Enhanced decomposition of vertical ground reaction forces via an improved off-loading model and optimised artificial
Yifan Zhang1, Bintian Lin2, Changqing Miao3
1School of Engineering, Newcastle University, Newcastle upon Tyne NE1 7RU, UK; School of Civil Engineering, Southeast University, Nanjing 210096, China.
Background:
Accurate quantification of the individual ground reaction forces (GRFs) generated by each foot during gait is fundamental to understanding human locomotion. A primary technical challenge arises during double support, when the measured GRF represents the summed contributions of both limbs; this issue is further amplified under non-standard conditions, such as uneven surface or variable terrain.
Research Question:
This study proposes an improved algorithm to refine the accuracy of vertical GRF decomposition across varying ground conditions and introduces an approach for identifying previously undetermined algorithmic constants, thereby promoting its practical application.
Methods:
Seven healthy adults completed gait trials under 14 walking and platform-vibration conditions. Based on the smooth-transition assumption, an improved two-term formulation was developed to estimate the off-loading component of vertical GRF. A simplified linear regression and artificial neural network (ANN) models were then used to predict the model constant from features extracted from pre-double-support GRF signals. A mind evolutionary computation algorithm was applied to initialise the ANN weights and biases. The method was evaluated using the experimental dataset and an open-access dataset including barefoot overground walking.
Results:
The proposed formulation achieved a relative root mean square error of 1.9% for the off-loading component, compared with 12.2% for the original smooth-transition model. Regression and ANN-based coefficient prediction produced errors of 3.8% and 3.3%, respectively. Similar trends were observed for the public dataset, with errors of 2.0%, 4.1% and 3.0%, respectively.
Significance:
The proposed method improved vertical GRF decomposition under the tested conditions including vibrating conditions and public barefoot dataset, providing a practical framework for estimating single-foot vertical GRF profiles from continuous force measurements.
