APPROXIMATE ANALYSIS OF STRUCTURAL GRILLAGES USING A NEURAL NETWORK.

The application of computers to structural analysis is now so well-established as to be almost taken for granted. Modern software is well-organized, fast in operation, comprehensive in options provided and generally easy to use. Educational attitudes have been steadily adapted to ensure that students are adequately trained in the reliable use of analysis software. There are, however, circumstances where 'exact' analysis is not strictly necessary and an approximate method may be acceptable in order to save central processing unit (CPU) time. Structural design optimization involves, coincidentally, the continuous re-analysis of the structure in line with changes in topology and structural properties. If the re-analysis is carried out by exact methods, then the CPU time needed for the optimization can be significantly increased. In these circumstances, approximate methods may offer an alternative to exact re-analysis. There are also other situations where access to a reliable and rapid approximate analysis would be an advantage, for example with highly standardized or regular structures. One such structure is the grillage, which is used in structural design as a conceptual representation of a concrete slab bridge. The bridge designer may want to experiment with a number of configurations of longitudinal and transverse members and with varying structural properties of the members. This paper describes a study of the application of a neural network-based method of approximate analysis to right-grillage structures and offers some observations on matters such as accuracy, network topology and training.