Predicting the transient burning of non-charring materials using physics-informed neural networks

Abstract

This paper explores the application of Physics-Informed Neural Networks (PINNs) for predicting transient burning rates in Non-Charring materials. By integrating physical principles with deep learning models, PINNs provide an efficient solution, requiring significantly fewer data points compared to traditional numerical methods. For instance, solving the problem numerically with 2000 spatial mesh points would require approximately 131 million data points, whereas the PINNs model used in this study reduced the data points to less than 60,000, while maintaining less than 2 % error in predictive accuracy. The results demonstrate that PINNs can effectively capture transient phenomena, such as burning rate overshoot and undershoot caused by abrupt changes in convective heat flux, offering critical insights into the behavior of materials under varying thermal conditions. The study also highlights opportunities for further improvements in model accuracy and stability, particularly in cases with sparse or noisy data. This methodology holds potential for broader applications, including pyrolysis analysis, combustion processes, and fluid dynamics, showcasing the flexibility and computational efficiency of PINNs in addressing complex dynamic problems.