Optimized Sustainable Thin-Film Thermoelectric Generator Design for Sensor Powering Using Machine Learning

This study explores the optimization and development of a thin-film thermoelectric generator (TEG) designed for sensor powering, with a primary emphasis on maximizing electrical power. Sustainable materials were chosen for the TEG legs, with Cu0.85Ag0.15FeS2 (CAFS) as the n-type material and Cu2SnS3 (CTS) as the p-type material, while soda lime glass (SLG) was implemented as the substrate. Utilizing these eco-friendly materials addresses a pressing need for sustainable solutions in TEG technology and aligns with modern environmental and cost-reduction goals. Machine learning techniques, specifically polynomial regression and Differential Evolution implemented in Python, were applied to identify the optimal geometric parameters of the TEG, aiming for peak electrical power. Experimental correlations for the temperature-dependent electrical and thermal properties were derived and integrated into a simulation model created in COMSOL Multiphysics. These simulations were validated with prior experimental data, showing excellent accuracy, which supports the model's reliability in predicting TEG performance. The simulation examined a range of parameter values, revealing that a greater width of the n-type CAFS legs, coupled with a thinner SLG substrate, significantly enhances power output. Additionally, configuring the TEG with an extended length of the hot surface and a shorter insulated section further increased performance. With the optimized parameters, the TEG achieved a power output of 2.372 nW—an improvement of approximately 17.68 % compared to the maximum power from the initial optimization data. The optimized thin-film TEG design generates power solar-assisted sensor, utilizing a concentrated solar system for the hot side and an air-based heat sink for the cold side to create the required temperature differential.