Prediksi Evolusi Korosi Multi-Pit Pada Baja SS304 Terlindungi Lapisan Kitosan Dan Kitosan/PTA Menggunakan SHARP Physics Informed Neural Networks (SHARP-PINNs)

Agung, Yudhadarma Rizqi Prawira (2026) Prediksi Evolusi Korosi Multi-Pit Pada Baja SS304 Terlindungi Lapisan Kitosan Dan Kitosan/PTA Menggunakan SHARP Physics Informed Neural Networks (SHARP-PINNs). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5011221077-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (15MB) | Request a copy

Abstract

Kegagalan infrastruktur akibat korosi sumuran pada baja SS304 lebih sering dipicu oleh interaksi antar lubang korosi (multi-pit) yang tumbuh berdekatan di bawah lapisan pelindung yang mengalami cacat mikro, daripada oleh satu Pit tunggal yang terisolasi. Kompleksitas distribusi ion klorida pada kondisi multi-pit dengan variasi pelapis kitosan belum sepenuhnya dapat dimodelkan secara efisien menggunakan metode numerik konvensional berbasis mesh. Penelitian ini bertujuan menganalisis pengaruh koefisien difusi efektif pelapis kitosan dan variasi topologi multi-pit terhadap distribusi konsentrasi klorida, fenomena Shielding effect, dan waktu penyatuan morfologi Pit, sekaligus mengevaluasi akurasi dan efisiensi komputasi arsitektur SHARP Physics Informed Neural Networks (SHARP-PINNs). Pemodelan dilakukan pada domain dua dimensi dengan material baja SS304 dalam elektrolit NaCl 3,5%, menggunakan tiga variasi pelapis (Naked SS, Kitosan/Chi, dan Kitosan/PTA/Chi/PTA) dan empat konfigurasi topologi Pit (Single, Double, Triple, dan Penta-pit). Persamaan difusi Fick II heterogen diselesaikan oleh arsitektur SHARP-PINNs dengan Fourier Feature Embedding, Modified MLP enam lapis, dan strategi Curriculum Learning. Finite Difference Method (FDM) dengan harmonic averaging digunakan sebagai solusi referensi. Akurasi dievaluasi menggunakan Relative L₂ Error, MARE, dan MaxAE pada tiga variasi titik kolokasi (Nf = 20k, 30k, 40k). Hasil simulasi menunjukkan bahwa pelapis Chi/PTA dengan koefisien difusi efektif paling rendah (Dcoat mult = 0,001) memberikan hambatan difusi terkuat, menekan laju penetrasi korosi hingga 0,0111 mmPY atau sekitar 72% lebih rendah dibandingkan Naked SS (0,0396 mmPY), serta memperpanjang waktu mencapai kedalaman kritis dari 178,6 hari menjadi 491 hari. Peningkatan jumlah Pit memperluas zona korosi aktif dan mempercepat coalescence, di mana Double-pit Naked SS mencapai penyatuan pada hari ke-8,2 sedangkan Chi/PTA menundanya hingga hari ke-40,9. Fenomena Shielding effect teridentifikasi melalui penurunan gradien konsentrasi antar-pit hingga 0,096 M/µm pada konfigurasi Triple-pit dengan pelapis Chi. Resolusi Nf = 30k dipilih sebagai kondisi optimal karena menghasilkan L₂ Error pada rentang 0,4-6,41% dengan waktu komputasi 187-1.159 detik, dibandingkan FDM yang hanya memerlukan 6-48 detik namun tidak memiliki fleksibilitas geometri mesh free. Kesimpulan penelitian ini menegaskan bahwa parameter difusi pelapis menjadi faktor penentu utama perkembangan Pit, peningkatan kompleksitas topologi meningkatkan interaksi difusi antar-cacat secara non-linear, dan arsitektur SHARP-PINNs mampu mempertahankan akurasi di bawah 10% pada seluruh konfigurasi sebagai pendekatan mesh free yang adaptif untuk pemodelan korosi multi-pit kompleks.
=======================================================================================================================================
Infrastructure failure due to Pitting corrosion in SS304 steel is more often triggered by the interaction between corrosion Pits (multi-pits) that grow close together under a protective layer that has micro defects, rather than by a Single isolated Pit. The complexity of chloride ion distribution in multi-pit conditions with variations in Kitosan coatings has not been fully modeled efficiently using conventional mesh-based numerical methods. This study aims to analyze the effect of the effective diffusion coefficient of Chitosan coatings and variations in multi-pit topology on the distribution of chloride concentration, the Shielding effect phenomenon, and the coalescence time of Pit morphology, while evaluating the accuracy and computational efficiency of the SHARP Physics Informed Neural Networks (SHARP-PINNs) architecture. Modeling is carried out on a two dimensional domain with SS304 steel material in 3,5% NaCl electrolyte, using three coating variations (Naked SS, Chitosan/Chi, and Chitosan/PTA/Chi/PTA) and four Pit topology configurations (Single, Double, Triple, and Penta-pit). The heterogeneous Fick II diffusion equation is solved by SHARP-PINNs architecture with Fourier Feature Embedding, six-layer Modified MLP, and Curriculum Learning strategy. Finite Difference Method (FDM) with harmonic averaging is used as the reference solution. Accuracy is evaluated using Relative L₂ Error, MARE, and MaxAE at three variations of collocation points (Nf = 20k, 30k, 40k). Simulation results show that Chi/PTA coating with the lowest effective diffusion coefficient (Dcoat mult = 0.001) provides the strongest diffusion barrier, suppresses the corrosion penetration rate to 0,0111 mmPY or about 72% lower than Naked SS (0.0396 mmPY), and extends the time to reach critical Depth from 178,6 days to 491 days. Increasing the number of Pits expands the active corrosion zone and accelerates coalescence, where Double-pit Naked SS reaches coalescence at day 8,2 while Chi/PTA delays it until day 40,9. The Shielding effect phenomenon was identified through a decrease in the inter-pit concentration Gradien to 0,096 M/µm in the Triple-pit configuration with Chi coating. The resolution of Nf = 30k was chosen as the optimal condition because it produces L₂ Error in the range of 0,4-6,41% with a computational time of 187-1.159 seconds, compared to FDM which only requires 6-48 seconds but does not have the flexibility of mesh free geometry. The conclusion of this study confirms that the coating diffusion parameter is the main determining factor for Pit development, increasing topological complexity increases non-linear inter-defect diffusion interactions, and the SHARP-PINNs architecture is able to maintain accuracy below 10% in all configurations as an adaptive mesh free approach for complex multi-pit corrosion modeling.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Physics Informed Neural Networks (PINNs), Korosi Multi-pit, Baja SS304, Pelapis Kitosan, Distribusi Klorida Transien Physics Informed Neural Networks (PINNs), Multi-pit Corrosion, SS304 Steel, Chitosan Coating, Transient Chloride Distribution.
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TA Engineering (General). Civil engineering (General) > TA418.74 Corrosion and anti-corrosives T Technology > TA Engineering (General). Civil engineering (General) > TA467 Iron and Steel Corrosion and protection against corrosion T Technology > TA Engineering (General). Civil engineering (General) > TA491 Metal coating.
Divisions:	Faculty of Industrial Technology > Material & Metallurgical Engineering > 28201-(S1) Undergraduate Thesis
Depositing User:	Yudhadarma Rizqi Prawira Agung
Date Deposited:	20 Jul 2026 02:23
Last Modified:	20 Jul 2026 02:23
URI:	http://repository.its.ac.id/id/eprint/135618

Actions (login required)

View Item