Rotary torque sensors, static reaction torque sensors, and non-rotating strain-gauge units form the three primary buying families, with rated capacity spanning roughly 0.05 N·m to 200,000 N·m depending on construction [S2].
Buyers in 2026 are weighing capacity range, accuracy class (commonly ±0.1% to ±1% FS), rotational speed ceiling, and output protocol (analog mV/V, 4-20 mA, CAN, EtherCAT, USB) before shortlisting a supplier [S2].
Sensor families and what each one actually measures
Rotary torque sensors sit in-line on a rotating shaft and output torque in N·m through an analog or digital channel; the Simscape Torque Sensor (AB) block, introduced in MATLAB release R2026a, models this device as a series element in an angle-based rotational network with two conserving ports (B base, F follower) and a physical signal output t in N·m [S2]. Static reaction sensors mount on a fixed frame and measure the reaction torque on a bearing housing, which means the rotating shaft does not need sliprings — useful above ~10,000 rpm where slip-ring life dominates downtime. Non-rotating beam-type sensors use strain gauges bonded to a calibrated shaft and read torque as a bending or torsional strain; these are common in hand-tool and bench-stand applications where the rotational speed is zero or near-zero [S1].
Reaction-type units typically cap at lower rotational speeds because the body is fixed, but they remove the slipring/brush wear mode and tolerate harsher IP-rated environments. In-line rotary units add slip-ring or telemetry (rotary transformer) electronics, so datasheet maximum speed (commonly 5,000–25,000 rpm) is a hard procurement line, not a soft target. For robotics and cobot joints, a six-axis force-torque sensor is often used instead of a pure torque sensor because the joint loads combine force and moment vectors — see ROS driver packages such as the KTH dumbo force-torque stack for reference integration patterns [S3].
Selection criteria: capacity, accuracy, speed, and output
Four numeric parameters decide 80% of buy/no-buy outcomes. (1) Rated capacity must sit at roughly 1.2× to 2× the maximum expected operating torque, with overload tolerance usually 150% of capacity for static and 120% for rotary. (2) Accuracy class is normally quoted as a percentage of full scale (%FS) or combined non-linearity + hysteresis; industrial cells often land at ±0.1% to ±0.5% FS, while workshop-grade units run ±1% to ±3% FS. (3) Rotational speed, in rpm, is a hard ceiling — sensors with integrated slip-rings are commonly limited to 5,000–15,000 rpm, while telemetry (non-contact) rotary units extend to 20,000 rpm and beyond. (4) Output protocol drives PLC integration cost; analog 4-20 mA or ±10 V is the lowest-friction path for retrofit, while CANopen, EtherCAT, and PROFINET dominate new machinery builds [S2].
For bench-top torque auditing, buyers cross-reference the sensor against a torque wrench tester on the same calibration chain so that wrench readings and inline sensor readings can be traced to a common standard. When the sensor is mounted into a moving stage, the linear guide and crossed-roller guide carrying the reaction frame must be specified for the reaction moment, not just the axial load — a common procurement miss.
Comparison of the three main types against decision criteria
Side-by-side on the four criteria: a rotary in-line torque sensor (e.g. the drivetrain class used for EV and turbine test stands) typically delivers ±0.1% to ±0.5% FS accuracy, 5,000–25,000 rpm speed ceiling, mV/V or EtherCAT output, and the lowest mechanical footprint per N·m of capacity. A static reaction torque sensor sits at ±0.1% to ±0.5% FS, 0 rpm (static frame), 4-20 mA or CAN output, and tolerates the harshest washdown/IP environments. A strain-gauge beam or hex sensor reads ±0.5% to ±2% FS, low rpm, mV/V or USB, and is the cheapest path for hand-tool and bench-stand calibration [S2][S1].
The decision rule: pick rotary in-line when you must measure torque on a live rotating shaft and the speed ceiling is acceptable; pick reaction when the shaft is already bearing-mounted and you want a non-invasive retrofit; pick beam/strain when the test is static or sub-100 rpm and budget is the binding constraint.
Who a torque sensor is for, and who should not buy one
It is for: EV powertrain dynamometer cells (e.g. 50–500 N·m at 0–18,000 rpm), wind turbine gearbox end-of-line stands, fastener torque auditing on a torque wrench tester, e-bike pedal-assist calibration, and robotics joint torque feedback where a six-axis force-torque head is overkill. [S3]
For a process line where the actual deliverable is fluid handling rather than shaft torque, a diaphragm pump selection guide is the more relevant procurement document; for valve-actuator torque on a piped service, the binding spec is usually the ball valve selection guide class rating, not the sensor itself.
Integration, calibration, and failure modes
Calibration chain: most industrial torque sensors are shipped with a factory calibration certificate traceable to a national standard (NIST, NPL, PTB), and on-site verification against a torque wrench tester is the standard audit step. Recalibration interval is typically 12 months for industrial cells and 6 months for audit-grade cells used on fastener lines. Installation pitfalls: misalignment between the sensor coupling and the driven shaft adds bending moment, which most torque sensors cannot reject and which the operator will read as zero-shift or noise; a flexible coupling or a piloted coupling is mandatory above 3,000 rpm. [S2]
Common failure modes: slip-ring brush wear on rotary in-line units (lifetime rated in millions of revolutions, so 24/7 operation at 10,000 rpm reaches end-of-life inside 12–18 months), strain-gauge fatigue on static reaction cells driven beyond their rated moment, and EMC-induced zero-drift on analog-output cells routed next to VFD cables — a documented failure pattern addressed by shielded cable, ferrite, and a grounded reaction frame.
Sourcing and 2026 supplier landscape
Procurement in 2026 is being shaped by three forces: the migration from analog mV/V to digital fieldbus (EtherCAT and CANopen) on new machinery, the continued use of analog 4-20 mA on retrofit and brownfield cells, and the growth of USB-output bench-top cells for lab and field audit. Buyers should request the calibration certificate scope (capacity points, applied load type, torque direction), the overload rating as a multiple of capacity, the IP rating for reaction-frame cells (commonly IP65 to IP67), and the CE/UKCA/UL file number where the cell ships as part of a CE-marked machine. [S2]
Trackable signals over the next two quarters: more reaction-type sensors showing EtherCAT and IO-Link output options on datasheets, more strain-gauge beam units bundling USB-C and on-board shunt-cal resistors, and continued consolidation of slip-ring rotary cells into telemetry-rotary designs above 15,000 rpm. Treat any cell that does not publish a stated %FS accuracy with a defined loading direction as a hand-tool accessory, not a process instrument.