Handrim Reaction Force and Moment Assessment Using a Minimal IMU Configuration and Non-Linear Modeling Approach...

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Machine Learning

Background:

• Manual wheelchair propulsion causes shoulder joint injuries in spinal cord-injured individuals due to repetitive strain.
• Shoulder joint load is directly correlated with handrim forces and moments during propulsion.

Purpose of the Study:

• To estimate handrim reaction forces and moments during wheelchair propulsion using a single inertial measurement unit per hand.
• To compare the efficacy of Hammerstein-Wiener (HW) modeling and a BiLSTM recurrent neural network for this estimation task.

Main Methods:

• Two approaches were employed: Hammerstein-Wiener (HW) modeling and a BiLSTM neural network.
• Input data included linear acceleration and angular velocity from wrist-mounted inertial sensors.
• Eleven subjects performed a linear propulsion protocol, with forces and moments measured by a dynamic platform.

Main Results:

• Both BiLSTM and HW models demonstrated comparable accuracy in estimating horizontal, vertical forces, and sagittal moments.
• Mean Average Errors (MAE) for horizontal force were 6.10 N (BiLSTM) and 4.30 N (HW).
• MAE for vertical force were 5.91 N (BiLSTM) and 7.59 N (HW), and for sagittal moment were 0.96 Nm (BiLSTM) and 1.09 Nm (HW).

Conclusions:

• Both HW and BiLSTM models provide accurate estimations of handrim forces and moments, comparable to dynamic platform uncertainties.
• BiLSTM offers an advantage by not requiring prior knowledge of subject-specific force patterns, making it superior for time-series prediction.
• The study demonstrates the feasibility of measuring dynamic handrim forces in ecological settings using inertial sensors, crucial for injury prevention research.