Adaptive sampling assisted surrogate modeling of initial failure envelopes of composite structures

Composite structures are increasingly analyzed in a multiscale way, where lower scale structures are homogenized and effective properties such as structural stiffness are passed to upper scales. However, strength properties are rarely treated this way to construct a complete failure envelope for the structural element. In this work, we propose to use the Gaussian process to approximate the overall shape of the initial failure envelope in the space of sectional forces and moments of composite beam cross-sections. Tsai-Wu failure criterion is used for the strength analysis at the material level and strength ratio is used as the output to be fitted. To better select training data without guessing and reduce the number of samples, adaptive sampling technique is used. Predicted variance scaled by predicted mean is used as the measure of the score to select the most promising point as the new sample for the next training iteration. Two sets of numerical examples are provided to demonstrate and test the proposed method.

Automated summary

In this work, the Gaussian process is proposed to approximate the overall shape of the initial failure envelope in the space of sectional forces and moments of composite beam cross-sections, using Tsai-Wu failure criterion for strength analysis at the material level. The adaptive sampling technique is used to select training data and the predicted variance scaled by predicted mean is used as a measure of the score to select the most promising point as the new sample for the next training iteration.