Surrogate Modeling of High-Fidelity Fracture Simulations for Real-Time Residual Strength Predictions

A surrogate-model methodology is described for real-time prediction of the residual strength of flight structures with discrete-source damage. Starting with design of experiment, an artificial neural network is developed that takes discrete-source damage parameters as input and then outputs a prediction of the structural residual strength. Target residual-strength values used to train the artificial neural network are derived from three-dimensional finite-element-based fracture simulations. A residual-strength test of a metallic integrally stiffened panel is simulated to show that crack growth and residual strength are determined more accurately in discrete-source damage cases by using an elastic-plastic fracture framework rather than a linear-elastic fracture-mechanics-based method. Improving accuracy of the residual-strength training data would, in turn, improve the accuracy of the surrogate model. When combined, the surrogate-model methodology and high-fidelity fracture simulation framework provide useful tools for adaptive flight technology. [PUBLICATION ABSTRACT]