Manufacturing Shop-Floor Supercomputer for Distributed Simulation and Control

该赠款为开发低成本超级计算机提供资金，这些超级计算机可以使用可重新配置的计算机分布在整个制造车间。 可重构计算机在芯片上包含数千（100，000）个逻辑门，其互连可以通过写入其内部存储器来即时配置，以作为大规模并行计算机，其中算法可以“硬连线”。 因此，可重新配置的计算机保持了传统的基于软件的计算系统的可编程性，但能够显著地改进计算。 车间超级计算机将用于： (a)消除离散事件模拟的复杂性和开销的分布式时间尺度模拟算法;以及（B）用于调度离散事件的连续分布式控制算法，其消除组合方法的计算复杂性。一个实验性的车间超级计算机原型将被构建并与工业ERP系统集成。将研究使用可重构计算和千兆网络加速分布式时间尺度仿真的新架构。如果成功，在短期内，这将导致工业ERP系统的制造协处理器的开发，以解决制造车间的计算挑战。 这种面向企业系统的芯片将利用仿真和控制算法固有的大规模并行/分布式特性，并利用可重构计算所实现的“硬连线”能力。所得到的实时自适应时间缩放算法将利用连续变量分布式控制算法和时间缩放分布式仿真相结合所产生的协同作用。 模拟的速度和准确性将得到优化，以有效利用原始计算能力和原始通信网络带宽。 车间超级计算机将实现超快速分布式仿真（10，000 x实时），用于离散事件调度的反馈控制。 这项工作的潜在长期影响将重塑制造车间计算和决策的未来。

This grant provides funding for the development of low cost supercomputers that can be distributed throughout the manufacturing shop floor using reconfigurable computers. A reconfigurable computer contains thousands (100,000) of logic gates on a chip whose interconnections can be instantaneously configured by writing to its internal memory to serve as massively parallel computers in which algorithms can be 'hardwired'. Therefore, a reconfigurable computer maintains the programmability of a conventional software-based computing system, yet enables phenomenal improvements in computing. The shop-floor supercomputer will be used for: (a) Distributed time-scaled simulation algorithms that eliminate complexity and overhead of discrete event simulations; and (b) Continuous distributed control algorithms for scheduling discrete-events that eliminate the computational complexity of combinatorial approaches. An experimental shop-floor supercomputer prototype will be constructed and integrated with an industrial ERP system. Novel architectures using reconfigurable computing and gigabit networks for accelerating distributed time-scaled simulation will be investigated. If successful, in the short term, this will lead to the development of a manufacturing co-processor for industrial ERP systems that addresses computational challenges on the manufacturing shop-floor. Such a chips to enterprise system would utilize the inherently massively parallel /distributed nature of simulation and control algorithms and take advantage of 'hardwiring' capability enabled by reconfigurable computing. The resulting real-time adaptive time-scaling algorithms would utilize the synergy resulting from combining continuous variable distributed control algorithms and time-scaled distributed simulation. Speed-up and accuracy of simulations would be optimized to effectively utilize raw computing power and raw communication network bandwidth. The shop-floor supercomputer will enable ultra-fast distributed simulations (10,000 x real-time) for feedback control of discrete-event scheduling. The potential long-term impact of this work will be reshaping the future of computing and decision making on the manufacturing shop floor.