Design of near and far-field thermophotovoltaic energy converters using bayesian optimization

The environmental crisis has been one of the biggest challenges facing humanity. The release of carbon dioxide (CO2) and waste heat are among the root causes of this situation through global warming. Thermophotovoltaic (TPV) energy conversion, in which the photons thermally generated by an emitter are harvested using a photovoltaic cell, is a promising method because TPVs can utilize both renewable solar energy and waste heat. This thesis presented the design of optimized TPV systems in both the near-field and far-field. Near-field refers to sub-wavelength distances where radiative transfer significantly exceeds the estimate of the Planck law and is explained by the theory of fluctuational electrodynamics. A novel approach is adopted in both regimes by examining the optimal dielectric functions of the constituent materials to establish material selection criteria for maximum performance. Bayesian optimization, a machine learning-based method, is used in both analyses, in which a surrogate model imitates the objective function, while an acquisition function explores and exploits the input space with a limited number of objective function evaluations. In the far-field design, a multilayer selective emitter is utilized to approach the ideal spectral emittance identified by a parametric TPV analysis, a method known as the inverse design. In the near-field design, the evanescent coupling between a metal emitter and a thin cell that eliminates the bulk recombination losses on a metal substrate is analyzed in detail for maximum efficiency. The efficiency and power density values obtained in the far-field and near-field designs are 17.5% and 4.88 kW/m2, and 31.1% and 18.3 kW/m2, respectively. These findings open up a new perspective on the design of sustainable energy conversion technologies as well as spectrally selective optoelectronic and photonic systems.