Torque is a derived quantity. A torque sensor measures strain in a shaft and converts it into newton-metres using a calibration factor, and that factor is only as good as the calibration that produced it. The uncertainty statement behind that factor is what ISO/IEC 17025 asks a calibration laboratory to document, and it is the first thing to read on a certificate. The classic method — a known mass on a known lever arm — is simple in principle and full of small errors in practice, most of them in the arm length.
The deadweight method, done properly
A lever arm of length L is fixed to the sensor shaft, a mass m is hung at the end, and the applied torque is m × g × L. For a 0.5 m arm and a 20 kg mass, the torque is about 98 N.m. The uncertainties that matter:
| Source | Typical size | How to control it |
| Arm length | 0.1 mm on 500 mm is 0.02 percent | Measure from the shaft axis to the load line, not to the arm end |
| Mass value | 0.01 percent with a calibrated mass | Use calibrated masses, not shop weights |
| Gravity | 0.05 percent across a country | Use local g, not 9.80665, unless the error is negligible |
| Arm horizontality | 1 degree is 0.015 percent | Set with a spirit level and keep the load line vertical |
| Bearing friction | Can be 0.2 percent or more | Load and unload; friction shows as hysteresis |
Adding the terms in quadrature (root sum square) puts a well-run deadweight calibration at roughly 0.05 to 0.1 percent. That is better than the sensor. The practical floor of 0.5 percent quoted in most datasheets is not a calibration limit; it is the combined effect of cross-talk, creep and temperature on the sensor in service.
Cross-talk is the reason a torque reading drifts
A torque transducer is designed to respond to torsion and to ignore everything else. In practice it responds a little to bending moments, axial force and even to the position of the load on the shaft. Cross-talk is quoted as a percentage of full scale and is typically 0.5 to 2 percent on industrial sensors.
Why it matters: a test rig that applies torque through a belt drive also applies a radial load. The sensor reads torque plus a fraction of the radial load, and the fraction depends on belt tension, which changes as the belt warms. The result is a reading that wanders by 0.5 percent over a shift with no change in the actual torque.
- Mount the sensor so that the drive imposes no radial load, or use a double-bearing arrangement that carries the radial load outside the sensing element.
- Couple with a flexible element that accommodates misalignment but transmits torque cleanly. A rigid coupling transmits every misalignment as a bending moment.
- Check cross-talk by applying a known radial load with no torque and recording the indicated torque. If it is more than the datasheet figure, the mounting is wrong.
Creep and the minute nobody waits
Apply a torque and hold it. The reading rises quickly, then continues to creep for several minutes. The datasheet creep figure is typically 0.05 percent of full scale over 30 minutes. On a 100 N.m sensor that is 0.05 N.m, which is small until the measurement is a friction torque of 1 N.m.
For static measurements, wait for the reading to settle within the resolution needed, and record the time at which it was taken. For dynamic measurements, the creep is short compared with the event and does not matter.
Temperature and the zero
Torque sensors are temperature compensated, but the zero drifts. A sensor that reads 0.00 N.m in the morning reads 0.03 N.m after the rig has run for two hours. Two habits remove the problem:
- Zero the sensor in the mounted condition, at working temperature, with no load applied.
- Re-check the zero after the first hour of operation, and at the end of the session. A zero that has moved by more than 0.1 percent of full scale indicates a mounting problem rather than a sensor problem.
What a calibration certificate should contain
Five items, and a certificate without them cannot be used to defend a measurement: the applied torque values and the indicated values, the uncertainty at each point, the temperature during calibration, the mounting arrangement, and the direction of loading. The last one matters because clockwise and anticlockwise calibrations differ, and a sensor calibrated in one direction only should not be used to measure the other.
The torque sensor range lists rated capacity, cross-talk and creep for each body size; the product index links the couplings and readouts used in a calibration rig.
