SOLVING TRANSIENT INVERSE HEAT TRANSFER PROBLEMS IN COMPLEX GEOMETRIES USING PHYSICS-GUIDED NEURAL NETWORKS (PGNN)

Abstract

Temporally and spatially unstable thermal conditions lead to transient or inhomogeneous thermo-elastic behavior of workpieces during manufacturing or geometric inspection. Temperature monitoring by means of sensors consign transient temperature fields, but do not yield information about the heat flow acting as thermal boundary condition, which is a relevant input parameter for nearly any thermal simulation. Addressing the need for efficient methods, the authors propose an approach to solve inverse heat transfer problems in complex geometries. In the presented study, locally acting heat loads are experimentally investigated based on virtual demonstrators running in FEM. The conducted method shows high potential for transient heat flow modelling in terms of accuracy and computational efficiency.