Presence sensors — is the part there?

Detecting a part sounds trivial until you meet the part. Aluminium reads at roughly a third the distance of steel on an inductive sensor. A clear PET bottle is invisible to a diffuse photoelectric. A shiny machined face throws a retroreflective beam straight back and makes the sensor think it is looking at its own reflector. A dusty room defeats optics; a wet room defeats capacitive.

So the choice is never "which sensor is best" but "which failure mode can this machine live with". The four families below cover almost every industrial detection job, and each one fails in a way you can predict before you buy it.

An inductive proximity sensor runs an LC oscillator at its face. Metal entering the field induces eddy currents, damping the oscillator, and the sensor switches. No contact, no moving parts, no light path to keep clean — which is why it is on nearly every machine built.

The first catch is the reduction factor. The rated sensing distance Sn on the datasheet is for mild steel (Fe360). Other metals read less: stainless roughly 0.7×, brass roughly 0.5×, aluminium roughly 0.4×, copper roughly 0.3×. Buy for the metal you have, not the number on the box. Factor-1 sensors exist and read all metals at full range, and they cost more for exactly that reason.

The second catch is flush versus non-flush. A flush (shielded) sensor can be buried level with a metal bracket; a non-flush (unshielded) one senses sideways too and must have a clear zone around it — typically 3× the diameter of free space, and its own diameter of clearance from a facing metal surface. Mount a non-flush sensor flush in a steel block and it triggers permanently on the block.

Through-beam is the strongest mode by a wide margin: emitter one side, receiver the other, the longest range and the best tolerance of dirty air. It costs two housings and two cable runs, and it needs the target to break the line.

Retroreflective puts emitter and receiver in one housing and a prismatic reflector opposite. Half the wiring, still long range. The trap is a shiny target: a polished can lid can return the beam like the reflector does, so the sensor never sees it. Polarised retroreflective models solve this — the prism rotates the light 90°, and the receiver only accepts rotated light, so a plain shiny surface no longer counts as a reflector.

Diffuse (proximity mode) has no reflector at all and reads the light bounced back off the target itself. Convenient, and the most temperamental: a black rubber part returns a fraction of the light a white carton does, so range collapses with target colour. Background suppression (BGS) models fix the other half of the problem by triangulating distance rather than intensity, which lets them ignore a conveyor bed 20 mm behind the part.

A capacitive sensor detects a change in the dielectric in front of its face, so it will pick up plastic, wood, powder, water — and, importantly, product inside a non-metallic tank wall. That is its unique job: point-level detection of pellets or liquid through a 3–5 mm plastic wall, without a hole in the vessel.

It is also the least stable of the four. Humidity, condensation and product film all shift the trigger point, which is why capacitive sensors carry a sensitivity potentiometer and why people spend their lives adjusting them. If build-up is likely, prefer another technology or budget for a wipe-down schedule.

Ultrasonic sensors time an echo, so they are colour-blind and see clear objects perfectly — the two cases that break optics. Their weaknesses are foam (absorbs the pulse), strong temperature gradients (bend the speed of sound) and angled surfaces (deflect the echo away). Keep the target face within about ±3° of perpendicular and stay outside the blind zone right in front of the transducer.

Normally open closes on detection; normally closed opens on detection. For plain sequence detection the choice is arbitrary. For anything protective or interlocking, use normally closed, because a cut cable and a lost sensor then look identical to a detected fault — the machine stops.

Beyond that, three-wire DC sensors follow the same PNP/NPN rules as chapter 1, and two-wire DC sensors deserve a mention of their own: they always pass a small leakage current (typically 0.5–2 mA) even when off. On a high-impedance input that leakage can hold the input at a level the card reads as on. A bleeder resistor across the input fixes it, but a three-wire sensor avoids the argument entirely.

Most presence-sensor callouts are mechanical, not electrical. A bracket flexing 0.5 mm under load, a sensor slowly walking out of a nut because nobody used a lock washer, weld spatter cooking an optical face, a chip pile bridging an inductive face — these are the everyday causes.

Design for them: stainless brackets stiff enough that a person leaning on the guard does not move them, sensors set back out of the impact line, air purge or a recessed pocket in a spatter zone, and mutual interference spacing between sensors of the same type (typically 2× the diameter for flush inductives, more for ultrasonics, which will happily hear each other).