Publication: Development of a triaxial biomechanical force plate based on uniaxial load cells and deep learning
| datacite.subject.fos | oecd::Engineering and technology::Mechanical engineering | |
| dc.contributor.author | Yeo, Ying Heng | |
| dc.date.accessioned | 2025-12-03T02:58:47Z | |
| dc.date.available | 2025-12-03T02:58:47Z | |
| dc.date.issued | 2024-08-01 | |
| dc.description.abstract | Custom force plate developed from half-bridge strain gauge load cells is a potential low-cost alternative to expensive laboratory-grade force plates. Nevertheless, the measurement accuracy has not been thoroughly validated. The inability to quantify bilateral ground reaction force (GRF) prevents the utilization of the low-cost force plate in biomechanical analysis. In this study, a low-cost custom force plate has been developed by using uniaxial half-bridge strain gauge load cells. Based on Poisson effect, the load cells could produce readings even when the off-axis bilateral GRF which was orthogonal to the primary axis vertical GRF was applied. These readings were used as features to infer the bilateral GRF measured with laboratory-grade force plate. The mapping of uniaxial load cell readings to bilateral GRF was carried out using deep learning models. The validity of the custom force plate in measuring three-dimensional GRF, centre of pressure (CoP), and clinical metrics derived from vertical GRF and CoP was evaluated. The custom force plate validity in vertical GRF and CoP measurement for all tasks was indicated by mean absolute error of lower than 9.90 N and 6.29 mm, and high Pearson correlations (ρ), coefficient of determinations (R2), and intraclass correlation coefficients (ICC) of more than 0.94, 0.88, and 0.94 respectively. In acquiring clinical metrics, the custom force plate achieved ρ, R2, and ICC of greater than 0.98, 0.96, and 0.98 respectively. The recorded ρ and ICC were higher than that achieved in five previous studies which investigated other low-cost force plates. Autoencoder and U-net models were trained to receive time series or Short-Time Fourier Transformed (STFT) vertical GRF (acquired from the individual single-axis load cells of a custom force plate) as input and generate bilateral GRF as output. Different models were trained with Adam optimizer under the implementation of early stopping and hyperparameter tuning. The most accurate model was U-net model that accepted STFT-transformed input. Apart from the mediolateral GRF measured during sit-to-stand, the model predicted the bilateral GRF in the test dataset with root mean squared error (RMSE), and relative RMSE of less than 1.95% of body weight and 14.17%, and ρ, R2, and ICC of more than 0.89, 0.79, and 0.88, respectively. The values of ρ were greater than that obtained in six previous works that studied the bilateral GRF prediction methods with devices other than low-cost force plate. The result comparison with previous works highlighted the good measurement performance of the custom force plate. Hence, the custom force plate could potentially be a low-cost solution to measure GRF, CoP, and clinical metrics. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/23306 | |
| dc.language.iso | en | |
| dc.title | Development of a triaxial biomechanical force plate based on uniaxial load cells and deep learning | |
| dc.type | Resource Types::text::thesis::doctoral thesis | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |