Determining thermomechanical response of thin disk lasers with physics-informed neural networks

Abstract

Acquiring structural behavior related to thermal effects is essential to the design of high-power thin disk lasers. However, quantifying the thermomechanical response of a thin disk laser crystal (TDLC) is challenging since these variables of interest are always highly tensorial. Herein, we developed a method to directly quantify the thermo-induced structural behavior of a TDLC from scalar temperature by a physics-informed neural network (PINN). The PINN is established to approximate the unknown states (thermal displacement, strain, and stress fields) using the loss functions encoded with sparse temperature and thermal mechanics equations. Steady-state and transient results of thermo-induced structural responses are precisely obtained from the temperature field by training PINN. In addition, transfer learning is demonstrated to enhance the computation efficiency of PINN for TDLCs with various materials and geometric properties. Implanting PINNs into thin disk lasers paves the way for precise quantification of highly tensorial thermal effects of lasers in-situ using easily accessible scalars.