A Neural Differentiable Framework for Efficient Simulation and Rapid Optimization of Energy Harvesters

In response to the rapid growth of energy consumption, renewable energy harvesters have emerged as a powerful technique for providing sustainable energy and achieving carbon neutrality, including turbines, hydrofoils, flapping foils, flow-induced vibrations (FIV) energy harvesters, etc. The harvesters are usually under strong fluid-structure interactions (FSI), which requires effective simulation of FSI at design and optimization in operation. However, classical computational fluid dynamics (CFD)-based solvers are too expensive to tackle the increasing large-scale and long-period simulation demands. The ever-increasing data availability and rapid developments in deep learning (DL) have opened new avenues to tackle these challenges by integrating deep neural networks (DNN) into traditional numerical solvers to enable effective data-driven modeling. The seamless integration of DL and classical CFD, allowing end-to-end AD through the entire differentiable programming (DP) framework, can leverage the advantages of both techniques and significantly improve the data-driven modeling performance. In this regard, we established a fully differentiable programming framework for simulating FSI problems based on JAX. The fluid is solved by direct numerical simulation (DNS), and the solid is immersed in the fluid field through the direct forcing method. Specifically, the velocity inside an immersed solid is interpolated by a sinusoidal function. As the framework is entirely built in JAX with auto-differentiation (AD) capability, different DNN models can be easily integrated and optimized within the numerical solver as a whole in an end-to-end manner. Taking the FIV energy harvester as an example, it is simulated to demonstrate the merit and potential of the proposed method for efficient modeling of energy harvesters. In addition, due to the environmental impacts and community acceptance is essential for making project investment and implementation about renewable energy harvesters. Based on the simulated results, we analyze several acceptance factors following a stakeholder survey, and rapidly find an optimized resolution regarding to the acceptance-related factors by integrating DNN to this framework.