Modelling / Simulation
The precise measurement of ground reaction forces and moments (GRF/M) usually requires stationary equipment and is, therefore, only partly feasible for field measurements. In this work we propose a method to derive GRF/M time series from motion capture marker trajectories for cutting maneuvers (CM) using a long short-term memory (LSTM) neural network. We used a dataset containing 637 CM motion files from 70 participants and trained two-layer LSTM neural networks to predict the GRF/M signals of two force platforms. A five-fold cross-validation resulted in correlation coefficients ranging from 0.870 to 0.977 and normalized root mean square errors from 3.51 to 9.99% between predicted and measured GRF/M. In future, this method can be used not only to simplify lab measurements but also to allow for determining biomechanical parameters during real-world situations.
Fohrmann, Dominik; Mundt, Marion; David, Sina; Koeppe, Arnd; Markert, Bernd; and Potthast, Wolfgang
"CREATING VIRTUAL FORCE PLATFORMS FOR CUTTING MANEUVERS FROM KINEMATIC DATA BASED ON LSTM NEURAL NETWORKS,"
ISBS Proceedings Archive: Vol. 38:
1, Article 109.
Available at: https://commons.nmu.edu/isbs/vol38/iss1/109