High-power fiber components, such as signal–pump combiners, demand rigorous thermal and optical performance to support high-power fiber lasers and amplifiers. This work introduces an integrated modeling framework that combines the beam propagation method (BPM), an axisymmetric thermal model, and a physics-informed neural network (PINN) to analyze and predict the performance of an active side-pumping fiber combiner. The combiner, based on double-clad Yb-doped large-mode-area fibers and multimode pump fibers, is modeled to capture optical power distribution, heat generation in the doped core and cladding, and bend-induced coupling losses critical for packaging. To mitigate the high computational cost of BPM simulations, a PINN surrogate model is developed to rapidly estimate conversion efficiency with minimal accuracy loss. This hybrid physics–ML approach accelerates design cycles, reduces computation, and preserves predictive fidelity, offering a scalable pathway for optimizing next-generation high-power fiber systems.
Qiu et al. (Thu,) studied this question.