A one-dimensional flow model enhanced by machine learning for simulation of vocal fold vibration.

Area of Science:

• Fluid dynamics
• Biophysics
• Computational modeling

Background:

• Accurate modeling of unsteady, viscous flow in the larynx is crucial for understanding voice production.
• Existing one-dimensional (1D) models often lack precision in capturing complex laryngeal dynamics.
• Physics-based models require enhancement to incorporate detailed flow characteristics.

Purpose of the Study:

• To enhance a physics-based 1D unsteady and viscous flow model of the larynx.
• To integrate machine learning for determining unknown model parameters.
• To develop a faster and more accurate simulation tool for laryngeal fluid-structure interaction (FSI).

Main Methods:

• Derived a 1D unsteady, viscous flow model from conservation equations.
• Utilized a machine learning approach, specifically sparse identification, to train the 1D model.
• Performed 3D fluid-structure interaction (FSI) simulations on idealized and subject-specific laryngeal models to generate training data.

Main Results:

• Obtained analytical expressions for entrance effects and pressure loss in the glottis through machine learning.
• Successfully incorporated these expressions into the 1D flow model to account for vocal fold vibration.
• Demonstrated robust performance of the enhanced 1D model in FSI simulations of various laryngeal geometries.

Conclusions:

• The enhanced 1D flow model provides a fast and accurate simulation tool for laryngeal dynamics.
• This approach offers a significant improvement over previous 1D models for FSI analysis.
• The method is applicable to both idealized and patient-specific laryngeal geometries.