Neural Network Technique For Predicting Instability Failure Of Single-layer Latticed Domes

This paper establishes an artificial intelligence model for predicting the instability expressed in failure pattern and failure load of Shell Kiewitt 8 in the TAFT loads, using the technique of artificial neural networks (ANN). The failure mode and failure load of the shell are based on the results of ANSYS numerical simulation in which the time history analysis method is used and the seismic load is applied step by step until the destruction of shell model. For the cellular automata (CA) numerical model of sell, the criterion for identifying the failure region in the shell is applied, that is, if a region failure in the shell, the similar region in other shells is also identified as the failure region. As the number of nodes in different shells, the number of CA lattice is certainly different. But, the BP neural network input requires a fixed dimension of lattice cells; thus, this paper proposes a concept of generalized shell. The concept of generalized shell can transfer the digital matrixes of failure modes of different-size shells into the same-size matrixes, in which the inductive method makes use of the mean method.This paper treats the failure mode of shell as the training and testing data of the ANN model and takes the normalized failure load of shell as the output data of the ANN model. Then, the optimal hidden layers in the architecture of BP neural networks and the learning rate are obtained through fully training. Finally, the predicted failure loads are compared with their corresponding the numerical failure loads. The results show that the proposed ANN model could be a feasible way to predict the failure loads of shells through their corresponding failure modes, which enriches the content of directly predicting failure loads of shells through their failure modes using artificial intelligence techniques.