Desain Sistem Artificial Intelligence Berbasis GraphRAG untuk Perancangan TEG Contactor dan Instrumentasi Kontrol pada Dehidrasi Gas Alam dengan Pendekatan Evaluasi Teknoekonomi

Indrasandi, Yumna Almira (2026) Desain Sistem Artificial Intelligence Berbasis GraphRAG untuk Perancangan TEG Contactor dan Instrumentasi Kontrol pada Dehidrasi Gas Alam dengan Pendekatan Evaluasi Teknoekonomi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009221002-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Perancangan peralatan proses seperti TEG contactor memerlukan banyak perhitungan iteratif dengan parameter yang saling bergantung. Kondisi ini dapat meningkatkan risiko kesalahan, ketidakkonsistenan asumsi, dan lamanya proses revisi. Penelitian ini mengembangkan TEG.AI sebagai platform berbasis web yang mengintegrasikan Graph Retrieval-Augmented Generation atau GraphRAG dengan modul sizing TEG contactor, evaluasi hidraulika, analisis teknoekonomi, optimisasi Genetic Algorithm, dan generasi diagram P&ID dalam satu antarmuka percakapan. Knowledge graph dibangun menggunakan Neo4j, sedangkan LLaMA-3 70B melalui Groq API digunakan sebagai model generatif. Validasi dilakukan terhadap pengenalan intent, perhitungan sizing dan hidraulika, analisis teknoekonomi, optimisasi GA, generasi flowsheet, serta kualitas retrieval. Evaluasi GraphRAG menggunakan framework RAGAs terhadap 20 pertanyaan mengenai dehidrasi TEG. Hasil menunjukkan bahwa deviasi geometri dan neraca massa berada di bawah 5%. Deviasi entalpi lean TEG mencapai 7,297%, sedangkan deviasi rata-rata profil net liquid mencapai 29,089%. Perbedaan tersebut terutama disebabkan oleh perbedaan definisi fase cair dan tingkat kompleksitas model kesetimbangan. Modul teknoekonomi menghasilkan deviasi maksimum 0,75% terhadap Excel. Optimisasi GA menurunkan TAC sebesar 26% dari kondisi awal. Variabel keputusan berbeda kurang dari 0,5% dibandingkan MATLAB, tetapi nilai TAC berbeda 17,23% karena basis komponen biaya belum sepenuhnya sama. GraphRAG mengungguli RAG konvensional pada seluruh metrik, termasuk peningkatan context recall sebesar 90% dan context precision sebesar 58,3%. Modul P&ID juga berhasil menambahkan elemen kontrol PC dan LC secara bertahap tanpa mengganggu diagram sebelumnya. Secara keseluruhan, TEG.AI mampu mengintegrasikan basis pengetahuan graf dan komputasi teknik dalam satu platform pendukung perancangan proses.
=====================================================================================================================================
Process equipment design, including TEG contactors, requires iterative calculations involving multiple interdependent parameters. This complexity may increase the risk of calculation errors, inconsistent assumptions, and prolonged design revisions. This study developed TEG.AI, a web-based platform that integrates Graph Retrieval-Augmented Generation (GraphRAG) with TEG contactor sizing, hydraulic evaluation, techno-economic analysis, Genetic Algorithm (GA) optimization, and P&ID generation within a conversational interface. The knowledge graph was developed using Neo4j, while LLaMA-3 70B, accessed through the Groq API, served as the generative language model. Validation was conducted for intent recognition, sizing and hydraulic calculations, techno-economic analysis, GA optimization, flowsheet generation, and retrieval quality. GraphRAG performance was evaluated using the RAGAs framework on a dataset of 20 questions related to TEG dehydration. The results showed that deviations in equipment geometry and mass balance were below 5%. The deviation in lean TEG enthalpy reached 7.297%, while the average deviation in the net-liquid profile reached 29.089%. These differences were mainly attributed to variations in liquid-phase definitions and the complexity of the equilibrium models. The techno-economic module produced a maximum deviation of 0.75% compared with Excel calculations. GA optimization reduced the Total Annual Cost by 26% relative to the initial condition. The optimized decision variables differed by less than 0.5% from MATLAB results, although the TAC differed by 17.23% because the cost-component bases were not fully equivalent. GraphRAG outperformed conventional RAG across all evaluated metrics, including improvements of 90% in context recall and 58.3% in context precision. The P&ID module also successfully added PC and LC control elements incrementally without disrupting previously generated diagram components. Overall, TEG.AI successfully integrated graph-based knowledge retrieval and engineering computation into a unified process-design support platform.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Artificial Intelligence, Genetic Algorithm, GraphRAG, knowledge graph, Large Language Model, TEG contactor, RAGAs
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T58.64 Information resources management T Technology > TJ Mechanical engineering and machinery > TJ212 Control engineering systems. Automatic machinery (General)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Yumna Almira Indrasandi
Date Deposited:	04 Aug 2026 01:09
Last Modified:	04 Aug 2026 01:09
URI:	http://repository.its.ac.id/id/eprint/142518

Actions (login required)

View Item