*3.2. Performance Results*

Results are presented using data of a single sensor of the four-sensor array as a proxy for the behavior of the overall system. The calculated relative errors are shown in Figure 3. The bias of the mean (cross marker) indicates accuracy while the standard deviation (error bar) represent precision. Because of the limited influence of test rig velocity on sensor performance [16], velocities were combined for each distance. A bias of less than 0.9% was achieved for the *x*-direction at 40 mm. The *y*-direction achieved biases within 1.43% for uni-axial motion and 4.76% for diagonal motion at 40 mm. The standard deviation at 40 mm was less than 0.9% for uniaxial as well as diagonal movements.

**Figure 3.** Mean (cross marker) and standard deviation (error bar) of the relative error for one sensor from (**a**) uni-axial and (**b**) diagonal movement. Velocities were combined at each distance. Calibration factors were *k*x = 2.018 and *k*y = 1.987.

The calculated relative errors demonstrate comparable results in both accuracy and precision to the functional model [16]. Although the behavior among sensors was comparable, differences were noted in the required calibration factors with values for the four sensors ranged between 1.976 and 2.018 for *k*<sup>x</sup> and between 1.915 and 1.987 for *k*y. As the thickness of the liner patch varied by approximately 1 mm along the diagonal, these differences likely stemmed from offset variations between the individual sensors and the liner. The system's effective resolution depends upon the necessary calibration factor and is in the low micrometer range (the chosen sensor resolution of 8200 cpi corresponds to 3.1 μm per count).

Sampling frequencies per sensor were confirmed for four sensors between 25 Hz and 275 Hz in 25 Hz increments. Across all frequencies, the difference between set sampling period and the average sampling period calculated from the recorded timestamps was 0.01 ± 0.15 μs.

Errors were found to occur only rarely. Out of 839 measurements, seven cases of transient behavior in both *x* and *y* displacements were recorded. Since this error affected only individual sensors and not all four, the source was unlikely to have been due to external factors, e.g., a bump to the test-rig. These errors were easily identifiable and localized, i.e., they did not affect subsequently recorded data. In cases where measurement duration approached or exceeded ten minutes, data recording at the PC-end was delayed, resulting in a loss of data if the measurement was ended too soon after the period of interest. Given that the duration of measurements is intended to be around two minutes for different dynamic walking tasks, this problem is unlikely to be relevant in practice.
