Data parallel neural dynamics model for integrated design of large steel structures

We have two objectives in creating novel design theories and computational models: automation and optimization. These two aspects are particularly important in design of complex and large engineering structures. In this article, a robust data parallel neural dynamics model is presented for discrete optimization of large steel structures based on the AISC ASD or LRFD specifications. The computational model has been implemented on a CM-5 supercomputer and applied to integrated minimum-weight design of two steel high-rise building structures. The largest example is a 144-story modified tube-in-tube super-high-rise building structure with 20,096 members. Optimization of such a large structure subjected to the highly nonlinear constraints of actual design codes, such as the AISC LRFD code, where nonlinear second-order effects have to be taken into account, has never been attempted before. The computational model developed in this research finds the minimum-weight design for this very large structure subjected to multiple dead, live, and wind loadings in three different directions automatically. This research demonstrates how a new level in design automation is achieved through the ingenious use of a novel computational paradigm and new high-performance computer architecture.