Aneural-network-based reanalysis method for integration with structural design

The analysis of displacements and stress resultants is essential in the process of structural design. It is usual to prepare an initial design before proceeding to any formal analysis of the structure. This first analysis will generally lead to a revision of the design which, in turn, will require a reanalysis producing a circular operation of repeated design and analysis. It would clearly be an advantage if the separate operations of design and analysis could be integrated in some convenient fashion. A neural-network type of computational device is shown to be suitable for continuous updating of the analysis with provision for a wide range of structural modifications, including geometry, topology, nodes, loads, supports, and material properties. The aim is to incorporate the analysis in the design environment.
The method described will be most useful in circumstances where the design is focused on a well-defined structural type, where the design changes are at the level of detail, rather than involving a major change in structural type. However, more substantial changes can be accommodated at the cost of additional processing.

Professor W. M. Jenkins, BSc, PhD, DTech, CEng, FIStructE, FICE University of Leeds (formerly, University of Hertfordshire)