An ANN based surrogate model for real-time simulations and track location control supports of retractable roof structures

This paper presents an application of artificial neural networks (ANNs) to predict the behaviour of a large-span retractable roof structure subjected to variable transverse and longitudinal horizontal forces resulting from movements of mobile segments with different moving speeds, variable wind speeds, and positions of the mobile roof segments (open, semi-open, and closed). Firstly, geometrically nonlinear displacements of the rail track for two movable segments located on two parallel spatial truss arches with a span of 224 m are analysed using a finite element method. Subsequently, ANNs are trained and tested to replace the nonlinear finite element analysis and predict the required structural response characteristics. Results obtained using the proposed ANNs are compared to the results obtained using the standard nonlinear finite element method to demonstrate the effectiveness and robustness of artificial intelligence techniques. A significant conclusion is that the developed feedforward multilayer perceptron neural network with the 42-79-42-42 topology can adequately predict the behaviour of the retractable roof structure subjected to variable conditions. Results confirm the mathematical correctness and physical relevance of the presented approach. The neural network model can improve the control of the nonlinear behaviour of retractable roof structures.