整合機器視覺與機器手臂快換裝置之虛實整合技術應用於彈性組裝任務學習

Abstract

因應現今大量客製化且彈性化的製造流程，工廠中的生產線必須時常進行任務的轉換，而大量花費人工與時間進行機械手臂實機調教的方法難以達到高效率生產目標，因此本論文提出一套虛實整合技術應用於機械手臂的任務學習中，透過虛擬環境中以軟體示教方式完成機械手臂高階任務決策模型的學習。學習演算法的開發主要是透過任務樹(Task Tree)決策模型進行復雜任務的自動規劃，此決策模型先從虛擬環境機械手臂的動作中學習完成複雜的組合任務，並輸出相對應任務的動作命令來控制機械手臂。該決策模型可即時在手臂端進行決策思考，並且當有新的組裝任務時，也可快速將新的操作步驟加入至原本模型中。本論文所開發之虛實整合技術具備以下特點：步驟合理性分析、快速新增新任務以及物件特性分析，首先導入支持向量機演算法(Support Vector Machine，SVM)使能夠判斷物件擺放狀態及夾取物件的種類，將物件特性考慮至決策模型中，決策模型可決策該物件合適的夾爪並且透過快換裝置進行治具的快速替換。在實驗驗證方面，透過機械手臂搭配快換裝置進行加工及自動組裝燈具之操作任務來展示本論文所提出之虛實整合技術可應用於彈性製造的解決方案。Due to increasing heavy demands for the customized and flexible manufacturing process, the production line in the factory has to handle a variety of different products from task to task. The traditional method of using many robotic arms that requires lots of human resources and is time-consuming on machine tuning is no longer applicable. To this end, this thesis follows the concept of cyber-physical systems (CPS) for developing a task learning approach for the robotic arm to achieve high-level decision-making through learning from demonstration in the virtual environment. The task tree algorithm is employed for the automatic planning of complex tasks, in which a decision-making model is presented to generate complex task sets from a large number of actions. Then, the decision-making model can export tasks and control the robotic arm to yield the correct operation in real-time when different tasks are endowed to a robotic arm by means of arranging a new task or new steps to the existing model. The decision-making model proposed in our thesis can analyze the rationality of model steps, quickly add new tasks, and perform object analysis. A support vector machine (SVM) algorithm is used to identify the state of an object and a suitable gripper for the object. By considering the characteristics of the relevant object, the decision-making model can then select a suitable gripper and switch between grippers quickly. The decision tree algorithm can be applied to complex tasks and thus can replace expert systems used for adjustment. The conducted experiments demonstrate the machining task and assembly task to investigate the proposed system capability towards the feasible solution to the flexible manufacturing.