Abstract
Transient conjugate heat transfer in coupled solid–fluid systems is governed by diffusion, advection, and heat transfer processes that can make repeated simulations expensive and time-consuming in the design phase. This study develops a physics-informed neural network for one-dimensional coupled solid–fluid heat transfer. The model takes space, time, and operating parameters as inputs and predicts the temperature fields in both the solid and fluid domains. The governing equations, heat transfer, boundary conditions, and initial conditions are incorporated into the training through differentiable residual losses, reducing the reliance on high-fidelity training data. The proposed approach is formulated in nondimensional form and evaluated using metrics that quantify conservation, boundary conditions satisfaction, and prediction accuracy. Particular attention is paid to the model behavior within the training range and when operating conditions extend beyond it. The resulting framework provides a basis for the thermal–hydraulic assessment of the performance of superconducting coils for fusion applications in different operating scenarios under evaluation during the design optimization phase using the physics-informed neural network approach. The analyzed case study is the Central Solenoid of the DTT facility currently under construction at ENEA Frascati in Italy.