PolygonLogix 1.5.0 ships with a new nesting engine. We call it Logix. It nests denser than our previous engine on real production work, it does it in a fraction of the time, and it honors the part spacing you set on every sheet. It is now the default engine in both Quoting and Production.
This post is about what changed, why it works, and how we measured it. We have tried to keep the claims to things we can show you numbers for.
Why Nesting Engines Plateaued
Almost every nesting engine on the market, including the one we shipped until last week, works the same way. It sorts the parts, places the first one in a corner, slides the next one up against it, and keeps going until the sheet is full. Once a part is placed, it never moves again. The engine’s cleverness goes into the order it tries and the geometry it uses to find where the next part fits.
That approach has a ceiling, and the industry hit it years ago. The first part placed decides the shape of the whole sheet. A bad early choice can’t be undone, so engines compensate by trying several orders and keeping the best, which is slow, and the result still depends on a guess made before the engine knew what the rest of the sheet would look like.
It also explains a quiet problem most shops never see: when the geometry says a part almost fits, some engines let the gap shrink below the spacing you set rather than lose the placement. The part cuts fine. The number on the screen just isn’t the number on the sheet.
How Logix Nests
Logix works the other way around. It starts by putting all the parts on a sheet that is deliberately too small, overlapping. Then it separates them: part by part, it moves whatever is colliding to a spot that collides less, until nothing overlaps. Then it shrinks the sheet and does it again. Every part is free to move at every step, so an early placement never locks the layout in.
The method comes from the open research literature, where it has set most of the recent records on the standard nesting benchmarks. We built our own implementation of it in PolygonLogix, on our own geometry, with the things a production shop needs that a research solver doesn’t have: fixed sheet sizes, multi-sheet jobs, parts nested inside the cutouts and bays of other parts, edge margins, kerf, and common-line cutting.
A few details matter for what you see on screen:
- Parts fit into each other. Logix looks for places inside the cutouts and concave pockets of parts already on the sheet, and it remembers pairs of parts that interlock well so it can use them again on the next job with those parts.
- Every sheet is compacted, not just the last one. Parts slide toward the closed end of the sheet, so what’s left over comes off as one clean drop on the last sheet instead of a strip of scrap on every sheet.
- Spacing is exact. Logix treats your part spacing as a rule, not a target. Every nest is checked against the true part geometry before it leaves the engine, and a layout with a single violation is rejected.
- It uses the time you give it. Set the nest time to 10 seconds for a quote or 60 for production, and Logix keeps improving the layout until the clock runs out, drawing the sheets as it goes.
What It Means on Real Work
We measured Logix against our previous engine on 92 real production jobs from our own shop, 154 material groups, every nest run three times with different seeds and the median taken. Same sheets, same spacing, same rotation rules, same machine.
- Fewer sheets. At a 60-second nest time, Logix used 474 sheets where the previous engine used 487, about 2.7% less plate. On most jobs the sheet count is the same; the savings concentrate on the big multi-sheet jobs, where Logix regularly drops one or two sheets.
- Cleaner drops. Because every sheet is compacted, the leftover material comes off as one piece on the last sheet.
- Faster. On a synthetic set of 500 jobs at a 10-second budget, Logix finished all 500 and reached a valid nest in 0.28 seconds median; the previous engine finished 410 of them and took 3.4 seconds median.
- Honest spacing. Under a strict check that fails any nest with a gap under the spacing setting, Logix passed every one of the 154 groups. The previous engine, which we had been cutting with for a year, failed 57 of them, with gaps as tight as 0.38 inches on a 0.5-inch setting.
None of these numbers were produced by the engine grading itself. They come from an independent validator that checks the exact part geometry of every nest, the same one that scores our previous engine, and that we now run on every change before it ships.
How It Compares on the Standard Benchmark
Nesting research uses a standard set of ten test instances, maintained by the EURO Special Interest Group on Cutting and Packing, and reports density: how much of the used plate is part. Scores there don’t translate directly to a shop (they’re strip-packing problems with no sheet boundaries), but they’re the one place different engines can be compared on identical inputs.
At a 60-second budget, Logix averages about 86% density across the ten. The best open-source research solver we know of, sparrow from KU Leuven, averages about 86.7% on the same machine at the same budget, and it is the engine that holds most of the published records. Our previous engine, and the popular open-source nester most shops have tried, both sit near 73%.
So on the academic scoreboard Logix is within a point of the state of the art and about thirteen points ahead of where we were a week ago. We’ll keep working on that last point. The numbers above are from our own rig, every layout checked by our validator, and we’ll publish the full table when it’s a table we’re proud of.
How We Tested It
The rule for this release was simple: nothing ships unless it is no worse on real jobs, and every nest is validated against the true geometry. In practice that meant a bench of 92 real jobs and 500 synthetic ones, run three times each at 10 and 60 seconds, for every change we tried. Changes that helped the benchmark but cost a sheet on real work were thrown out, and several were.
We applied the same standard to cut paths, which had never had an independent check before. We wrote a validator for them too: every contour must have its lead-in and lead-out, no lead or rapid may cross another part or leave the sheet, every setting must do what it says, and the exported NC must match the path on screen. On the first run it found 12,913 defects across those 92 jobs. After this release’s fixes, it finds 4.
Along the way it caught things we would not have found by eye: an arc-direction repair that could have turned a bezel’s edge into a 3,400-inch circle, lead-outs that silently went missing on large parts, a pathing setting that did nothing, and corner rounding that skipped any corner touching an arc. It also made text engraving usable: a stencil part with 45,000 vertices went from 7.7 seconds and 32 defects to 1.6 seconds and none, and a 12,000-plate job went from 13 minutes to 11 seconds with identical output.
What’s in 1.5.0
- Logix is the default engine in Quoting and Production. Turbo, the previous engine, is still in the dropdown; it also handles common-line cutting jobs, where the two engines now work together and the tighter result wins.
- Nest time. A new control sets how long the engine works: 30 seconds by default, remembered per screen. More time, denser nest.
- Options that tell the truth. When Logix is selected, the settings it doesn’t use are grayed out, and the same for Turbo. Six pathing settings that did nothing have been grayed with a note on what governs them instead. A Reset to Defaults button sits beside the engine selector.
- Live preview. In Production, the sheets redraw as Logix compacts them, so you can watch the nest tighten.
- Cut paths. Leads are always present and never cross another part; text and dense sheets generate in seconds; corner rounding now works on arc corners; exported NC follows the drawn path exactly.
- Quoting fix. The quoting screen now nests with the engine you select. Previously it always used Turbo regardless of the dropdown.
Updated post-processor packages for every supported machine are included with the release. They change only one thing, rapids between parts now follow the routed path the App shows, and we recommend importing them one machine at a time and watching the first program on each.
What’s Next
Logix is a week old in production and it already has a list. Nesting memory will grow with every job you run: good interlocks found once get reused forever, which matters most for the parts a shop cuts again and again. Native common-line cutting inside Logix will replace the hand-off to Turbo. And there’s a point of density left on the benchmark that we intend to go and get.
If you run PolygonLogix, update to 1.5.0, set the nest time to 60 seconds on your biggest job, and count the sheets. That’s the test that matters, and it’s the one we’d like to hear about.