Steer Torque Accuracy

I’ve been checking over the accuracy of my steer torque measurements in detail to make sure that the measurement error is low. If my measurements are poor, then the hope of validating a bicycle model with the data may be impossible. I measure the torque in the steer column between two sets of bearings (I have an upper and lower headset). I chose a Futek TFF350 torque sensor and paid extra to have it factory calibrated along with the amplifier (CSG-110). So far I’ve been trusting that the factory calibration gives the “true” values of torque in the steer column. I’ve run a couple of simple checks to see if the torque sensor is giving me reasonable values. I attached a fairly accurate 15-75 in-lb (1.7-8.5 Nm) torque wrench to the top of the steer column. This allows me to apply a “known” reference torque to the steer column. I put known in quotes because the torque wrench output has some error associated with it. At first I simply held the front wheel still while the bicycle was sitting on the ground and applied a series of 1 Nm step increases in torque. There was a slight (~-0.2 Nm) bias on the torque and that may had to due with the fact that the steer tube was not vertical and the torque due to the torque wrench hanging in gravity. I shifted the data up by the bias amount to produce the following graph. 

I applied a stepwise torque to the torque wrench at 1 Nm increments. At torques above 4 or 5, it was increasingly difficult to hold a steady torque value by hand. The bicycle was sitting on the ground normally and I held the wheel in place with my legs and/or feet while applying the torque.

Torques 3 Nm and below look pretty good, but the difference in the torque wrench reading and the sensor is apparent at higher torques. There is some non-linearity in the torque differences. I then mounted the bicycle with it’s headtube vertical and the front wheel off the ground so that the weight of the torque wrench wouldn’t cause any bias. I also applied the torques by affixing a rope to the wrench and locking it in place at specific torque values. We took 3 second samples at 200 hertz at each torque level and plotted the mean resulting torque from the sensor signal versus the applied “known” torque from the wrench.

The torque sensor output plotted as a function of the torque wrench reading. The black line is the 1 to 1 line and the dotted black lines are estimates of the 3 sigma error in the torque wrench readout based on the manufacturer’s provided calibration data.

This graph gives me better trust in the torque sensor factory calibration. The blue dots should fall within the grey dotted lines (estimates of the torque wrench error), but there is still some error on the positive torque. This may be due to the fact that the torque wrench dial has to be adjusted for positive and negative readings and that it has different error characteristics in the two directions. The torque wrench manufacturer must meet a ±3.0 tolerance on the torque wrench output (i.e. the error in torque wrench output to true torque applied must be less than 3 percent of the reading). The grey dotted lines are plotted based on ~1.5% tolerance bounds which more closely matched the error shown the manufacturer’s calibration sheet. Anyhow these two tests are not sufficient enough to truly test the accuracy of the torque sensor. In particular, I question the accuracy and linearity of the torque wrench which I’m using as the standard. I’ll need to be able to apply a known torque more precisely for a better check. There may be other errors too that I haven’t thought of. But, overall I feel pretty good about the accuracy of the sensor.