Matrix Framework for Discrete Event Control of Manufacturing Systems and Sensor Network

by F. L. Lewis, IEEE Fellow, Professional Engineer- Texas, Head, Advanced Controls Group, Associate Director for Research, University of Texas at Arlington Automation and Robotics Research Institute

 :  11 Apr 2005 (Mon)
 :  11:00am - 12:30pm
Venue  :  Room 3412, 3/F (Lift 17-18) HKUST

To remain competitive in a global economy companies have gone away from old style fixed flow manufacturing systems and are now using Flexible Manufacturing Systems (FMS), where the same machines and robots can produce different products depending on the sequence of jobs performed. Requirements for fast reprogramming of FMS, dynamic job sequencing, and conflict resolution for shared resources have placed increased demands on the FMS supervisory controller. Similar demands exists in control and configuration of deployable wireless sensor networks.

A new matrix framework for the design and analysis of discrete event (DE) supervisory controllers has been developed at The University of Texas at Arlington. A U.S. Patent has been received on this novel formulation. It has been discovered that any machine-generated plan can be represented in terms of four matrices-- two for job sequencing and two for resource assignment. Given these four Task Plan Matrices, it is straightforward to configure a rule-based DE controller that can sequence the jobs and assign shared resources dynamically in real-time. The job sequencing matrices contain information like that in the Bill of Materials and assembly tree. The second pair of matrices contains information like that in the resource requirements matrix. The discrete event matrix controller plus the Petri net transition equation together provide a complete dynamical description for DE systems. This makes it straightforward to simulate DE systems using software such as MATLAB in a similar fashion to standard ODE integrators for the simulation of continuous-state systems.

Deadlock is a condition where operation of an FMS is indefinitely delayed and production halted; it can occur if incorrect dispatching is performed for assigning shared resources. Rigorous matrix techniques are given for analyzing the circular wait structure of the FMS. In terms of certain associated 'critical siphons', techniques are given for shared resource dispatching that are guaranteed to avoid deadlock. To avoid deadlock, the Work in process is limited in certain 'critical subsystems'. This MAXWIP technique amounts to a generalized kanban sort of feedback control that offers the best performance possible while
avoiding deadlock.

Applications are given for robot workcell control and supervisory control of mobile wireless sensor networks.


*** All are Welcome ***