Framework Pemodelan dan Simulasi Konduksi Panas 2D Steady-State Menggunakan Physics-Informed Neural Networks: Aplikasi Thermal-Barrier Coating dan Komposit Serat-Matriks dengan Hambatan Termal Kapitza

Yusuf, Saifulloh (2026) Framework Pemodelan dan Simulasi Konduksi Panas 2D Steady-State Menggunakan Physics-Informed Neural Networks: Aplikasi Thermal-Barrier Coating dan Komposit Serat-Matriks dengan Hambatan Termal Kapitza. Other thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 5011221102-Undergraduate_Thesis.pdf] Text
5011221102-Undergraduate_Thesis.pdf
Restricted to Repository staff only

Download (9MB) | Request a copy

Abstract

Penelitian ini mengembangkan kerangka Physics-Informed Neural Network untuk mensimulasikan konduksi panas dua dimensi steady-state pada domain dengan diskontinuitas sifat material pada antarmuka. Kerangka divalidasi pada sistem Thermal Barrier Coating multilayer dengan antarmuka sinusoidal, kemudian diterapkan pada studi parametrik komposit serat-matriks dengan hambatan termal Kapitza. Implementasi menggunakan arsitektur jaringan terpisah per subdomain dengan Fourier feature encoding dan strategi curriculum learning tiga fase diikuti optimasi L-BFGS. Model menunjukkan akurasi baik pada ketiga kasus validasi. Pada kasus TBC, L2 Relative Error berkisar 0,463%–1,804% untuk variasi amplitudo antarmuka sinusoidal. Pada kasus single-fiber, error berada di rentang 0,0002%–1,9285% untuk seluruh kombinasi rasio konduktivitas κ dan parameter Kapitza α. Pada kasus multifiber, akurasi sangat tinggi untuk α = 0 (di bawah 0,136%), namun meningkat hingga 4,83% pada α = 10 seiring meningkatnya kompleksitas diskontinuitas suhu pada keempat antarmuka. Studi parametrik menunjukkan bahwa peningkatan α menurunkan κ eff secara konsisten hingga 67,05% pada multifiber (κ=100), dengan hambatan Kapitza tinggi terbukti menekan kontribusi konduktivitas serat sehingga κ eff mendekati nilai matriks terlepas dari besarnya κ.
=========================================================================================================================================
This study develops a Physics-Informed Neural Network framework to simulate two-dimensional steady-state heat conduction in domains with material property discontinuities at interfaces. The framework was validated on a multilayer Thermal Barrier Coating system with a sinusoidal interface, then applied to a parametric study of fiber-matrix composites with Kapitza thermal resistance. The implementation employs a separate network architecture per subdomain with Fourier feature encoding and a three-phase curriculum learning strategy followed by L-BFGS optimization. The model demonstrates good accuracy across all three validation cases. For the TBC case, the L2 Relative Error ranges from 0.463% to 1.804% across sinusoidal interface amplitude variations. For the single-fiber case, the error falls within 0.0002%–1.9285% across all combinations of conductivity ratio κ and Kapitza parameter α. For the multifiber case, accuracy is very high at α = 0 (below 0.136%), but increases to 4.83% at α = 10 as the complexity of temperature discontinuities across the four interfaces grows. The parametric study reveals that increasing α consistently reduces κ_eff by up to 67.05% in the multifiber case (κ=100), with high Kapitza resistance suppressing fiber conductivity contributions such that κ_eff approaches the matrix value regardless of κ magnitude.

Item Type: Thesis (Other)
Uncontrolled Keywords: Physics-Informed Neural Networks, Konduksi Panas, Thermal Barrier Coating, Komposit Serat-Matriks, Hambatan Termal Kapitza.
Subjects: T Technology > TA Engineering (General). Civil engineering (General) > TA418.9 Composite materials. Laminated materials.
Divisions: Faculty of Industrial Technology and Systems Engineering (INDSYS) > Material & Metallurgical Engineering > 28201-(S1) Undergraduate Thesis
Depositing User: Saifulloh Yusuf
Date Deposited: 23 Jul 2026 03:48
Last Modified: 23 Jul 2026 03:48
URI: http://repository.its.ac.id/id/eprint/136362

Actions (login required)

View Item View Item