Industrial IoT sensors: what sensor does a machine need?

Frequently asked questions
Which sensor detects whether a machine is running or a part is present?
It depends on the available signal. If the PLC or equipment already exposes running/stopped status, reading it is enough. If not, a current sensor, presence detector, limit switch or optical barrier can be assessed. The choice depends on the variable you actually need and the decision you will make with that data.
What is the difference between monitoring and alerting?
Monitoring observes and records status or trend. Alerting turns a deviation into a notification with an owner and an expected action. If the data does not change a decision, it is not a useful alert yet. Alert design, channels and escalation are covered in [industrial IoT alerts](/en/blog/industrial-iot-alerts).
In this article
The question “what sensor do I need?” often appears before anyone describes which decision the sensor should improve. That order can lead to buying hardware that produces data without solving an operational need.
A machine can measure temperature, vibration, current, pressure, flow, level, position, presence, cycles or states. The right variable depends on what you need to know, the quality of the data already available and the action someone will take afterward.
Industrial IoT sensors: types and protocols
There is no universal sensor. The variable, environment, and decision define the measurement and how it is transmitted.
| Variable | Typical sensors | What to review |
|---|---|---|
| Temperature | RTD, thermocouple, infrared | Range, contact, environment, and calibration |
| Vibration | Accelerometer, velocity sensor | Mounting, frequency, range, and diagnostics |
| Current | Current transformer, clamp | Range, power, and electrical safety |
| Level | Radar, ultrasonic, float, pressure | Material, foam, tank, and maintenance |
| Flow | Magnetic, mass, differential | Fluid, diameter, range, and calibration |
| Pressure | Pressure transmitter, differential | Range, fluid, overpressure, and purging |
| Presence / cycle | Inductive, photoelectric, counter | Distance, material, dirt, and bounce |
Typical connectivity options are 4–20 mA, Modbus, OPC UA, MQTT, and LoRaWAN. The choice depends on distance, available network, criticality, data volume, and who will maintain the installation.
Start with the decision, not the catalog
Before choosing a sensor, complete a concrete sentence:
I need to know ___ so I can ___ when ___ happens.
For example:
- I need to know whether the machine is running so we can intervene when it remains stopped too long.
- I need to know the temperature to prove that a process stayed within a range.
- I need to compare consumption and output to identify deviations between shifts.
- I need to record cycles to understand real availability and capacity.
The sentence helps distinguish a useful measurement from a collection of data with no owner or response.
Check which information already exists
Many machines already expose data through a PLC, controller, drive, meter, HMI or electrical output. The signal may not reach a management system, but that does not necessarily mean a new sensor is needed.
Before instrumenting, review:
- which controller or electronics the equipment uses;
- which variables it shows locally;
- whether electrical or technical documentation exists;
- which interfaces or outputs are available;
- whether the signal can be read without altering the control logic;
- whether its frequency and accuracy are enough for the decision.
Reusing a signal can simplify installation. It may also show that the missing piece is an interface, gateway or acquisition stage rather than a sensor.
If a sensor is needed, define the complete variable
“Measure temperature” or “need vibration” is not enough. Selection should consider:
- measurement range;
- accuracy and resolution;
- sampling frequency;
- mounting and power;
- dust, humidity, temperature and vibration in the environment;
- distance to the acquisition point;
- calibration and maintenance;
- measurement criticality;
- what happens if the sensor fails or loses communication.
Two sensors measuring the same variable can behave very differently in a plant. That is why universal compatibility with machines, PLCs or protocols should not be claimed without reviewing the actual case.
From measurement to usable data
The sensor is only one part of the chain. A monitoring architecture may include:
- Source: new sensor, existing signal, PLC, controller or meter.
- Acquisition: interface, gateway, datalogger or other equipment that receives and organizes the signal.
- Connectivity: local network, Ethernet, Wi‑Fi, cellular, radio or another option suited to the site.
- Persistence: local storage, buffer, database or later synchronization.
- Visualization: status, history, trends, reports or a dashboard with permissions.
- Action: alert, ticket, intervention, maintenance order or integration with another system.
If the last step is undefined, the project can end with a dashboard nobody checks. Our page on industrial IoT, sensors and monitoring explains how to move through these layers without instrumenting the whole plant on day one.
What if the internet connection fails?
The answer depends on the criticality of the data and process. Some applications need to keep capturing information when the external connection is interrupted. In those cases, local temporary storage, retries and later synchronization may be worth assessing.
Not every variable needs to be transmitted every second. An event, periodic summary or appropriately timed sample may be enough. Connectivity should be designed around location, power, data volume, latency and maintenance.
An older machine can be a good first pilot
Imagine a machine that works correctly but whose output is recorded in a spreadsheet. Before buying sensors, we would check whether a cycle, running/stopped or state signal can be read safely.
If the signal does not exist, we could assess an external measurement of presence, current or another variable that helps infer the state. Then we would define what the team needs to see: perhaps current status, downtime duration, units per shift and an alert when the stop exceeds a threshold.
The pilot should answer whether the data is reliable, communication is sufficient and a decision changes. If one machine does not produce a useful action, multiplying sensors across ten assets will not solve the problem.
Monitoring is not automatically predictive maintenance
Recording vibration, temperature or hours of use can help observe conditions. It does not turn the system into predictive maintenance by itself.
Prediction requires enough history, data quality, a variable related to the failure mode and a method that can be evaluated. In many cases, a first condition-based maintenance stage — with thresholds, trends and owners — is more honest and useful than promising prediction without evidence.
How to choose the next step
A first conversation should make five things clear:
- which problem or decision motivates the measurement;
- which asset and variable are involved;
- which data exists and which installation constraints apply;
- what response is expected when an event occurs;
- how the pilot will be validated before scaling.
When data must enter an ERP, management system or maintenance workflow, technology integration becomes part of the design. If it is still unclear which technology fits, technology and operations diagnosis helps structure the decision before buying.
If you want to review a concrete monitoring and alerting option, you can also see the early warning system. The final selection should start with the operation, the variable and the expected response.