Solution of Nonsteady Inverse Coefficient Problems for a Thermoelastic Layer by Machine Learning

A method of solving inverse coefficient problems for a thermoelastic layer is proposed on the basis of a physics-informed neural network (PINN). The classical thermoelasticity equations are adopted, with appropriate initial and boundary conditions; observable data are employed. A loss function allows the remainders of the equations and the initial and boundary conditions to be taken into account, along with errors in the specified data. Numerical analysis of a model problem indicates high accuracy of the method in recovering unknown coefficients of the material. The PINN method proves effective and flexible in solving complex continuum mechanics problems.