Web defect marking 20 m downstream: encoder or tacho delay

An infrared camera watches a film web for missing adhesive; a marker 20 m downstream must ink exactly that stretch so the cutter removes it. The web runs up to 120 m/min, the speed drifts, and the cabinet already offers a 0–10 V tacho voltage proportional to speed. The decision is whether to turn that voltage into a delay, or to count pulses from the incremental encoder that voltage usually comes from. Count the pulses. What follows runs both error budgets with the arithmetic shown, gives the wiring and logic for the encoder route, and states where each route still wins.

Which camera finds the defect before you mark it?

For transparent adhesive on transparent film, a longwave infrared camera is the detector: the adhesive lane leaves a thermal footprint where visible contrast is zero. A line scan camera takes over only when the defects show optically. Either way the marking arithmetic is worthless if detection cannot localize, so size the detection first.

At a 50 Hz frame rate one frame spans 20 ms, and 20 ms at 2 m/s — the 120 m/min top speed — is 40 mm of web. That 40 mm is the position granularity of the detection stage, it is common to both marking routes, and it is comfortable when the defects worth cutting run hundreds of millimetres. Processing latency is mostly constant; the constant part trims out of the count target, and only the jitter stays in the budget. Sizing the thermal pixel's footprint on the web follows the same reasoning as infrared camera selection by IFOV for electrical inspection, and camera fps and shutter speed for moving web inspection covers the frame-rate side. For the thermal route, start among the infrared cameras; the LS-IRC-0044 is a 640 × 512 LWIR camera with an 8–14 μm band, a 50 Hz frame rate, a GigE interface and a C-mount.

Why does a fixed time delay miss the mark?

Because time only becomes distance through speed: at 120 m/min the 20 m delay is exactly 10.0 s, and each 1% of speed error moves the mark 200 mm — a ±1.25% speed band is all a ±25 cm tolerance buys.

The arithmetic is one line:

Film lines ramp, dwell at splices and jog; a 10% excursion during the 10 s delay is a 2 m miss, and the wrong section gets cut while the defect ships. A fixed, calibrated timer is defensible only on a line that contractually holds speed inside ±1% for the whole delay, which film lines rarely do. On constant-speed conveyors a fixed trigger offset after a sensor is normal practice — camera trigger delay timing for offset inspection sensors works that case through — but on a drifting web it is the first option to delete.

How far off can the 0–10 V tacho route be?

Integrating a 10-bit 0–10 V tacho sampled every 20 ms lands the mark 200–300 mm off at full speed, worst case, and the analog channel's gain error contributes almost all of it. Three terms add up:

• Scale resolution: 1024 codes across 10 V is 9.8 mV per code, or 1.95 mm/s at full scale. Held as a bias for the 10 s delay that is ±20 mm.
• Gain error: an untrimmed analog input commonly sits within ±1% of full scale, which is 20 mm/s here, and over 10 s that is 200 mm. A two-point trim takes the residual toward ±0.5%, or 100 mm.
• Scan latency: the alarm arriving and the marker firing each wait up to one 20 ms scan, which is 40 mm each at full speed. Trimming removes the average 40 mm; the other half remains as jitter.

Re-integrating the speed every scan does handle speed changes inside the delay — a 0.4 m/s step within one scan costs 0.4 × 0.020 = 8 mm — so the age of the sample is not the problem; the scale of the value is. Buying a 12- or 16-bit input shrinks the 20 mm quantization term and leaves the 100–200 mm gain term untouched.

How does counting encoder pulses make the delay speed-independent?

Latch the encoder count when the alarm arrives and fire the marker when it has advanced 20 m worth of pulses: the delay becomes a counted distance, and speed, acceleration and scan time leave the equation. On this class of line the tacho voltage usually comes from a converter fed by the line's incremental encoder, so the pulses already exist and the analog conversion is where the accuracy is spent.

Worked example: 1024 pulses per revolution on a measuring roller of 0.2 m circumference (about 64 mm diameter) gives 1024 ÷ 0.2 = 5,120 pulses per metre. The 20 m target is 20 × 5,120 = 102,400 counts. At full speed the pulse rate is 5,120 × 2 = 10,240 Hz, far inside any counter input, and the quantization is ±1 count, or 0.2 mm. If the line halves speed mid-delay the mark still lands on the same web coordinate, because nothing in the target is time. This is the same counting discipline encoder ticks per pixel for line scan uses to hold image scale true.

Which route holds a ±25 cm tolerance?

Budget the tacho route at ±300 mm untrimmed and ±200 mm calibrated; budget the encoder route at ±80 mm worst case, about ±40 mm of jitter once the fixed offset is trimmed out.

The decisions fall out directly. At ±50 cm or looser, a trimmed tacho integration is acceptable and needs no new hardware. At ±25 cm, use the encoder: the trimmed tacho spends 80% of the budget before the marker adds its own repeatability, while the encoder holds a 3× margin. Below ±10 cm, keep the encoder and move the compare into hardware — a counter card output that fires without waiting for the logic scan. Detection jitter of ±40 mm at 50 Hz is common to every route and trims out where it is constant.

How do I wire and program the encoder route?

One splitter, one counter input, one compare: split the encoder signal so the drive and the counter each see clean edges, scale it to pulses per metre, latch on the alarm, fire on the count.

1. Split the encoder signal through a signal splitter or isolator so the existing drive input and your counter each get a proper load; never parallel a second input straight onto the pair.
2. Match the signal standard — 24 V HTL, 5 V TTL and RS-422 line driver all exist, and the counter input must match. Level adapters are ordinary catalogue parts.
3. Take A and B in quadrature where available, so a direction reversal during a jog or splice dwell decrements the count instead of double-counting; encoder-triggered line scan capture wiring for quadrature inputs shows the same wiring at a camera.
4. Compute the scale at commissioning: pulses per metre = pulses per revolution ÷ circumference, here 1024 ÷ 0.2 = 5,120.
5. Program the logic: on the alarm, snapshot the counter; fire the marker output when the live count minus the snapshot reaches 102,400; hold the output for 1,024 counts.
6. Trim once: run a known defect, measure the mark-to-defect offset on the stopped web, and subtract it from the target. The spread over ten runs is your true accuracy.
7. Tie the encoder supply, the counter and the marker supply to one 0 V reference — floating references are the classic cause of phantom counts, the same failure a common trigger reference diagnoses on trigger lines.

What fires the marker, and for how long?

Fire for a counted window, not a timed one: 1,024 counts at 5,120 pulses per metre is a 0.2 m patch at any speed, whereas 100 ms prints 200 mm at full speed but only 40 mm at a fifth of it. The patch length is part of the cut decision, so lock it to counts.

Marker hardware is an ink-jet head, a spray valve or a label applicator. A small solenoid valve is electrically the same class as a strobed LED load, so a built-in light channel on a vision controller drives it once the channel's current rating is checked; anything bigger gets its own relay and keeps only the signal from the controller. When several defects can be in flight inside 20 m — at full speed, anything within 10 s — the logic queues alarms first-in, first-out and fires them in order, the pattern drive controllers have long used for FIFO defect marking with encoder pulse duration tracking. The MV-SVC2506-16G60-128G provides 4 camera GigE ports, 1 configuration GigE port, 1 communication GigE port and 4 built-in light channels, so the camera link, the line handoff and the marker output share one box; the handoff itself is what machine vision PLC integration on the production line covers. Compare port and channel counts across the vision controllers range for your line.

Automating the mark also takes hands off the line: nobody flags a defect by reaching into a running web, which justifies the project on its own.

What else does the same encoder buy you?

The pulse train also paces acquisition: a line scan camera triggered once per encoder increment images the web at constant samples per millimetre, so every defect inherits a web coordinate you can mark, map and cut. At full speed with 0.2 mm per line the camera needs 2 ÷ 0.0002 = 10,000 lines per second, and 4,096 pixels across a 1,000 mm web give 0.24 mm per pixel — a matched, square sample. Whether the camera or the frame grabber takes the encoder is the choice line scan camera encoder trigger, software or hardware walks through, and the line scan camera line rate calculator does the division for your speed and sample size.

For a detection stage with optical contrast, the LS-LCM-0080 is a 4K GigE line scan camera with a global shutter, an M42 mount and 4096 × 8 effective pixels. The MV-CL042-91GC-V2 is a 4K GigE colour line scan camera with 7 μm pixels, a 29 mm image circle and an isolated version. Browse the line scan cameras range for the web width you image.

Each alarm logged with its count and speed becomes a roll map: the customer sees where every defect sat, and the cutter gets coordinates instead of judgement.

When is the encoder the wrong answer?

When the encoder measures the drive instead of the web: nip slip of 1% on a driven roller silently becomes 200 mm over 20 m, and the count looks self-consistent the whole time. Check what the encoder actually rides before trusting it, in this order.

• Drive-mounted encoder through a gearbox: web slip and stretch break the count-to-distance relation. Move to a measuring wheel riding the web, or to an idler with enough wrap and tension that slip is negligible, then prove it by rolling 10 m of marked web and reading the counter — 5,120 per metre predicts 51,200 counts, and a reading of 51,000 means your scale, and every mark, is off by 0.4%.
• No encoder can be fitted at all: a voltage-to-frequency converter turns the 0–10 V into pulses and the same counter logic runs unchanged. Cycle-time and acceleration immunity survive; the tacho's scale error survives with them, so at ±0.5% trimmed you still carry ±100 mm over 20 m. This is a ±25 cm-class answer, not tighter. Drives make the opposite trade — replacing a damaged pulse encoder with an analog tacho on a DC drive is standard speed-regulation practice — because a drive tolerates what a cut position does not.
• Tolerance below ±10 mm: the compare must fire from hardware — a counter card's comparator output or the camera's position-synchronized output — since a 20 ms logic scan is 40 mm of web at full speed on its own.

Close these out before ordering:

• Encoder counts the web path, not the motor, proven with a rolled 10 m.
• Signal standard matches the counter input; A and B taken for direction.
• Pulse rate at maximum speed — 10,240 Hz here — inside the counter's rating.
• Marker minimum on-time fits the count window at minimum line speed.
• Detection jitter measured once, offset trimmed, spread recorded.
• Tolerance agreed with the customer in millimetres of web, not in seconds.

Pick the detection head in the camera finder, run the system configurator for the line, and bring the tolerance in millimetres to the contact page — the marking route follows from that one number. More line-side builds sit under the applications topic.