The remaining steps after architectural design are the "conventional" detailed design and production/procurement (of control hardware and software), integration and testing of these parts to make the control system, and finally integration and testing of the control system into the overall robot system. This is by no means underestimated, but from its principles is so well known that it is not discussed any further here. For each step, the CDM offers support in the form of generic architectural schemes (reference models), guidelines on their use, and even computer-based tool assistance. The two most important tools are an activity analysis method (ActAM) for Step 1 and a functional reference model (FRM) for Step 2. They will be introduced below. 2.3 Hierarchical activity analysis The activity analysis method of the CDM is a systematic procedure to produce refined user requirements suitable for further system and subsystem analysis. From the analysis of the process to be automated, a hierarchical description of robot system activities to be performed is obtained. The CDM proposes a three-layer hierarchy of activities: • The very top level, robotic missious, represents the highest level of activity for which a robot system is responsible (e.g. "SERVICE a life science experiment", "REPAIR a satellite"). • Each mission can be decomposed into tasks, defined as the highest level of activity performed on a single subject (e.g., "OPEN a door", "INSTALL a sample in a processing furnace", "POLISH a surface", "WELD a seam"). • Finally, each task can be decomposed into actions (e.g., "GRIP a sample container", "DISPLACE tool to a position", "MOVE the container to the freezer", "INSERT the container in the port", "SLIDE a drawer", "RUB along a path on a surface", "FOLLOW a seam", "TRACK a part on a conveyor"). For each action, a realization with a particular control concept can be established (e.g., free continuous path control for MOVE, or impedance control for INSERT and SLIDE) such that there are well-defined criteria for identifying actions within a task. The CDM offers a "catalog" of frequently occurring tasks and actions, together with templates on which action attributes should be defined. As an example, a template for the RUB action in a polishing application is reported in Section 3. 2.4 Functional requirements analysis For control system requirements analysis, the key concept is to start with a purely functional analysis. Even the most perse applications use the same fundamental control functions, but in possibly quite different realizations.
The control functions are cast in the same overall structure as the activities, namely a three-layer hierarchy responsible for achieving robot missions, tasks, and actions, and a general framework (logical model) for robot control functions, called the functional reference model (FRM) is defined. The top-level view of the FRM is shown in Fig. 2. The global goal (the robot's mission objectives) is successively broken down into sub-goals: the tasks and actions, each handled by control functions on the appropriate layer, until the elementary "control outputs" can be issued to the robotic devices (e.g., currents to the joint motor drives). The "vertical" branches of the FRM are based on the concept of feedback. The center branch is called forward control (FC). FC functions are responsible for activity decomposition, execution planning and control by taking the most appropriate a priori information known to each layer into account. The left branch consists of nominal feedback (NF) functions for the refinement and update of a priori knowledge ("world models"), based on the actual, but essentially expected, evolution of the process, and consequently the formulation of controlled adjustments of the FC. By FC and NF, "cascaded" control loops are closed on action, task, and mission layers and are equipped with everything necessary for the "nominal" course of events. Besides FC and NF, the FRM foresees a third branch of "non- nominal feedback" (NNF) functions responsible for the monitoring of discrepancies between actual and allowable states in both the FC and NF functions, diagnosis of their origins, and generation of directives (including recovery strategies and constraints) for FC. A much more refined definition of the FRM is given in (Putz and Mau, 1992). It should also be acknowledged that the FRM concept and structure have been heavily and beneficially influenced by the NASREM architecture developed at the US NIST (Albus, et al., 1989). 2.5 Application reference model The FRM is by its nature still very generic, and it applies to general automation control systems. To be more specific for the frequent application to "classical" robot control, the CDM has elaborated a more detailed reference model for this application class, called the application reference model (ARM). The control concepts relevant to an industrial robot are completely located in the action layer of the FRM. In fact, the robot controller has to implement the control functions relevant to action planning, execution and control, while mission- and task-level functions are in the charge of the robot and workcell engineer, who has to program a mission execution as a suitable sequence of tasks, and tasks as sequences of actions. In the same way, it is a human operator that has to intervene to "rescue" unexpected, non- nominal situations. Therefore the ARM is a refinement of the FC and NF functionalities in the action layer. In FC, one will thus find functions like path preparation and interpolation, path servo control, inverse kinematic transformation, and joint servo control. In NF, functions include proprioceptive and exteroceptive sensor data processing (SDP), and process-, or device-, and control-oriented data processing (PODP, CODP, respectively). 工业机器人的结构设计英文文献和中文翻译(2):http://www.751com.cn/fanyi/lunwen_34681.html