This paper describes an efficient computational method for estimating the probabilistic properties of the maximum microscopic stresses in a unidirectional fiber-reinforced composite material against microscopic random variations of fibers locations. Some microscopic geometrical random variations will cause a large variation of the microscopic stresses, even if the influence on the homogenized elastic properties is small. The random variation of the microscopic stresses will have a significant influence on the apparent strength of composites, and therefore, estimation of the random variation will be important for reliability-based design of a composite structure. Further, for more precise analysis, a unit cell containing many inclusions should be employed. When the number of random variables becomes large, a multipoint approximation-based approach will not be appropriate. Therefore, a computational approach with a local surrogate constructed by a successive sensitivity analysis is proposed in this paper. The realizations of the microscopic stresses are estimated with the successive sensitivity-based local surrogate, and the probabilistic properties of the stresses are estimated with using the approximated realizations in the Monte Carlo simulation. As an example, the multiscale stochastic stress analysis of a unidirectional fiber-reinforced composite plate under unidirectional tensile load along the transverse direction is performed with considering randomness in fibers locations. For this problem, probabilistic properties as the expectation and coefficient of variation of the maximum microscopic stresses in resin are estimated. From comparisons between the direct Monte Carlo simulation and the proposed method, validity and effectiveness of the proposed approach are discussed.
Application of Monte Carlo Simulation to Mesomechanics of Fiber-reinforced Composite Materials.
