Physics-Informed Neural Network Untuk Simulasi 2d Steady-State Heat Conduction Pada Material Berpori Dengan Variasi Boundary Conditions

Destrio, Mohammad Kelvin (2026) Physics-Informed Neural Network Untuk Simulasi 2d Steady-State Heat Conduction Pada Material Berpori Dengan Variasi Boundary Conditions. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5011221070_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (14MB) | Request a copy

Abstract

Material berpori memiliki mikrostruktur yang menyebabkan jalur konduksi panas menjadi tidak homogen, sehingga distribusi temperatur dan fluks panas sangat dipengaruhi oleh geometri pori dan tingkat porositas. Penelitian ini mengkaji kinerja Variational Physics-Informed Neural Network untuk menyelesaikan persoalan konduksi panas tunak dua dimensi pada media berpori dengan konduktivitas termal heterogen k(x,y), serta membandingkannya terhadap solusi Finite Element Method (FEM) sebagai acuan. Variasi mikrostruktur dilakukan menggunakan empat geometri pori (A–D) dan tiga tingkat porositas (30%, 40%, 50%), dengan beberapa skenario kondisi batas (Kasus I–III). Formulasi vPINN dibangun dalam bentuk weak form melalui integrasi residual dengan fungsi uji, sehingga lebih stabil untuk koefisien yang dapat berubah tajam antar-fasa. Evaluasi kuantitatif dilakukan menggunakan metrik Mean Square Error (MSE), Mean Absolute Error (MAE), Maximum Absolute Error (MaxAE), dan Relative L2 Error (RelL2), serta analisis kualitatif melalui peta temperatur dan peta fluks panas. Hasil menunjukkan vPINN mampu mereplikasi solusi FEM dengan baik pada media homogen sebagai baseline, kemudian mempertahankan konsistensi pola medan temperatur pada media berpori dengan nilai error global pada Kasus I = 1.09% - 2.19%, Kasus II = 1.41% - 9.19% , dan Kasus III = 3.27% - 7.12%. Secara umum, peningkatan porositas dan kompleksitas geometri memperkuat pembelokan jalur fluks serta meningkatkan error lokal terutama di area gradien tinggi dan dekat batas. Dari sisi komputasi, waktu vPINN didominasi proses pelatihan dan meningkat dengan penambahan collocation point.
========================================================================================================================================
Porous materials have microstructures that make heat-conduction paths non-uniform, so temperature distribution and heat flux are strongly influenced by pore geometry and porosity level. This study examines the performance of a Variational Physics-Informed Neural Network to solve the two-dimensional steady-state heat-conduction problem in porous media with heterogeneous thermal conductivity k(x,y), and compares it with the Finite Element Method (FEM) solution as a reference. Microstructure variations are generated using four pore geometries (A–D) and three porosity levels (30%, 40%, 50%), with several boundary-condition scenarios (Case I–III). The vPINN formulation is constructed in a weak form by integrating the residual with test functions, making it more stable for coefficients that can change sharply across phases. Quantitative evaluation is carried out using the Mean Square Error (MSE), Mean Absolute Error (MAE), Maximum Absolute Error (MaxAE), and Relative L2 Error (RelL2) metrics, along with qualitative analysis through temperature maps and heat-flux maps. The results show that vPINN can replicate the FEM solution well for homogeneous media as a baseline, and then maintain consistent temperature-field patterns in porous media with global error values of Case I = 1.09%–2.19%, Case II = 1.41%–9.19%, and Case III = 3.27%–7.12%. In general, increasing porosity and geometric complexity strengthens flux-path deflection and increases local error, especially in high-gradient regions and near boundaries. From a computational perspective, vPINN time is dominated by the training process and increases with the addition of collocation points.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	vPINN, FEM, konduksi panas, media berpori, mikrostruktur, weak form, Physics-Informed Neural Network, Heat Conduction, Porous Materials, Boundary Conditions, Numerical Simulation
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA371 Differential equations--Numerical solutions Q Science > QC Physics > QC173.4.C63 Composite materials Q Science > QC Physics > QC320 Heat transfer
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Material & Metallurgical Engineering > 28201-(S1) Undergraduate Thesis
Depositing User:	Mohammad Kelvin Destrio
Date Deposited:	27 Jan 2026 06:24
Last Modified:	27 Jan 2026 06:24
URI:	http://repository.its.ac.id/id/eprint/130406

Actions (login required)

View Item