Learning aerodynamics with neural network.

Area of Science:

• Computational fluid dynamics
• Aerodynamics
• Machine learning

Background:

• Accurate prediction of airfoil lift coefficient is crucial for aerodynamic design.
• Existing neural network (NN) architectures face challenges in achieving high accuracy with computational efficiency.

Purpose of the Study:

• To introduce a novel neural network (NN) architecture, the Element Spatial Convolution Neural Network (ESCNN), for airfoil lift coefficient prediction.
• To demonstrate the ESCNN's superior performance compared to state-of-the-art NNs.

Main Methods:

• Development of the Element Spatial Convolution Neural Network (ESCNN) architecture.
• Utilizing standard convolution layers within the ESCNN.
• Investigating the network's ability to extract physical patterns from aerodynamic data.

Main Results:

• The ESCNN achieved higher prediction accuracy than existing state-of-the-art NNs.
• The ESCNN demonstrated superior parameter efficiency, requiring two orders of magnitude fewer parameters.
• Analysis revealed the ESCNN's capacity to extract and reflect physical aerodynamic patterns within its layers.

Conclusions:

• The ESCNN effectively learns physical laws and equations of aerodynamics from simulation data.
• The proposed architecture offers a computationally efficient and accurate approach to airfoil lift coefficient prediction.
• ESCNN's ability to learn underlying physical principles from data represents a significant advancement in applying NNs to fluid dynamics.