Predicting The Cracking Pattern Of Masonry Wall Panels With Opening Using Artificial Neural Network And Support Vector Machine

Because masonry is complex in its working behavior with great variability,the existing analytic methods are difficult to achieve satisfied results in the analyzing of masonry structures,which is restricting the development of masonry structures.The Finite Element Method(FEM)is the popular technique for evaluating the cracking patterns of masonry wall panels,but in many cases,when compared with the experimental results of laterally loaded masonry wall panels,the results of the FEM fail due to inaccurate prediction of masonry working behavior.Despite the long history and related research on the masonry wall panels,there is still discrepancy over the suitability of techniques for anticipating the cracking patterns of masonry wall panels subjected to wind and other lateral loads,particularly for wall panels with opening.Some researchers have tried to apply artificial intelligence techniques(AIT),for instance,cellular automata(CA)and artificial neural networks(ANNs),to resolve the problems in the analysis of masonry structures.The existing application of artificial intelligence technology(AIT)in structural analysis has not been able to produce better or more reliable and accurate results.Therefore,it is necessary to further develop artificial intelligence technology in order to promote the development of masonry structures and provide reference for other fields of structural analysis.This dissertation conducts the following studies based on the concept of machine learning techniques.(1)It introduces the cellular automata model for calculation of state value of masonry wall panels with opening.Masonry wall panels failure pattern is converted into numerical matrix by dividing a panel into zones of equal size to be used as a label data for training of model.State value of all the panels are calculated using Moore neighborhood model.Failure load of all the wall panels is normalized for better prediction of result.(2)It introduces artificial neural network for the prediction of cracking pattern of unseen masonry wall panels using training data of seen panel.Selection of neural network,number of hidden layers and number of hidden neurons in each hidden layer is made using trial and error method.The data is trained on seen data using features as input and numerical matrix as a label.(3)It introduces support vector machine(SVM)model for the prediction of cracking pattern of unseen masonry wall panels using training data of seen panel.Support vector machine model is designed using different kernel functions.Kernel function is selected using error and trial method.Hyperparameter optimization is performed using Bayes optimization technique.Predicted result of both the designed model are compared with experimental cracking pattern and sensitivity analysis of predicted result is performed.It has been found that SVM prediction accuracy is much better than artificial neural network.The obtained results imply that ANN and SVM model combining with the CA numerical model has a capacity of modeling the variation in material properties of the structure,which opens a way to handle similar issues in both theoretical analyses of many engineering problems.