Busbar nesting and scrap: building a cost model that survives audit

Most busbar shops know their scrap rate to the nearest guess. The rate is "about ten percent," copper is expensive, and nesting software is supposed to help. What is usually missing is a defensible number for the finance director's desk. This text presents that number, shows its origin, and is honest about the part of it that cannot be captured.

The cost of scrapped copper is dominated by the conversion premium paid to the mill and not recoverable, not by the metal price. The achievable yield is set by the length distribution of the parts in the cutting pool, not by the quality of the nesting solver.

What a 6000 mm bar actually gives you

Start with the geometry, because the arithmetic is unforgiving and most people have never written it down.

A mill-supplied Cu-ETP flat bar arrives nominally 6000 mm long. Before a saleable part is cut, the following is lost:

• A lead crop: mill ends are rarely square, often slightly bruised from handling, and the first clamp position needs clean metal. Ten to twenty millimetres is typical.
• Cut allowance: on a guillotine shear the cut is chipless, so one stroke produces the trailing end of part n and the leading end of part n+1, and no length is consumed. On a circular cold saw the kerf is lost, typically 2.5 to 3.5 mm per cut depending on blade; on a bandsaw, roughly 1.3 to 1.6 mm. Over a bar cut into six parts a saw therefore throws away 15 to 21 mm before a single mistake is made. That is why shearing rather than sawing is the default on a busbar line, and the reason common-line cutting is free on a shear and impossible on a saw.
• A tail remnant: whatever is left when no remaining part will fit.

The tail is where the money is. Consider a 6000 mm bar with a 10 mm lead crop, giving 5990 mm usable, cut into parts of one length only:

Same machine, same software, same operator, same shift. Yield ranges from 83.3% to 99.3% purely because of the relationship between the part length and 6000.

Consider the 1000 mm row. Six 1000 mm parts fit a 6000 mm bar exactly. Taking 10 mm off the front for a lead crop results in five parts, and 990 mm goes in the bin. A single decision about crop length moved the yield by nearly seventeen points. This is the most common self-inflicted scrap loss in the trade.

Measure the incoming bar, do not assume it

Mill bar is not exactly 6000 mm. Depending on supplier and order it arrives at 6000 mm with a plus tolerance, or as random mill lengths in a band. EN 13601 governs the section dimensions and tolerances for copper rod, bar and wire for electrical purposes; the cut length received is a commercial term between the shop and the supplier, so the purchase spec should be consulted to determine what was actually agreed.

If the bars are actually 6040 mm and the nest assumes 6000, 40 mm is left on every bar, and in the 1000 mm case above, a whole part is left. If they are 5975 mm and the nest assumes 6000, the last part of every bar comes up short and gets scrapped, costing the tail and the part.

The fix is dull and effective: measure incoming bar lengths on receipt, by section and by heat where the supplier changes, and feed the measured length into the nest rather than the nominal. Machines with a bar-position encoder can do this at load. It costs a few seconds per bar.

Why the remnant length distribution beats the algorithm

One-dimensional cutting stock is a well-studied problem, and modern solvers get very close to optimal on a given input. That is precisely why the solver is not the constraint.

A nest stops adding parts to a bar when the remaining length is shorter than the shortest part still waiting to be cut. The tail on any bar therefore lies somewhere between zero and the shortest available part length. If remainders are roughly evenly spread across that interval, the expected tail per bar is about half the shortest part length. That gives a usable ceiling:

Best achievable yield ≈ 1 − (shortest part length) / (2 × 6000)

If the shortest part in the pool is 400 mm, the expected tail is around 200 mm per bar and the ceiling sits near 96.7%. If the shortest part is 1200 mm, the expected tail is around 600 mm and the ceiling drops to 90%. No solver beats that, because there is nothing left to cut.

Two consequences follow, and both can be acted on faster than any software setting can be changed.

First, short parts are yield currency. A shop that cuts its earth bars, link bars, droppers and terminal pads on the same line as its main runs will out-yield a shop that batches long runs separately, because the short parts fill the remainders. Sending short work to a different machine or a different day throws away the filler.

Second, section proliferation destroys yield faster than anything else. Every distinct combination of width, thickness and temper is a separate cutting pool. Ten sections means ten independent nests, ten sets of tails, and ten times fewer opportunities for a short part to rescue a remainder. Consolidating a catalogue from fourteen sections to eight usually buys more yield than buying nesting software, and costs nothing but engineering discipline. There is a counter-argument: rationalising upwards means some circuits carry more copper than they need, which costs metal too. Work it both ways using an ampacity calculation before committing.

The remnant library, and the problem of actually finding an offcut

A remnant is only an asset if it can be found, verified and loaded inside a few minutes. Otherwise it is a decorative object leaning against a wall.

The software side is the easy half. SMARTNEST BUSBAR maintains an oddment library alongside the stock list and nests the whole project rather than one drawing at a time, so remnants generated by job A are candidates for job B. The machine side sets the floor: the EMAC-BP-40 and EMAC-BP-60 publish a minimum oddment length of 55 mm, and the IMAC-CENTER machines 70 mm, which is the shortest stub the feed can still clamp and position. Below that it goes to the bin regardless of what the software thinks.

The organisational half is where these systems die. Four failures account for most of it.

The offcut is not put back the same shift. If the remnant leaves the machine and does not reach an addressed rack location before the operator goes home, it will not be in the system tomorrow. Making return-to-rack part of the cut cycle, not part of end-of-shift tidying, solves this.

The label falls off, or never existed. Every remnant needs a durable label carrying section, temper, length and a unique ID, printed at the machine at the moment of cutting. Handwritten tape is not adequate; copper offcuts get handled with oily gloves.

The rack is organised by job, not by section. An operator looking for 1100 mm of 100 × 10 half-hard needs a rack addressed by section, with a bin per section and lengths sorted within it. Organising by originating job guarantees nobody will ever find anything.

Two nests claim the same remnant. Reservation and consumption have to be atomic. If the software allocates a remnant to a nest but does not remove it from the available pool until the part is cut, a second nest run twenty minutes later will plan against metal that no longer exists. The operator then improvises, and improvisation always cuts new bar.

There is also a threshold question: below what length is a remnant not worth keeping? Work it from the money. A 100 × 10 mm section is 1000 mm² and copper is about 8.9 g/cm³, so the bar runs 8.9 kg per metre. A 300 mm remnant is 2.67 kg. Using the net loss figure derived below (about US$3,700 per tonne), keeping it is worth roughly US$10. If retrieving, verifying and loading it takes an operator four minutes at a loaded rate of US$25/hour, that costs US$1.70. Comfortably worth it. At 150 mm the value halves and the handling does not, and by the time you are down near the machine minimum the economics have gone. Most shops end up with a threshold between 300 and 500 mm and a written ageing rule: anything unused after six months gets sold as scrap rather than occupying rack space forever.

Aggregation windows: yield against lead time

Nesting yield rises with the size of the pool nested across. Nesting one drawing produces the single-length table above. Nesting a whole switchboard starts filling the tails with short parts. Nesting a week of released work across all projects, yield climbs further.

The cost is lead time and work in progress. Holding a job for four days waiting for company in the nest adds four days to that job's cycle, and cut parts are held for jobs that will not be assembled for another fortnight. A sortation problem is also created: parts from twenty jobs come off the machine interleaved and someone has to kit them, which needs bench space, labelling discipline and time that nobody costed.

A workable compromise looks like this:

• Aggregate by section and temper within a rolling window of three to five working days. That is usually enough to collect a useful spread of lengths without pushing the schedule.
• Allow hot jobs to break the window and be cut on their own, knowingly at worse yield. Price that decision so the planner sees it: at the numbers below, a bar's worth of tail is roughly US$25 to US$30 of net loss, which is cheap insurance on a late shipment and expensive as a habit.
• Kit at the machine, not later. Parts get labelled and binned by job as they come off, or the sortation cost eats the yield gain.

On a machine with high positioning speed and a large tool magazine, such as the IMAC-CENTER 80 with 24 tool stations and ±0.05 mm positioning, mixed nests across many jobs cost very little in setup, which pushes the sensible window wider. On a line that needs a tooling change between part families, the window has to respect the setup, and that constraint belongs in the nest, not in the operator's head.

The annual copper number

Here are the assumptions. They are stated so they can be replaced with others, and every figure below scales linearly with them.

Assumption 1 (market, verifiable): LME cash copper traded at record levels around US$14,500 per tonne in mid-August 2026. Copper is volatile; use today's number.

Assumption 2 (assumption): conversion premium for bare Cu-ETP rectangular bar in common sections, delivered, of US$2,000 per tonne. Delivered bar therefore US$16,500 per tonne. The premium depends on section, order size and region.

Assumption 3 (assumption, within a typical range): clean, sorted, single-grade copper offcuts sell to a merchant at around 88% of exchange price, so US$12,760 per tonne. Merchant payout on top-grade copper scrap typically runs somewhere between 85% and 95% of exchange, varying with region, volume and how well the material is sorted.

Assumption 4 (assumption): the shop ships 180 tonnes of finished copper busbar a year.

Net loss per tonne scrapped is the difference between what was paid and what is recovered: 16,500 − 12,760 = US$3,740 per tonne.

Now run the two cases against a fixed output of 180 t of good parts.

The delta is US$46,750 a year, which is 12.5 tonnes of copper no longer bought at US$3,740 of net loss each. As a share of net material spend it is 1.5%, which sounds unimpressive until compared to what a nesting licence costs.

Two things this figure deliberately excludes. It excludes the freight, goods-in handling and working capital on 12.5 tonnes no longer bought and stored. And it excludes the cost of a mis-cut part, which scraps a part length rather than a tail length and is a first-pass-yield problem rather than a nesting problem.

The sensitivity that matters

Run the same model with copper at US$9,000 per tonne. Delivered bar becomes US$11,000, scrap credit becomes US$7,920, and net loss per tonne falls to US$3,080. The annual saving becomes US$38,500.

Copper fell 38% and the saving fell 18%. The reason is that most of what is lost when binning an offcut is the conversion premium, which the scrap merchant does not pay for, and that premium moves far less than the metal price. The idea that scrap only matters when copper is expensive is contradicted by this arithmetic.

Tonnage scales straight through: 90 t/year of finished output gives about US$23,000, 360 t/year about US$94,000.

What is actually achievable

A shop cutting to drawing with no remnant strategy, no aggregation and no measured bar lengths typically scraps around a tenth of the bar it buys. That is the honest starting point and it matches the table at the top: single-length nesting on 6 m stock averages somewhere in the high eighties to low nineties, and then rejects, mis-cuts and unrecorded offcuts take the rest.

The landing point after fixing it depends almost entirely on job mix, and the spread is wide:

• Long, uniform busway runs with two or three part lengths and one section: 2% to 3% is realistic, because the lengths can be chosen to divide the bar.
• Switchboard work with a broad length distribution and a handful of sections: 4% to 6% is a fair target.
• High-mix jobbing with many sections, small quantities and short lead times: below 6% is hard work, and the aggregation window is usually what stops it.

A single number quoted without asking about part length distribution and section count is a guess. The question of what the shortest routinely-cut part is is key, and if the reason it matters is unknown, the arithmetic above explains it.

The order of work, to achieve yield without the argument: measure incoming bars, cut the lead crop deliberately rather than by habit, put the short parts in the same pool as the long ones, count how many sections are really needed, and only then worry about which solver is running.