Multiscale reliability analysis of composite structures based on computer vision

The geometric uncertainties of composite structures are considered in the multiscale reliability problem. A convolutional neural network (CNN) is developed and trained using deep learning to link geometric uncertainties and the randomness of structural responses or performances. The CNN training set includes graphical samples and corresponding stress components and strength characteristics of the lamina. A method for generating a graphical sample is developed, which integrates the stochasticity of the fibre shape, misalignment, arrangement, volume fraction, matrix voids, and stacking sequences of the laminates. The corresponding stress components are simulated using the Kernal-Hashin model and laminate plate theory. A reliability analysis procedure is developed using a Monte Carlo simulation. Numerical cases are presented to demonstrate the validity of the proposed method.

Automated summary

This study considers the geometric uncertainties of composite structures and uses deep learning to develop a convolutional neural network (CNN) that can link geometric uncertainties and the randomness of structural responses or performances. The proposed method is validated through numerical simulations and analysis.