A flow feature detection method for modeling pressure distribution around a cylinder in non-uniform flows by using a...

Area of Science:

• Fluid Mechanics
• Computational Engineering
• Machine Learning

Background:

• Accurate prediction of load conditions around structures is crucial in engineering.
• Flow feature detection offers a promising avenue for load estimation.

Purpose of the Study:

• To develop a data-driven method for predicting pressure on a cylinder using wake flow velocity distributions.
• To evaluate the efficacy of a convolutional neural network (CNN) for this prediction task.

Main Methods:

• A deep learning neural network combining convolutional and fully-connected layers was designed.
• The network extracts features from velocity data to predict pressure coefficients.
• Computational Fluid Dynamics (CFD) results were used as reference values for validation.

Main Results:

• The CNN accurately predicts pressure coefficients for trained Reynolds numbers and various inlet flow profiles.
• Transfer learning, by freezing convolutional layer parameters, maintained high precision with reduced computational cost.
• The study demonstrates the potential of CNNs in fluid mechanics for load prediction.

Conclusions:

• The developed CNN provides a precise and efficient method for predicting cylinder pressure from flow data.
• Transfer learning enhances the practicality of this approach by lowering computational demands.
• This work highlights the significant potential of deep learning in advancing fluid mechanics and structural load analysis.