Prediction of steady flows passing fixed cylinders using deep learning.

Area of Science:

• Fluid dynamics
• Machine learning
• Computational science

Background:

• Computational fluid dynamics (CFD) simulations are computationally expensive.
• Deep learning and machine learning offer potential cost reductions.
• Predicting fluid flow around multiple cylinders is a complex challenge.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting steady fluid flows around multiple fixed cylinders.
• To assess the accuracy of the predicted velocity field and fluid forces.
• To investigate the model's generalization capabilities.

Main Methods:

• A deep learning model was designed to predict the x and y components of the velocity field.
• The model takes the cylinder arrangement as input.
• Accuracy was evaluated by comparing predicted velocity profiles and forces against ground truth.

Main Results:

• The deep learning model accurately predicts flow fields for cylinder numbers close to the training data.
• Extrapolation to fewer cylinders resulted in prediction errors, potentially due to internal fluid friction.
• The model demonstrated good generalization performance for predicting fluid forces on systems with more cylinders.

Conclusions:

• Deep learning shows promise in reducing the computational cost of fluid dynamics simulations.
• Model accuracy is dependent on the similarity of the cylinder arrangement to the training dataset.
• The model exhibits effective generalization for predicting fluid forces in more complex, larger cylinder systems.