Ensemble latent assimilation with deep learning surrogate model: application to drop interaction in a microfluidics...

Area of Science:

• Microfluidics
• Fluid Dynamics
• Data-Driven Modeling

Background:

• Predicting and controlling drop interactions in microfluidics presents a significant challenge.
• Existing methods often face computational limitations for real-time analysis.

Purpose of the Study:

• To develop a reliable data-driven model for forecasting drop dynamics in microfluidic devices.
• To improve the accuracy and efficiency of predicting microfluidic drop interactions.

Main Methods:

• Utilized reduced-order modeling to compress experimental image data into low-dimensional representations.
• Employed recurrent neural networks to create a surrogate model learning dynamics in the reduced-order space.
• Integrated the surrogate model with real-time observations via an ensemble-based latent data assimilation algorithm.

Main Results:

• Achieved high-fidelity predictions of drop interactions, outperforming previous approaches in accuracy.
• Demonstrated the model's ability to learn complex drop dynamics from compressed variables.
• Validated the model's performance against experimental video data not used during training.

Conclusions:

• The developed data-driven approach enables reliable and accurate prediction of microfluidic drop interactions.
• The ensemble-based latent assimilation scheme offers improved predictive capabilities.
• The methodology is generalizable to other dynamical systems beyond microfluidics.