Calibration of an instrumented treadmill using a precision-controlled device with artificial neural network-based error corrections

• A precision-controlled device calibrated in situ vertical loads and COP for an IT.
• IT-measured vertical loads and COP were highly non-linear and speed-dependent.
• An ANN was effective in correcting errors in vertical loads and COP for the IT.
• The calibration method may help determine when a factory re-calibration is needed.

Instrumented treadmills (ITs) are used to measure reaction forces (RF) and center of pressure (COP) movements for gait and balance assessment. Regular in situ calibration is essential to ensure their accuracy and to identify conditions when a factory re-calibration is needed. The current study aimed to develop and calibrate in situ an IT using a portable, precision-controlled calibration device with an artificial neural network (ANN)-based correction method. The calibration device was used to apply static and dynamic calibrating loads to the surface of the IT at 189 and 25 grid-points, respectively, at four belt speeds (0, 4, 6 and 8 km/h) without the need of a preset template. Part of the applied and measured RF and COP were used to train a threelayered, back-propagation ANN model while the rest of the data were used to evaluate the performance of the ANN. The percent errors of Fz and errors of the Px and Py were significantly decreased from a maximum of −1.15%, −1.64 mm and −0.73 mm to 0.02%, 0.02 mm and 0.03 mm during static calibration, respectively. During dynamic calibration, the corresponding values were decreasing from −3.65%, 2.58 mm and −4.92 mm to 0.30%, −0.14 mm and −0.47 mm, respectively. The results suggest that the calibration device and associated ANN will be useful for correcting measurement errors in vertical loads and COP for ITs.