Failure by Fatigue and Stress Rupture in Fiber-Reinforced Composites

Abstract

Composite durability under conditions where the material survives the initial application of load but then deteriorates with time is a major issue in most engineering applications of composites. The degradation can be caused by the propagation of many damage modes: delamination, matrix cracking, interface degradation, creep, and fiber degradation, among others. In many situations, ultimate and total failure is associated with failure of the fiber bundle supporting the dominant portion of the applied load. Here, the lifetime of a composite due to fiber degradation is investigated for two important modes of fiber degradation: fatigue crack growth under cyclic loading and slow crack growth under stress rupture conditions of constant load. In both cases, individual fiber failure is caused by the growth of pre-existing cracks to a critical size via fatigue or rupture, which is modeled here by a Paris law. A simulation model is used to determine the time-dependent stresses on a bundle of uniaxial brittle fibers, each containing an initial distribution of cracks corresponding to a Weibull strength distribution. As individual fibers fail, their stresses are transferred to nearby fibers, which increases the rate of degradation of the nearby fibers. The damage evolution thus accelerates locally, culminating in the very rapid growth of damage across the entire specimen at the failure time. The average failure time for a set of nominally identical specimens is primarily a function of the fiber Weibull modulus, the Paris law exponent in the fatigue or rupture model, the initial fiber strength, and the rate coefficient of the Paris law. Guidance for the major dependencies of the failure time is obtained by considering an analogous “Global Load Sharing” (GLS) model in which broken fibers transfer load equally to all remaining fibers in the same cross-section. The statistical distribution of failure times at fixed composite size and the size scaling of the mean failure time with increasing size, neither of which can be obtained from the GLS model, are presented and discussed.