Blog / Connectivity
Connectivity · Guide

Part Counting Sensors: What to Use, Where to Mount

OL By OEE Lab Editorial |Updated 2 November 2026 |7 min read
Short answer

Use an inductive proximity sensor for metal parts or machine strokes, and a photoelectric sensor for most packages and mixed materials. Mount one where output is single file and a second after the reject point, so you get total and good counts, and prove both against a hand count.

Start from the part, not the catalog

OEE needs two numbers from the floor: how many parts came out, and how many were good. Total count feeds performance, and good count feeds quality.

Whether a sensor can count your part depends on its material, shape, speed and spacing. Those four facts choose the sensor for you.

SensorBest at countingMain limit
Inductive proximityMetal parts, or a metal cam on a moving shaftMetal only, with a short sensing range
Photoelectric, through-beamOpaque items on a conveyor, over longer gapsNeeds an emitter and a receiver, one on each side
Photoelectric, retro-reflectiveCartons and bottles, with wiring on one side onlyShiny or clear items need a suitable model
Photoelectric, diffuseItems passing close to the sensor, with no reflectorColor and surface finish affect the range
CapacitiveNon-metal items, and some contents through a thin non-metal wallMoisture and buildup can affect it
UltrasonicClear or shiny items, and some dusty placesOften slower than optical sensors
A starting guide. Check range, speed and enclosure rating on the data sheet for your exact case.

Other options fit special cases. A light grid can catch small parts falling from a press or moulding machine, and a machine's own cycle signal can stand in for a sensor when each cycle makes a fixed number of parts.

A camera can also count, which helps when items do not travel in single file. See computer vision for OEE.

Count the cycle or count the part?

Counting cycles is often the easy path. A proximity sensor on a press ram or a cam gives one pulse per stroke, with no contact with the product.

But a cycle is not a part. Empty strokes count as parts, and a multi-cavity tool makes several parts per pulse.

  • Cycle count: quick to fit, immune to product shape, but blind to empty cycles.
  • Part count: measures real output, but depends on stable spacing and clean optics.

Example: a 4-cavity mould gives 4 parts per pulse, as long as all 4 cavities are running. If one is blocked, the pulse count overstates output until someone updates the multiplier.

Where to mount them

Mount points matter more than brand. This layout works on most single machines and short lines.

MachineTotalCheckGoodRejectsto scrap bin
Fit any two of the three points. The third follows: good = total minus rejects.
  • Total count: just after the machine you are measuring, before anything can be removed.
  • Good count: after the last automatic reject, so that it counts only what moves on.
  • Reject count: at the chute or bin, when a second conveyor point is hard to reach.
  • On a line: count at the constraint, and again at the end of the line. See line OEE vs machine OEE for how to use those numbers.

Parts removed by hand after the last sensor are invisible. Add a simple entry for them, or the quality rate will look better than it is, as what counts as a good count explains.

Practical mounting rules

  • Find single file. Items side by side count as one, so mount where a guide or a speed-up belt has spread them out.
  • Create a gap if you need one. A short faster belt opens gaps between touching packages. For round bottles that touch, a beam at neck height can often see the gap that the bodies hide.
  • Stay away from accumulation. If items stop in the beam, one long block can read as a single item.
  • Fix the bracket. Vibration moves a sensor, and it then drifts off target. Use a rigid mount and a guard against knocks.
  • Shield it from light and dirt. Sunlight, glare and dust cause both missed and false pulses.
  • Mind the wiring. Most industrial sensors of this kind use 24 V DC with a PNP or NPN switching output, which must match your input card.
  • Work safely. Fit sensors with the machine locked out, keep every guard in place, and leave connections inside a cabinet to a qualified electrician.

A failed or dirty sensor can look exactly like a stopped line. Plan for that with the checks in sensor fault stops.

Check the timing before you buy

A sensor pulse is only useful if the input can see it. Both the on time and the off time must be longer than the input's filter and read time.

Belt speed = 0.5 m/s = 500 mm/s
Part 70 mm + gap 30 mm = 100 mm pitch
Rate = 500 / 100 = 5 parts/s = 300 parts/min
Beam blocked = 70 / 500 = 140 ms
Beam clear = 30 / 500 = 60 ms
Hypothetical conveyor. The shorter of the two times, here the 60 ms gap, sets the limit for the input.

In this example, a PLC program that reads an ordinary input only every 100 milliseconds could miss the 60 ms gap and merge two parts. A fast counter input, or a shorter read cycle, avoids this.

Also check the sensor's own switching frequency on the data sheet, which must comfortably exceed your rate. Here the rate is only 5 pulses a second, so most sensors will cope.

Stop double counts

A part that wobbles in the beam can create two pulses. So can a label flap, a gap in an odd-shaped part or a sliding item that bounces back.

The cure is a lockout: ignore new pulses for a set time after each one. The time must be long enough to cover the wobble, and shorter than the fastest time between parts.

Example: on the conveyor above, a part arrives every 200 milliseconds and blocks the beam for 140. A 150 ms lockout outlasts a bounce as the part leaves, and it ends before the next part arrives.

Prove it with a hand count

Do this before the counts reach any report. It takes about an hour, and it is an easy check to skip.

Example: a team compares a sensor against a hand count for a batch of 1,000 parts, three times, and changes one thing each time.

First mount: 960 counted4.0% low
After re-aiming: 988 counted1.2% low
After adding lockout: 1,002 counted0.2% high
A hypothetical test against 1,000 hand-counted parts. Bars show the error, on a scale where a full bar is 5%.

Agree your own tolerance, and write it down. Repeat the test on each new product shape, because a sensor tuned for one carton may miss the next.

Turning counts into OEE

Example: a machine ran for 400 minutes with an ideal rate of 50 parts a minute. The sensor shows 17,200 total, and 16,900 good.

  • Performance = 17,200 / (400 x 50) = 86.0%
  • Quality = 16,900 / 17,200 = 98.3%

Run time comes from a separate state signal, and the full method is in how to calculate OEE performance and how to calculate OEE quality. For a machine with no readable controller, pair the counter with a clamp-on current sensor.

See it live in Fabrico

Counts help most once they reach a live OEE number. Fabrico is a light MES for OEE that takes counts, speed and stops from PLCs or external IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and shows OEE per line and shift. See how Fabrico captures counts.

What people ask next

What is the best sensor for counting clear bottles?

Standard beams can pass through clear items, so use a photoelectric model made for clear objects, or an ultrasonic sensor. Test it on your real bottle at line speed. Glare and bottle shape can affect the result.

Should I count at the machine or at the end of the line?

Count at both when you can. The machine count shows what it made, and the line-end count shows what survived. Line OEE vs machine OEE explains how they differ.

Why does my sensor count differ from the ERP or the shipped quantity?

The two often measure different points. Rejects, hand removals, rework and sampling all happen between them. Map each point where a part can leave, then reconcile for one batch, using downtime data accuracy as a guide to checking records.

Can I get cycle time from a part counting sensor?

Yes. The time between pulses is the cycle, and pulses per minute give speed. Compare it with your ideal cycle time to see speed loss, and remember that an average can hide slow cycles.

Do I need a separate sensor for rejects?

Not always. If a good count exists after the last reject point, then rejects are total minus good. A separate reject sensor is simply a way of getting the same number where the conveyor is hard to reach.

See OEE per line and shift on your own machines

Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.

Book a Fabrico demo

← More OEE Lab guides · What counts as a good count →

OEE essentials: OEE calculator · What is OEE · Six big losses · Availability · TEEP · Cost of downtime · OEE benchmark · Free OEE tracker · Free OEE software · How to track downtime · How to implement OEE · Light MES vs full MES