SavedAdd your journal

Journals / Türkiye İnsansız Hava Araçları Dergisi / 2026 / Sayı: 8

Multi-Agent Architecture in AI-Assisted Decision Support Systems: PRISM-UAV for the Automation of UAV Design Processes

Yapay Zekâ Destekli Karar Destek Sistemlerinde Çoklu-Ajan Mimarisi: İHA Tasarım Süreçlerinin Otomasyonu için PRISM-UAV

Ömer Çağrı Yavuz(ATATÜRK ÜNİVERSİTESİ)0000-0002-6655-3754
Pages
—
Publication Date
2026-10-30
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

All reported performance values were obtained with the LoRA adapter at the 74% training-progression checkpoint (2,266 / 3,062 steps; epoch ≈ 0.74) and therefore represent an interim — not a final-trained — evaluation.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Unmanned Aerial Vehicle (UAV) design is a complex, interdisciplinary domain that requires Multi-Criteria Decision Making (MCDM) processes.
📊
Key Findings & Data:This study presents the PRISM-UAV (Pareto-optimal Reasoning with Integrated Symbolic Modules) system, which transforms complex design requests into a structured Bill of Materials (BOM).
💡
Conclusion & Impact:All reported performance values were obtained with the LoRA adapter at the 74% training-progression checkpoint (2,266 / 3,062 steps; epoch ≈ 0.74) and therefore represent an interim — not a final-trained — evaluation.
🔬Design:Araştırma Makalesi
📍Scope:Comprehensive Sample
⚙️Variables:İnsansız hava araçları (İHA'lar) · orman yangını erken algılama · ÇKKV

Abstract

Unmanned Aerial Vehicle (UAV) design is a complex, interdisciplinary domain that requires Multi-Criteria Decision Making (MCDM) processes. This study presents the PRISM-UAV (Pareto-optimal Reasoning with Integrated Symbolic Modules) system, which transforms complex design requests into a structured Bill of Materials (BOM). The system features the coordination of 28 expert agents (21 main and 7 defense-focused), Neuro-Symbolic validation mechanisms (BEMT and Peukert correction), and compliance verification against international export regulations such as ITAR and EAR. Pareto-optimal results are produced in an eight-dimensional solution space using the NSGA-II algorithm, while hallucinations are mitigated by leveraging academic knowledge bases through a FAISS-based Retrieval-Augmented Generation (RAG) architecture. The hybrid linguistic architecture (Gemini 2.5 and a LoRA fine-tuned DeepSeek-R1-Distill-Qwen-7B) enables domain-specific reasoning with Turkish natural-language inputs. In an evaluation across ten benchmark scenarios with the adapter at 74% training progression, the system achieved an average constraint-detection recall of 96.1%, a design-completeness score of 0.99, a validation pass rate of 96.5%, and an average inference latency of 113 seconds while producing structured BOM outputs. A comparison with the Gemini 2.5 Flash baseline on five scenarios indicates that PRISM-UAV provides complementary contributions through export-compliance verification, symbolic validation, and quantified component-selection layers. All reported performance values were obtained with the LoRA adapter at the 74% training-progression checkpoint (2,266 / 3,062 steps; epoch ≈ 0.74) and therefore represent an interim — not a final-trained — evaluation.

Özet

İnsansız Hava Aracı (İHA) tasarımı, çok-kriterli karar verme (MCDM) süreçleri gerektiren karmaşık ve disiplinler arası bir alandır. Bu çalışma, karmaşık tasarım taleplerini yapılandırılmış bir Malzeme Listesine (BOM) dönüştüren PRISM-UAV sistem önerisini sunmaktadır. Sistem, 28 uzman ajanın (21 ana ve 7 savunma odaklı) koordinasyonu, Neuro-Sembolik doğrulama mekanizmaları (BEMT ve Peukert düzeltmesi) ve ITAR/EAR gibi uluslararası ihracat regülasyonlarına uyum kontrolleriyle öne çıkmaktadır. NSGA-II algoritmasıyla sekiz boyutlu bir çözüm uzayında Pareto-optimal sonuçlar üretilirken, FAISS tabanlı RAG mimarisiyle akademik bilgi tabanından yararlanılarak halüsinasyonlar minimize edilmiştir. Hibrit dil mimarisi (Gemini 2.5 Flash ve LoRA ile ince ayarlanmış DeepSeek-R1-Distill-Qwen-7B) sayesinde Türkçe doğal dil girdileriyle alan-spesifik akıl yürütme başarıyla gerçekleştirilmektedir. On benchmark senaryosunda %74 eğitim ilerlemesi düzeyindeki adaptör ile yapılan değerlendirmede sistem; ortalama %96,1 kısıt tespit geri çağırma oranı, 0,99 tasarım tamlık skoru, %96,5 doğrulama geçiş oranı ve 113 saniyelik ortalama çıkarım gecikmesi ile yapılandırılmış BOM üretmiştir. Sunulan tüm performans değerleri, eğitim süreci %74 düzeyinde (2.266/3.062 adım; epoch ≈ 0,74) tamamlanmış LoRA adaptörü ile yürütülen ön-final ara değerlendirmeden elde edilmiştir.

🔖

Cite This Article

Yavuz, Ö. Ç. (2026). Multi-Agent Architecture in AI-Assisted Decision Support Systems: PRISM-UAV for the Automation of UAV Design Processes. Türkiye İnsansız Hava Araçları Dergisi, Sayı: 8. https://doi.org/10.51534/tiha.1961991

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (30)

  1. De Raedt, L., Dumančić, S., Manhaeve, R., & Marra, G. (2020). From statistical relational to neuro-symbolic artificial intelligence. Proceedings of the 29th International Joint Conference on Artificial Intelligence (IJCAI), 4943–4950. https://doi.org/10.24963/ijcai.2020/688
  2. Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. (2002). A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2), 182–197. https://doi.org/10.1109/4235.996017
  3. DeepSeek-AI. (2025). DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning (arXiv:2501.12948). arXiv. https://arxiv.org/abs/2501.12948
  4. Dettmers, T., Pagnoni, A., Holtzman, A., & Zettlemoyer, L. (2023). QLoRA: Efficient finetuning of quantized LLMs. Advances in Neural Information Processing Systems, 36.
  5. Doerffel, D., & Sharkh, S. A. (2006). A critical review of using the Peukert equation for determining the remaining capacity of lead-acid and lithium-ion batteries. Journal of Power Sources, 155(2), 395–400. https://doi.org/10.1016/j.jpowsour.2005.04.030
  6. Drone Industry Insights. (2025). Global UAV market report 2025. Drone Industry Insights GmbH. Retrieved May 15, 2026, from https://droneii.com/the-drone-market-report
  7. Güdek, B., Ahmayev, S., & Peker, S. (2025). A bibliometric analysis on the use of unmanned aerial vehicles: A comparative study of Russia and Türkiye. Turkish Journal of Unmanned Aerial Vehicles, 7(1), 1–15. https://doi.org/10.51534/tiha.1660404
  8. Hinton, G., Vinyals, O., & Dean, J. (2014). Distilling the knowledge in a neural network. Proceedings of the NeurIPS Workshop on Deep Learning and Representation Learning.
  9. Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W. (2022). LoRA: Low-rank adaptation of large language models. International Conference on Learning Representations (ICLR).
  10. Hwang, C. L., & Yoon, K. (1981). Multiple attribute decision making: Methods and applications. Springer-Verlag.
  11. INCOSE. (2023). INCOSE systems engineering handbook (5th ed.). International Council on Systems Engineering. John Wiley & Sons.
  12. Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), 1–38.
  13. Johnson, J., Douze, M., & Jégou, H. (2021). Billion-scale similarity search with GPUs. IEEE Transactions on Big Data, 7(3), 535–547. https://doi.org/10.1109/TBDATA.2019.2921572
  14. Kasapoglu, C. (2022). Techno-geopolitics and the Turkish way of drone warfare. Atlantic Council. https://www.atlanticcouncil.org
  15. Leishman, J. G. (2006). Principles of helicopter aerodynamics (2nd ed.). Cambridge University Press.
  16. Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W., Rocktäschel, T., Riedel, S., & Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33, 9459–9474.
  17. Malkov, Y. A., & Yashunin, D. A. (2020). Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(4), 824–836. https://doi.org/10.1109/TPAMI.2018.2889473
  18. Meurer, A., Smith, C. P., Paprocki, M., Čertík, O., Kirpichev, S. B., Rocklin, M., Kumar, A., Ivanov, S., Moore, J. K., Singh, S., Rathnayake, T., Vig, S., Granger, B. E., Muller, R. P., Bonazzi, F., Gupta, H., Vats, S., Johansson, F., Pedregosa, F., … Scopatz, A. (2017). SymPy: Symbolic computing in Python. PeerJ Computer Science, 3, e103. https://doi.org/10.7717/peerj-cs.103
  19. Nonaka, I., & Takeuchi, H. (1995). The knowledge-creating company: How Japanese companies create the dynamics of innovation. Oxford University Press.
  20. Opricovic, S. (1998). Multicriteria optimization of civil engineering systems [Doctoral dissertation, University of Belgrade, Faculty of Civil Engineering].
  21. Reimers, N., & Gurevych, I. (2019). Sentence-BERT: Sentence embeddings using siamese BERT-networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), 3982–3992.
  22. Roskam, J. (1985). Airplane design: Part I — Preliminary sizing of airplanes. Roskam Aviation and Engineering Corporation.
  23. Saaty, T. L. (1980). The analytic hierarchy process: Planning, priority setting, resource allocation. McGraw-Hill.
  24. Sałabun, W. (2015). The characteristic objects method: A new distance-based approach to multicriteria decision-making problems. Journal of Multi-Criteria Decision Analysis, 22(1–2), 37–50.
  25. Schweter, S. (2020). BERTurk — BERT models for Turkish [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.3770924
  26. Trendyol Tech. (2024). Trendyol-LLM-7b-base-v1.0 [Computer software]. Hugging Face. https://huggingface.co/Trendyol/Trendyol-LLM-7b-base-v1.0
  27. Wang, W., Wei, F., Dong, L., Bao, H., Yang, N., & Zhou, M. (2020). MiniLM: Deep self-attention distillation for task-agnostic compression of pre-trained transformers. Advances in Neural Information Processing Systems, 33, 5776–5788.
  28. Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022). Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35, 24824–24837.
  29. Wu, Q., Bansal, G., Zhang, J., Wu, Y., Zhang, S., Zhu, E., Li, B., Jiang, L., Zhang, X., & Wang, C. (2023). AutoGen: Enabling next-gen LLM applications via multi-agent conversation (arXiv:2308.08155). arXiv. https://arxiv.org/abs/2308.08155
  30. ytu-ce-cosmos. (2024). Turkish-Llama-8b-v0.1 [Computer software]. Hugging Face. https://huggingface.co/ytu-ce-cosmos/Turkish-Llama-8b-v0.1