Voice production is due to a pair of vocal fold whose vibration is induced by the glottal airflow.  The vibration involves aerodynamic interaction of a pulsatile glottal jet and the soft vocal fold tissue stretched between the laryngeal cartilages.  Certain voice disorders, e.g., unilateral vocal fold paralysis, may require surgery for voice restoration. Due to many factors such as surgical techniques, patient variations, and experience of surgeons, the revision rates of the surgeries can vary widely, and there are sometimes inconsistent or even undesirable clinical outcomes. It is necessary to develop high-fidelity computational modeling of the fluid-structure interaction (FSI) during the vocal fold vibration for surgical planning. In the past, we have developed a 3D FSI model that incorporates patient-specific anatomy of the larynx based on medical imaging data. However, unknown tissue properties of the vocal fold limit the fidelity of such models. It would be too expensive to perform parameter identification directly using 3D FSI simulations due to their computational cost. In our work, we aim to develop a reduced-order FSI model that balances between computational cost and accuracy to achieve the fast simulation. More specifically, we coupled a machine-learning enhanced 1D flow model with a 3D nonlinear tissue model for simulation of vocal fold vibration.  

Bernoulli based 1D flow equations have long been used for vocal fold vibration. It has some obvious limitations, such as no-loss assumptions along the glottis and sensitivity to the separation location. To overcome these limitations, we have developed a 1D flow model based on the momentum equation with pressure loss and entrance effects. To further enhance this reduced-order flow model, we use machine learning (MR) to determine the uncertain terms in the equation such as pressure loss and entrance effect. The training data are taken from 3D FSI simulations of an idealized larynx.  More specifically, five non-dimensional geometry parameters and the Reynolds number are used as inputs in MR, while pressure loss and entrance effect at several specific locations are the outputs. Sparse identification of nonlinear dynamics (SINDY) method has the strong capability to discover the governing equation from data, and it is employed here to identify the relations between inputs and outputs. We study six idealized vocal fold cases with two different medial thicknesses and three different tissue properties to create the training data set for MR. Four more cases are used for additional validation, where we vary the subglottal pressure and medial thickness.  In all ten cases, we compare vibration amplitude, frequency, and phase delay between the reduced-order FSI and the full 3D FSI.  Results from the reduced-order model agree well with those from the 3D FSI for all cases.  The same reduced-order model enhanced by MR is then used to solve the vocal fold vibrations in five subject-specific cases, where the laryngeal geometry is generated from MRI scan of rabbit larynx. The simulated frequency and normalized amplitude from the reduced-order model are consistent with the in vivo experimental results for all the ﬁve samples despite signiﬁcant individual variations.  The reduced-order FSI model enhanced by MR thus is valid for both idealized and subject-specific cases. Therefore, this model has promising application in parameter identification of the actual vocal fold tissue.


Fluid-Structure Interaction Modeling of Vocal Fold Vibration Enhanced by Machine Learning

Division / Sponsors: 
Heat Transfer Division (HTD), Fluids Engineering Division (FED)

Heat Transfer Summer Conference
Paper Proceedings: https://asmedigitalcollection.asme.org/HT/HT2020/volume/83709

18th International Conference on Nanochannels, Microchannels, and Minichannels
Paper Proceedings: https://asmedigitalcollection.asme.org/ICNMM/ICNMM2020/volume/83693

Fluids Engineering Division Summer Meeting
Paper Proceedings: https://asmedigitalcollection.asme.org/FEDSM/FEDSM2020/volume/83716

HT-ICNMM2020 - Virtual Conference, SHTC - Virtual Conference, FEDSM2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.