Optimal design parameter discovery for nonlinear energy harvesters using neural optimization machine

The growing need for microsensing technologies with less reliance on depletable batteries has attracted interest in the conversion of the vibrational energy available from many sources and in different forms into electrical power using piezoelectric transduction. Considering that the geometry of the vibration source limits the available volume to place the harvester, it is essential to determine the suitability of a specific harvester configuration to meet a specific power requirement. Toward this objective, we develop a neural optimization approach to search for the design parameters resulting in optimal performance of piezoelectric energy harvesters defined in terms of power density under specific excitation characteristics and constraints. A nonlinear harvester configuration, namely the magnetopiezoelastic energy harvester, is considered. Details of the numerical solution are presented. The solution is validated using previously published experimental data. The training of neural networks and the implementation of the optimization procedure are based on data generated from numerical simulations of hundreds of configurations. The details of setting up the neural optimization machine are presented. Differences in the training models and constraints required for regular and irregular excitations are pointed out. The significance of optimized configurations in each of the two cases is discussed.