5 Prototyping and Bench Testing

While simulation can be helpful for the preliminary design and assessing feasibility, prototyping and bench testing is necessary for verifying actual performance. Simulation excludes unrealized variables and therefore does not have all the necessary parameters to exactly match real-world test cases. Bench tests reveal some of the possible discrepancies between simulations and a fabricated system with manufacturing and assembly tolerances.

The fabricated diametric design presented in this application note is shown in Figure 5-1 along with the setup for benchmarking the performance. In this setup, a Newport URS50BCC rotation stage rotates the Eshifter stick, while a python program collects the Newport angle and the TMAG5170D magnetic flux measurements and angle calculations. After the data is collected, the first angle collected from the TMAG5170D is compared against the first angle reported by the newport and the difference between these two values is subtracted from all subsequent calculated angles from the TMAG5170D for a 1 point calibration. Subsequently the difference between newport and the shifted TMAG5170D angles is calculated to determine the 1 point calibration angle error in degrees as shown in Figure 5-2.

Through plotting the 1 point calibrated data versus the newport's angle, which is treated as the absolute angle point of reference, the data in Figure 5-3 appears to be linear and similar to the reference yet does not have an exact 1 to 1 relationship as shown in Figure 5-2. To minimize the difference between the absolute angle source and the angle procured from the TMAG5170D, a two or more point calibration can be done. For this demo, the movement bounds were used for the two point calibration. The difference between the bound points where then used to determine the coefficients for the equation describing error versus angle calculated from the TMAG5170D as indicated in Figure 5-4.

From the two point calibration, Figure 5-5 was obtained. These results show the max error being between -0.5° and 0.12° as opposed to the previous error bounded between -0.8° and 0.63°. This shows that the max error can be calibrated to be as low as ±0.5° and the spread of error reduced by nearly 60%. Extending the calibration to 3 pt, Figure 5-6 was obtained. These errors for this set are -0.15° to 0.14°, providing 80% reduction in the error spread from a single point calibration.

As E-shifters need to be redundant, error between sensors is also a very important metric. Figure 5-7 through Figure 5-10 show how many degrees of difference were observed between the sensors on each device. Inspection reveals that no calibration had error bounded to roughly ±0.6 degrees while 1 point calibration offset the error range from -1.2 to 0.2. This makes sense from the standpoint that both sensors start angle are zeroed in the 1 pt calibration. Calibrations involving more points that compensate for the error exhibit smaller ranges of error with each added calibration point.