Batch Control Systems: Design, Application, and Implementation
Batch Control Systems: Design, Application, and Implementation
A study in applying case-based reasoning to engineering design: Mechanical bearing design
Artificial Intelligence for Engineering Design, Analysis and Manufacturing
| Hi-index | 0.00 |
Batch manufacturing poses unique challenges to schedulers. The manufacturing processes are unpredictable, the environment is dynamic, and the required task and resource models are complicated. The Honeywell Batch Scheduler uses constraint envelope scheduling to address these needs, offering support for both schedule modifications and rescheduling. The authors are currently reimplementing the scheduler as production quality software. Batch manufacturing is a challenging domain for automatic scheduling systems because, unlike other types of manufacturing, processes are neither continuous (there is no steady inflow of raw materials resulting in a steady product outflow) nor discrete (there is no manufacture or assembly of individual items). Moreover, the plant environment is constantly changing-equipment breaks down, new orders come in-and shared resources may be both local (process units) and plantwide (labor, process heat).In continuous manufacturing, process optimization is largely a matter of understanding and exploiting plant dynamics in the face of, say, daily temperature fluctuations. In batch manufacturing, the scheduling problem changes and must be re-solved with each new set of orders. Once an initial schedule has been constructed, say, for a week's production, the scheduler must support rescheduling when the situation changes. To meet the unique needs of batch manufacturing, we have developed the Honeywell Batch Scheduler, which we are currently reimplementing as production-quality software. The scheduler uses constraint envelope scheduling, a least-commitment approach to constraint-based scheduling, which we developed and have applied in many scheduling domains. Constraint-envelope scheduling uses a partial order representation of schedules similar to the partial order representations used in AI planners. We implemented this representation by placing a task and resource model on top of a temporal constraint graph, a highly optimized network for propagating temporal constraints. We coupled the partial order representation with advanced search techniques to rapidly find answers to scheduling problems or, when no answer is possible, to identify constraints that must be relaxed. The approach we have developed has several advantages: ý Flexibility. Activities need not be tied to exact starting (and ending) times. Users can thus construct a schedule by specifying only what is needed, such as "between 10 a.m. and noon," rather than naming a starting time that satisfies those requirements. ý Quick reaction to events. We draw a distinction between schedule modification and rescheduling, thus avoiding the classic conundrum of choosing between two extremes: rigid schedules that rapidly become useless as events change and schedules that have so much slack that they are nearly useless to begin with. ý More information and traceability. The explicit linking of constraints with scheduling choices and requirements supplies users with a great deal of information when scheduling fails. They can trace infeasibilities to a set of requirements, some of which must be relaxed if scheduling is to be successful. In an environment where scheduling authority is distributed, conflicting decisions can be traced to the responsible parties, making it easier to resolve conflicts. For manufacturers, some of whom spend as much as $1,000,000 on operations management and scheduling, this feature means more insight into and control over current and future operations. Better control, in turn, translates to better use of available capacity. Manufacturing cycle times are lower, which means the manufacturer can become more responsive to its customers.In this article, we describe the process of building the scheduler and lessons learned during this process. We hope this account will further the state of practice in scheduling and intelligent manufacturing.