Identification of constant-posture EMG-torque relationship about the elbow using nonlinear dynamic models

Area of Science:

• Biomechanics
• Neuroscience
• Biomedical Engineering

Background:

• Surface electromyogram (EMG) signals reflect muscle activity, crucial for understanding movement.
• Accurate elbow torque estimation is vital for biomechanical analysis and prosthetic control.
• Current methods for relating EMG to torque have limitations in precision.

Purpose of the Study:

• To enhance elbow torque estimation accuracy by optimizing surface EMG signal processing and model structures.
• To compare the performance of different EMG amplitude estimation processors, model types, and system identification techniques.
• To identify optimal parameters for dynamic, nonlinear models for precise torque prediction.

Main Methods:

• Collected surface EMG data from biceps and triceps muscles of 33 subjects during elbow movements.
• Employed advanced EMG amplitude (EMGσ) estimation processors, including whitened, multiple-channel signals.
• Utilized linear and nonlinear model structures, including polynomial, Hammerstein, and Wiener models, with pseudoinverse and ridge regression for parameter determination.
• Evaluated models using varying training data durations (26 s vs. 52 s).

Main Results:

• Torque estimation accuracy improved with advanced EMGσ processors, longer training sets (52 s), and pseudoinverse/ridge regression parameter estimation.
• Dynamic, nonlinear parametric models incorporating second- or third-degree polynomial functions of EMGσ outperformed linear and Hammerstein/Wiener models.
• A minimum error of 4.65 ± 3.6% maximum voluntary contraction (MVC) flexion was achieved with a third-degree polynomial, 28th-order dynamic model using the pseudoinverse method.
• Comparable performance (4.67 ± 3.7% MVC flexion error) was obtained with a second-degree, 18th-order ridge regression model.

Conclusions:

• Advanced EMG signal processing and nonlinear dynamic models significantly improve elbow torque estimation accuracy.
• Third-degree polynomial models with pseudoinverse parameter estimation and second-degree polynomial models with ridge regression offer high precision.
• These findings advance the capability for accurate EMG-based torque prediction in biomechanical applications.