Development of Proficient Machines for Manufacturing

在本报告中，首先分析了刮削操作，以调查人类的熟练过程。刮削是从粗略切割的表面生成平面的过程。发现刮削的熟练程度包括对表面形状的理解，规划刮削顺序以及去除不必要的零件。其次，开发了一个熟练机床的原型。熟练的机床可以通过工艺仿真设计加工工艺，并通过实际加工更新工艺模型。也就是说，它具有从操作经验中学习的能力。在学习过程中，通过过程仿真从多个过程模型中选择一个理想的过程模型。在实际铣削后，采用数学规划的方法从测量到的传感器信号中得到新的工艺模型参数。利用基于记忆的学习方法，对与当前条件相对应的过程模型进行调整和记忆。结果表明，由于机床获得了丰富的加工经验，可以提高过程仿真预测的准确性。第三，开发了熟练机械手的原型。它由传感器反馈机械手、测量人体熟练动作的装置和操作模型构建和操作规划工作站组成。研究了将人体灵巧动作移植到机械手上。在将可变形软管插入插头期间，通过使用位置和力传感器来测量人类的动作，并根据软管与插头之间的接触进行分析。研究发现，在此过程中，拉拔运动对于将可变形软管成功插入堵头是有效的。在机械手程序中嵌入有效的拉入运动，机械手能够顺利完成插入。

In this report, scraping operation is firstly analyzed in order to investigate the proficient process of human. The scraping is a process to generate flat surfaces from roughly cut surfaces. It is found that the proficiency of the scraping consists of the understanding of surface shpes, planning scraping sequences, and the removal of unnecessary parts.Second, a prototype of a proficient machine tool is developed. The proficient machine tool can design machining process via process simulation and can update the process model through actual machining. Namely, it has a capability of learing from operation experience. In the learning, a desirable process model is selected among multiple process models via process simulation. After actual milling, new process model parameters are derived from measured sensor signals using a mathematical programming method. The process model correponding to current conditions is adjusted and is memorized by use of a memory-based learning. It is found that prediction correctness of the process simulation can be improved as the machine tool obtains much experience of the machining.Third, a prototype of a proficient manipulator is developed. It is consists of a sensor-feedback manipulator, a device capable of measuring human skillful motion, and a workstation for operation model construction and operation planning. Transplanting human skillful motion to a manipulator is examined. Human action during the insertion of a deformable hose into a plug is measured by use of the position and the force sensors, and analyzed with regard to the contact between the hose and the plug. It is found that the pull motion during the process is effective for the successful insertion of a deformable hose into a plug. Embedding effective pull motion in a manipulator program, it turns out that the manipulator can perform the insertion successfully.