Diagnosing and Predictive Maintenance Systems for Abnormal Behavior of Power Scheduling Loading and Its Applications to Robotics System
           

- 指導教授 黃漢邦 博士 研究生 吳思嫻

- Advisor :Dr.Han-Pang Huang Student : Shih-Hsien Wu

Lab. of Robotics., Department of Mechanical Engineering National Taiwan University Taiwan

Abstract:

Economic development is dependent on power supply. Production, livelihood, and government departments rely on continuous and steady power supply to proceed for economic activities. Through research and development of intelligent robots, design bottlenecks have emerged particularly in the scheme process of emotional sensors of modernized robots or electrical vehicles. Batteries supplied to robots discharge too quickly. Under unstable discharge conditions in a heavy-duty platform, reliable cycle-lifespan is shortened and cannot be assured. Abnormal behavior may influence robot’s demonstration time. Thus, awareness of the battery status and the time for charging are important. In order to save power and promote efficiency, steady power supply depends on power loading management and abnormal behavior diagnosis to construct suitable supply of power system. One of the aims of this thesis is to construct a diagnosing and analyzing system. We collect the data of each motor’s operating condition through assigned scheduling until rescheduling is triggered. To establish an abnormal behavior model that can be applied at an appropriate time to conduct basic rescheduling in accordance with dispatching rules and facilitate better performance, the collected data are classified by two methods: classification tree and self-organizing feature maps (SOM). The second aim is to construct a predictive maintenance system that uses fuzzy inference to predict the of battery power supply level using the collected information of residual power and temperature, and considering the power loss of using cycle and rising temperature of battery. The administrator can easily observe the operating condition of operating robot and battery through the constructed generic message-passing platform (GMPP) web service. Once the diagnosing and analyzing system discovers any abnormal behavior or once the predictive maintenance system detects low level of battery supply, GMPP will send active warning notifications to the engineers to conduct management and repair of the robot’s motors and battery. Finally, we discuss the combination of batteries in a single package considering the limits of real application to make the functions of optimized package meet the purpose of reaching the maximum duration of usage and power.





中文摘要:


電力與經濟發展之關係極為密切,不論是生產部門、家計部門或政府部門,均需仰賴電力持續穩定的供應,以進行正常的經濟活動。而智慧型機器人經歷不斷研發並 加裝情緒感測等愈來愈精密多元的光電儀器,放電量大且靠電池供電的機器人,一方面異常動作的發生會影響機器人展示時間,另一方面得知電池狀態與何時需充電 更為重要。


而電力的穩定供應,則有賴電源管理系統與異常診斷,建立診斷分析系統指出錯誤並修正,使排程時時重新建置以盡可能提高效能。本論文欲建 構一診斷與分析系統,機器人依照排程運作,系統收集機器人身上各部位馬達的放電功率狀態直到再排程觸發前,資料蒐集後送進分類樹 (Classification Tree)與自我組織特徵映射圖網路(Self-organizing Feature Maps, SOM)兩種分類法做分類,建立異常行為分析模組,應用在適當的時間點做重新排程的依據,使得績效更佳。在預測維修系統方面,藉由收集而來的殘電量與溫度 資訊,並考慮電池因使用次數和升溫造成的電量損失,使用模糊推論(Fuzzy Inference)來預測電池供電良好程度並以此作為預測維修(Predictive Maintenance)基礎。此外,系統管理者可以藉由GMPP(Generic Message-passing Platform)網路服務,得知機器人運作狀態和電池狀態。一旦發現有機器人運作異常或電池供電性低之情勢,系統會主動傳送警告訊息給相關人員,藉此做 機器人與電池控管並進行問題修復。


最後, 我們考慮實際應用上的限制來探討單一電池組內電池組合的最佳化,以達到最大的使用時間與電量。