Meshfree formulation for inverse frequencies analysis incorporating artificial neural network

Abstract

This study concerns on the formulation of meshfree method for the inverse frequency analysis of damage detection incorporating Artificial Neural Network (ANN). Meshfree is used in developing the characteristic dynamic equation of the structure where as ANN is utilized to obtain the damaged stiffness matrix through the inverse process. Input data for the process are the frequencies and the mode shapes of the structure. The architecture used in developing the artificial neural network model is multilayer perceptron and the supervised learning process applied in this study is back propagation algorithm. The training data are randomly generated to cover all possible damage areas by using meshfree method. Output from the process is the stiffness reduction ratio. The performance of meshfree is first assessed against finite element method. Classical plate theory based on Kirchoff assumption has been discretized by both methods. The results are in good agreement between both methods, thus validate the meshfree formulation. The parametric study is then conducted to obtain the optimum value for the parameters of meshfree method. These parameters are shape parameters which involved; dimensionless shape parameter, ac and q, dimensionless size of support domain, as , background cell, gauss points and polynomial basis. The optimum values of these parameters are then used in constructing the random input and output data for the training process of artificial neural network model. The performance of the ANN model is verified against previous work based on the generated data. It can be concluded that ANN exhibits excellent performance thus highly potential as an analysis method in the study of inverse frequency analysis for damage detection incorporating Artificial Neural Network