A Behavior Oriented Control System for Machining

This paper describes a new approach for controlling an intelligent machining workstation. The control system is built in behavior achieving layers, which allows a machine tool to operate at increasing levels of competence. Each layer focuses on a specific simple control task for machining processes. Qualitative reasoning is used to augment the system at the highest layer, giving the system the ability to foresee patterns of behavior that lead to failure. We discuss how qualitative models can be selected, setup and used to adjust control parameters. 1.0 INTRODUCTION One focus of manufaduring research has been the development of integrated, self-adjusting manufacturing systems that are capable of machining varied parts without human supervision. Machining process automation has already been achieved for some routine operator functions such as the loading and unloading of work pieces and tools, parts scheduling and distributing, and initiating NC programs. The functions which remain to be developed include monitoring machining operations, ensuring safe and effiaent metal removal rates, and taking corrective actions in the event of process disturbances or failures. Traditional manufacturing control systems are based on a serial flow of information from sensors to actuators (see Figure 1.1). These contml systems usually process one step at a time, which can make a machining system susceptible to unexpected events. In addition, a "serial controller" is plagued by a reliability bottleneck, io., every processin step has to work reliably in order for the whole system to advance to the next step f Bourne and Wright 1988 I. As an alternative to a serial model of control, Brooks has proposed a parallel organization that is based on natural behavior [Brooks 1%]. This architecture controls complex amalgamations of simpler behaviors, and has lead to a tractable reformulation of the problem of combining multiple goals. We considered this wntrol organization by translating Brooks' parallel mobile robot behaviors into pardel machine tool behaviors (see Figure 1.2). Figure 1.k Traditional Flow of Information in a Control System [see Brooks 19851 The architecture in Figure 1.2 was W i t up with a conservative view of preserving the machine first, the tools second and then various aspects of the part being produced.