Design of Experimental System for Simulation of Heat Transfer Phenomenon Based on Physics‐Informed Neural Network

The heat transfer equation is a fundamental mathematical model in physics that describes heat transfer phenomena. However, due to the extremely complex solving process and difficulty in visualizing results when analyzing practical problems, as well as the insufficient application of solving methods for forward and inverse problem analysis, making it difficult to meet the experimental teaching needs effectively. To address these issues, a numerical experimental system for simulating heat transfer phenomena based on Physics‐Informed Neural Network (PINN) was developed using PySide2 and Python. Specifically, the PINN solvers were implemented in PyTorch (via DeepXDE) and integrated into the GUI, training points were generated by Latin hypercube sampling (pyDOE). This system implements PINN algorithms for three different heat transfer equations, providing visualizations of the results for both forward and inverse problems across varying parameters through color‐mapping charts. Additionally, it compares analytical solutions with numerical results and reports residuals, mean square error (MSE), and root mean square error (RMSE). To evaluate the instructional value of the PINN‐based simulation platform, a randomized controlled teaching experiment () was conducted. On the posttest, the experimental group outperformed the control group ( vs. ), and larger pre–post gains were observed (). Self‐assessed understanding (5‐item Likert, 25‐point total) likewise improved more in the experimental group (post vs. ), whereas control‐group gains were not significant (tests: ; Likert: ). System usability and acceptance were rated highly (), indicating strong perceived ease of use and usefulness. Overall, the system was found to provide an interactive and effective learning environment for teaching heat transfer equations and to improve both objective performance and perceived understanding.