IoT Sensor Selection Guide: Choosing the Right Sensor

Buying a sensor is easy; selecting one is not. The market offers thousands of devices that all claim to measure temperature, power, or vibration, and most will do exactly what they claim. The difference between a good selection and a bad one shows up a year later, in the quality of the decisions the data supports - whether chiller staging works, whether submetering matches the utility bill, whether the maintenance team trusts the alarm or ignores it. Selection is really an exercise in matching measurement characteristics to the decisions they feed, and doing that deliberately beats browsing a catalog.

Start from the Decision, Not the Device

Every sensor purchase should trace back to a decision that needs data. Write the decision first:

• “We will reset the supply air temperature setpoint based on zone demand” - which requires sensors that reflect zone load, with response times fast enough to follow occupancy changes.
• “We will allocate energy costs to three tenants” - which requires metering with billing-grade accuracy and interval recording.
• “We will trigger maintenance on the fourth-floor AHU before it fails” - which requires vibration monitoring with the right axis, bandwidth, and alarm thresholds.

The decision defines the required accuracy, response time, range, and reliability. Everything else is negotiation.

The Sensor Classes and What Each One Can Tell You

Temperature Sensing

Temperature is the most common IoT measurement and the easiest to get wrong by over-specifying:

• Thermistors are cheap, sensitive, and accurate over a narrow range - excellent for room comfort sensing, roughly 10 to 40 degrees Celsius.
• RTDs (resistance temperature detectors) trade cost for stability and linearity over a wider range - the choice for process and cold-chain applications where repeatability matters.
• Thermocouples cover extreme ranges at low cost but with lower accuracy and more drift - appropriate for high-temperature industrial monitoring, not comfort control.

The specification that matters most is not peak accuracy but drift rate and behavior in the actual installation range. A thermistor rated at +/-0.5 degree in its sweet spot is more valuable than a thermocouple accurate to +/-1.0 at 600 degrees no one will ever reach.

Humidity Sensing

Capacitive humidity sensors are the common choice - stable, low cost, good enough for comfort monitoring. The pitfalls:

• Range mismatch - a sensor rated 0-100% RH behaves poorly near saturation; condensation destroys many sensors, so cold-chain applications need units designed for repeated exposure.
• Temperature dependence - humidity sensors are temperature-sensitive, and compensation quality varies between manufacturers.
• Drift in dirty environments - sensors in kitchens, laundries, and process areas foul faster than clean-room ratings suggest.

Power and Energy Metering

Power sensing has a hard rule: accuracy class follows the decision.

• Revenue and billing require verified accuracy (typically 0.5s or 1.0 class per IEC 62053) and calibrated current transformers.
• Submetering for allocation and benchmarking can work with slightly lower classes if boundaries are clear and consistent.
• Operational monitoring often needs less absolute accuracy but more granularity - per-circuit CTs to separate loads, and interval data (1-15 minutes) to catch peak demand.

The most common mistake is confusing what the meter measures with what the circuit carries. A panel CT that wraps a feeder carrying two loads cannot separate them, no matter how good the meter is.

Vibration Sensing

Vibration monitoring splits into two uses:

• Overall-level monitoring (detecting gross problems) can use compact accelerometers reporting overall RMS or peak values on a schedule.
• Condition-based diagnostics (bearing fault frequencies, imbalance, misalignment) need higher-bandwidth sampling, multiple axes, and waveform or envelope capture.

The driver is the maintenance model: a team that acts on “this pump is degrading” needs only overall-level sensing; a team diagnosing specific faults needs more instrumentation. Paying for diagnostics you will not perform is wasted budget; buying sensing too coarse produces alarms nobody trusts.

A Comparison Framework

Criterion	Environmental	Power	Vibration
Primary accuracy need	Stability + low drift	Class per billing/use	Repeatability + bandwidth
Response time	Minutes	Seconds to minutes	Milliseconds (waveform)
Power source	Battery-friendly	Mains-fed	Wired or high-capacity battery
Communication	LoRaWAN/Zigbee/BLE	Modbus/CT interface	Modbus/IO-Link/BLE
Key failure mode	Drift, contamination	CT orientation, burden	Mounting, loose coupling

Practical Selection Criteria Beyond the Datasheet

The datasheet answers “what does it do in the lab?” The criteria that predict field performance are different:

• Long-term stability and drift specs - a vendor that publishes drift rates is telling you more than one that only publishes accuracy.
• Calibration traceability - does it ship with a calibration certificate, and can you renew it?
• Power architecture - for battery devices, ask for sleep current, not the battery-life claim. A “5-year battery life” without the reporting interval and payload size is meaningless.
• Communication robustness - does it handle network loss gracefully, buffer data, and resync?
• Interoperability - what protocols does it speak natively? Devices requiring proprietary gateways add fragile dependencies.
• Environmental ratings - IP rating, temperature range, and chemical resistance must match the installation.

Sizing the Fleet and the Budget

Selection includes quantity, and quantity follows decisions:

1. Decide the minimum sensors needed to support the decisions.
2. Price the best-in-class option for that set - the one whose accuracy, drift, and reliability match the decision.
3. Compare against a budget option and quantify what the difference actually costs: replacement visits, drift-based maintenance, lost confidence.
4. Scale only where decisions improve; leave headroom in the platform, not in unnecessary sensors.

For most facilities, fewer, better-specified sensors in the right places outperform more, cheaper sensors in convenient places - at equal or lower total cost once installation and maintenance are counted.

Common Selection Mistakes

• Chasing the accuracy number - buying a +/-0.1 degree sensor for a decision that tolerates +/-1.0, and paying five times more.
• Ignoring mounting and integration cost - a sensor needing a custom bracket, adapter, and proprietary gateway is never cheap.
• Assuming the protocol works end to end - a Modbus sensor on a BACnet network needs a bridge, and bridges are where reliability dies.
• Forgetting the data plan - rich data with no reliable path to the platform is a collection hobby, not a monitoring program.

Conclusion

Sensor selection is a matching problem: match measurement technology, accuracy, power design, and integration path to the decision the data will serve. Done well, it produces a fleet that is smaller than the salesman suggested, more accurate than the budget feared, and trusted by the teams that rely on it. The cheapest sensor is the one you never have to re-buy or re-trust; the most expensive is the one whose readings quietly steer the facility the wrong way.

Integrar IoT’s ecosystem spans environmental, power, and vibration sensors across multiple protocols and partner devices, so selection can start from the decision and end with a platform that already speaks the sensor’s language.