Parts | Days | Selection | Search | Updates | Downloads | Help

AKE: Arbeitskreis Energie

AKE 1: Innovative Contributions for the Energy System Transformation

AKE 1.3: Talk

Tuesday, March 17, 2026, 11:45–12:00, KH 01.022

Comparison of Surrogate Models Trained with Small Data Sets for a Thermoelectric Energy Harvester — •Eugen Vambolt¹, Niklas Pöpel¹, Johannes Wieczorek², Mahla Moazamigoudarzi², Johannes Rothmayr², and Jan Lohbreier¹ — ¹Technische Hochschule Nürnberg Georg Simon Ohm — ²Fraunhofer Institute for Integrated Circuits (IIS) Nürnberg

Energy harvesting technologies can be used to cover the energy consumption of wireless sensor nodes. This enables the use of sensor nodes in locations without a wired power supply. It also eliminates the need to charge or replace batteries, which can quickly become an economic obstacle when a large number of sensors are installed. For energy harvesting technologies to work, there must be a sufficient energy source near the sensor node. For example, temperature differences are essential for the use of thermoelectric generators (TEG) so that electrical energy can be harvested. Locations where such temperature differences can be found typically occur on supply lines or on machines that generate waste heat. The size of the required energy harvesting system depends on the energy consumption of the sensor node and the amount of available ambient energy. The design of a TEG can be optimized for a specific application to meet the energy requirements of the sensor node. To do this, a sensitivity analysis is performed to determine the factors that have the greatest influence on, for example, performance. A surrogate model is usually used in this step, as a large number of predictions have to be calculated. In this thesis, three approaches for surrogate models are investigated and compared with each other.

Keywords: Surrogate models; Polynomial Chaos Expansion; Gaussian Progress Regression; Deep Neural Network; Thermoelectric Generantor