進階搜尋


   電子論文尚未授權公開,紙本請查館藏目錄
(※如查詢不到或館藏狀況顯示「閉架不公開」,表示該本論文不在書庫,無法取用。)
系統識別號 U0026-1807201918475300
論文名稱(中文) 應用自組織映射圖於四軸無人機飛行數據分析之診斷系統
論文名稱(英文) Application of Self-Organizing Map on Flight Data Analysis for Quadcopter Health Diagnosis System
校院名稱 成功大學
系所名稱(中) 航空太空工程學系
系所名稱(英) Department of Aeronautics & Astronautics
學年度 107
學期 2
出版年 108
研究生(中文) 鄭德力
研究生(英文) De-Li Cheng
學號 P46061225
學位類別 碩士
語文別 中文
論文頁數 69頁
口試委員 指導教授-賴維祥
口試委員-楊憲東
口試委員-李君謨
中文關鍵字 四軸無人機  失效診斷  振動分析  自組織映射圖  機器學習 
英文關鍵字 UAV  Quadcopter  Fault Detection  Health Diagnosis  Self-Organizing Map 
學科別分類
中文摘要 隨著多軸無人機蓬勃的發展,無人機健康的診斷是無人機發展的重要課題。本研究以無人機動力機構為基底並設定馬達底座鬆脫和螺旋槳破損這兩種失效狀態做為本研究主要探討的因素。首先收集無失效、馬達底座鬆脫和螺旋槳破損的飛行數據並擷取其中的振動訊號分別做方均根、標準差和取樣熵的時域特徵擷取。並以三種狀態之數據做為訓練資料透過自組織映射圖(Self-Organizing Map, SOM)演算法訓練出SOM模型。SOM為非監督式的機器學習演算法同時也是類神經網絡的一種。完成訓練的SOM模型能找出高維度訓練資料中的規則並以神經元的方式保留其拓樸性質,藉此我們便能以保留訓練資料特性的SOM模型並搭配最近鄰居法(K-Nearest Neighbor, KNN)進行失效狀態的分類和失效等級的辨識。
初步的模型對類似的測試資料有非常好的分類效果其準確率甚至高達96%,但後續當加入輕微損傷模型卻無法準確的辨別。因此本研究提出階層SOM模型,透過距離比較,篩選訓練資料,並以此訓練第一層模型,而第一層模型僅處理失效分類的問題;第二層模型僅使用無失效和單個失效狀態之訓練資料訓練而成的,其用做後續失效等級的辨識。藉由階層模型,本研究成功的增加了失效分類的範圍,且第二層模型使用失效狀態的招回率也能做為失效等級的一種判斷指標。
英文摘要 The development of drones is booming day by day, it brings convenience and benefit but it also makes a lot of risk of safety. Therefore, health diagnosis is a crucial issue about drones. This study dedicates to health diagnosis system of structure of quadcopter. Loosening of motor mount and propeller broken are the mainly discussed fault conditions used in this study. In the beginning of the research, the data of the undamaged, loosening of motor mount and propeller broken are acquired. Then, the features of vibration signal are extracted by three methods, root mean square, standard deviation and sample entropy respectively. Next, Self-Organizing Map (SOM) model can be trained by using features which are extracted by the vibration signal. SOM is an unsupervised machine learning method and it’s a type of neuron network. After training by SOM model, the regulation of high dimensionality data can be found, and neurons of SOM can also preserve the topological property of data. Then, KNN (K-Nearest Neighbor) is used to apply SOM model, and do fault classification and fault level recognition. The first result shows the good performance of model which can exceed 96% of precision if the test data is similar with train data. However, the model can’t classify slight fault condition truly. Because of that, this study proposed hierarchical diagnosis model. In the first layer, distance comparison is used to find low gap train data, and SOM model is trained by selected data. First layer model only does fault classification. In the second layer, SOM model is trained by only two conditions, undamaged and one fault condition. Without the interaction of other fault condition, this model can recognize the fault level by the fault recall rate. By hierarchical diagnosis model, the new classification model has the bigger classification range, and the fault recall rate of second layer can be an indicator as the fault level.
論文目次 中文摘要 I
英文摘要 II
誌謝 VI
目錄 VIII
表目錄 XIII
圖目錄 XIV
符號表 XVII
第一章 緒論 1
1.1 前言 1
1.2 文獻回顧 2
1.2.1 多軸無人機重要特徵分析 2
1.2.2 多軸無人機損傷狀態分析 5
1.2.3 自組織映射圖 7
1.2.4 振動分析診斷系統 7
1.2.5 熵 8
1.2.6 文獻回顧總結 9
1.3 研究動機與目的 10
1.4 研究方法與流程 11
1.5 論文架構 13
第二章 無人機振動訊號蒐集與處理 15
2.1 實驗設備 15
2.1.1 四旋翼無人機 15
2.1.2 飛行控制板 16
2.1.3 地面站 17
2.2 飛行實驗 19
2.2.1 實驗之限制條件 19
2.2.2 實驗流程 20
2.3 失效狀態之定義 20
2.3.1 無失效 20
2.3.2 失效狀態之定義 21
2.3.3 馬達底座鬆脫 21
2.3.4 螺旋槳破損 22
2.4 特徵擷取 23
2.4.1 資料前處理 23
2.4.2 方均根 25
2.4.3 標準差 25
2.4.4 取樣熵 26
2.4.5 特徵擷取比較 28
第三章 演算法之介紹與相關應用 31
3.1 自組織映射圖之理論 31
3.2 自組織映射圖演算法之應用 32
3.3 KNN演算法之介紹 38
3.4 KNN輔助SOM分群之模型 39
3.4.1 神經元分群 39
3.4.2 KNN輔助模型之失效判別 41
3.4.3 神經元篩選門檻與KNN交互問題 41
3.5 混淆矩陣之介紹 42
第四章 模型建立與分析 44
4.1 建立分類模型 44
4.1.1 極值去除 44
4.1.2 SOM模型建立 44
4.1.3 混淆矩陣分析 46
4.2 階層診斷模型 50
4.2.1 階層診斷 50
4.2.2 第一層分類模型 51
4.2.3 第二層損傷程度判別模型 56
第五章 實驗結果與分析 57
5.1 模型比較 57
5.1.1 原SOM模型 57
5.1.2 資料篩選後SOM模型 59
5.1.3 各模型分析 61
5.2 損傷參考等級 61
5.2.1 損傷等級 61
5.2.2 馬達底座鬆脫損傷等級 62
5.2.3 槳葉損傷等級 62
第六章 結論與未來工作 65
6.1 結論 65
6.2 未來工作 66
參考文獻 68

參考文獻 [1] Anaya, M., Ceron, H., Vitola, J., Tibaduiza, D. and Pozo, F. “Damage Classification Based on Machine Learning Applications for an Un-Manned Aerial Vehicle Structural Health Monitoring,” 2017.
[2] 江衍函, “間接式切削力與刀具壽命估測技術之研發,” 國立中正大學機械工程研究所碩士論文, 2018.
[3] Ghalamchi, B. and Mueller, M., “Vibration-Based Propeller Fault Diagnosis for Multicopters,” International Conference on Unmanned Aircraft Systems (ICUAS) Dallas, TX, USA, June 12-15, 2018.
[4] Kandaswamy, G. and Balamuralidhar, P., “Health monitoring and Failure Detection of Electronic and Structural Components in Small Unmanned Aerial Vehicles,” World Academy of Science, Engineering and Technology International Journal of Mechanical and Mechatronics Engineering Vol:11, No:5 ,2017.
[5] Kohonen, T., “The Self-Organizing Map. New York, NY, USA: Springer,” 1995.
[6] 林育新, “滾動軸承智能診斷與剩餘壽命預估技術之研發,” 國立中正大學機械工程研究所碩士論文, 2016.
[7] Misra, P., Kandaswamy, G., Mohapatra, P., Kumar, Kriti., Balamuralidhar P., “Structural Health Monitoring of Multi-Rotor Micro Aerial Vehicles,” Embedded Systems and Robotics, TCS Research and Innovation, TATA Consultancy Services Ltd. Bangalore and Chennai, India, 2018.
[8] Olson, I. and Atkins, E. M., “Qualitative failure analysis for a small quadrotor unmanned aircraft system,” In AIAA Guidance, Navigation and Control (GNC) Conference, 2013.
[9] Radkowski S., Szulim P., “Analysis of Vibration of Rotors in Unmanned Aircraft,” 19th International Conference on Methods and Models in Automation and Robotics (MMAR), pp.748–753, Międzyzdroje, Poland, September, 2–5, 2014.
[10] Richman, J. S. and Moorman, J. R., “Physiological Time-Series Analysis Using Approximate Entropy and Sample Entropy,” American Journal of Physiology-Heart and Circulatory Physiology, Volume 278, pp. H2039-H2049, 2000.
[11] Vesanto, J. and Alhoniemi, E., “Clustering of the Self-Organizing Map,” IEEE Transactions. Neural Networks, Volume.11, pp.586–600, 2000.
[12] Yong Keong Yap, “Structural Health Monitoring for Unmanned Aerial Systems,” Technical Report UCB/EECS-2014-70. Electrical Engineering and Computer Sciences University of California at Berkeley, 2014.
[13] Ardupilot, http://ardupilot.org/
[14] Mission Planner, http://ardupilot.org/planner/
[15] 圖3-1取自http://inspirehep.net/record/1273422/plots#0
論文全文使用權限
  • 同意授權校內瀏覽/列印電子全文服務,於2022-08-01起公開。
  • 同意授權校外瀏覽/列印電子全文服務,於2022-08-01起公開。


  • 如您有疑問,請聯絡圖書館
    聯絡電話:(06)2757575#65773
    聯絡E-mail:etds@email.ncku.edu.tw