A torque sensor converts torsional mechanical input into a proportional electrical output, with two principal topologies — reaction (static) and rotary (dynamic) — covering nearly every industrial use case [S1][S2]. Selection reduces to four numeric decisions: nominal capacity with margin, accuracy class (% FS), rotational speed limit (RPM), and output type (mV/V, 4–20 mA, USB, RS-232, RS-485, or wireless) [S1][S2].
Reaction torque sensors measure stationary torque on a non-rotating member; rotary torque sensors measure dynamic torque on a spinning shaft and require slip rings, rotary transformers, or wireless telemetry to get the signal off the shaft [S1][S2]. Mistaking one for the other is the single most common spec error on rotating driveline applications.
Reaction vs Rotary: Topology Decision
Reaction torque sensors are flange- or shaft-mounted transducers that measure the reaction force on a stationary housing, making them the correct choice for nut-runner auditing, bolt-joint research, valve-actuator testing, and any application where the shaft does not rotate more than a fraction of a turn [S1][S2]. They have no rotational speed limit in the rotary sense, no slip-ring wear, and typically cost 30–60% less than an equivalent-capacity rotary unit.
Rotary torque sensors must rotate in-line with the shaft and use either slip rings, rotary transformers, or wireless telemetry (battery + 2.4 GHz / 5 GHz RF modules are now common) to transmit the signal out of the spinning reference frame [S1][S2]. They are specified for motor test stands, turbine auditing, pump and gearbox efficiency testing, and torque feedback control loops on servodrives. The mechanical envelope must accommodate the maximum RPM of the driveline plus a service-factor derating; overspeed destroys the strain-gauge bridge.
Capacity, Overload, and Speed Margin
Capacity selection follows the same rule used for load cells: rated torque should be ≥1.25× the worst-case sustained operating torque, with intermittent peaks kept below the published overload rating (commonly 150% to 200% of capacity, depending on model) [S2]. For shock-loaded applications such as fastener rundown or clutch engagement, peak torque transients must be checked against the sensor's ultimate-overload spec, not the rated capacity.
RPM limits for rotary units are set by the bearing package, the slip-ring or telemetry electronics, and the dynamic balance grade. Standard industrial units are typically rated to 3,000–6,000 RPM; high-speed laboratory units reach 15,000–25,000 RPM; specialist turbo-machinery sensors are specified above that [S2]. If the duty cycle involves frequent direction reversal or high-cycle fatigue, derate capacity by an additional 10–20% and confirm fatigue-life curves with the vendor.
Accuracy Class, Repeatability, and Calibration
Industrial torque sensors are commonly specified at 0.1% FS to 0.5% FS nonlinearity, with combined error (nonlinearity + hysteresis + non-repeatability) usually quoted between 0.1% and 1.0% FS depending on price tier [S2]. Strain-gauge units based on metal-foil gauges — the most common construction — dominate the 0.1% class; piezoelectric units excel at high-bandwidth dynamic measurements but drift on static signals.
Calibration must be NIST-traceable (or equivalent national institute) and should be re-verified annually for production test stands, every 6–12 months for laboratory use, and after any event that may have mechanically overloaded the sensor [S2]. For audit-grade applications — fastener torque auditing on safety-critical joints, calibration of torque wrenches, or torque-wrench tester verification — pair the sensor with a torque wrench tester traceable to the same standard to keep the measurement chain consistent.
Signal Output, Bandwidth, and Integration
Native output is typically mV/V from a Wheatstone-bridge strain gauge, which requires an external amplifier or indicator before reaching a PLC or DAQ [S1][S2]. Integrated electronics packages offer 4–20 mA (with HART on higher-end units), ±5 V or ±10 V analog, USB, RS-232, RS-485, CAN, and Ethernet; wireless rotary units increasingly use 2.4 GHz digital protocols with battery life of 8–40 hours per charge depending on sample rate [S2].
Bandwidth is set by the natural frequency of the shaft, the sample rate of the telemetry, and any digital filtering in the indicator. For control-loop closure on servo-driven machines, total system bandwidth (sensor + telemetry + amplifier + controller input) should be at least 5× the highest frequency of interest in the torque signature; otherwise phase lag will destabilise the loop. For acceptance tests on a flow sensor calibration rig or pump-efficiency map, lower bandwidth is acceptable but DC accuracy becomes critical.
Environmental, Material, and Mounting Constraints
Ingress protection, operating temperature, and shaft material must all be specified against the actual installation. Standard industrial torque sensors are rated 0 °C to +70 °C with IP65 shaft seals; harsh-environment and washdown versions reach IP67 or IP68 with stainless-steel shaft options [S2]. For subsea, chemical, or hygienic applications, confirm the wetted materials and elastomer compatibility against the process fluid and cleaning chemistries.
Shaft-to-shaft alignment is the most overlooked field failure mode: even 0.1 mm of parallel misalignment will skew the reading, accelerate bearing wear, and may exceed the sensor's side-load rating. Use flexible couplings, check the published maximum moment and side-load ratings, and verify that the driveline's critical speed is well above the operating RPM band. For hydraulic power unit test stands, the same alignment discipline applies on both the input motor coupling and the output pump coupling.
When Not to Use a Reaction or Rotary Torque Sensor
Do not specify a reaction sensor for any application where the shaft must rotate continuously — the bearings are not designed for sustained rotation and will fail within hours. Do not specify a rotary sensor on a sub-RPM application where the reaction variant would be cheaper and more accurate; rotary units pay a price and accuracy penalty for the slip-ring or wireless telemetry that a stationary sensor does not need. [S2]
Do not specify a strain-gauge torque sensor for very high-bandwidth dynamic measurements (kHz-range torque ripple on electric drives) where a piezoelectric torque sensor is the correct technology — strain-gauge bandwidth is set by the shaft's first torsional mode and is typically a few hundred Hz to low kHz. Do not pair a high-accuracy laboratory sensor with an actuator whose own torque-control loop is only ±2% — the sensor cost is wasted, and a capacitive sensor-based or lower-grade strain unit will deliver equivalent system accuracy at a fraction of the price.
Selection Checklist and Sourcing Signals
Before issuing a PO, confirm seven items on the datasheet: (1) capacity ≥ 1.25× peak operating torque; (2) accuracy class (% FS nonlinearity + hysteresis) matched to the test requirement; (3) RPM rating with margin; (4) output type compatible with the DAQ or PLC; (5) overload rating in % of capacity; (6) IP rating and operating-temperature envelope; (7) calibration certificate with NIST-traceable (or equivalent) accreditation and stated uncertainty [S1][S2].
For a related comparison of selection criteria on a different transducer type, see the displacement sensor reference page.