Neural Network-Based Optimization of an Acousto Microfluidic System for Submicron Bioparticle Separation.

Area of Science:

• Microfluidics and Lab-on-Chip Technologies
• Acoustic Particle Manipulation
• Biotechnology and Biomedical Engineering

Background:

• Conventional sub-micron particle isolation is being replaced by cost-effective lab-on-chip platforms.
• Acoustic-driven separation offers label-free, biocompatible particle manipulation.
• Designing acoustofluidic devices is challenging due to complex physics and performance prediction limitations.

Purpose of the Study:

• To develop an automated design approach for acoustofluidic devices.
• To integrate machine learning and multi-objective optimization for enhanced device performance.
• To overcome limitations in predicting performance for iterative and resource-intensive design processes.

Main Methods:

• Developed a neural network to predict resonator frequency response based on interdigitated transducer geometry.
• Employed multi-objective optimization to identify optimal design features for device performance.
• Utilized 3D finite element simulations to analyze power loss and bandwidth effects on acoustic pressure distribution.

Main Results:

• The proposed methodology significantly improves interdigitated transducer designs.
• Achieved minimum power loss and maximum working frequency range for acoustofluidic devices.
• Demonstrated enhanced device performance by addressing constraints in bandwidth, acoustic radiation force, and pressure distribution.

Conclusions:

• The integrated machine learning and optimization approach enables efficient design automation of acoustofluidic devices.
• This method overcomes key challenges in optimizing acoustic transducers for particle manipulation.
• The findings pave the way for more robust and predictable lab-on-chip separation technologies.