Phoenix S Leigh, Newman William I
Department of Theoretical and Applied Mechanics, Cornell University, Ithaca, New York 14853, USA.
Phys Rev E Stat Nonlin Soft Matter Phys. 2009 Dec;80(6 Pt 2):066115. doi: 10.1103/PhysRevE.80.066115. Epub 2009 Dec 17.
Fiber bundle models (FBMs) are useful tools in understanding failure processes in a variety of material systems. While the fibers and load sharing assumptions are easily described, FBM analysis is typically difficult. Monte Carlo methods are also hampered by the severe computational demands of large bundle sizes, which overwhelm just as behavior relevant to real materials starts to emerge. For large size scales, interest continues in idealized FBMs that assume either equal load sharing (ELS) or local load sharing (LLS) among fibers, rules that reflect features of real load redistribution in elastic lattices. The present work focuses on a one-dimensional bundle of N fibers under LLS where life consumption in a fiber follows a power law in its load, with exponent rho , and integrated over time. This life consumption function is further embodied in a functional form resulting in a Weibull distribution for lifetime under constant fiber stress and with Weibull exponent, beta. Thus the failure rate of a fiber depends on its past load history, except for beta=1 . We develop asymptotic results validated by Monte Carlo simulation using a computational algorithm developed in our previous work [Phys. Rev. E 63, 021507 (2001)] that greatly increases the size, N , of treatable bundles (e.g., 10(6) fibers in 10(3) realizations). In particular, our algorithm is O(N ln N) in contrast with former algorithms which were O(N2) making this investigation possible. Regimes are found for (beta,rho) pairs that yield contrasting behavior for large N. For rho>1 and large N, brittle weakest volume behavior emerges in terms of characteristic elements (groupings of fibers) derived from critical cluster formation, and the lifetime eventually goes to zero as N-->infinity , unlike ELS, which yields a finite limiting mean. For 1/2<or=rho<or=1 , however, LLS has remarkably similar behavior to ELS (appearing to be virtually identical for rho=1 ) with an asymptotic Gaussian lifetime distribution and a finite limiting mean for large N. The coefficient of variation follows a power law in increasing N but, except for rho=1, the value of the negative exponent is clearly less than 1/2 unlike in ELS bundles where the exponent remains 1/2 for 1/2<rho<or=1. For sufficiently small values 0<rho1, a transition occurs, depending on beta , whereby LLS bundle lifetimes become dominated by a few long-lived fibers. Thus the bundle lifetime appears to approximately follow an extreme-value distribution for the longest lived of a parallel group of independent elements, which applies exactly to rho=0. The lower the value of beta , the higher the transition value of rho , below which such extreme-value behavior occurs. No evidence was found for limiting Gaussian behavior for rho>1 but with 0<beta(rho+1)<1, as might be conjectured from quasistatic bundle models where beta(rho+1) mimics the Weibull exponent for fiber strength.
纤维束模型(FBMs)是理解各种材料系统失效过程的有用工具。虽然纤维和载荷分担假设很容易描述,但FBM分析通常很困难。蒙特卡罗方法也受到大尺寸纤维束严重计算需求的阻碍,当与真实材料相关的行为开始出现时,这种需求就变得难以承受。对于大尺寸尺度,人们继续关注理想化的FBMs,这些模型假设纤维之间要么是等载荷分担(ELS),要么是局部载荷分担(LLS),这些规则反映了弹性晶格中真实载荷重新分布的特征。目前的工作重点是在LLS条件下的一维N纤维束,其中纤维中的寿命消耗在其载荷上遵循幂律,指数为rho,并随时间积分。这种寿命消耗函数进一步体现在一种函数形式中,导致在恒定纤维应力下寿命服从威布尔分布,威布尔指数为beta。因此,除了beta = 1的情况外,纤维的失效率取决于其过去的载荷历史。我们利用在我们之前的工作[《物理评论E》63, 021507 (2001)]中开发的计算算法,通过蒙特卡罗模拟验证了渐近结果,该算法大大增加了可处理纤维束的尺寸N(例如,在10^3次实现中有10^6根纤维)。特别是,我们的算法是O(N ln N),而以前的算法是O(N^2),这使得这项研究成为可能。发现了(beta, rho)对的区域,对于大N产生了对比行为。对于rho > 1和大N,从临界团簇形成导出的特征元素(纤维组)方面出现了脆性最弱体积行为,并且随着N趋于无穷大,寿命最终趋于零,这与ELS不同,ELS产生有限的极限均值。然而,对于1/2 <= rho <= 1,LLS与ELS具有非常相似的行为(对于rho = 1似乎几乎相同),具有渐近高斯寿命分布和大N时的有限极限均值。变异系数在增加N时遵循幂律,但除了rho = 1外,负指数的值明显小于1/2,这与ELS纤维束不同,在ELS纤维束中,对于1/2 < rho <= 1,指数保持为1/2。对于足够小的值0 < rho < 1,会发生转变,这取决于beta,由此LLS纤维束寿命由少数长寿命纤维主导。因此,纤维束寿命似乎大致遵循一组独立元素并行组中最长寿命的极值分布,这恰好适用于rho = 0。beta值越低,rho的转变值越高,低于该值会出现这种极值行为。没有发现对于rho > 1但0 < beta(rho + 1) < 1的极限高斯行为的证据,这可能是从准静态纤维束模型推测出来的,其中beta(rho + 1)模拟纤维强度的威布尔指数。