Bottle label orientation vision for servo alignment lines

Bottles leave the conveyor randomly spun, and every label must end up facing the same direction before the packer or case packer sees them. This guide designs the whole loop: a camera that measures the label angle, a light that keeps a glossy cylinder readable, a PLC that receives the angle over Ethernet, and a servo or belt stage that turns the bottle by the shortest correction. You will leave with the accuracy you can hold (±1° in one pass, ±0.5° with the loop closed), a lens and strobe worked example, the shortest-turn math for the PLC, and a decision between a single spin station and a multi-stage correction belt.

What angle accuracy can a vision system hold on a bottle label?

Plan on ±1° after one measurement and ±0.5° once a verify camera closes the loop before the final correction; the camera resolves the label edge about forty times finer than that, so the mechanics and the tracking spend the budget, not the pixels.

Put the tolerance in millimetres first. A Ø60 mm bottle has a circumference of π × 60 = 188 mm, so 0.5° of rotation is 188 × 0.5 / 360 = 0.26 mm of surface travel. At the 0.035 mm per pixel scale derived below, a subpixel edge tool holding 0.2 px locates the label edge to 0.007 mm. The line around the camera is where the degrees go:

The last row only counts when you must align the bottle rather than the label. If the target is a label that faces front, rotate until it does, and the label's placement error is irrelevant. If a later station keys on the bottle — a wrap labeler timed to a mold seam — that ±1.9° lands in your budget, so measure the seam or the base gate mark, never the label.

Converting an edge position into an angle is one line of trigonometry:

Here s is the pixel scale in mm/px, R the bottle radius in mm, and x the edge's pixel distance from the bottle centerline. Sine compression destroys resolution near the silhouette edges, so measure label edges inside the central ±60° of the visible arc. The verify camera closes the loop after each turn, so repeatable errors — datum offset, tracking bias — cancel, and the ±0.5° that remains is per-bottle spread.

For scale: cartoning tolerates ±2-3°, retail presentation wants ±1-2°, and print registration wants ±0.5°. This set-up holds the tightest of the three.

Which camera should measure the label angle?

A 5 MP class global-shutter GigE area scan camera is the standard choice; a smart camera that runs the measurement on the device does the same job with no PC, and 12 MP only pays when one camera must also read codes or cover two lanes.

Global shutter is the one non-negotiable spec when you strobe bottles in motion. A strobe freezes the bottle only if every pixel row exposes at the same instant; a rolling shutter turns a single 50 µs flash into bright bands and a sheared label edge. Rolling shutter stays usable where the bottle stops at a station under continuous light.

For a PC-based line, the MV-CU020-90GC is a 2.5 MP camera with a global shutter, GigE and a C-mount (this listing is the color version; filter the finder for mono if you prefer). GigE carries images 100 m on standard cabling to a cabinet PC, and what GigE Vision guarantees for software compatibility saves an integration surprise. To drop the PC, the MV-SC5060XM-08M-WBN is a 6 MP mono smart camera with an 8 mm focal-length lens in an M12 mount and GigE. Its rolling shutter rules out on-the-fly strobing: trigger it where the bottle stops, or run continuous diffuse light on a slow belt. The MV-CH120-11UC brings 12 MP, a global shutter, USB3 and a C-mount to the verify station, where the extra pixels also serve OCR.

Three routes cover most lines:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Browse the area scan cameras and smart cameras categories with global shutter set as a filter. For a wider walk-through of cameras on label lines, read selecting a vision system for bottle label inspection.

How do I light a round, glossy bottle so the label edges show?

Use a diffused ring light with a crossed polarizer, strobed for 50-100 µs, mounted square to the bottle; that kills the specular band that runs down a glossy cylinder and leaves the label edges measurable at subpixel level.

A cylinder reflects any compact source into one bright stripe along its length, and the stripe lands somewhere new on every bottle. The fix is light from many angles at once — a diffuser ring around the lens — plus a polarizer on the light and an analyzer on the lens; how cross polarization removes surface glare shows the geometry, and the same glare drives the missed matches covered in bottle recognition lighting. Cross-polarization costs two to three stops of light, so run the strobe near full current instead of raising camera gain.

When the label is white stock on a white or clear bottle, use its physical edge rather than its print. Label stock stands 0.05-0.15 mm proud of the wall, and a low-angle ring lights that step as a bright line while the smooth wall stays dark; surface lighting versus silhouette lighting decides which effect you get, and which LED colour gives the best contrast fixes the wavelength.

At the 300 mm working distance of the next section, a 100-120 mm diffused ring covers a Ø60-80 mm bottle evenly. Keep the ring square to the bottle axis; a few degrees of tilt moves the glare band. The LS-RNG-0143 is a 100 mm ring light with a diffuser at a 30° angle, 24 V, on a 3-pin flat connector.

The ring lights category filters to diffused and polarized types in strobe-capable 24 V versions; strobed intensity drifting as the LEDs warm is the repeatable-image problem closed loop lighting control solves.

How do I choose the lens, working distance and exposure?

For a Ø60 mm bottle with a 60 mm label band, a 25 mm lens at 300 mm working distance on a 1/1.8-inch-type sensor gives an 86 × 65 mm field of view; a 50 µs strobe holds blur to 25 µm at 0.5 m/s, under one pixel.

Work the numbers in this order:

1. Fix the targets: Ø60 mm bottle, 60 mm tall label band, ±12 mm of lateral wander, sensor 7.2 × 5.4 mm (a 1/1.8-inch type), working distance 300 mm set by the guard rail.
2. Vertical coverage governs: m = 5.4 / 60 = 0.09.
3. Focal length: f = 300 × 0.09 / 1.09 = 24.8 mm — order a 25 mm C-mount lens.
4. Verify: at 300 mm, a 25 mm lens gives m ≈ 25/300 = 0.083, so the field of view is 7.2 / 0.083 = 87 mm wide by 5.4 / 0.083 = 65 mm high. That swallows the bottle plus the wander.
5. Pixel scale: a 5 MP camera puts about 2,500 pixels across the width, so 87 / 2500 = 0.035 mm per pixel, and 0.5° of label travel (0.26 mm) is 7.4 pixels of edge motion.
6. Aperture f/5.6. Depth of field ≈ 2 × CoC × N × (1 + m) / m² = 2 × 0.07 × 5.6 × 157 ≈ 120 mm total, covering the bottle's 60 mm front-to-silhouette depth and the wander. At f/8 the Airy disc grows to about 10.7 µm (at a 0.55 µm wavelength), past two pixels on a 3 µm-pitch sensor — f/5.6 is the ceiling.
7. Exposure: blur = speed × pulse. At 0.5 m/s, a 50 µs pulse gives 500 mm/s × 0.00005 s = 0.025 mm = 0.7 px. At 1 m/s, halve the pulse to 25 µs.

Run your own bottle through the same arithmetic:

How does the measured angle reach the PLC?

The vision side writes one record per bottle — ID, signed angle in tenths of a degree, quality flag — into registers over EtherNet/IP, PROFINET or Modbus TCP, and the PLC applies the shortest-angle correction before the bottle reaches the rotator; budget 60-100 ms from strobe to register.

Smart cameras and vision controllers expose those result registers natively; wiring results, triggers and heartbeats into ladder logic is laid out in vision controller Modbus communication, and the pattern carries to the other two protocols. Keep the record to three fields: bottle ID, signed angle, quality. The shortest-turn math is one rung:

Measured 350° against a 0° target gives ((0 − 350 + 540) mod 360) − 180 = +10°, the short way. Measured 170° gives −170°, also the short way. The result always lands between −180° and +180°, with the sign giving direction.

Latency is harmless when you track by position. Sixty to a hundred milliseconds of exposure, processing and network at 0.2 m/s is 12-20 mm of travel — a 1-2° error on a timer, zero on an encoder count. The full shutter-to-actuator path is budgeted in image based control latency.

If the barcode is that landmark, the reader's rotation handling is covered in barcode image rotation. When quality falls below threshold, write a reject record and let the PLC divert the bottle; the reject-rung patterns are in PLC programming for label detection. Never spin on the previous bottle's angle.

How does the PLC know which bottle to turn, and by how much?

Track every bottle with an encoder wheel rubbing the belt within about 0.5 m of the rotator — 1000 pulses per revolution on a Ø50 mm wheel resolves 0.16 mm of travel, so the 1 mm position error that costs 1.9° of final angle never accumulates.

Read the belt, not the drive shaft: belt stretch under varying load makes shaft-derived positions drift by several millimetres. A Ø50 mm wheel turns 157 mm of belt per revolution, which at 1000 ppr gives the 0.16 mm resolution above. Trigger the camera from the same encoder every fixed pulse count — one bottle pitch, say 150 mm — so every image lands at the same belt position with no trigger jitter.

The rest is bookkeeping. Divide the count by the pitch to get a bottle ID, store the correction against that ID when the result arrives, and fire the spin when the ID's count reaches the rotator position minus the grip-and-ramp lead. Teach the camera-to-rotator offset once with a marked bottle turned by a known 90° command; the verify camera's residual is the offset. Conveyor encoder calibration troubleshooting covers the count resets that corrupt this bookkeeping, and sharing one encoder across stations avoids re-measuring the belt per station.

This is also the safety boundary. The rotator is moving tooling: if the tracked position and the real bottle disagree, the gripper strikes the bottle sideways. Guard the station so a mistrack cannot reach an operator, gate every spin command on a bottle-present sensor that confirms the encoder count, and keep that interlock in the safety controller, independent of the vision path.

Should I use a servo spin station or speed-differential belts?

For near-aligned infeed (±30°) on a slow line, one servo spin station that stops or grips the bottle is simplest; for fully random bottles at line speed, run six to eight speed-differential belt stages of about ±25° each, with a verify camera before the last stage.

A spin station drives the bottle about its axis between a side belt or bottom wheel and rollers that clamp cap and base. A servo at 360°/s closes the worst-case 180° in 0.5 s; with grip, settle and release the cycle is about 1.2 s, so one station sustains roughly 50 bottles per minute. Homing, the registration move and the data hand-off for exactly this station are walked through in configuring servo registration for bottle orientation.

Filled bottles need the servo, not a stepper: fluid slosh shifts the mass, so grip high and low, rotate about the bottle axis, and ramp at a few hundred °/s² so the liquid follows instead of tipping the bottle out of the rollers. Empty bottles index happily on a stepper.

Belt stages never stop the bottle. Narrow belts along one side run at line speed plus or minus a differential; each stage trims about ±25°, and through a differential gearbox a stationary motor runs at line speed while a small speed delta adds the rotation. Eight stages span ±200°, enough for fully random infeed, and the verify camera before the last stage replaces dead reckoning with a fresh measurement.

When is vision the wrong tool for label alignment?

When the bottle has no distinguishable feature at the alignment point — label applied later, plain glass, no mold mark — vision has nothing to measure, and the fix is upstream: a bottle unscrambler, a twist-rail orienter, or aligning to the base gate mark from below.

The common trap is labeling order. If the label goes on after alignment, the camera must key on the mold parting seam or the base gate pip, read from below in dark field — a different lighting problem from a paper label. With none of those features, orient mechanically upstream and spend nothing on cameras.

The second trap is label width. One side camera works only while the label covers at least 180° of the circumference; a narrow front label is out of view half the time. Below 180°, fit two cameras 90° apart, or one camera above a tapered bottle where the label shows from the top.

The third is invented tolerance. Cartoners index on the bottle diameter and tolerate ±2-3°; twist rails reach that from a ±20° infeed with no camera at all. Ask what the next station actually senses before buying vision for exact alignment.

What should I check before ordering parts?

Run this list before you buy: every item is a commissioning failure mode on orientation lines, and each takes an afternoon to test with a sample bottle.

• Photograph 20 bottles of each SKU, wet and dry, under the proposed ring light and polarizer: film thickness and condensation move the glare band.
• Measure the label's circumference coverage. Under 180°, plan the second camera position — bracket and cable route — now, even if you buy one camera first.
• Confirm global shutter wherever you strobe bottles in motion; rolling shutter only at a stop station.
• Order polarizer and analyzer film with the light and test at full strobe current, because cross-polarization costs two to three stops.
• Reserve the encoder wheel mount within 0.5 m of the rotator, reading the belt surface.
• Write the no-read rule into the PLC specification: reject, never spin.
• Reserve the verify camera position and wiring in phase one; it is the part that buys you ±0.5°.
• Rate the camera and light enclosures for the line's washdown; a beverage line is wet, and condensation on the window reads as a defect.

Walk the camera, lens, light and controller steps in order for your bottle:

The camera finder filters by shutter and interface, and requesting a quote with your bottle diameter, line speed and label samples gets the set reviewed before you commit.