Lessons · Lesson 4 of 6
- 01 · What a pattern carries that a drawing does not
- 02 · A wrong rule is not a wrong pattern
- 03 · What the nester optimises, and what it was never told
- 04 · Automatic against a marker maker: where each one wins
- 05 · The block library nobody maintains
- 06 · What survives a handover, and what silently does not
Automatic against a marker maker: where each one wins
Compare an automatic nest against a marker maker on real markers, price the difference in both directions, and decide which markers should be made by which.
Lesson 4 of 6 · 18 min
The week of 12 May, the marker room
Karnaphuli has two marker makers and one nesting licence. The argument in this room is the wrong argument almost everywhere: whether the software is better than the person. It is a question with an answer, the answer is different for every marker on the list, and both answers are cheap to measure.
Six markers were made twice that week, once by each method, on the same pattern set and the same stated width. Consumption throughout is pattern area divided by efficiency times cuttable width — course 4.5 derives it, and it is used here rather than derived again. CH-482's pattern area is 1.612 m² a pair and Karnaphuli cuts this twill at 1.46 m.
| Marker | Nester | Time it ran | Marker maker | Time it took |
|---|---|---|---|---|
| CH-482 main, 3 pairs | 84.7% | 20 min unattended | 85.2% | 55 min |
| CH-482 top-up, 2 pairs | 79.4% | 15 min unattended | 82.6% | 40 min |
| CH-482 small parts | 88.9% | 8 min unattended | 89.1% | 25 min |
| Width-band re-mark at the low width | 84.1% | 20 min unattended | not attempted | |
| Checked shirt, matched, 6 garments | not possible | 76.9% | 3 h 10 min | |
| CH-482 sample marker, 1 pair | 71.2% | 4 min unattended | 74.8% | 20 min |
The marker maker's hour is almost never the expensive thing
Take the top-up marker, made at the end of the order for the 1,850 pairs left over after the main lays.
at 79.4%: 1.612 / (0.794 x 1.46) = 1.3906 m a pair
at 82.6%: 1.612 / (0.826 x 1.46) = 1.3367 m a pair
difference = 0.0539 m a pair
1,850 pairs x 0.0539 = 99.7 m, at USD 3.95 = USD 393.82The marker maker took 25 minutes longer than the machine. At Karnaphuli's loaded CAD rate of USD 7.40 an hour, those 25 minutes cost USD 3.08.
Turn that into a break-even and the argument closes. At 0.0539 m a pair and USD 3.95 a metre, the human work pays for itself at under 15 pairs. The main marker is the same shape: half a point, USD 748.13 on 24,600 pairs, against 35 extra minutes worth USD 4.32.
So the machine is not there because a marker maker is expensive. On these markers a marker maker is almost free. The machine is there for something else.
What the machine is actually for
The machine has two real jobs. Only the first of them shows up in that table.
The first is the width-band re-mark. Course 4.5 argues for sorting rolls into width bands instead of marking everything at the contractual minimum width. Banding means a second and a third marker for every style, at every width, and nobody makes those by hand:
at 84.1%: 1.3129 m a pair
at 84.7%: 1.3036 m a pair
difference x 24,600 pairs = 228.8 m, at USD 3.95 = USD 903.76USD 903.76 for twenty minutes of a machine nobody was using. By hand it is 55 minutes of a marker maker who has 34 other markers on the week's list. So it does not happen, and the low-width rolls get cut on the marker made for the high-width ones.
The second job is not in the table at all. It is quieter and larger. A nester improves with the time you give it, and that time is unattended. The CH-482 main marker reached 84.7% in 20 minutes and 85.4% when it was left running for three hours overnight.
at 84.7%: 1.3036 m a pair
at 85.4%: 1.2929 m a pair
difference x 24,600 pairs = 263.2 m, at USD 3.95 = USD 1,039.64For nothing. The nester's currency is unattended time, which is the cheapest input in the building, and almost nobody spends it, because the marker looked acceptable at eight in the morning.
Where the marker maker wins outright, and why
Three of the rows go the other way, and the pattern in them is not about skill.
The matched check. The nester at Karnaphuli cannot express a constraint between two pieces at a fabric repeat. So it does not produce a worse marker. It produces nothing. Three hours and ten minutes of a person is not an expensive answer to that; it is the only answer.
The top-up and the sample marker. Both are few pieces, large and awkward, with little to nest against. This is exactly where the gap is widest: 3.2 points on the top-up and 3.6 on the single-pair sample, against 0.5 on the main marker and 0.2 on the small parts.
That ordering is the finding. It is a statement about the shape of the problem rather than about the software. The gap narrows as pieces get smaller and more numerous. It widens as they get larger, fewer and more constrained. Small pieces give a search algorithm many nearly-equal moves, and it finds a good one quickly. Six large panels have few good arrangements and a great many bad ones, and a person who has nested that shape a thousand times goes straight to one of the few.
| Send to the nester | Keep for a marker maker |
|---|---|
| every style at every width band | anything matched at a repeat |
| every re-mark after a ratio change | top-ups and odd leftover quantities |
| markers with many small pieces | single-garment sample markers |
| overnight runs on anything already released | any marker with a constraint not in the file |
The last line on the marker maker's side of that table is the one people forget. A marker maker who knows the fusible pieces come off a 0.90 m roll, or that this buyer's inspector always looks at the back yoke, is applying a constraint that is not in the file and cannot be. Until somebody puts it in the file, that marker belongs to a person.
Check yourselfYour nester gets 83.9% in fifteen minutes and your marker maker gets 85.1% in an hour on the same marker. The order is 30,000 pieces, pattern area 1.94 m² a piece, cuttable 1.52 m, cloth USD 4.20. Which do you use?Show the answer
Consumption is 1.94 / (0.839 x 1.52) = 1.5213 m and 1.94 / (0.851 x 1.52) = 1.4998 m, so 0.0215 m a piece, 645 m over the order, USD 2,709. Forty-five extra minutes of a marker maker, at any rate you will ever pay, is under USD 10. Use the marker maker, and the answer would still be the marker maker at a tenth of the quantity. Then ask the second question, which is the one that actually matters: how many other markers did not get made this week because that hour went here. If the honest answer is more than one, the machine should be running overnight on all of them, and the marker maker should be spending the hour on whichever one is matched.