Using neural networks in the mapping of mixed discrete/continuous design spaces with application to structural design

Abstract : The objective of this task was to extend recent research efforts in order to evaluate the suitability of using artificial neural networks to provide quantitative design space representations for structural concepts with combined continuous and discrete design variables. For a simple structural problem containing both discrete and continuous design variables it was demonstrated that the character of the design space could be well represented with a relatively small amount of training data. It was also shown that design methods are available which can be used for mixed discrete/continuous design variable problems. These methods are however limited by the discontinuous nature of the discrete design problem and by the ability to predict system characteristics in an efficient manner. Using an approach with feed-forward, back-propagation neural networks an efficient method for the design of mixed discrete/continuous systems can be obtained. Neural networks, Structural design, Design variables, Finite, Element analysis, Optimization