Adaptive Fourier Feature Physics-Informed Neural Networks for Nonlinear Cattaneo–Vernotte Heat Conduction with Temperature-Dependent Properties
محل انتشار: مجله مکانیک کاربردی محاسباتی، دوره: 57، شماره: 4
سال انتشار: 1405
نوع سند: مقاله ژورنالی
زبان: انگلیسی
مشاهده: 31
فایل این مقاله در 24 صفحه با فرمت PDF قابل دریافت می باشد
- صدور گواهی نمایه سازی
- من نویسنده این مقاله هستم
استخراج به نرم افزارهای پژوهشی:
شناسه ملی سند علمی:
JR_JCAM-57-4_005
تاریخ نمایه سازی: 22 آبان 1405
چکیده مقاله:
In this study, Physics-Informed Neural Networks (PINNs) are employed as mesh-free solvers for nonlinear non-Fourier heat conduction governed by the Cattaneo–Vernotte (CV) equation with temperature-dependent thermal conductivity. The CV formulation accounts for finite thermal relaxation effects and leads to a hyperbolic heat conduction model, for which analytical solutions are generally unavailable in the nonlinear regime. Three PINN architectures are investigated: a standard PINN (S-PINN), a Fourier Feature PINN (F-PINN), and a novel Adaptive Fourier Feature PINN (AF-PINN). The models are systematically evaluated in both linear and nonlinear regimes and validated against analytical and high-resolution numerical reference solutions. The results demonstrate that fixed Fourier feature mappings substantially improve accuracy compared to standard PINNs; however, their performance is highly sensitive to the choice of spectral bandwidth and may deteriorate under non-optimal initialization, particularly in nonlinear cases. The proposed AF-PINN overcomes this limitation by dynamically adapting the Fourier bandwidth during training, yielding robust convergence across a wide range of initial conditions. Statistical analyses based on multiple random initializations confirm consistently low prediction error and reduced variance. Further examination of the optimization dynamics highlights the critical role of decoupled learning rates for the spectral scale and network weights in enabling recovery from adverse high-frequency initializations. Overall, the adaptive framework enhances the accuracy, stability, and robustness of PINN-based solvers for nonlinear non-Fourier heat transfer and provides a reliable foundation for modeling wave-dominated and multiscale thermal phenomena.
کلیدواژه ها:
نویسندگان
- -
Mechanical Engineering Department, Amirkabir University of Technology, Tehran ۱۵۸۷۵-۴۴۱۳, Iran
- -
Department of Automotive Engineering, Faculty of Engineering, Atilim University, ۰۶۸۳۰, Ankara, Türkiye
- -
Mechanical Engineering Department, Amirkabir University of Technology, Tehran ۱۵۸۷۵-۴۴۱۳, Iran
مراجع و منابع این مقاله:
لیست زیر مراجع و منابع استفاده شده در این مقاله را نمایش می دهد. این مراجع به صورت کاملا ماشینی و بر اساس هوش مصنوعی استخراج شده اند و لذا ممکن است دارای اشکالاتی باشند که به مرور زمان دقت استخراج این محتوا افزایش می یابد. مراجعی که مقالات مربوط به آنها در سیویلیکا نمایه شده و پیدا شده اند، به خود مقاله لینک شده اند :