Device Controller Synthesis for Systems of Interacting Discrete-State Components

对于复杂系统，单一模型变得不切实际，必须通过子系统和组件对其进行建模。这种系统的控制逻辑的开发也变得复杂，除非子系统和组件的结构能够以系统的方式加以利用。在这个建议中，我们考虑了使用离散状态模型组件的结构来综合控制逻辑和诊断信息的方法的发展。这些方法是自动化的，生成可分析的形式化控制逻辑模型，生成可执行的控制软件，以及生成可提供故障诊断的诊断模型。这种自动化方法将减少控制器开发时间，与手动开发的控制器相比，将提供更高的可靠性，并将随着基础系统设计的修改而自动重新配置。在这项研究中，我们建议在先前研究的基础上开发这样的工具和技术。在我们以前的研究中，我们开发了以下技术：对制造系统的公共组件进行建模；将这些组件组装在一起以表示定制的系统设计；合成被称为“任务块”的控制逻辑以控制这些组件；以及将这些任务块顺序地和层次地组装在一起以达到给定的规范。开发了软件工具，用于以图形方式输入组件模型和规范，并将合成的控制逻辑自动转换为C++代码。然后，一个自动合成的软件“监控器”协调多个并发控制活动，引导系统进入不需要的状态。我们考虑的模型是离散事件系统的一种形式，综合技术建立在离散事件控制方法的基础上。目前的方法有几个未解决的问题，将在拟议的研究中解决。这项工作的一个主要重点是识别和利用与系统组件的多层交互。建模框架和我们目前的分析方法非常适合于这种交互组件的模型，但它们必须进行扩展，以便紧凑地表示具有大状态空间的组件。第二个问题涉及系统的定时动态。一些活动可能是时间关键的，例如在一定的时间段内或在接收到感测后立即关闭执行器。综合控制软件应考虑这样的时间问题，并应确保在各种可能同时进行的活动中确定适当的优先顺序。最后，我们建议扩展我们目前的软件工具，成为其他研究人员和教育工作者的资源。这将允许其他人使用我们的图形系统编辑器、代码合成工具和硬件接口技术来测试、开发和评估其他技术。将使用制造和嵌入式控制应用程序来演示自动控制合成和诊断合成技术。

With complex systems, monolithic models become impractical and it becomesnecessary to model them through subsystems and components. Development of controllogic for such systems also becomes complex, unless the structure of the subsystemsand components can be exploited in a systematic way. In this proposal, we consider thedevelopment of methods that use the structure of discrete-state model components forsynthesizing control logic and diagnostic information. These methods are automated,generating formal control logic models that can be analyzed, generating control softwarethat can be executed, and generating diagnostic models that can provide fault diagnosishypotheses. Such automated methods would reduce controller development time, wouldprovide enhanced reliability compared to manually developed controllers, and would beautomatically reconfigured as underlying system designs are modified. In this research,we propose to develop such tools and techniques by building upon our prior research. Inour previous research, techniques were developed for: modeling common componentsof manufacturing systems; assembling such components together to represent customsystem designs; synthesizing control logic called "taskblocks" for control of thosecomponents; and assembling those taskblocks together sequentially and hierarchicallyin order to achieve given specifications. Software tools were developed for graphicalentry of the component models and the specifications, and for automatically convertingthe synthesized control logic into C++ code. An automatically synthesized software"supervisor" then coordinates the multiple concurrent control activities, steering thesystem around undesired states. The models we consider are a form of discrete eventsystem, and the synthesis techniques build upon discrete event control methods.There are several unresolved issues with this current approach that will be addressed inthe proposed research. One major emphasis of this work is recognizing and exploitingthe multiple layers of interaction with components of the system. The modeling frame-work and our current analysis methods are ideally suited for such models of interactingcomponents, but they must be extended in order to compactly represent componentswith large state spaces. A second issue involves the timed dynamics of the system.Some activities may be time critical, such as turning off an actuator within a certainperiod of time or immediately upon receipt of a sensing. The synthesized controlsoftware should consider such time issues, and should ensure appropriate prioritiesamong various potentially concurrent activities. Finally, we propose to extend our currentsoftware tools to become a resource for other researchers and educators. This wouldallow others to test, develop, and evaluate other techniques using our graphical systemeditor, our code synthesis tool, and our hardware interface techniques. The automatedcontrol synthesis and diagnostic synthesis techniques will be demonstrated using bothmanufacturing and embedded control applications.