A Deep Learning Approach to an Inverse Problem of Heat-Source Detection in a Piezoelectric Semiconductor Plate

A heat source in a piezoelectric semiconductor plate can result in an electric potential well, which blocks the electric current and change its electric behaviors. On the contrary, the measured current fields can be inversely mapped to the position and magnitude of the corresponding heat source, which can be applied to heat-source detection. This heat-source detection problem is difficult to be calculated using traditional computation methods, such as finite element analysis. In this paper, deep learning (DL) methods are investigated to solve the heat-source detection problem. DL models are trained on datasets that contain pairs of induced current fields and heat sources to learn the mapping relationship. The trained DL models can predict the position and magnitude of heat sources given the input current fields. This learning capacity of DL models provides an effective tool for heat-source detection.