以裂縫自動探索為導向之無人機整合飛控系統研發

Abstract

在鋼筋混凝土建築中，裂縫不僅是結構健康與安全的重要指標，更是地震後評估與補強工作中至關重要的信息來源。地震發生時，結構工程師需迅速評估裂縫的尺寸、形態及位置，以判斷是否存在即時的結構危害，並制定相應的補強策略。因此，裂縫檢測在現代社會中具有重要意義，這項技術不僅影響到建築物的長期穩定性，也直接關乎公眾的安全與福祉。近年來，隨著硬體技術的迅速進步及人力成本的提高，無人機技術開始被廣泛應用於裂縫檢測領域。儘管學術界已有大量相關研究，但許多應用仍限於無人機作為資料收集平台，未能實現裂縫的自動捕捉及自主飛行。本研究針對此問題，致力於開發一種新型無人機飛行系統，專為實現裂縫的自動捕捉和強化自主飛行能力而設計。首先，我們採用深度強化學習方法對模型進行精細訓練，使其能夠基於局部裂縫特徵有效判斷最佳飛行路徑。其次，為實現無人機的自主飛行，我們運用MAVLINK無人機通信協議編寫了專屬的飛行控制指令，以確保系統在各種環境條件下能夠靈活適應並執行任務。同時，我們在系統設計中融合了先進的軟體與硬體技術，保證無人機在飛行過程中能夠即時進行邊緣運算，實現高效的全自動操作。實驗結果表明，該系統成功捕捉並分析了62%的可見裂縫，檢測準確率為86%。這些發現突顯了該系統在提高結構檢測的可靠性和效率方面的潛力。

In reinforced concrete structures, cracks serve not only as crucial indicators of structural health and safety but also as vital sources of information for post-earthquake assessment and reinforcement planning. During seismic events, structural engineers must swiftly assess the size, morphology, and locations of cracks to determine immediate structural hazards and devise appropriate reinforcement strategies. Hence, crack detection holds significant importance in modern society, impacting both the long-term stability of buildings and public safety and well-being directly. In recent years, with rapid advancements in hardware technology and the increasing costs of human labor, unmanned aerial vehicle (UAV) technology has found widespread application in the field of crack detection. Despite extensive academic research, many applications still confine UAVs to data collection platforms, failing to achieve automated crack detection and autonomous flight. Addressing this gap, this study focuses on developing a novel UAV flight system designed specifically for automated crack detection and enhanced autonomous flight capabilities. Firstly, we employ a deep reinforcement learning approach to finely train the model, enabling it to effectively determine optimal flight paths based on local crack features. Secondly, to achieve autonomous flight, we utilize MAVLink UAV communication protocol to develop tailored flight control commands, ensuring the system's adaptability and mission execution across diverse environmental conditions. Concurrently, our system design integrates advanced software and hardware technologies to facilitate real-time edge computing during UAV flight, thereby achieving efficient fully automated operations. Experimental results indicate that the system successfully captured and analyzed 62% of visible cracks, with a detection accuracy rate of 86%. These findings underscore the system's potential in improving the reliability and efficiency of structural inspections.