Sensors in manufacturing convert physical conditions, temperature, pressure, position, vibration, flow, and images, into electrical signals a controller or monitoring system can read. They are the foundation of both automation and monitoring: every control decision a PLC makes and every OEE number a plant reports starts life as a sensor reading.
It is easy to talk about smart factories, analytics, and AI as if they were the interesting part. But none of it exists without sensors. A dashboard is just a pretty way of showing what some sensor measured, and an algorithm is only as good as the readings feeding it. This guide walks the main sensor types, what each measures, how they get their data to a controller, and how to choose and deploy them so the readings are worth trusting.
What is a sensor in manufacturing?
A sensor is a device that detects a physical quantity and turns it into a signal a machine can use. It sits at level 0 of the Purdue model the point where the physical world meets the control system. A temperature sensor turns heat into a voltage; a proximity sensor turns "something is here" into an on/off signal; a pressure transmitter turns force per area into a standardized current. In every case the job is the same: translate physics into numbers.
The counterpart to a sensor is an actuator, which turns a signal back into physical action, a valve opening, a motor starting. Sensors sense, actuators act, and the PLC decides between them. Everything above the floor, from SCADA screens to analytics, is ultimately reading and reacting to what the sensors report.
Two words worth separating are sensor and transmitter. The sensing element produces a small, raw signal, a few millivolts from a thermocouple, a tiny resistance change from a strain gauge. A transmitter conditions and standardizes that raw signal into something a controller can read reliably over distance. In many field devices the two are packaged together, which is why people use the terms loosely, but the distinction matters when you are diagnosing why a reading looks noisy or wrong.
What are the main types of sensors?
Manufacturing runs on a handful of sensor families, each measuring a different physical quantity. Here is what each one measures and where it shows up.
| Sensor | Measures | Typical use |
|---|---|---|
| Temperature | Heat (thermocouple, RTD, thermistor) | Ovens, motors, process control |
| Pressure | Force per unit area | Hydraulics, steam, pneumatics |
| Proximity | Presence/position, no contact | Part detection, counting, end-of-travel |
| Vibration | Acceleration/oscillation | Bearing and rotating-equipment health |
| Flow | Rate/volume of liquid or gas | Water, chemicals, fuel, coolant |
| Level | Height of material in a vessel | Tanks, hoppers, silos |
| Vision | Images (via camera + processing) | Inspection, reading codes, guidance |
| Current/power | Electrical draw | Load, run-state, energy monitoring |
A few notes that matter in practice. Temperature has three common technologies: thermocouples (wide range, rugged), RTDs like the PT100 (accurate, stable), and thermistors (sensitive over a narrow range). Proximity sensors come in flavors for the target, inductive for metal, capacitive for almost anything, photoelectric and ultrasonic for distance. And vibration sensors, usually piezoelectric accelerometers, are the backbone of predictive maintenance because a bearing starts vibrating differently long before it fails.
How do sensors send their data?
Through a small set of standard signals, and knowing them explains most of what you will see wired into a panel. The oldest and most common is the 4-20 mA analog loop: the sensor represents its reading as a current between 4 and 20 milliamps. It is popular because current resists electrical noise over long cable runs, and because "live zero" at 4 mA means a reading of 0 mA signals a broken wire rather than a real zero. Voltage signals (0-10 V) do the same job over shorter distances.
On top of analog, digital protocols carry far more. HART layers a digital signal over the 4-20 mA loop so a device can report diagnostics alongside its primary reading. IO-Link (standardized as IEC 61131-9) is a point-to-point digital link to a single sensor, carrying the value plus device identity, parameters, and health over an ordinary three-wire cable up to about 20 m. And fieldbuses like Profibus and Profinet network many devices together. The trend is unmistakable: from a single analog number toward rich digital data that includes the sensor's own condition.
How do sensors feed monitoring and predictive maintenance?
By streaming a continuous record of the machine's condition that software can watch for patterns. A single reading tells you the state right now; a stream of readings tells you the trend, and trends are what warn you before a failure. A bearing's vibration signature drifting upward, a motor's current climbing week over week, an oven cycling wider than it used to, none of these trip an alarm on any single reading, but all of them are visible in the history.
This is why sensors are the foundation of machine monitoring and predictive maintenance rather than just control. Control uses the instantaneous value; monitoring uses the pattern over time. The same vibration accelerometer that a controller might use for a shutdown interlock becomes, when its history is captured and analyzed, an early-warning system. The value multiplies when the reading is captured as real-time data instead of a number someone jots down once a shift.
There is a catch that trips up a lot of monitoring projects: a trend is only meaningful if the readings behind it are consistent. If a sensor drifts out of calibration, or if it is sampled once an hour when the event you care about lasts seconds, the pattern you think you are watching is partly noise. Good monitoring is as much about disciplined, well-calibrated, adequately sampled sensing as it is about the algorithm on top. The cleverest analytics cannot recover information the sensor never captured in the first place.
How do you choose and deploy sensors?
You choose from the measurement backward, start with what you need to know, not with a sensor you saw in a catalog. A disciplined selection avoids the two classic mistakes: buying precision you do not need, and mounting a good sensor where it reads garbage.
- Define the measurement. What quantity, over what range, to what accuracy, in what environment? This narrows the type and technology before anything else.
- Match the technology to the conditions. Metal target means inductive; wet or dusty means a rugged temperature probe; long cable run means 4-20 mA over voltage.
- Get the mounting right. A vibration sensor on the wrong spot or a thermocouple with poor thermal contact reads confidently and wrongly. Placement is half the accuracy.
- Choose the signal for the destination. Analog for a simple loop, IO-Link or fieldbus when you also want device identity and health data.
- Plan calibration. Sensors drift. Decide up front how and how often each critical sensor is verified, or the readings quietly become fiction.
- Connect it to something that uses it. A reading that stops at a local display is wasted. Route it where it drives control, monitoring, or analytics.
By the numbers
The reason sensors from different makers interoperate is a stack of shared standards. The 4-20 mA current loop has been the analog workhorse for decades, HART layers digital diagnostics over it, and IO-Link (IEC 61131-9) defines the point-to-point digital link now common on modern sensors. National measurement bodies underpin the accuracy itself: the NIST Sensor Science Division maintains the physical measurement standards that industrial sensors are ultimately traceable to. Standards are what let you trust that a reading means the same thing on every line.
Where Harmony fits: sensors generate the signal, but the signal only becomes valuable when it reaches a decision. Harmony connects directly to PLCs and sensors, captures their data in real time, computes true OEE from those source signals rather than estimates, and joins them with orders, quality, and downtime so a reading turns into context and action, with the control layer untouched. No rip-and-replace (see the connected systems module), and a real deployment is described in how CLS unified its floor.
Where do sensors fit in the bigger picture?
Sensors are the ground floor everything else is built on. Their signals travel up over networks like Profinet and Profibus through the layers described in SCADA vs MES vs ERP into the IIoT systems that turn readings into insight. The identifiers that RFID reads are just another kind of sensor event on the same timeline. Get the sensing right, the right type, well mounted, calibrated, and connected, and every layer above it inherits data it can trust. Get it wrong, and no amount of analytics will save you, because the whole tower is only as honest as the reading at the bottom.