We present a novel parameter sensitivity analysis framework for damage and failure modeling of particulate composite materials subjected to dynamic loadings. The proposed framework is employed to quantify the contributions of model parameters to the uncertainty in the failure response of energetic materials. The parameter sensitivity is investigated through the global sensitivity analysis (GSA) which decouples the uncertainty or variance in the failure response as a function of uncorrelated parameters in the proposed framework. In view of the high computational cost of conducting microscale simulations over the entire parameter space to capture the characteristics of response surfaces, a discontinuous Gaussian Process (GP) surrogate model, which is based on Gaussian kernel and approximates the failure response surface with a low-cost mapping function, is embedded into the framework. In order to capture the discontinuity in response surfaces, the GP model is integrated with a classification algorithm that identifies the parameter subspace separated by discontinuities in response surfaces. A support vector machine (SVM) classifier is trained to obtain a separating hyperplane on the condition that the number of parameter subspace is known or estimated. On the condition that the number of parameter subspace (or, the number of local surrogate models) cannot be determined directly, the Elbow method is utilized to estimate its value.
The proposed framework is employed to quantify the contribution of material properties and morphological parameters, which define the microscale structure, to the variability of the failure response of particular energetic materials with a polymeric binder (PBX) and polycrystalline particles. The failure behavior is characterized with thermo-mechanical crystal plasticity analyses incorporating the viscoelastic binder and polycrystal particles at the scale of the material microstructure. Particular emphasis is placed on the identification of pa