Rancang Bangun Sistem AI Berbasis Graph-RAG untuk Desain Packed-Bed Reactor serta Sistem Instrumentasi dan Kontrol pada Proses Naphtha Hydrotreating dengan Pendekatan Teknoekonomi

Shafi, Hafizh Syahsia (2026) Rancang Bangun Sistem AI Berbasis Graph-RAG untuk Desain Packed-Bed Reactor serta Sistem Instrumentasi dan Kontrol pada Proses Naphtha Hydrotreating dengan Pendekatan Teknoekonomi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009221092-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Perancangan awal packed-bed reactor pada proses naphtha hydrotreating memerlukan integrasi pengetahuan proses, perhitungan rekayasa, penyusunan flowsheet, sistem instrumentasi dan kontrol, serta evaluasi ekonomi. Penelitian ini bertujuan merancang dan mengevaluasi sistem kecerdasan buatan berbasis Graph-RAG untuk mendukung tahapan tersebut pada reaction section dengan reaksi hidrodesulfurisasi thiophene. Sistem mengintegrasikan Large Language Model, Knowledge Graph berbasis Neo4j, modul kalkulasi Python, Microsoft Visio, dan Genetic Algorithm. Knowledge Graph menyediakan konteks terstruktur bagi LLM, sedangkan modul kalkulasi menjalankan sizing, validasi, dan optimisasi berdasarkan data serta persamaan yang tersimpan. Hasil pengujian menunjukkan retrieval precision, recall, dan F1-score masing-masing sebesar 0,26, 0,96, dan 0,39, sedangkan generation precision, recall, dan F1-score sebesar 0,90, 0,89, dan 0,89. Evaluasi Knowledge Graph menghasilkan tingkat kelengkapan rata-rata 88,89%. Pada skala industri, sistem menghasilkan volume bed 103,411 m³, massa katalis 77.558,330 kg, diameter bed 2,891 m, panjang bed 15,752 m, dan pressure drop 1,997 bar untuk mencapai konversi thiophene 80%. Hasil sizing menunjukkan kesesuaian terhadap perhitungan manual, data referensi, dan simulasi Aspen HYSYS dengan deviasi dalam batas penerimaan. Sistem juga menghasilkan flowsheet reaction section yang secara topologi identik dengan acuan serta dilengkapi elemen instrumentasi dan kontrol. Optimisasi menurunkan Total Annual Cost sebesar 0,76% dan meningkatkan profit tahunan sebesar 3,62%. Dengan demikian, sistem mampu mendukung desain awal packed-bed reactor secara terstruktur, dapat ditelusuri, terverifikasi, dan lebih ekonomis.
=======================================================================================================================================
The preliminary design of a packed-bed reactor in the naphtha hydrotreating process requires the integration of process knowledge, engineering calculations, flowsheet development, instrumentation and control systems, and economic evaluation. This study aims to design and evaluate a Graph-RAG-based artificial intelligence system to support these activities in the reaction section for thiophene hydrodesulfurization. The system integrates a Large Language Model, a Neo4j-based Knowledge Graph, a Python calculation module, Microsoft Visio, and a Genetic Algorithm. The Knowledge Graph provides structured context for the LLM, while the calculation module performs sizing, validation, and optimization using stored data and equations. Testing produced retrieval precision, recall, and F1-score values of 0.26, 0.96, and 0.39, respectively, while generation precision, recall, and F1-score reached 0.90, 0.89, and 0.89. The Knowledge Graph completeness evaluation yielded an average score of 88.89%. At industrial scale, the system generated a bed volume of 103.411 m³, a catalyst mass of 77,558.330 kg, a bed diameter of 2.891 m, a bed length of 15.752 m, and a pressure drop of 1.997 bar to achieve 80% thiophene conversion. The sizing results agreed with manual calculations, reference data, and Aspen HYSYS simulations within the acceptance limits. The system also generated a reaction-section flowsheet that was topologically identical to the reference and included instrumentation and control elements derived from the Knowledge Graph. Genetic Algorithm optimization reduced the Total Annual Cost by 0.76% and increased annual profit by 3.62% compared with the baseline design, with a maximum deviation of 0.9% from manual calculations. Therefore, the developed system can support preliminary packedbed reactor design in a structured, traceable, verifiable, and more economical manner.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Naphtha Hydrotreating, Hidrodesulfurisasi, Packed-Bed Reactor, Kecerdasan Buatan, Rekayasa Desain, Hydrodesulfurization, Artificial Intelligence, Engineering Design
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TJ Mechanical engineering and machinery > TJ212 Control engineering systems. Automatic machinery (General) T Technology > TP Chemical technology > TP155.5 Chemical plants--Design and construction
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Hafizh Syahsia Shafi
Date Deposited:	02 Aug 2026 15:14
Last Modified:	02 Aug 2026 15:43
URI:	http://repository.its.ac.id/id/eprint/141983

Actions (login required)

View Item