透過您的圖書館登入
IP:3.133.160.156
  • 學位論文

以類神經網路技術建構飛機故障自動分類系統

Using Neural Network technology to struct Aircraft Maintenance’s Automatic Classified System

指導教授 : 李維平
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


由於跨地域業務往來的需要,藉由飛機做為交通運輸工具的人也越來越多,故對於以空中運輸為主要業務的航空公司而言,確保航機能夠準時且安全的運作即為首要之務,而航空公司的維修部門則肩負了這項工作的重責大任。傳統上,修護工作係基於人工經驗來檢視及排除故障,此方式須依靠豐富的修護經驗而造成修護人員養成不易,且在需要修護效率的機坪維護工作上,經常無法滿足需求而造成飛機的延誤,進而影響航空公司的營運成本,因此,若能建立一故障自動分類系統,將會對飛機的維修人員與維修效率的提升有很大的幫助。 目前在飛機維修知識推薦的相關研究方面,我們發現過往的研究主要以最近鄰居(K-nearest neighbor)法來建立分類模型。使用此法建立的飛機故障自動分類系統,在推薦維修知識時的正確率為75.8%,我們認為這樣的正確率若能再提升,將會使得此系統的實用性大為提高。 飛機維修知識推薦為一個分類的研究問題,在分類技術的相關領域中,類神經網路(Neural Network)經常被用來解決分類上的問題。本研究嘗試將類神經網路的技術運用在飛機維修資料上,希望探討類神經網路對於文字類型維修資料的適用性,並期望進而提升分類模型的正確率。 在研究設計上,分別使用類神經網路技術中,屬於監督式學習的倒傳遞類神經網路及屬於非監督式學習的競爭式學習網路,來作為建構維修故障自動分類模型的主要方法。我們實際使用了數萬筆的真實維修資料作實驗,實驗結果發現,以本方法所建立的飛機故障自動分類系統,比過去的研究在正確率上提升了約9%,達到84.9%的推薦正確率,因此,本研究的成果將相當適合相關維修故障自動分類系統建構之參考。

並列摘要


Because of international business affair, more and more people take airplane as transportation tool. For any airlines, to guarantee on-time and safety during the flight are very important. Therefore , maintenance division of an airlines became the responsible department for this work. Conventionally, the maintenance work of an aircraft is based on human experience to inspect and solve system faults. This work have to depend on enough experience so that the maintenance technicians are not easy trained to deal with his work well. In addition, the maintenance efficiency can not be satisfied with on-line requirement for airlines. It results an airlines have to spend more costs to the daily operation. Therefore, we try to build an automatic classification system of to maintenance efficiency for maintenance work. According to related work in aircraft knowledge maintenance. The method used was applied K-nearest neighbor method to building the classification system. By using this system model, the accurate rate for the test data is equal to 75.8%. In this paper, we propose an new approach to increase the accurate rate. Because the defect recommendation of an aircraft system is a classified problem, we apply Neural Network to solve the classification problem. In this research, we devoted Neural Network technology apply to Aircraft maintenance data. We also discuss text repair data using by Neural Network and improve accurate rate of the classification system model. In the simulation model, we use the Back-propagation Network of supervised learning network and Competition Network of Unsupervised learning network to construct the classification system model. We use ten thousands of the data from an aircraft maintenance database to train this simulation model. As a result, Aircraft Maintenance’s Automatic Classified System promoted 9% admeasure rate compared with the past system model. This model achieved to 84.9% admeasure rate. So, this model is very satisfied with Aircraft Maintenance’s Automatic Classified System.

參考文獻


[8] 黃天鴻,以知識地圖為基礎發展的網路全迴路學習,中原大學資訊工程學系碩士論文,民92年
[3] 吳澤民,應用資料探勘技術於維修知識推薦之研究-以飛機維修資料庫為例,中原大學資訊管理學系碩士論文,民91年
[20] 魏健宏,應用人工神經網路於先進高速公路交通管理系統之研究,國科會
[15] 楊啟良,發動機氣路多重故障診斷法,國立成功大學航空太空工程學系碩士論文,民90年
[37] Peter Turney and Michael Halasz, “Contextual Normalization Applied to

被引用紀錄


溫智超(2006)。資料採礦技術應用於國內航空市場之顧客價值分析研究〔碩士論文,國立高雄餐旅大學〕。華藝線上圖書館。https://doi.org/10.6825/NKUHT.2006.00007
謝宏其(2014)。應用資料探勘技術對肩關節攝影檢查鑑別診斷差異性之探討-以某區域教學醫院為例〔碩士論文,國立中正大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0033-2110201614000271

延伸閱讀