Where the money hides in a corrugator's numbers
Two 110-inch corrugators, seven million square feet a day each, three shifts, five days a week. A dashboard of their performance — and the part that matters, which is what you do with it. A report of numbers finds nothing. Someone has to ask where the money is.
The data is synthetic, built to look like two real machines over a year with a few real problems in it. The dashboard is live: pick a plant, pick a window, hover anything. Then read on for what it's hiding and how you'd find it.
Corrugator performance
table view — weekly numbers
What the dashboard is for
Every corrugator already has a report. It comes out of the MES or a spreadsheet every morning: volume, waste percentages, downtime, a top-five reasons list. People glance at it, nod, and go to the floor. It changes nothing, and the reason is structural: a number with no comparison isn't information. 3.4% side trim is neither good nor bad until you know it was 3.2% in June, that the other plant runs 3.0%, and that the gap is worth a specific number of dollars a year.
So the dashboard is built around three comparisons, and nothing goes on it that doesn't support one:
- Against last week — every tile carries a week-over-week delta, color-coded against the direction that's good for that metric. This is the "what changed" question.
- Against the trend — weekly lines over 13, 26, or 40 weeks, so a slow creep is visible as a slope rather than as noise. This is the "when did it start" question.
- Against the other machine — Dallas and Chicago on the same axes. Two machines of the same width and roughly the same mix are each other's best benchmark. This is the "is this normal" question.
The opportunities panel at the bottom does the last step that the morning report never does: it converts gaps into dollars, using the board cost and margin assumptions you can edit. Those small percentage points are where the money is, and the panel says how much.
What to look at, in order
| Metric | What it tells you | Where to look when it moves |
|---|---|---|
| Side trim % | Width you bought and threw away. Pure paper cost — the most expensive waste on the machine per point. | Order width mix vs. roll stock widths (are you running 87" orders on 98" paper?); slitter/scorer setup; trim allowance in the scheduling system. A creep after a rebuild is a setup problem, not a mix problem. |
| Shear % | Board cut out at order changes and splices. Scales with stops and order count, not with volume. | Stops per day; order sequencing (grouping like widths and grades); splice success. If shear rises with stops, fix the stops. |
| Speed without stops | True machine capability when it's running. Your ceiling. | Grade mix (double wall runs slow), steam, glue, operator setpoints. If this is flat while you expected a speed project to lift it, the project isn't working. |
| Speed with stops | Capability divided by reliability. What you actually got. | The gap between this and speed-without-stops is downtime, nothing else. If the ceiling rose and this didn't, you bought capacity and spent it on stops. |
| Downtime % and top reasons | Where the gap goes. The Pareto is the whole value here. | The top reason by minutes, then by count. Many short stops (stacker, knife) and few long ones (splicer, power) need different fixes — a 30-second stacker fault 40 times a day is a maintenance item; a 95-minute outage is insurance. |
| Minutes per stop | Separates chronic from catastrophic. | Rising min/stop with flat stops = longer recoveries (training, spares). Rising stops with flat min/stop = a new fault source. |
| Dry-end waste % — bond, warp, misalignment | Board that made it through the machine and failed at the end. Each category points at a different cause. | Bond → starch, temperature, glue gap. Warp → moisture balance, grade mix, ambient (watch Mondays and summer). Misalignment → web guides, edge sensors. Never read dry-end as one number. |
| Bale weight % | Total waste reconciled by the scale: everything baled over everything produced. The honest number. | If bale % is much higher than trim + shear + dry-end, waste is leaving the machine uncounted — roll ends, rejected units, floor sweep. The gap is a measurement problem before it's a process problem. |
What this dataset is hiding
I seeded five things into it. Try to find them on the dashboard before reading this — switch plants, widen the window, hover the Pareto.
- Dallas side trim is creeping. It's been climbing since early August, about a tenth of a point a week. The WoW delta never looks alarming because each week is only slightly worse than the last; the 26-week view makes it a slope. The knife/slitter downtime rose at the same time. Both point at the slitter rebuild — setup drift, not mix.
- Chicago had a splicer problem in September. Missed-splice minutes roughly tripled for five weeks, stops went up, and shear followed. On the 13-week Pareto it's the top reason; on YTD it's buried at third. A YTD report would never have shown it.
- Chicago's speed project worked — and didn't. Speed-without-stops rose through the year. Speed-with-stops barely moved. The capacity was real; the stops ate it. That's the single most common way a capital project "fails" on paper while succeeding on the machine.
- Chicago warps on Mondays. Dry-end warp is higher on the first day after the weekend, and higher in summer. That's moisture — cold board, humid plant — not a process setting. You'd never see it in a weekly average; it's in the day-of-week breakdown behind the hover.
- Dallas stops more, for less time each. Compare the two plants on the stops panel: Dallas logs more stops per week at fewer minutes per stop. That signature — many short stops — is the stacker, which is third or fourth on the minutes Pareto and so never gets the work order. Chronic beats catastrophic for dollars over a year, and the Pareto by minutes hides it.
None of these needs machine learning. They need the right comparison and someone who knows that a slitter rebuild happened in week 31. That knowledge isn't in the data; it's in the superintendent. The dashboard's job is to make the question sharp enough that the superintendent can answer it — which is exactly the design behind the shift investigator: the tool surfaces the finding and the question, the person supplies the cause.
The arithmetic
First, the speed, because it's the number everyone argues about. Seven million square feet a day on a 110" machine netting 96" after trim: 7,000,000 ÷ 8 ft = 875,000 lineal feet a day. Over three shifts — 1,440 scheduled minutes — that's an average of 608 fpm with every stop counted. With 10% downtime the machine has to actually run at about 675 fpm to hit it. That's why the dashboard shows run speeds in the high 600s and shift speeds around 600, and why the gap between them is the single biggest number on the opportunities panel: at these speeds a minute of downtime is 675 lineal feet, 5.4 MSF, gone.
Seven million a day per machine, five days a week, is about 1.8 billion square feet a year each — 3.6 million MSF across both. At a paper cost around $55/MSF, one tenth of a point of side trim is roughly $200,000 a year. Dallas's creep is six tenths. A single stop costs its minutes times the run speed times the width: a 110-minute outage at 675 fpm on 96" is about 600 MSF of board not made, and the margin on that is gone for good because the corrugator is the bottleneck — the LP in the optimization post put a shadow price of over $1,200 an hour on exactly that constraint. The opportunities panel runs these numbers on the live data with your own assumptions.
That's the post in one sentence: the report tells you the trim was 3.4%. The analysis tells you it used to be 3.2%, the other plant is at 3.0%, it started in week 32, the slitter was rebuilt in week 31, and the gap is over a million dollars a year. Only one of those gets a work order written.