Robust Bayesian Framework for Model Class Selection and Model Parameter Identification in Piezoelectric Energy Harvesters.
Initiation Project (2019-2021)
The collection of energy by means of piezoelectric devices has been of great interest especially for the development of sensors with remote applications, for example in the monitoring of difficult access structures such as antennas, bridges, or structures located in isolated regions. The design of these devices has been particularly complex due to the large dispersion with which the electro-mechanical characteristics of the piezoelectric materials are reported. Despite this, the trend in the design of these devices has been the use of deterministic analytical/numerical models, that is, without considering the dispersion or variability introduced by the aforementioned characteristics at the risk of designing devices that operate outside the optimum range. In this sense, the tuning/calibration of the models is made compulsory by direct comparison with experimental results. However, there is no rigorous methodology for selecting the analytical/numerical model that should be used to make such a comparison. The awarded project seeks to propose a methodological framework based on Bayesian inference that allows the appropriate selection of the model (and its parameters) that must be used in the tuning/calibration process. It should be noted that Bayesian inference techniques adapted to this type of problem would allow the selection of models based on the parsimonious model principle, which would correspond to the selection of the model of less complexity and greater precision. The project involves experimental tests of linear and non-linear devices, stochastic simulations, and generation of predictive numerical models oriented towards topological optimization.