Estimation of vertical ground reaction force during running using neural network model and uniaxial accelerometer

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Wearable technology offers potential for unconstrained human motion capture during activities like running.
• Existing methods for measuring forces during running are often complex, requiring extensive kinematic and kinetic data.
• Vertical Ground Reaction Force (VGRF) is a critical parameter in running analysis.

Purpose of the Study:

• To investigate the efficacy of a neural network (NN) model combined with accelerometer data for estimating VGRF.
• To simplify the process of VGRF estimation in unconstrained running environments.
• To reduce computational demands and the number of required wearable sensors.

Main Methods:

• An experimental study was conducted to gather data for training, validation, and testing of the NN model.
• A single uniaxial accelerometer was used in conjunction with the NN model.
• Estimated VGRF values were compared against measurements from an instrumented treadmill.

Main Results:

• The NN model achieved high accuracy in VGRF estimation, with an average root mean square error (RMSE) below 0.017 Body Weight (BW) and a cross-correlation coefficient above 0.99.
• The model could differentiate and estimate impact and active force components with average errors between 0.10 and 0.18 BW across various running speeds.
• The combined NN and accelerometer approach demonstrated significant simplification and reduced computational requirements.

Conclusions:

• A neural network model utilizing uniaxial accelerometer data provides a simplified and computationally efficient method for estimating VGRF during running.
• This approach reduces the reliance on complex biomechanical measurement systems and multiple sensors.
• The findings support the use of accessible wearable technology for accurate running dynamics analysis.