Predicting net joint moments during a weightlifting exercise with a neural network model

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Accurate assessment of Net Joint Moments (NJM) is crucial for understanding weightlifting biomechanics and injury prevention.
• Traditional inverse dynamics methods require extensive motion capture equipment, limiting their application in field settings.
• Exploring machine learning for predicting biomechanical variables offers a potential solution for more accessible analysis.

Purpose of the Study:

• To develop and train a Neural Network (NN) capable of predicting hip, knee, and ankle NJM during weightlifting.
• To utilize barbell mass and motion data as inputs for the NN model.
• To assess the accuracy of the NN in predicting NJM compared to traditional inverse dynamics.

Main Methods:

• Seven weightlifters performed cleans at 85% of their maximum effort.
• Ground reaction forces and 3-D motion data were collected for inverse dynamics calculations of NJM.
• Barbell mass and kinematic data (vertical and horizontal motion) were used to train a NN.
• K-fold cross-validation was employed to rigorously test NN performance and accuracy.

Main Results:

• The NN achieved high prediction accuracy for joint-specific NJM, with coefficients of determination (r²) ranging from 0.79 to 0.95.
• The percent difference between NN-predicted and inverse dynamics-calculated peak NJM was between 5% and 16%.
• The NN successfully predicted both the spatiotemporal patterns and peak values of hip, knee, and ankle NJM.

Conclusions:

• It is feasible to predict lower extremity NJM in weightlifters using only barbell mass and motion data via a NN.
• The developed NN demonstrates reasonable accuracy, suggesting potential for practical biomechanical analysis.
• Future research should investigate the use of low-cost technology with NN models for field-based NJM prediction during training and competition.