Design Optimization of Sandwich Composite Armors for Blast Mitigation Using Bayesian Optimization with Single and Multi-Fidelity Data

The most common and lethal weapons against military vehicles are the improvised explosive devices (IEDs). In an explosion, critical cabin’s penetrations and high accelerations can cause serious injuries and death of military personnel. This investigation uses single and multi-fidelity Bayesian optimization (BO) to design sandwich composite armors for blast mitigation. BO is an efficient methodology to solve optimization problems that involve black-box functions. The black-box function of this work is the finite element (FE) simulation of the armor subjected to blast. The main two components of BO are the surrogate model of the black-box function and the acquisition function that guides the optimization. In this investigation, the surrogate models are Gaussian Process (GP) regression models and the acquisition function is the multi-objective expected improvement (MEI) function. Information from low and high fidelity FE models is used to train the GP surrogates. The high fidelity model considers the nonlinear behavior of each layer of the composite armor while the low fidelity model only considers the elastic behavior. The sandwich composite is made of four layers: steel, carbon fiber reinforced polymer (CFRP), aluminum honeycomb (HC) and an additional layer of CFRP. The design variables are the thickness of each layer and the fiber orientations of the CFRP laminas. The optimization problem includes constraints over the maximum thickness of the sandwich composite and fiber orientations. Two objective functions are minimized, the cabin’s penetrations and the reaction force at the armor’s supports. The results show that sandwich composites are an excellent alternative to generate lightweight armors. In terms of penetration, an optimized 100-kg composite armor performs similarly to a 195-kg steel armor. The results also show that the multi-fidelity BO approach is an appealing alternative if the number of function evaluations (of the high fidelity FE model) to perform optimization is low (less than 25). However, if the optimization stage can use a large number of function evaluations (larger than 80), the single fidelity BO approach produces better designs.