Deep-Learning-Based Reduced-Order Model for Power Generation Capacity of Flapping Foils

Area of Science:

• Fluid dynamics
• Renewable energy technologies
• Computational science

Background:

• Oscillating foils present a promising avenue for harnessing wind and water energy.
• Traditional reduced-order models (ROMs) often face limitations in long-term prediction accuracy for complex fluid flows.

Purpose of the Study:

• To develop and validate a novel proper orthogonal decomposition (POD)-based reduced-order model (ROM) integrated with deep neural networks for power generation prediction.
• To enhance the accuracy and predictive capability of ROMs for fluid dynamics simulations, particularly for long time durations.

Main Methods:

• Numerical simulations of incompressible flow past a flapping NACA-0012 airfoil at Re=1100 using the Arbitrary Lagrangian-Eulerian approach.
• Construction of pressure POD modes from simulation snapshots to form a reduced basis.
• Development and application of long-short-term neural network (LSTM) models to predict temporal coefficients of POD modes.

Main Results:

• The LSTM-based ROM accurately predicts temporal coefficients for extended durations, surpassing the accuracy of traditional ROMs.
• Hydrodynamic forces and moments, crucial for power computation, are accurately reconstructed using the predicted coefficients and POD modes.
• The model demonstrates superior performance in capturing flow physics over long time intervals compared to conventional ROMs.

Conclusions:

• The proposed hybrid POD-LSTM model offers a significant advancement in accurately predicting power generation from flapping airfoils.
• This approach provides a more reliable method for long-term forecasting in fluid dynamics simulations, overcoming limitations of traditional ROMs.
• The study highlights the potential of integrating deep learning with physics-based models for efficient and accurate renewable energy analysis.