Neural network-based interface reconstruction algorithm for two-phase fluid flow

Area of Science:

• Computational fluid dynamics
• Interface reconstruction algorithms
• Two-phase flow modeling

Background:

• Surface tension causes curved vapor-liquid interfaces in two-phase flow.
• Accurate interface reconstruction is crucial for resolving bubbles and droplets in numerical simulations.
• Existing methods like PLIC, ELVIRA, QUASI, and CIR have limitations in accuracy and efficiency.

Purpose of the Study:

• To propose novel curve reconstruction algorithms based on artificial neural networks (ANN) and convolutional neural networks (CNN) for two-phase fluid flow.
• To enhance the precision and reliability of interface reconstruction, particularly for circular shapes.
• To evaluate the performance of the proposed algorithms against established methods in terms of accuracy and computational cost.

Main Methods:

• Development of CIR-ANN and CIR-CNN algorithms utilizing ANN and CNN architectures.
• Implementation of a strict mass conservation strategy to ensure prediction reliability.
• Comparative analysis against PLIC, ELVIRA, QUASI, and CIR algorithms for static and dynamic interface reconstruction.

Main Results:

• CIR-CNN demonstrated superior accuracy in static interface reconstruction compared to all benchmark algorithms.
• Both CIR-ANN and CIR-CNN achieved significant reductions in computational time for reconstructing circular interfaces.
• The advantage of the proposed algorithms diminished in complex flow fields with intricate fluid volume fraction distributions.

Conclusions:

• The proposed CIR-CNN algorithm offers significant advantages for static interface reconstruction in two-phase flow simulations.
• ANN and CNN-based approaches provide efficient and accurate alternatives for interface reconstruction.
• Further research is needed to optimize algorithms for complex dynamic flow scenarios.