BP and RBF Neural Networks for Predicting Minimum Weight of Double-Layer Space Grids

In recent years, extensive and increasing use has been made of space trusses, especially in the form of double-layer space grids (DLGs). These structures have various geometrical configurations and proved to be very economic in terms of weight and ease in construction. The Space grid competitiveness to other structural systems depends much on the final weight and relative stiffness, which in turn depends on the decisions taken at preliminary design stage. An equally important goal at this stage is to determine fairly accurately the weight of the double-layer grids, because preliminary designs form the basis for competitive bidding and weights are often decisive in winning contracts. As analysis and design of such structures are normally time consuming and the configuration processing and data generation process is complicated since a large number of nodes and members are involved, approximate methods are needed at the preliminary stage of design.

Artificial Neural Networks (ANNs) are amongst the AI tools in which the human creativity, intuition and past experience can be incorporated in the network training process. Like human experts, ANNs learns from experience and examples. One of the techniques to reduce the resources and the time required for the design process is to store many optimum designs and train a neural network for the design. Thus, the neural network will come up with a design based on its training rather than conducting a full design from scratch.

In this study, most commonly used five kinds of DLGs like SOS, SOLS, SOD, DOS and DOD have been investigated. MultiLayer Perceptron (MLP) with Backpropagation (BP) algorithm and Radial Basis Functions (RBF) networks are employed for training efficient neural networks for evaluation of minimum weight of DLGs. The performance obtained of neural networks for pilot investigation shows the potentiality of application of ANN technology for preliminary design of DLG structures.