AI Vision Rebar Counting for Bar Mills and Bundle Shipping

By Josh Brook on October 23, 2026

ai-vision-rebar-counting-bar-mills-bundle-shipping

Every rebar bundle that leaves a bar mill carries a promise: the count and weight on the tag match what is inside. One missed bar becomes a short count claim at the customer's yard. One extra bar is steel given away for free. At line speed, with small sizes, touching bars and ends that never sit perfectly flush, people and counting screws both slip. AI vision rebar counting checks every layer and every bundle end before the tag is printed. See it on real bar mill images in a short bundle counting demo.

AI VISION · BAR MILL FINISHING END

Count Every Bar, Tag Every Bundle, Ship the Right Number

iFactory Vision Counting & Classification counts bars on the cooling bed transfer and at the bundle end, checks both against the order, and holds any bundle that disagrees. Short counts and over counts stop at the mill, before the tag is printed.

Bundle 24-1187 · 16 mm × 12 mRecount
Line 2 tying station · B500B · heat 51734
Order count120
Bundle end count119
Transfer layer count120 (4 layers × 30)
Likely causeOne recessed end, partly hidden
Tag printOn hold
Image sent to shift lead · crew recount requested
BAR MILL FINISHING END

Where Bars Are Counted Between the Last Stand and the Truck

After the finishing stand, bars are cut to multiple lengths and laid on the cooling bed. From there they move in layers to the cold shear, which cuts them to commercial lengths such as 12 m. The counting station then forms each layer and builds the bundle, which is tied, weighed and tagged before it goes to the bay. Many mills still count with mechanical counting screws that must be changed for each bar size. The count is fixed at one step, but it is checked at none of the later ones. Map your own finishing end with our bar mill team.

1Cooling bedMultiple-length bars cool and move across in steps
2Cold shearLayers are cut to commercial length
3Count and formBars are separated, counted and built into layers
4Bundle and tieLayers stacked, headed up and tied with wire
5Weigh and tagBundle weighed, tag printed and fixed
6Bay and dispatchBundles stored, picked and loaded by order
WHY COUNTING IS HARD

Five Things That Make a Bar Easy to Miss

A bundle of 40 mm bars is easy to count by eye. A bundle of 10 mm bars, packed tight and seen end on, is not. Research on end-face counting, including a Daido Steel technical paper, traces most errors to uneven brightness on the end faces and to bars that sit back from their neighbours and are partly hidden. Bring your hardest product sizes to a counting trial review.

01
Touching and overlapping barsBars in a layer touch or ride over each other, so two can look like one
02
Small diametersAn 8 or 10 mm end face covers only a few pixels unless the camera is close
03
Ends that are not flushBars sit back from the face of the bundle and are partly hidden
04
Changing sizesEach product change brings a new diameter, rib pattern and bundle size
05
End-face conditionShear burrs, scale, paint and glare change how each end looks
BAR BUNDLE COUNTING CAMERA

Two Counts That Check Each Other

A single camera gives one number and no way to know if it is wrong. iFactory uses two independent views. The first counts bars in each layer as they pass the transfer after the cold shear. The second reads the bundle end face after tying. When the two counts agree with the order, the tag prints. When they disagree, the bundle is held and the image goes to the crew. Plan both camera points with our support engineers.

LAYER COUNT AT THE TRANSFER
  • Camera above the transfer chains after the cold shear
  • Bars seen side by side, before they are stacked
  • Best for small sizes and touching bars
  • Gives a count per layer and a total per bundle
BUNDLE END COUNT AFTER TYING
  • Camera facing the bundle end at the tying or weighing station
  • Counts every end face in the finished bundle
  • Catches bars added or lost during forming
  • Image kept with the bundle record as proof of count

A Second Count Before the Tag Is Printed

Most short counts are found by the customer. iFactory finds them at the tying station, where a recount takes a minute and nothing has been invoiced yet. The crew sees the image, the reason and the two counts on one screen.

Finishing end · Shift lead and iFactory
Shift leadWhy is bundle 1187 on hold?
iFactoryTransfer count was 120 and the end count was 119. One bar end sits back and was only partly visible. The image is on your screen.
Shift leadCrew recounted. It is 120.
iFactoryLogged as an end-count miss. Tag released with 120, and the image is added to the weekly review set.
SHORT COUNT CLAIMS AND STEEL GIVEAWAY

Why the Count Matters Under Every Way Rebar Is Sold

Sales terms vary by market and customer. Some mills invoice the weighed bundle, some invoice theoretical weight worked out from count, length and nominal mass per metre, and some buyers order pieces. A common shortcut for nominal mass is d²/162 kg per metre, with d in mm. Under every one of these terms a wrong count costs money. Check how your contracts treat it with our dispatch specialists.

By theoretical weight
The count is the invoiceWeight is count × length × nominal mass per metre. One missing bar is a direct short-weight claim.
By piece count
The tag is the contractSite crews count bars at unloading. A short bundle stops placing and starts a dispute.
By weighed bundle
The count still shows upA wrong count makes the tag disagree with the scale and puts heat and order records in doubt.
Mass tolerance is not count tolerance

ASTM A615 allows the measured mass of a deformed bar to be as low as 94% of nominal, and overweight is not a cause for rejection. That tolerance covers rolling variation per bar. It does not cover a missing or extra bar in the bundle.

REBAR BUNDLE TAGGING

What the Bar Says, What the Tag Says, What the Camera Confirms

Rebar carries rolled-on marks. Under ASTM A615 and A706 the order is producer, size, type of steel (S for A615 carbon steel, W for A706 low-alloy) and grade, shown as a number or as lines. BS 4449 bars carry rolled-on marks for the maker and grade. Tag content is usually set by the customer or a local regulator. Dubai Municipality, for example, requires a tag on every bundle with maker, origin, standard, grade, size, heat number and its own approval mark. Review your tag layout with our support team.

Tag field
Source
What vision checks
If it fails
Bar count
Order and bundle recipe
Layer count and end-face count agree with the target
Hold for recount
Size and length
Rolling schedule
End-face diameter class matches the product on the mill
Hold, check mix
Heat number
Melt shop and billet tracking
Bundle sits in the heat window the mill reports
Flag at heat change
Standard and grade
Order and rolled-on marks
Tag text read back and matched to the order
Reprint tag
Tag present
Tagging station
Tag seen on the bundle end before it leaves
Stop at bay entry
COUNTING ACCURACY VALIDATION

Prove the Count Before You Trust It

Counting accuracy should be measured on your bundles, by your crews, before the system releases tags on its own. The cleanest method is a manual recount trial scored two ways. At bundle level, ask how often the count was exactly right. At bar level, split errors into missed bars and phantom bars, the same idea as a confusion matrix. The numbers below are an illustration of the arithmetic, not a result. Run a trial plan with our validation team.

Bundles recounted200 bundles of 150 bars, so 30,000 bars by hand
Exact bundle counts195 of 200, so 97.5% of bundles exactly right
Missed bars4 bars in 3 bundles were in the bundle but not counted
Phantom bars2 bars in 2 bundles were counted but not there
Bars found29,996 of the 30,000 real bars were counted
Counts that were real29,996 of the 29,998 bars counted were real bars
MANUAL RECOUNT PROTOCOL

A Recount Checklist Your Crews Can Run

A recount is only useful if it is more reliable than the count it checks. Two people counting independently, with marked ends, beats one tired person with a chalk stick. Tick the steps here or print them for the tying station. Ask our support team for a version with your bundle sizes.

01Before the trial
02During each recount
ROUND BAR COUNTING SYSTEM

Same Method for Rounds, Pipe Bundles and Other Long Products

The two-count method is not limited to rebar. Plain rounds count like rebar without the ribs. Pipe and tube bundles are often packed in hexagons, and the camera counts open ends instead of solid faces. Each product family gets its own trained model and its own trial. Discuss your product mix in a product scoping call.

RB
RebarRibbed ends, wide size range, tight round bundles from the bar mill
RD
Plain round barSmooth, bright or black ends, often cut cleaner than rebar
PB
Pipe and tube bundlesOpen ends in hexagonal or square packs, counted ring by ring
SQ
Squares and flatsEdges touch fully, so the model learns corners instead of circles
MILL DISPATCH ACCURACY

A Bundle Record That Answers Any Claim

When a customer reports a short bundle weeks later, the question is simple: what did the mill ship? A bundle record with both counts and the end-face image answers it in minutes. These are the fields iFactory keeps for each bundle. Get a sample record from our support team.

Bundle ID24-1187, barcode on tag
Product16 mm × 12 m, B500B
Heat number51734
Order count120 bars
Layer count120 (4 × 30)
End-face count119, then 120 on recount
Weigh scaleRecorded mass vs theoretical
EvidenceEnd-face image, recount by crew B
COST AND ROI

A Giveaway Worksheet With Your Own Inputs

Over counts are quiet losses, because no customer complains about an extra bar. The worksheet sizes that loss with assumed inputs only. Swap in your own figures. Short count claims, freight for replacement bars and time spent on disputes come on top. Build the full case with our value team.

Bar size and length16 mm × 12 m, assumed
Nominal mass per bar16² ÷ 162 × 12 = 18.96 kg
Bundles per day500, assumed
Bundles with one extra bar2%, assumed, so 10 per day
Steel given away per day10 × 18.96 = 189.6 kg
Steel given away per year189.6 kg × 330 days, assumed = about 62.6 t

Multiply by your own price per tonne. Measure your real over count rate in the validation trial before you use this worksheet for a budget.

TURNKEY AI DEPLOYMENT

Cameras, Counting Models and Edge Server in One Package

iFactory delivers bar counting as a turnkey system. Cameras are fitted at the transfer and the tying station, and a pre-configured NVIDIA AI server ships racked and ready with counting models loaded, so images are processed on site at line speed. You rack it, plug in power and Ethernet, and the AI is live. Scope it for your finishing end with a turnkey quote.

What's Included
  • Pre-configured NVIDIA AI edge server, racked and ready
  • Industrial cameras and lighting for transfer and bundle end
  • Counting models trained on your sizes and bundle shapes
  • PLC link to the counting station and tying machine
  • Hold and release logic for the tag printer
  • Bundle records with images, linked to heat and order
  • 24×7 remote monitoring of models and hardware
Live in 6–12 weeks
Wk 1–4
Survey and Install

Camera points agreed, lighting and server installed, PLC and tag printer signals linked.

Wk 5–8
Train and Validate

Models trained on your sizes, then scored against crew recounts while tags print as usual.

Wk 9–12
Go-Live and Hand Over

Hold and release switched on, crews trained, weekly accuracy review started.

HOW iFACTORY FITS

Vision Counting Built for the Finishing End

iFactory Vision Counting & Classification is designed for scale, glare and fast product changes, where a count must be right before the bundle moves on. Walk through a pilot plan in a short demo.

1
Counts on the cooling bed transfer. Every layer is counted after the cold shear, before stacking.
2
Counts at the bundle end. The tied bundle is counted again from its end face.
3
Tags only good bundles. Short and over counts are held for a recount before the tag prints.
4
Keeps proof for every bundle. Counts and images are stored with heat, order and dispatch.
1000+
Industrial clients running iFactory
99.9%
Platform uptime for continuous mills
6-12 wks
From racked server to live bundle counts
FREQUENTLY ASKED QUESTIONS

What Bar Mill Managers Ask About Rebar Counting

Where should the counting cameras go?
Most mills use two points. One camera looks down on the transfer after the cold shear and counts each layer while bars lie side by side. A second faces the bundle end at the tying or weighing station and counts the finished bundle. Two independent counts make it clear when one of them is wrong.
Can it count small sizes like 8 and 10 mm?
Small sizes are the hardest case, because each end face covers only a few pixels and the bars pack tightly. Camera distance, lens and lighting are chosen for your smallest size, and the layer count on the transfer helps most here because bars are not yet stacked.
What happens when the two counts disagree?
The bundle is held before the tag prints. The crew sees both counts and the end-face image with the suspect area marked, recounts by hand and confirms the number. The tag is then released, and the case is added to the review set used to improve the model.
How do we know the system is accurate enough?
Run a recount trial on your own bundles across all sizes before switching on hold and release. Score it at bundle level, as the share of bundles counted exactly right, and at bar level, as missed bars and phantom bars. Repeat the check after major product or camera changes.
Does it read the rolled-on marks on the bars?
The counting system looks at bar ends, not the rolled-on marks along the bar. Grade and size come from the rolling schedule and order, and the system checks the tag against them. A separate OCR camera can read tag text back before the bundle leaves the station.
Can the same system count pipe bundles?
Yes. Pipe and tube bundles are counted from their open ends, usually in hexagonal or square packs, using a model trained for that product family and checked with its own recount trial. Our support team can help you scope a pipe line.

Ship the Count on the Tag, Every Time

Count each layer on the transfer, count each bundle at the end, and hold anything that disagrees before it is tagged. The full system is delivered turnkey and goes live in 6 to 12 weeks.

Example shift summary · Bar mill line 2
Bundles taggedEach with two counts on record
Bundles heldReleased only after recount
EvidenceEnd-face image per bundle
ReviewWeekly accuracy by bar size
Every bundle traceable by count, heat and order

Share This Story, Choose Your Platform!