IoT sensors: the main types and how to choose one
IoT sensors explained: the main sensor types by what they measure, how analog, digital and wireless outputs differ, and how to choose one for a project.
This article covers what an IoT sensor is, the main types grouped by what they measure, the three kinds of output you will meet, and a method for choosing one. It does not cover industrial safety-rated instrumentation, calibration against reference standards, or the electronics of signal conditioning in detail.
Every connected system starts with a sensor, and the choice of sensor decides most of what follows: what the data means, how it gets off the device, what powers it, and how often anyone can trust it. This article maps the territory. It groups IoT sensors by what they measure, explains the three kinds of output you will meet, and gives a way to choose one that holds up whether the project is a greenhouse, a pump room or a warehouse.
What an IoT sensor is
A sensor converts a physical quantity into a signal that electronics can read: a temperature into a resistance, a pressure into a voltage, a movement into a change in capacitance. An IoT sensor is a sensor whose reading leaves the device over a network. That second half is the whole difference. A thermometer on a wall is a sensor; a thermometer that reports to a dashboard, a controller or a phone is an IoT sensor.
The United States National Institute of Standards and Technology’s IoT device guidance (NISTIR 8259) describes the device side of that in more formal terms. In practice the phrase covers three different things that get sold under one name. There is the bare sensing element: a chip or a probe with a few pins that a microcontroller reads. There is the sensor module: the element on a small board with the regulator, the bus interface and sometimes a display, ready to plug into a development board. And there is the sensor node: the element, a microcontroller, a radio and a power source in one enclosure, which sticks to a wall or a machine and reports on its own. Which of the three you are buying changes the price, the work involved and the skills needed, so the first thing to settle is which one the project needs.
The main types of IoT sensor
Grouping sensors by what they measure is more useful than grouping them by technology, because a project starts from a quantity and works back to a part.
Temperature. The most common measurement in any IoT system. Thermistors and resistance thermometers change resistance with temperature and need an analog input; thermocouples generate a small voltage and suit high temperatures; digital sensors put the element and the conversion on one chip and hand over a number. For rooms, enclosures and equipment casings a digital sensor is the easy answer (the Bosch BME280 and the Sensirion SHT3x family are the usual examples, each with a public datasheet), and the comparison of the DHT22, SHT31 and BME280 covers the three most people reach for.
Humidity. Almost always measured with a capacitive element whose capacitance changes with the water content of the air, usually on the same chip as a temperature sensor because the reading depends on temperature. Relative humidity is the number reported; absolute humidity and dew point are calculated from it.
Pressure and level. Barometric sensors measure air pressure for weather and altitude. Gauge and differential pressure sensors measure the pressure in a pipe or across a filter, which is how air handling systems know a filter is clogged. Level in a tank is measured either by pressure at the bottom, by an ultrasonic or radar sensor looking down from the top, or by a float switch that only reports full or empty.
Motion, vibration and position. Accelerometers measure acceleration along one or more axes, which serves both for detecting movement (a door, a parcel, a fall) and, sampled fast enough, for measuring vibration on a machine. Gyroscopes add rotation; magnetometers add heading; together they form an inertial measurement unit. Vibration monitoring on motors and pumps is its own discipline, covered in the predictive maintenance article.
Presence and occupancy. Passive infrared sensors detect the heat of a moving body and are the standard for lighting and security. Microwave and millimeter-wave radar sensors detect movement through some materials and can sense a still person breathing, which infrared cannot. Ultrasonic and time-of-flight sensors measure distance to whatever is in front of them. Simple magnetic reed switches and Hall effect sensors report whether a door or window is open.
Light and color. Photodiodes and ambient light sensors report illuminance for daylight harvesting and display dimming; color and spectral sensors go further for horticulture and quality control.
Gas and air quality. Electrochemical cells for specific gases such as carbon monoxide, non-dispersive infrared sensors for carbon dioxide, metal-oxide sensors for volatile organic compounds, and optical particle counters for particulate matter. This group has the widest gap between a cheap part and an accurate one, and the cheap parts often report an index rather than a concentration.
Current, voltage and power. Current clamps around a supply cable and shunt resistors in series with a load measure how much electricity a machine draws, which is a surprisingly good proxy for what it is doing. Dedicated energy-metering chips do the arithmetic for power and consumption.
Flow. Turbine and paddle-wheel meters for liquids, thermal mass sensors for gases, ultrasonic clamp-on meters that need no cut in the pipe. Flow sensing is one place where the mechanical part matters as much as the electronics.
Sound. Microphones for noise levels and for detecting specific events such as breaking glass or an alarm; acoustic sensors for leak detection in pipes.
Location. GNSS receivers for outdoor position; indoors, position is inferred from radio signal strength or timing between beacons rather than from a single sensor.
Most projects use two or three of these together, and the interesting information is often in the combination: temperature and humidity give dew point, current and vibration together say more about a motor than either alone.
Analog, digital and wireless sensor outputs
The second axis is how the reading leaves the sensor, and it decides what has to sit next to it.
Analog. The sensor produces a voltage or a current proportional to the quantity. A microcontroller reads it with an analog-to-digital converter, and the reading is only as good as that converter, the wiring and the reference voltage. The industrial standard is the four to twenty milliamp current loop: four milliamps means the bottom of the range, twenty the top, and zero means a broken wire, which is why it has survived for decades. Analog is simple, cheap and tolerant of long cable runs; the cost is that every step of the conversion is yours to get right.
Digital. The sensor converts internally and sends a number over a bus. In small projects that bus is usually I2C or SPI, both of which can carry several sensors on the same few pins; in industrial settings it is a fieldbus or an industrial Ethernet protocol. Digital sensors remove the conversion problem, often include calibration data, and can report status as well as value. The microcontroller comparison covers which boards read which buses.
Wireless. The sensor is a node with its own radio and sends readings as messages. The radio decides range, power and data rate: Zigbee and Thread (both on the IEEE 802.15.4 radio) and Bluetooth Low Energy for short range in buildings, LoRaWAN for long range and low data, cellular for anything that moves or sits far from your own infrastructure. The IoT protocols article lays those out layer by layer. Wireless is the easiest to install and the hardest to power, which is why battery life dominates node design. At the far end, ambient IoT tags drop the battery altogether and report only when they have harvested enough energy.
| Output | What sits next to the sensor | Strengths | Costs |
|---|---|---|---|
| Analog (voltage, 4 to 20 mA) | An analog input on a controller or gateway | Simple, cheap, long cable runs, decades of practice | Conversion accuracy is yours; no status information |
| Digital bus (I2C, SPI, fieldbus) | A microcontroller or controller with that bus | Calibrated numbers, status codes, many sensors per bus | Short cable runs for I2C and SPI; needs a reader nearby |
| Wireless node | Nothing; a gateway or coordinator somewhere in range | Installs anywhere, no wiring | Power budget, radio planning, a network to run |
How to choose an IoT sensor
Start from the decision the reading will support, and work backwards. The method below has six questions, and each one rules out a set of parts.
- What decision does the reading feed? A fan that switches on above a threshold needs a repeatable sensor, not an accurate one. A compliance log needs traceable accuracy. A trend for predictive maintenance needs stability over months more than precision on any one day. Write the decision down before opening a catalog.
- What range and resolution does that decision need? A cold store needs a narrow range read finely; an outdoor weather node needs a wide range read coarsely. Datasheets specify both, and a sensor built for one is usually poor at the other.
- How fast does the quantity change, and how fast must the sensor follow? Response time is the specification people forget. A humidity sensor that takes a minute to settle cannot catch a door opening for ten seconds. Sampling interval follows from this and sets the power budget.
- What is the environment? Condensation, dust, vibration, chemicals, sunlight and temperature extremes each kill particular sensor types. An enclosure rating, a conformal coating or a sensor designed for the environment costs less than replacing a failed one on a roof.
- How will the reading leave? Choose the output from the section above based on what is already there: a controller with spare analog inputs, a microcontroller with an I2C bus, or nothing at all, which points to a wireless node.
- What will power it? Mains if there is a socket; otherwise the sensor’s own current draw and its response time set how long a battery lasts, as the article on powering sensors works through.
Two habits improve every choice. Read the datasheet for the exact part number, because manufacturers revise parts and second-hand numbers go stale. And buy one before buying fifty: a week of readings from a single unit in the real location will tell you more than any specification.
Where IoT sensors are used
The same handful of sensor types recurs across very different settings, which is why learning the types once pays off.
In a home, temperature, humidity, presence, door contacts and light sensors drive heating, lighting and security. They are almost always wireless nodes on a low-power mesh, chosen for battery life and ease of pairing, and the smart home hub covers how they connect.
In a small business, the same sensors appear with two additions: temperature loggers in fridges and cold rooms for food safety, and current sensors on equipment to know when it runs. Wireless nodes still dominate, often with a single gateway.
In a plant, vibration and temperature on rotating machinery, pressure and flow in process lines, current on drives, and level in tanks. Analog and fieldbus sensors wired to controllers remain the norm, with wireless nodes added for retrofit condition monitoring. The industrial IoT hub describes how those readings reach a dashboard.
In agriculture, soil moisture, air temperature and humidity, leaf wetness and tank level, spread over distances where only a long-range radio makes sense.
In logistics, position, temperature and shock inside vehicles and containers, on cellular links because the asset moves.
Common mistakes with IoT sensors
Measuring the enclosure. A temperature sensor mounted next to a warm microcontroller board, or inside a sealed box, reports the box. Put the element on a lead, in the air or on the surface you actually care about.
Buying accuracy you cannot use. A sensor specified to a fraction of a degree, read through a poor analog input, over a long cable, on a board with a noisy regulator, delivers nothing like its datasheet figure. The measurement chain is only as good as its weakest part.
Ignoring response time. Slow sensors sampled fast produce smooth, wrong data. Match sampling to response time and to how fast the quantity really changes.
Forgetting drift. Gas sensors, humidity elements and some pressure sensors drift over months. Plan to check them against a reference or a second unit, and expect to replace the cheap ones.
Skipping the placement test. Wireless nodes are cheap enough that the temptation is to fit them all at once. Fit one, watch a week of readings and signal strength, then fit the rest.
Choosing the sensor before the question. The most common mistake of all. A sensor is the answer to a question about a quantity; a project that starts from a part number usually ends up measuring something nobody needed.
The sensor is the smallest part of an IoT system and the one everything else depends on. Group by quantity, choose the output from what is already in place, budget the power honestly, and test one in place before buying more. Get those four right and the rest of the system has something worth carrying.
Written by the IoT Crest editorial team. Facts in this article follow the editorial policy: only well-established behavior is stated and no figures are invented.