Multicomputer networks for smart structures

TR Number
Journal Title
Journal ISSN
Volume Title
Virginia Tech

A crucial element of a smart structure is the computer system that processes data collected by sensors and determines an appropriate response. Multicomputers possess many capabilities that are required in computer systems for smart structures. This research examines the implementation and use of multicomputers for distributed processing in smart structures.

The research begins by examining previous research and showing the suitability of multicomputers for distributed processing in smart structures. Appropriate cost and performance metrics for evaluating multicomputer architectures are defined. The cost metrics are the number of processors, the number of communication links, and the length of fiber required to embed the network in the structure. The performance measures are the algorithm cycle time and the mean and standard deviation of message latency in the network. The scalability of these metrics is also examined. A key issue in the examination of these metrics is how their application to smart structures differs from their application in traditional systems.

The research continues by using a three-processor testbed network to identify general characteristics of algorithms that may be executed in smart structures. The testbed network uses fiber optic sensing, the MIL-STD-1773 communication protocol, and several different assignments for partitioning the necessary computations among the processing nodes to determine the shape of a triangular structure. The effects of math coprocessing on performance and the viability of hybrid links, in which a single optical fiber is used simultaneously for sensing and communication, are also demonstrated.

Simulation models of a damage detection, location, and estimation algorithm implemented in VHDL, a hardware description language, are used to examine and compare the performance of multicomputer interconnection network topologies. The topologies examined in this research are a binary hypercube, a custom planar topology, and a custom hierarchical topology. The ability of hierarchical architectures to limit cost while providing acceptable performance is demonstrated. The simulations also examine the effects of background message traffic and the ratio of communication time to processing time on performance. The combined results of the testbed and simulation experiments show the importance of process assignment and scheduling.