Hanafi, Thoriq Akbar (2026) Rancang Bangun Sistem AI Berbasis Knowledge Graph dan Graph Retrieval-Augmented Generation (Graph-RAG) Untuk Desain Teknoekonomi, Instrumentasi, Dan Kontrol Kolom Destilasi. Other thesis, Intitut Teknologi Sepuluh Nopember.
This is the latest version of this item.
![[thumbnail of 5009221022-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5009221022-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) |
Abstract
Tahap Front-End Engineering Design (FEED) pada industri proses kimia memerlukan keterpaduan antara desain proses, evaluasi teknoekonomi, serta perencanaan instrumentasi dan kontrol. Pelaksanaan ketiga domain tersebut secara terpisah dapat menimbulkan inkonsistensi antardokumen dan memperpanjang siklus iterasi. Penelitian ini bertujuan merancang, membangun, dan memvalidasi AI Distillation Flowsheet Generator (ADFG), yaitu sistem kecerdasan artifisial hibrida neuro-simbolik untuk membantu desain konseptual kolom distilasi secara terintegrasi. Arsitektur ADFG terdiri atas empat lapisan makro yang direalisasikan melalui tujuh subsistem fungsional, meliputi antarmuka pengguna, layanan API dan orkestrasi, agen AI dan layanan Knowledge Graph–Graph-RAG, mesin kalkulasi deterministik, pembangkitan keluaran, serta penyimpanan data. Mesin rekayasa menerapkan metode Fenske–Underwood–Gilliland–Kirkbride, simulasi MESH Wang–Henke dengan algoritma matriks tridiagonal, perancangan hidraulika, evaluasi teknoekonomi, optimisasi algoritma genetika, rekomendasi kontrol, dan pembangkitan P&ID. Validasi dilakukan melalui enam jalur menggunakan studi kasus Benzena–Toluena dan Propana–Isobutana dengan acuan perhitungan manual, Aspen HYSYS, dan literatur. Pada kasus Benzena–Toluena, deviasi maksimum parameter FUG terhadap HYSYS sebesar 3,89%, tetapi keputusan jumlah tray fisik dan lokasi umpan identik, yaitu 27 tray dan tray umpan ke-17. Simulasi MESH menghasilkan MAE temperatur 0,4341 K dan MAE laju cair 0,0528 kmol/jam, dengan penutupan neraca massa di bawah 0,1%. Optimisasi pada 30 inisialisasi acak menurunkan TAC rata-rata sebesar USD 105.385/tahun atau 4,32% terhadap desain heuristik. Ekstraksi parameter bahasa alami mencapai akurasi 86,83% pada 729 bidang evaluasi. Audit KG menunjukkan bahwa 1.671 relasi pengetahuan memiliki metadata provenansi lengkap dan 1.133 relasi aktif memiliki penunjuk sumber yang teresolusi, masing-masing sebesar 100%. Kondisi KG-ON mengungguli KG-OFF pada dimensi factual correctness sebesar 95,0%, groundedness 98,3%, dan directness 91,7%, sedangkan context precision RAGAS mencapai 0,922. Pengujian P&ID skema LV menghasilkan kepatuhan skema 100%, kecocokan tepat topologi dan kelulusan aturan keselamatan masing-masing 90%, skor F1 node 0,980, serta skor F1 relasi 0,970. Hasil penelitian menunjukkan bahwa ADFG layak digunakan sebagai sistem bantu desain konseptual dan FEED pada ruang lingkup kasus yang diuji, tetapi belum menggantikan simulasi komersial dan verifikasi rekayasa rinci.
======================================================================================================================================
Front-End Engineering Design (FEED) in the chemical process industry requires the integration of process design, techno-economic evaluation, and instrumentation and control planning. Performing these domains separately may create inconsistencies among design deliverables and prolong iteration cycles. This study aims to design, develop, and validate the AI Distillation Flowsheet Generator (ADFG), a hybrid neuro-symbolic artificial intelligence system for integrated conceptual distillation column design. ADFG is organized into four macro-layers implemented through seven functional subsystems covering the user interface, application programming interface and orchestration services, the AI agent and Knowledge Graph–Graph-RAG services, deterministic engineering calculations, output generation, and data persistence. The engineering engine implements the Fenske–Underwood–Gilliland–Kirkbride method, Wang–Henke MESH simulation using a tridiagonal matrix algorithm, hydraulic design, techno-economic evaluation, genetic algorithm optimization, control recommendations, and automatic P&ID generation. Validation was conducted through six complementary pathways using benzene–toluene and propane–isobutane case studies, with manual calculations, Aspen HYSYS, and literature data as references. For the benzene–toluene case, the maximum FUG parameter deviation from HYSYS was 3.89%; however, both approaches produced identical physical design decisions of 27 trays and feed placement at tray 17. The MESH simulation achieved a temperature MAE of 0.4341 K and a liquid-flow MAE of 0.0528 kmol/h, with mass-balance closure below 0.1%. Across 30 independent random seeds, optimization reduced TAC by an average of USD 105,385/year, or 4.32%, relative to the heuristic design. Natural-language parameter extraction achieved 86.83% accuracy across 729 evaluated fields. The KG audit showed complete provenance metadata for 1,671 knowledge relations and fully resolved source locators for 1,133 active retrieval relations. KG-ON outperformed KG-OFF in factual correctness (95.0%), groundedness (98.3%), and directness (91.7%), while RAGAS context precision reached 0.922. Agentic testing of the primary LV P&ID scheme achieved 100% schema compliance, 90% topology exact match, 90% safety-rule pass, a Node F1 score of 0.980, and an Edge F1 score of 0.970. These results indicate that ADFG is suitable as a conceptual-design and FEED assistance system within the validated scope, but it does not replace commercial simulation or detailed engineering verification.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Kecerdasan Buatan, Knowledge Graph-Graph RAG, Perancangan Rekayasa, Process Control. Artificial Intelligence, Engineering Design, Knowledge Graph-Graph RAG, Process Control |
| 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 > TP156 Crystallization. Extraction (Chemistry). Fermentation. Distillation. Emulsions. |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis |
| Depositing User: | Thoriq Akbar Hanafi |
| Date Deposited: | 03 Aug 2026 03:48 |
| Last Modified: | 03 Aug 2026 03:48 |
| URI: | http://repository.its.ac.id/id/eprint/142000 |
Available Versions of this Item
-
Rancang Bangun Sistem AI Berbasis Knowledge Graph dan Graph Retrieval-Augmented Generation (Graph-RAG) Untuk Desain Teknoekonomi, Instrumentasi, Dan Kontrol Kolom Destilasi. (deposited 03 Aug 2026 01:58)
- Rancang Bangun Sistem AI Berbasis Knowledge Graph dan Graph Retrieval-Augmented Generation (Graph-RAG) Untuk Desain Teknoekonomi, Instrumentasi, Dan Kontrol Kolom Destilasi. (deposited 03 Aug 2026 03:48) [Currently Displayed]
Actions (login required)
 |
View Item |