PHYSICS-INFORMED NEURAL NETWORK FOR INVERSE HEAT CONDUCTION PROBLEM

A physics-informed neural network is developed to infer the unknown heat flux in a 1D inverse heat conduction problem. This is achieved by training the neural network by physics constraints including the governing equation, boundary and initial conditions, and sampled temperature data. When the total training loss is small enough, the neural network can approximate the heat conduction and the heat flux can be obtained from the neural network. The prediction performances of the physics-informed neural network have been examined using different network structures, different activation functions, and different forms of unknown heat flux. The results show that the physics-informed neural network has an overall satisfactory performance in predicting the unknown heat fluxes of different forms and predicting heat fluxes using temperature data with random errors. The present work demonstrates that the physics-informed neural network is a promising approach for solving inverse heat conduction problems with good accuracy and fast efficiency.