Artificial neural networks prediction of in-plane and out-of-plane homogenized coefficients of hollow blocks masonry wall

International audience; A masonry wall is a composite structure characterized by a large variety in geometrical and material parameters. The determination of the effective macroscopic properties, through the homogenization scheme, depends on a great number of variables. Thus, in order to replace heavy numerical simulation, in this paper, the use of artificial neural networks (ANN) is proposed to predict elastic membrane and bending constants of the equivalent Love-Kirchhoff plate of hollow concrete blocks masonry wall. To model the ANN, a numerical periodic homogenization in several parameters is used. To construct the model, five main material and geometrical input parameters are utilized. Multilayer perceptron neural networks are designed and trained (with the best selected ANN model) by the sets of input-output patterns using the backpropagation algorithm. As a result, in both training and testing phases, the developed ANN indicates high accuracy and precision in predicting the equivalent plate of a hollow masonry wall with insignificant error rates compared to FEM results.