An Architecture for the Distributed Intelligent Control of Flexible Manufacturing Systems

With the development of ever-greater computational capabilities and the increasing need to address realworld problems, complex large-scale, discrete-event systems are being proposed and designed. Numerous examples can be cited, but certainly the list must include flexible/agile manufacturing systems, intelligent highway vehicle systems, and air-traffic-control systems. These systems have several features in common. They all rely on distributed computation to handle the complexity of their overall planning and control. And the interaction among these computers necessarily results in discrete-event system dynamics. Currently, little formalism guides the construction of planning and control architectures for these systems. In many cases, the planning and control architectures and the associated enabling software are constructed on an ad hoc basis using engineering intuition alone. The resulting architectures can be extremely complex and difficult to validate. In this paper, we present an integrated solution to the problem of simulation and control of distributed largescale discrete-event systems. This integrated approach first decomposes the overall control problem in a manner that distributes the overall control function among a hierarchy of smaller sub-controllers. Given the distributed set of controllers, distributed simulation is employed to project the system response. Finally the integrated approach provides algorithms that allow each controller to make intelligent decisions based on information generated from these simulations. INTRODUCTION In the field of discrete-event systems (DES) control refers to the process of allocating available resources so that the overall system performs its assigned tasks. For example, in an automated manufacturing cell, the primary controller sends commands to each machine to perform certain tasks upon given jobs. In making this assignment, the primary controller must also coordinate the movement of a job to the machine where the execution of the task has been assigned. In this case, the controller sends a command to the material handling system (MHS) to pick up a job at its current location and deliver it to the allocated machine. Once the job is delivered, the controller then tells the machine to unload the job from the MHS and perform a set of prescribed tasks upon it. It is obvious that the controller must make many decisions. Typically the controller can send several jobs to a machine, where they wait in an input queue until the machine is available to execute the assigned tasks upon a given job. After the machine finishes processing a given job, it must notify the primary controller so that the job can be removed from the machine and assigned to another machine for the execution of another set of processing tasks. The machine must then decide which waiting job will be processed next. Certainly the primary controller must have scheduled the order in which the jobs will be processed at a given machine. However, there are other tasks that must be addressed at the machine level in order to permit the processing of the set of assigned tasks waiting to be executed upon the job. In general, the cell-level or primary controller will not consider these additional tasks. Hence even though the cell-level controller may prescribe the order in which the jobs will be processed at the machine, additional planning is required at the machine level. Intelligent control refers to the process of integrated planning and control. The planning that occurs at a given intelligent controller depends upon the state of the subsystem that it manages, including the assigned tasks and goals set by the supervisor. Given the dependence of planning on the system-state, the decision that the intelligent controller is facing changes constantly with time. Therefore, the need for planning at each intelligent controller is ongoing. In order to implement the plan produced by the planner, a control law must be specified for the execution of tasks. This control law could include the specification of a priority scheme for deciding which of the assigned tasks would be addressed next when a given resource became available. It could also provide information that would modify the manner by which a given processing task was to be implemented. As each control action is taken, the state of the system is modified, which also leads to the modification of the decision that the controller faces. For example, as each task is completed, it no longer needs to be considered in the planning. Thus, the plan as well as the control law by which the plan is implemented must be updated. It is clear then that intelligent control necessitates the concurrent consideration of both planning and control. In addition, the complexity of the considered system further necessitates that this integrated planning and control function be distributed across a collection of several controllers. Given the necessity for distributed planning and control, the next issue is one of pulling together these distributed functions in order to provide a coordinated response for the entire system. This paper will provide an integrated solution to the problem of coordinating the distributed functions, first decomposing the overall system into its component subsystems. Then using this decomposition, a new modeling paradigm is employed to simulate or project the future performance of the system. Finally, this simulation capability is employed to define a generic form of the intelligent controller that is required to manage each of the subsystems. DESCRIPTION OF THE INTEGRATED SOLUTION (Davis et al. 1993; Davis et al.1997) present a conceptual framework for the intelligent control of flexible manufacturing systems (FMS) and other distributed DESs. Their integrated solution consists of three primary components : 1. The Recursive Object-Oriented Control Hierarchy (ROOCH) architecture, which defines the component subsystems making up the overall system and specifies where intelligent controllers are needed, 2. The Hierarchical Object-Oriented Programmable Logic Simulator (HOOPLS), which simulates the system response while operating under the defined control architecture by explicitly modeling the interactions among the controllers contained within the control architecture, and 3. The Intelligent Subsystem Controller (ISC), which defines the generic framework for the intelligent controller that is contained within each subsystem. THE RECURSIVE OBJECT-ORIENTED CONTROL HIERARCHY (ROOCH). The ROOCH was originally published in (Tirpak et al. 1992). The ROOCH introduces the coordinated object, CO, as the basic element for modeling an FMS or other DES (see Figure 1). Each coordinated object represents a basic hierarchical element for which intelligent control is to be addressed. It is assumed that each coordinated object contains one or more primary unit processes, Pn (n=1,..., N), which are to be allocated in order to execute processing tasks upon jobs residing within the coordinated object. Both jobs and supporting resources (e.g., tools, part kits, and processing information) enter the coordinated object through its input port and will eventually exit through the output port. The jobs and supporting resources are assumed to be under the control of the coordinated object while they reside within the coordinated object.