Dynamometer testing a robot joint: setup, measurements and errors

By Tolerance Labs engineering team · Updated 18 October 2026

A datasheet tells you what the maker measured under the maker’s conditions. A dynamometer bench measures the same joint under yours, and it is the only reliable way to get an efficiency map, a thermal limit and a backdrive torque for your own design.

The bench: absorber, transducer and couplings

A joint dynamometer has its parts in one line on a rigid base plate: the joint under test, clamped by its housing; a coupling; an in-line torque transducer; a second coupling; and an absorber that loads the joint.

• Absorber motor. A servo motor with its own drive, often through a gearbox that matches the joint’s output speed and torque. It can hold a speed while the joint pushes against it, or apply a torque while the joint holds a speed. Because it can both drive and absorb, it covers all four quadrants: it can backdrive an unpowered joint and load a joint that is regenerating.
• Brake absorber. A hysteresis, magnetic-powder or eddy-current brake turns the joint’s output into heat. Brakes are simpler, but they only absorb, so they cannot backdrive the joint. Eddy-current brakes lose torque at low speed, where most joints work.
• Torque transducer. A rotating strain-gauge shaft with contactless signal transfer is typical. Choose its range near the joint’s peak torque, because accuracy is usually stated as a percentage of full scale. A 0.1% FS transducer rated 200 N·m has an uncertainty of about 0.2 N·m at every load, which is 10% of a 2 N·m friction reading. Some benches use a second, smaller transducer for friction tests. Joint torque sensors explains strain-gauge accuracy, overload and drift.

Couplings should be torsionally stiff and tolerate small misalignment; bellows and disc couplings are common. Placing the transducer between two couplings, on its own bearings, keeps bending loads off its shaft. Align the shafts with a dial indicator or a laser tool for each new setup. Check that the couplings’ rated torque and the transducer’s overload limit both exceed the torque in a fault, when the joint or the absorber runs to its current limit.

What to measure

For Kt, the bench replaces the lever and scale in calibrating a robot joint and adds speed. Hold one current while the absorber turns the output slowly, first with the joint’s torque and then against it. The two readings differ by about twice the friction torque at that load.

Run the torque-speed curve at nominal voltage and at the lowest bus voltage the robot will see, such as a nearly empty battery. Pulse each point for a second or two, so that the winding does not heat and shift Kt.

For the thermal run, mount the joint as it will be mounted in the robot, with the same plate, enclosure and airflow. A common rule is to run until the winding and housing temperatures change by less than about 1 °C in 10 minutes, which can take an hour or more. Thermal management of robot joint actuators explains why.

For backdrive torque, measure the breakaway torque, then the running torque at several slow speeds, cold and warm. Backdrivability in robot joints explains what the numbers mean for control.

Worked example: efficiency at one operating point

A joint runs at a steady operating point on a 48 V bus. The bench records:

• Output torque from the transducer: 20.0 N·m.
• Output speed: 95.5 rpm. In rad/s: 95.5 × 2π ÷ 60 = 95.5 × 0.1047 ≈ 10.0 rad/s.
• Bus voltage at the joint’s input: 48.0 V. Bus current: 5.20 A.
• Mechanical power out: 20.0 × 10.0 = 200 W.
• Electrical power in: 48.0 × 5.20 = 249.6 W.
• Efficiency: 200 ÷ 249.6 ≈ 0.80, or 80%.
• Losses: 249.6 − 200 = 49.6 W, all of which becomes heat in the joint.

If the joint reports its phase current, you can split the losses further. Suppose it reports 7.0 A RMS per phase and the warm phase resistance is 0.20 Ω. Copper loss is 3 × 7.0² × 0.20 = 3 × 49 × 0.20 = 29.4 W. The other 49.6 − 29.4 = 20.2 W is driver loss, iron loss and friction in the reducer and bearings. A backdrive test at the same speed, with the driver off, separates the friction part.

Now check how good the 80% is. If the transducer is accurate to ±0.2 N·m, the torque reading is good to ±1%. Add a speed error of 0.1% and a power-meter error of 0.5%. A rough worst-case sum is 1 + 0.1 + 0.5 = 1.6%, and 1.6% of 80% is about 1.3 percentage points, so the result is 80% ± 1.3. At light load the same ±0.2 N·m is a much larger fraction of the reading, so efficiency at low torque is much less certain. Report each point with its uncertainty.

In the generating quadrant, efficiency is electrical power out divided by mechanical power in. Measure bus current with its sign.

Sources of error

Torque ripple and cogging show up as a periodic signal. Average over whole motor revolutions. A bench resonance can amplify ripple at some speeds; vibration and resonance testing covers how to find one.

Safety

• Guarding. Cover every coupling and shaft with a fixed guard. A coupling or a bolt that fails at speed throws parts. Keep hands and loose clothing out of the guard while anything can turn.
• Interlocks and stop. Wire an emergency stop that removes power from both the joint and the absorber, and interlock the guard so that the bench cannot run with it open.
• Software limits. Set torque and speed limits in both drives below the coupling, transducer and joint ratings, and a watchdog that stops the test if data stops.
• Energy on the bus. In the generating quadrant, the joint under test pushes energy back onto its bus. A bench supply that cannot sink current lets the bus voltage rise until a driver trips or a part fails. Use a supply that can sink current, a battery or a brake resistor; see regenerative braking and bus overvoltage.
• Stored charge. An absorber drive fed from the mains often has a DC bus of several hundred volts, and its capacitors stay charged after power-off. Wait the time given on the drive’s label, then measure, before you touch the terminals.

Checklist

• An absorber that covers the quadrants you need: a motor for backdrive and regeneration, or a brake for loading only.
• A transducer ranged near peak torque, between two stiff, aligned couplings.
• Bench drag measured and subtracted.
• Kt, torque-speed at two bus voltages, and an efficiency map in both quadrants.
• A thermal run in the real mounting, and backdrive torque cold and warm.
• Every result recorded with temperature and uncertainty.
• Guards, interlocks, an emergency stop, a path for returned energy, and a check for stored charge.

Browse robot joint modules and robot motors, and compare their stated torque, speed and efficiency in Robot Parts before you plan your own tests.