Development of An Intelligent Robot Health Diagnosis System

機器人經過日經月累地使用會有逐漸老化及機構磨損的現象，而這些因素會降低其性能的可靠度或產生潛在發生故障的因子。若能在機器人故障之前檢測診斷出來，且先行維修以避免更嚴重的損失，將可提高機器人的可靠度及壽命。 為了精確地檢測製程變化和機器人的惡化症狀，本論文發展一個機器人健康診斷系統，能夠診斷出多種有潛在相關連結的故障原因。此系統主要由四個部分組成:機器人系統、故障檢測、故障診斷、預測維修。機器人系統通常包含機器人本身、感測器以及控制單元，並使用運動規劃器來規劃機器人的運動以滿足任務要求。從龐雜的資料數據中判斷機器人的健康狀態變得非常重要，為有效進行故障檢測，乃使用以主成分分析為基礎的統計製程管制和Nelson規則方法。在故障診斷方面，使用支持向量機發展診斷策略，並延伸至可辨別多種狀態。除此之外，不論故障是否發生皆使用模糊邏輯計算機器人健康指數。另外，建立訊息傳送平台，用於預測維修通知，其中包含電子郵件和推播功能。最後，所提出的機器人診斷系統，經模擬及實驗驗證，成效良好。Robot gradually degrades as a result of aging and wear. It will decrease its performance and increase the potential of faults and failures. It is important to detect the robot performance to predict and prevent the robot failures before the occurrence of serious breakdown. In order to precisely detect process variation and deteriorating symptoms of robots, the intelligent robot health diagnosis system (IRHDS) is developed in this thesis to diagnose multi-kinds of potential connection-related faults in the robotic system. The structure consists of four parts: the robotic system, fault detection, fault diagnosis and predictive maintenance. The robotic system usually includes motion planner, physical robots, sensors and motion control unit. It becomes more important to determine the state of the robot health from the multifarious and disorderly sensor data. For effective and sensitive fault detection operation, PCA-based (Principal Component Analysis) Statistical Process Control (SPC) with Nelson rules is adopted in this system. Using the predominant statistical theory in small sample population and abundant nonlinear kernel function of SVM, the diagnosis strategy is developed and expanded to identify multi-kinds of states. In addition, the robot health index (RHI) is designed by using fuzzy logic to diagnose the performance of the robot at that time. Moreover, the message passing functions for maintenance notification are also established, including email and push notification. Finally, the proposed IRHDS has been justified and demonstrated through simulations and experiments. The results are promising