A stochastic model based on fiber breakage and matrix creep for the stress-rupture failure of unidirectional continuous fiber composites

Abstract

Stress rupture is a time-dependent failure mode occurring in unidirectional fiber composites under sustained tensile loads, resulting in highly variable lifetimes. Stress-rupture is of particular concern in composite overwrapped pressure vessels (COPVs) since it is unpredictable, and has catastrophic consequences. At the micromechanical level, stress rupture begins with the breakdown of individual fibers at random flaws, followed by local load-transfer to intact neighbors through shear stress in the matrix. Over time, the matrix creeps in shear causing lengthening overload zones around fiber breaks, resulting in even more fiber breaks, and eventually, formation of a catastrophically unstable break cluster. Current reliability models are direct extensions of classic stochastic breakdown models for a single fiber, and do not reflect such micromechanical activity. These models are adequate for modeling composite stress rupture under a constant load, however, they may be unrealistic under more complex loading profiles, such as a constant load that follows a brief ‘proof test’ at a load level up to 1.5 times this constant load. For carbon fiber/epoxy COPVs, current models predict a reliability, conditioned on survival of a proof test, that is always higher than the reliability without such a proof test. Concern exists that this is incorrect, and that a proof test may result in reduced reliability over time. While the failure probability during a proof test may be very low, overwrap damage occurs nonetheless in the form of a large number of fibers breaks that would not occur otherwise based on fiber Weibull strength statistics. This phenomenon of increased fiber breakage during a proof test is captured in the model we develop and that specifically builds on the micromechanical failure process described above. For typical proof-test load ratios, the model predicts conditional reliabilities for lifetime that are typically much lower than those calculated in the absence of a proof test.