The feasibility of predicting ground reaction forces during running from a trunk accelerometry driven mass-spring-damper model

• Published in: PeerJ (San Francisco, CA) CA), 2018-12, Vol.6, p.e6105-e6105, Article e6105
• Main Authors: Nedergaard, Niels J, Verheul, Jasper, Drust, Barry, Etchells, Terence, Lisboa, Paulo, Robinson, Mark A, Vanrenterghem, Jos
• Format: Article
• Language: English
• Subjects: Accelerometers; Anatomy and Physiology; Artificial intelligence; Bioengineering; Biomechanical loading; Biomechanics; Body-worn accelerometer; Force; Human mechanics; Kinematics; Kinesiology; Laboratories; Mass-spring model; Muscle function; Optimisation; Physiological aspects; Physiology; Running; Sensors; Team sports; Tissue adaptations; Training load monitoring
• ISSN: 2167-8359; 2167-8359
• DOI: 10.7717/peerj.6105

Monitoring the external ground reaction forces (GRF) acting on the human body during running could help to understand how external loads influence tissue adaptation over time. Although mass-spring-damper (MSD) models have the potential to simulate the complex multi-segmental mechanics of the human body and predict GRF, these models currently require input from measured GRF limiting their application in field settings. Based on the hypothesis that the acceleration of the MSD-model's upper mass primarily represents the acceleration of the trunk segment, this paper explored the feasibility of using measured trunk accelerometry to estimate the MSD-model parameters required to predict resultant GRF during running. Twenty male athletes ran at approach speeds between 2-5 m s . Resultant trunk accelerometry was used as a surrogate of the MSD-model upper mass acceleration to estimate the MSD-model parameters (ACC ) required to predict resultant GRF. A purpose-built gradient descent optimisation routine was used where the MSD-model's upper mass acceleration was fitted to the measured trunk accelerometer signal. Root mean squared errors (RMSE) were calculated to evaluate the accuracy of the trunk accelerometry fitting and GRF predictions. In addition, MSD-model parameters were estimated from fitting measured resultant GRF (GRF ), to explore the difference between ACC and GRF . Despite a good match between the measured trunk accelerometry and the MSD-model's upper mass acceleration (median RMSE between 0.16 and 0.22 g), poor GRF predictions (median RMSE between 6.68 and 12.77 N kg ) were observed. In contrast, the MSD-model was able to replicate the measured GRF with high accuracy (median RMSE between 0.45 and 0.59 N kg ) across running speeds from GRF . The ACC from measured trunk accelerometry under- or overestimated the GRF obtained from measured GRF, and generally demonstrated larger within parameter variations. Despite the potential of obtaining a close fit between the MSD-model's upper mass acceleration and the measured trunk accelerometry, the ACC estimated from this process were inadequate to predict resultant GRF waveforms during slow to moderate speed running. We therefore conclude that trunk-mounted accelerometry alone is inappropriate as input for the MSD-model to predict meaningful GRF waveforms. Further investigations are needed to continue to explore the feasibility of using body-worn micro sensor technology to drive simple human body models that would