Module-Level Thermophotovoltaics with Polaritonic Emitters by Machine Learning
Qixiang Wang,Zhequn Huang,Jiazhou Li,Guanyao Huang,Dewen Wang,Heng Zhang,Jiang Guo,Min Ding,Jintao Chen,Zihan Zhang,Wen Shang,Jiayue Xu,Jian Zhang,Junichiro Shiomi,Tairan Fu,Tao Deng,Steven G. Johnson,Hongxing Xu,Kehang Cui
DOI: https://doi.org/10.2139/ssrn.4001951
2022-01-01
Abstract:Thermophotovoltaic (TPV) generators provide reliable and high-efficiency power output by utilizing local thermal emitters to convert various energy sources such as concentrated solar, carbon-neutral fuel and waste heat, to thermal radiation matching the bandgap of photovoltaic cells (PV). Current TPV designs omit the thermal loss and the electrical loss in devices, both of which are significantly hindering the industry-scale deployment of TPVs. In this work, we screen a design space of 4.8×1020 candidates using hierarchical sequential learning, and experimentally realize a module-level 6-inch polaritonic thermal emitter with a record-high spectral efficiency of 65.6%. The significantly quenched thermal and electrical losses lead to a projected system heat-to-electricity efficiency of 54.7%, which is competitive with those of the state-of-the-art turbines, piston engines as well as fuel cells (fuel-to-electricity efficiencies). Moreover, we demonstrate excellent reliability of the device under long-term thermal cycling ageing operation at 1473 K. The compatibility with industry-scale PV modules lays the cornerstone for scale-up applications of TPV generators.