Component model synthesis using model updating with neural networks

Abstract

A component model synthesis (CMS) has been explored by carrying out model updating with neural networks. Structural system identification was achieved by applying the degree of freedom (DOF)-based reduction method and the inverse perturbation method. Experimental vibration data were restored to a full finite element model to update the numerical model. The experimental data were obtained using the specific sensor location selection method. The system identification was improved using the system equivalent reduction-expansion process by carrying out proper orthogonal decomposition. The proposed CMS was applied for a tank bracket model by simplifying the bolt assembly. Top and bottom parts of the tank bracket were each constructed to have a modally equivalent model. The convolutional neural network was adopted by training the density, Young's modulus and contact properties to improve the computational efficiency.