On the Joint Optimization of Energy Harvesting and Sensing of Piezoelectric Energy Harvesters: Case Study of a Cable-Stayed Bridge

TL;DR

This paper explores the simultaneous optimization of piezoelectric energy harvesters for both energy generation and vibration sensing, using a real-world bridge dataset and multi-objective optimization techniques.

Contribution

It introduces a bi-objective design framework for PEHs, balancing energy harvesting and sensing accuracy, validated through a case study on a cable-stayed bridge.

Findings

Trade-off identified between energy harvesting and sensing performance.

Optimized Pareto fronts demonstrate the design space for bi-functional PEHs.

Effective use of neural networks and surrogate modeling for optimization.

Abstract

Piezoelectric Energy Harvesters (PEHs) are typically employed to provide additional source of energy for a sensing system. However, studies show that a PEH can be also used as a sensor to acquire information about the source of vibration by analysing the produced voltage signal. This opens a possibility to create Simultaneous Energy Harvesting and Sensing (SEHS) system, where a single piece of hardware, a PEH, acts as both, a harvester and as a sensor. This raises a question if it is possible to design a bi-functional PEH device with optimal harvesting and sensing performance. In this work, we propose a bi-objective PEH design optimisation framework and show that there is a trade-off between energy harvesting efficiency and sensing accuracy within a PEH design space. The proposed framework is based on an extensive vibration (strain and acceleration) dataset collected from a real-world operational cable-stayed bridge in New South Wales, Australia. The bridge acceleration data is used as an input for a PEH numerical model to simulate a voltage signal and estimate the amount of produced energy. The numerical PEH model is based on the Kirchhoff-Love plate and isogeometric analysis. For sensing, convolutional neural network AlexNet is trained to identify traffic speed labels from voltage CWT (Continuous Wavelet Transform) images. In order to improve computational efficiency of the approach, a kriging metamodel is built and genetic algorithm is used as an optimisation method. The results are presented in the form of Pareto fronts in three design spaces.