Author(s): LIN, Y.C., HSU, W.1, TANG, L.1, LIN, Y.2, Institution: NATIONAL TAIWAN UNIVERSITY OF SCIENCE AND TECHNOLOGY, Country: TAIWAN, Abstract-ID: 1322

INTRODUCTION:
During the training of athletes, explosive power training is essential. This ability is often being trained or evaluated by countermovement jump(CMJ), jump shrug(JS), and hang power clean(HPC) [1]. In previous research, joint angles have been found related to energy absolution and the performance of weightlifting[2]. However, due to the cost of the analysis of biomedical indexes, underfunded athletes find it hard to obtain trustful results to benefit from.[3] Also, the embedded system and its algorithms developed in this study only aimed to detect good or incorrect squats. Converts inertial measurement unit(IMU) data to joint angles of lower limbs by using artificial neural networks(ANN) is required to reduce analysis costs and provide a trustable result [4]. We aim to use ANNs to provide a trustable low-cost analysis solution for joint angle feedback on the CMJ, JS, and HPC.
METHODS:
A total of 37 subjects have performed the CMJ, JS, and HPC, and the weight of the barbell is set at 60%, 75%, and 90% of their one-repetition HPC maximum. The data collection was completed within a motion laboratory, using a passive motion capture system(Qualisys AB, Göteborg, Sweden) and wireless physiological feedback systems(Delsys Trigno®, Boston, MA, USA). For ANN training, using a supervised algorithm with ground truth provided by the motion capture system, and prediction input from IMUs within physiological feedback systems, for validation using leave-one-subject-out cross-validation.
RESULTS:
During the training of the ANN model. The mean absolute error(MAE) and normalized root-mean-square error(NRMSE) have been monitored, indicating that the MAE and NRMSE maximum average didn’t exceed 6.6099(degree) and 15.2819%. Also, it shows a good correlation with ground-truth data that the minimum average wasn’t lower than 0.8626 for all the tasks.
CONCLUSION:
This study shows that our models achieve lower NRMSEs compared to recent literature findings. Future research should explore if our models generalize well to new, independent data, and show the potential of using ANNs for complex movements such as weightlifting training in athletes.
REFERENCES:
1. Ebben, W.P. and D.O. Blackard, Complex training with combined explosive weight training and plyometric exercises. Olympic coach, 1997. 7(4): p. 11-12.
2. Kipp, K., et al., Weightlifting Performance Is Related to Kinematic and Kinetic Patterns of the Hip and Knee Joints. Journal of strength and conditioning research / National Strength & Conditioning Association, 2011. 26: p. 1838-44.
3. Fonseca, S.T., et al., Considerations for working with professional athletes versus nonprofessional amateur athletes during Olympic events. Handbook of Sports Medicine and Science: Sports Therapy Services: Organization and Operations, 2012: p. 79-90.
4. Rapp, E., et al., Estimation of kinematics from inertial measurement units using a combined deep learning and optimization framework. Journal of Biomechanics, 2021. 116: p. 110229.

Acknowledgments
Funding: Thanks to the support from NSTC-111-2410-H-011-006-MY3, CTU-TAIWAN TECH-2022-06, and CTU-NTUST-2024-0