and it often surprises people: a line with a tidy fifteen-minute stop can lose the equivalent of thirty minutes to a scrap-heavy ramp that nobody was watching.
Rank changeovers with a Pareto chart the same way you rank any loss. Do it twice, once by time loss and once by scrap loss, because the changeover that costs the most downtime is frequently not the one that costs the most scrap. A color-change or allergen-change on a filling line might swap in five minutes but burn twenty minutes of flush and reject product. A die swap might take an hour but ramp clean. Same word, "changeover," two completely different loss profiles.
What is the order of operations for cutting changeover loss?
Run the reduction as a sequence, not a scramble. This order keeps you from spending capital on the smaller loss:
- Measure both losses on one changeover. Time the stop from last good part to first good part, and count every trial and reject part on the ramp. Put both in the same unit, minutes of lost production at rate.
- Pareto the changeovers twice. Rank by time loss and again by scrap loss. Pick the target where the total of the two is largest, not the one that merely feels slowest.
- Separate the swap from the ramp. Decide, for the target, whether the money is in the stopped window, the scrap, or both. Assign the right toolkit to each before touching anything.
- Fix the time loss with SMED. Sort internal versus external, stage the external work on a kitted cart, convert what you can, and streamline the rest with quick clamps and locating hardware.
- Fix the scrap loss with preset recipes and verification. Capture the last-good-run parameters as a recipe, recall them at setup, mistake-proof the load, and put a first-piece gate at the start of the ramp.
- Standardize the method as work. Write the swap and the ramp as standard work train every shift on it, and stop relying on one hero crew's memory.
- Re-measure and hold. Track both losses on every changeover permanently. Gains decay the day the cart stops getting kitted or the recipe stops getting followed.
How much of the loss is really recoverable?
Most of it, and most of it cheaply, because both losses are dominated by method rather than hardware. On the time side, the documented history of quick-changeover work shows reductions of ninety percent and more, with the first big gains coming from staging and task-sorting before any equipment spend. On the scrap side, preset recipes and first-piece verification routinely turn a long trial-and-error ramp into a handful of parts, because the machine no longer has to be discovered from scratch each time.
The prize is capacity you already own. U.S. manufacturing capacity utilization has run in the mid-70s percent range, most recently about 75.7 percent, according to the Federal Reserve's G.17 release (Federal Reserve, Industrial Production and Capacity Utilization, G.17). Setup-and-adjustment and startup rejects are two of the six equipment losses catalogued in the international KPI standard for manufacturing operations, ISO 22400-2, which defines OEE and its supporting metrics precisely (ISO 22400-2:2014). Every changeover hour and every reject part recovered on a constraint line is throughput added to a plant that is already running well under its own installed capacity.
How does changeover loss connect to batch size?
Both losses tax small batches, so both push the plant toward long runs it may not want. When a switch costs an hour of downtime and a bin of scrap, the finance math forces big campaigns to amortize the loss, which bloats inventory and slows the plant's response to demand. Cut both losses and small batches become affordable, which is the door to leveled scheduling and pull. Reduce only the time loss and the scrap can still make small batches uneconomical; the ramp reject on every short run eats the flexibility you were trying to buy. Increasing first-pass yield at startup is what finally makes frequent, small changeovers pay.
How do you sustain both reductions?
Treat changeover as a measured event with two numbers, not a black hole with one. That means a reason-coded record of every changeover that captures both the stopped-machine minutes and the ramp scrap, the standard method visible at the machine, and a review when either number drifts. This is where real-time capture earns its keep: paper logs get filled in at end of shift from memory and almost never record the scrap honestly, while tablet capture at the station timestamps the stop and logs the reject count as it happens, so a supervisor sees drift the day it starts. That shift from paper logging to real-time capture is exactly what CLS did across its shops (see the CLS case study), and it is the foundation a two-loss changeover program stands on. You can pressure-test the payback of a given reduction with the OEE calculator before you commit the crew's time. No rip-and-replace, just both numbers arriving fast enough to act on.