Innovative ANN Technique for Predicting Failure/Cracking Load of Masonry Wall Panel under Lateral Load

Abstract

This paper introduces an artificial intelligent technique for predicting the failure/cracking loads of laterally loaded masonry wall panels based on their corresponding failure/cracking patterns derived from the laboratory experiments. First, a lattice is made on a wall panel based on the dimension of the wall panel. Then, the numerical values, 0 or 1, are assigned to the cells in the lattice in order to describe the failure/cracking pattern. Thus, a numerical matrix is formed to show the failure/cracking pattern of the wall panel. Since the matrices for the wall panels with various sizes have different dimensions, the gray level cooccurrence matrix is innovatively used to transfer these matrices into the matrices whose dimensions are the same. Next, the numerical modes of failure/cracking patterns of experimental wall panels and the corresponding normalized failure/cracking loads can be used as the input and output of the artificial neural network (ANN) training data, respectively. Finally, three types of ANN models for predicting the failure/cracking load of the unseen wall panel are achieved by repeatedly training and adjusting so as to optimize its parameters. In a wide significance, this study opens a novel way to establish the relationship between the failure/cracking pattern and the failure/cracking load of the wall panel.