Estimating 3D ground reaction forces in running using three inertial measurement units

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Understanding running injuries requires insight into biomechanical loading.
• Ground reaction forces (GRFs) are crucial for this but typically measured in labs.
• Estimating 3D GRFs outdoors is essential for real-world running analysis.

Purpose of the Study:

• To estimate 3D GRFs using three inertial measurement units (IMUs).
• To develop and validate machine learning models for GRF estimation.
• To enable outdoor biomechanical load quantification for running injury research.

Main Methods:

• Twelve runners performed trials at varying speeds and stride frequencies on an instrumented treadmill.
• Data from IMUs on the pelvis and lower legs were used to train artificial neural networks (ANNs).
• A hybrid model combined ANNs with a physical model; ensemble models were created for improved accuracy.

Main Results:

• The hybrid model achieved normalized RMSE of 10.8% (ML), 7.8% (AP), and 6.8% (Vertical).
• Pearson correlation coefficients were 0.58 (ML), 0.91 (AP), and 0.97 (Vertical).
• Ensemble models demonstrated higher accuracy than individual models; performance was comparable across ML, hybrid, and physical models.

Conclusions:

• This study is the first to estimate 3D GRFs during continuous running using IMUs.
• It is feasible to estimate GRF in the posterior-anterior and vertical directions for outdoor settings.
• This advancement aids in understanding running injury development by quantifying biomechanical load in real-world environments.