Vision sensor vs photoelectric sensor for label and cap inspection

Photoelectric cells pass a bottle when the label sits where it sat at set-up, and trip when it does not. On a water line running 3,000 five-litre bottles per hour, three cells — label presence, label position, cap presence — drift with the room light and fire on gloss, so the output filter grows until it blocks real faults too. This guide gives you the decision rule between cells and a vision sensor, the field-of-view and exposure arithmetic for one camera doing all three checks in one image, the lighting that makes the result immune to the room, and the point at which a plain cell is still the right part.

Why do photoelectric cells false-trigger on labels?

A diffuse cell judges one light beam against one threshold, so when the label sits a few millimetres high or glass gloss crosses the beam, the cell trips on the edge instead of the label — the standard patch is an input filter measured in tens to hundreds of milliseconds.

That threshold moves with the room and with the part. Gloss off a wet bottle shoulder, a label lifted a few millimetres at the seam, a slightly translucent wall — each pushes the return signal across the trip point, and the cell reports a fault on a good bottle or passes a bad one. Three checks mean three cells, each with its own threshold and its own drift.

The usual countermeasure is an input filter or immunity time on the sensor or the PLC input; industrial sensors offer these from a few milliseconds up to seconds. The filter works only while edge glitches stay shorter than it and real bottles stay longer. At 3,000 bottles per hour the pitch is 1.2 s, so a 120 ms filter — 10% of the pitch — costs nothing. On the faster line that pushed you off cells in the first place, 45,000 bottles per hour is 45,000 ÷ 3,600 s = 12.5 bottles per second, an 80 ms pitch. A 50–100 ms filter now eats most of the window and still passes a shifted label whose reflection glitch is brief. No re-aiming fixes it, because the cell has no image: it cannot tell gloss from label, so tuning cannot make it better.

What does a vision sensor add for label and cap checks?

A vision sensor replaces three cells with one camera that checks label presence, label position to about ±1.5 mm and cap presence in a single image, with each check defined as a software zone you can move without touching any hardware.

The sensor takes one picture per bottle and runs software tools inside zones you draw on it: a contrast or blob tool answers presence, and an edge-pair tool measures how far the label edge sits from its taught position, in pixels converted to millimetres through the pixel scale of 49 µm per pixel that the layout below works out. Three products with three label colors live in three recipes; a color sensor reads red, blue and green artwork with the same geometry tools, because an edge tool follows the label boundary, not its color. Selecting a recipe is a binary code on two digital inputs, and moving a check window 5 mm is a drag of the mouse rather than a bracket.

Where packaging checks stay simple — a registration mark, a flap — cells can still carry them; the sensor-versus-camera trade for print mark detection on packaging walks that line. When the label also feeds a servo alignment stage, bottle label orientation vision for servo alignment lines shows the same zone-and-position logic extended to rotation. For a whole-line view of camera-based label stations, the guide on selecting vision systems for bottle label inspection covers the station options end to end.

Is one vision sensor fast enough at 3,000 bottles per hour?

Yes: 3,000 bottles per hour is one bottle every 1.2 s, and budgeting 200 ms for the whole onboard decision still leaves about 1.0 s of margin before the next bottle arrives.

Divide it out: 3,000 ÷ 3,600 s = 0.83 bottles per second, so a bottle arrives every 1.2 s. Spend it like this: trigger and strobed exposure about 1 ms, onboard processing of the presence, position and cap checks budgeted at 200 ms worst case, result output 10 ms. Total about 211 ms, leaving 989 ms of the cycle idle.

The processing runs on the sensor itself — the category exists for inspection on the device, with no PC — so nothing large crosses the wire; results are a few bytes on Fast Ethernet. The budget scales, too: at 9,000 bottles per hour, a 400 ms pitch, it still fits with 190 ms to spare. Past about 18,000 bottles per hour, a 200 ms pitch, this class of sensor no longer fits and the answer is a faster smart camera with strobing — not a return to filtered cells.

The trigger-to-exposure side of that budget, including how external pulse width and frame rate interact, is worked through in camera hardware trigger frame rate with external pulses.

Which camera and lens cover the label and cap in one shot?

For a 120 × 100 mm view that takes in the label panel and the cap edge together, a 5 MP sensor in the 1/1.45" format with a 12 mm M12 lens works at a 170 mm working distance and resolves 49 µm per pixel.

Three lines size the whole layout, where WD is the working distance and f the focal length:

The geometry of the window looks like this:

The common 5 MP sensor in the 1/1.45" format carries 2448 × 2048 pixels at a 3.45 µm pitch, so the array measures 2448 × 3.45 µm = 8.45 mm wide by 2048 × 3.45 µm = 7.07 mm tall. Work the numbers:

• Field of view: 170 mm × 8.45 mm ÷ 12 mm = 119.7 mm wide; 170 mm × 7.07 mm ÷ 12 mm = 100.2 mm tall — call it 120 × 100 mm.
• Pixel scale: 119.7 mm ÷ 2448 pixels = 49 µm per pixel.
• Label position: a ±1.5 mm placement tolerance is ±1.5 mm ÷ 0.049 mm = ±31 pixels of edge shift, against a two-pixel (0.1 mm) edge-tool spread — a 15:1 margin between resolution and tolerance.
• Window: draw the label zone over the lower 80 mm and the cap zone across the top 20 mm of the frame, so one exposure covers all three checks.

The MV-SC3050XC-16M-ATN-H is a 5 MP color vision sensor: 1/1.45" global-shutter sensor, M12 lens mount and Fast Ethernet interface.

Global shutter matters at this station: the bottle keeps moving during the 0.3 ms exposure, and a rolling shutter would shear vertical edges even at that speed.

If the requirement collapses to presence-only, the MV-SCB007EC-12S-WBN ships with a fixed 12 mm lens, manual focus and 1.6 MP color on a 1/2.9" sensor. Its 1/2.9" array is about 4.97 mm wide, so the same 170 mm working distance frames 170 mm × 4.97 mm ÷ 12 mm = 70 mm by about 53 mm — enough window for a cap and a label being present, while the 120 × 100 mm position layout belongs to the 5 MP field.

The two data sheets, side by side:

All 2 share: Sensor type CMOS, global shutter; Pixel size 3.45 μm × 3.45 μm; Mono/color Color; Data interface Fast Ethernet (100 Mbit/s); Humidity 20% RH to 95% RH (no condensation); Client software SCMVS; Certification CE, KC. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Change the window or the stand-off and the same trade re-runs in the focal-length calculator:

Both cameras sit in the smart cameras and vision sensors category; the camera finder filters them by resolution, sensor format and mount.

How much light control does the inspection need?

Shroud the station and strobe the light: at a 167 mm/s belt speed a 0.3 ms exposure holds motion blur under one pixel, and a shroud removes the morning-to-evening ambient swings that unsettle cells and cameras alike.

The belt gives you the exposure budget. Bottles 200 mm apart arriving every 1.2 s move at 200 mm ÷ 1.2 s = 167 mm/s, about 10 m/min. Hold blur to one pixel, 49 µm, and the exposure ceiling is 0.049 mm ÷ 167 mm/s = 0.29 ms — set the strobe to 0.3 ms. That pulse occupies 0.3 ms of every 1,200 ms, a 0.025% duty cycle, so mains ripple cannot band the image either: mains-driven lighting beats at 100 Hz on a 50 Hz supply, a 10 ms period, and a 0.3 ms exposure samples 3% of it. The flicker mechanisms and their fixes are laid out in camera image flickers: causes and fixes for mains beat and triggers.

Inside a matte-black shroud with baffles at the bottle openings, room light never reaches the sensor — the same discipline as strobed lighting ambient rejection, where you measure the residual ambient and improve it until the image stops moving. A red LED light matched with a bandpass filter that blocks ambient light strips most of what slips past the baffles.

Lock the light aim once tuned: the classic silent failure is an operator twisting a bar light that glares at them, after which the recipe fails on the next shift. Strobing also needs a programmable controller, not a wall supply — the reasons are set out in why your vision light needs a controller.

The LS-CTL-0264 is a 2-channel, 400 W digital controller with a network option and constant-voltage 24 V output. With strobe width and current set in software, the exposure you tune on the bench ships inside the recipe for each product. Pick channel count and power from the light controllers range, or start at the lighting step of the configurator from your light's load.

When does a photoelectric cell still win?

A cell wins when the only question is there-or-not: a cap on a fixed bottle, a tray flag, or the trigger duty itself — a diffuse-reflective sensor with a settable 10–350 mm range and an NPN output then does the job in a fraction of the space and setup time.

Position is the requirement cells cannot meet: no photoelectric cell measures how far a label slid, so label position forces the camera. Keep a cell in the station anyway — demoted to trigger. Mount a diffuse sensor just upstream of the window so its pulse marks each bottle, give it a short input filter of 5–20 ms (illustrative values that suit a 1.2 s pitch) to swallow gloss glitches, and wire it PNP or NPN to match the camera input; the sink-versus-source decision is spelled out in photoelectric sensor to camera trigger wiring: PNP or NPN.

The MV-PE5502 is a diffuse-reflective sensor with a red LED, a settable 10–350 mm sensing distance and an NPN output. In the trigger role its red spot sits on the bottle shoulder well before the camera window, and the vision sensor's own handshake takes over from there. On a cap-only line with no position check, the same part pointed down at the cap does the whole inspection. Browse the photoelectric sensors range, or set the trigger up at the configurator's trigger step.

How do the results reach the PLC?

Two digital outputs — pass and fail — driven once per bottle are enough for a reject gate, and a short TCP message per bottle adds the label-position value when you want drift data in the PLC.

Wire the station as: trigger in from the cell, two recipe-select inputs, pass and fail out to the reject. Sequence each bottle as trigger → strobe and exposure → onboard decision → output, all inside the 1.2 s window; the full handshake chain is worked through in machine vision trigger timing: sequence camera, light and reject gate. The reject gate needs its own timing: at a 1.2 s pitch a fixed delay is fine, while shorter pitches push you to track bottles between trigger and gate, as compared in vision system reject delay timer: fixed delay or bottle tracking.

When you want the numbers rather than just the bit, a TCP message per bottle — recipe, pass/fail, label offset in millimetres — lets the PLC log drift and alarm before a false accept escapes; the message pattern is built in smart camera to PLC over TCP: trigger and pass/fail. Use the drift value while it is small: a steadily growing offset points at the labeler, not at the vision station.

What do you check before you order?

Order against your worst bottle, not your best: sample all three labels plus a shifted label and a capless bottle, measure the true belt pitch, and confirm the shroud fits the line before you commit to a model.

• Sample set: good bottles of all three labels, one label shifted ±5 mm, one capless, one without a label. The station must reject the last three — the worst bottle defines the tools; the golden sample only sets them.
• Measured pitch: time 30 bottles. A nameplate 3,000 bottles per hour sometimes runs at 3,300, which is 3,600 s ÷ 3,300 = 1.09 s per bottle and trims 110 ms off your budget.
• Field of view at worst case: label panel and cap edge inside the 120 × 100 mm window with the bottle leaning both ways; re-check the 170 mm stand-off on the real guard bracket, not on paper.
• Shroud and service: matte-black interior, baffles at both bottle openings, light and camera on one rigid bracket with the aim locked, and a door you can actually open.
• Trigger placement: mount the cell far enough upstream that its pulse arrives while the bottle is fully inside the window — 100 mm at 167 mm/s gives 0.6 s, against a 211 ms budget.
• Trigger wiring: PNP or NPN to match the camera input, filtered 5–20 ms as above.
• Power budget: add up the light's draw and check it against the controller channel rating before ordering either part.
• Feasibility: run the three checks on your own bottles before committing; the vision sensor selection guide for label inspection on conveyor systems lists the shortlisting questions to ask.

So which sensor do you buy?

Buy the vision sensor: one 5 MP color unit with a 12 mm lens at 170 mm working distance, a shroud, a strobed red light on a 400 W-class controller, and a diffuse cell as trigger — that is the whole station for all three checks at 3,000 bottles per hour.

Three decisions follow from the numbers:

• All three checks, three label colors, up to about 18,000 bottles per hour: one color vision sensor in the 5 MP layout above — 12 mm lens, 170 mm stand-off, 120 × 100 mm window — in a shroud with a strobed red light, a controller to drive the strobe, and one diffuse cell as trigger. On the 3,000 bottles per hour line that leaves 990 ms of the 1,200 ms cycle unused.
• Presence-only checks, minimal station: one diffuse cell per check, filtered, accepting that a shifted label still reads as present — the failure mode that started this.
• The fast line at 45,000 bottles per hour, an 80 ms pitch: neither three filtered cells nor a basic vision sensor fit the window; step up to a faster smart camera with strobing and re-run the budget at 80 ms — presence still fits, position checks get tight.

Whichever case is yours, light control comes first: the shroud and the 0.3 ms strobe decide whether the checks survive the room, at any line speed. Send the bottle drawings through the contact page and we will check the window, the pitch budget and the light plan against them; more line-side reading sits on the applications topic page.