University of Twente Student Theses


Estimating vertical ground reaction forces for rearfoot runners using a minimal IMU setup : For different running speeds and step frequencies

Usta, Hazal (2021) Estimating vertical ground reaction forces for rearfoot runners using a minimal IMU setup : For different running speeds and step frequencies.

[img] PDF
Abstract:Ground reaction forces (GRFs) are important measures to identify human movement. Someone standing, walking or running generates a force on the ground and as a reaction the GRF is applied on the body. The GRFs represent the load working on the body and are, therefore, often studied in running activities. This force may vary from 2 to 3 times body weight during running and is associated with running injuries. The GRF consists of three components of which the vertical force is the largest and most important component when studying injuries. The vertical GRF (vGRF) consists of first an impact peak and then an active peak. The first peak is generated by the impact the foot makes with ground and the second peak is a reaction of the neuromuscular feedback and push off force of the foot applied on the ground. GRFs can be measured using force plates mounted in the floor or instrumented in treadmills, however a restriction to this method is that is it limited to the lab setting. GRFs can also be estimated from other quantities. This could be used to determine the load on the body outside the lab. Different algorithms are developed for estimating GRFs based on inertial measurement units (IMUs), however no research has been done on estimating at different speed and step frequencies for rearfoot runners. The aim of this study is to develop a generic algorithm to estimate the vertical component of GRFs of rearfoot runners, using a minimal IMU setup. Eight healthy experienced rearfoot strike runners (age: 33.75 ± 10.98 years; height: 1.77 ± 0.08 cm; mass: 70.46 ± 14.50 kg; gender: 5 males/ 3 females) participated in this study. Every participant ran at three different speeds (in random order 10, 12 and 14 km/h) and three different step frequencies (100%, 110% and 90% of preferred step frequency) for 90 seconds. IMU data of eight sensors was collected with the MVN Link system to be used for the estimation of the vGRFs. The estimation algorithm is based on Newtons second law and uses the accelerations of the sensors in global frame to estimate the vGRFs. Data of the instrumented treadmill was measured and used as a reference to fit and validate the algorithm. An optimization function was applied on the estimation algorithm to find the best fit to the measured GRF (mGRF). The function optimizes the filtering cutoff frequencies, the filtering orders and the weight factor (WF) of the accelerations of the sensors by searching for the lowest root mean squared error (RMSE). By optimizing for lowest RMSE, the optimization function managed to focus on the estimation of the total gait cycle. Hence, both the impact peak and active peak were estimated. The lowest RMSE was found for the pelvis - tibia configuration (0.129 BW). The corresponding cutoff frequencies, orders and WF were applied on the data set and the absolute maximum active peak error (AMAPE) and Pearson’s correlation coefficient (ρ) were calculated. For the pelvis - tibia configuration an AMAPE of 0.0873 (± 0.0602) BW and ρ of 0.99 was found. The lowest AMAPE was found for the pelvis - thighs configuration (0.0730 (± 0.0509) BW). For this configuration a RMSE of 0.180 BW and ρ of 0.98 was found. The pelvis - tibia configuration is validated for the three running speeds and three step frequencies and strong correlation (ρ >0.99) between the estimated and measured GRF was found. Sensitivity analysis showed that the proposed algorithm is able to estimate well for differences in cutoff frequencies of the filter and different participants. Small differences in filtering order and WF have quite a big influence on the RMSE and should be chosen precisely. In conclusion, estimation of the vGRF is possible using three sensors with the proposed estimation algorithm for the pelvis - tibia configuration. Sensor data should be filtered with a second order filter with pelvis ƒc of 7.4 Hz, a fourth order filter with tibia ƒc of 9.0 Hz and WF of 0.496 of the pelvis acceleration should be used.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:42 biology
Programme:Biomedical Engineering MSc (66226)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page