Neural Network Based Surrogate Optimization Tool for Design of Improved FIV Energy Harvester

Abstract

Machine learning-based optimization employs model-based algorithms to find the best possible solution to engineering problems, often by iteratively improving a system or process. Neural networks are common machine learning techniques in many applications due to their several advantages, such as model-free estimation and generalization abilities. Neural networks have proven highly effective across various domains but require careful design, tuning, and training to yield optimal results for specific tasks. Recurrent Neural Networks (RNNs) are powerful tools for processing sequential data, offering flexibility and effectiveness in capturing dependencies over time. The present study uses an RNN-based optimization study to find the optimal parameters for improving the energy harvesting capability of coupled vortex-induced vibration and galloping-based laminated composite beam energy harvesting system. For this purpose, the effect of wind velocity, the coupling coefficient of vortex-induced vibration (VIV) and galloping, the cutting angle of the bluff body, the wake angle behind the bluff body, diameter to length ratio, and height to length ratio of the bluff body are considered on the output voltage. The system is composed of a laminated smart cantilever beam on which piezoelectric materials are attached for energy harvesting purposes, and at the free end of the beam, a bluff body is attached. It is found that the RNN-based particle swarm optimization method is efficient in finding the optimal parameters of the system.