# Stochastic uncertainty analysis of damage evolution computed through microstructure-property relations

 Journal Probabilistic Engineering Mechanics 25 (2010) 198-205 Authors Erdem Acar, Kiran N. Solanki, Masoud Rais-Rohani, Mark F. Horstemeyerd Paper PDF File:Acar et al 2009.pdf

Abstract

Uncertainties in material microstructure features can lead to uncertainty in damage predictions based on multiscale microstructure-property models. This paper presents an analytical approach for stochastic uncertainty analysis by using a univariate dimension reduction technique. This approach is used to analyze the effects of uncertainties pertaining to the structure-property relations of an internal state variable plasticity-damage model that predicts failure. The results indicate that the higher the strain the greater the scatter in damage, even when the uncertainties in the material plasticity and microstructure parameters are kept constant. In addition, the mathematical sensitivity analysis results related to damage uncertainty are consistent with the physical nature of damage progression. At the beginning, the initial porosity and void nucleation are shown to drive the damage evolution. Then, void coalescence becomes the dominant mechanism. And finally, when approaching closer to failure, fracture toughness is found to dominate the damage evolution process.

## Introduction

Accurate predictions of damage progression and failure in ductile materials require the capturing of history effects and the modeling of correlations among the various physical scales present in the material, ranging from the atomic-level interactions and the microstructure features to the macroscale behavior at the continuum level. With the help of mathematical models that can capture the multiscale microstructure-property relations, it would be possible to more accurately relate the structural responses, such as stress, strain, and damage, to key material parameters such as grain size, particle size, interfacial strength, and porosity.

Previous works on multiscale microstructure-property modeling includes those of Bammann et al., [1] [2] Horstemeyer [3], Ganapathysubramanian and Zabaras [4] and Shilkrot et al. [5] . For a survey of recent progress in multiscale microstructure-property relationship modeling and simulations, the reader is referred to [6] .

Although microstructure-property relations enable the modeling of history effects as well as the damage progression and failure, the presence of uncertainties in material microstructure features can lead to considerable variation in failure predictions. Recently, Horstemeyer et al. [7] and Solanki [8] used a first-order Taylor series (FOTS) uncertainty analysis to investigate the effects of stochastic uncertainties in the microstructure features and the boundary conditions that characterize the damage evolution in AA356-T6 cast aluminum alloy. In particular, void-nucleation, void-growth, and void-coalescence equations were evaluated and quantified in terms of the sensitivity and stochastic uncertainty of various parameters in the constitutive equations. However, the accuracy of the Taylor expansion method largely depends on the scale of uncertainty in the random parameters and the nonlinearity in the corresponding random response.

This paper extends the work of Horstemeyer et al. [7] by performing a more accurate stochastic uncertainty analysis using the univariate dimension reduction (UDR) technique [9] . UDR is an additive decomposition technique that evaluates the multidimensional integral of a random function by solving a series of one dimensional integrals. As such, UDR offers an efficient approach for the evaluation of statistical moments such as the mean, variance, skewness, and kurtosis of a random response. To fully characterize the uncertainty in a random response such as damage, the estimated moments can be used to find its probability distribution using a suitable probability distribution fitting technique.

Commonly used distribution fitting techniques include the Pearson and Johnson families of distributions, saddlepoint approximations, and generalized lambda distributions. In this work, we use the extended generalized lambda distribution (EGLD), which combines the generalized lambda distribution (GLD) with the generalized beta distribution (GBD). The efficiency and accuracy of distribution fitting via the GLD and the GBD can be found in [10] [11] [12] [13] . After the first four statistical moments of damage are calculated, the parameters of the EGLD are estimated by minimizing the differences between the moments of the EGLD and those obtained through UDR.

The overall procedure of UDR+EGLD uncertainty analysis is as follows. First, damage calculations are performed at some selected points in the random variable space. Next, the first four statistical moments of damage are estimated using UDR. Then, the probability distribution of damage is approximated with the EGLD. The accuracy of this analysis depends on the accuracies of the estimated moments and distribution fitting. Acar et al. [14] have also successfully used the UDR+EGLD approach in the field of structural reliability.

The main focus of this paper is to analyze the effects of stochastic uncertainties in microstructure features of the material on the uncertainty in damage, which is calculated using the microstructure -property relations in a finite element analysis (FEA) of the selected component. Moreover, we aim to quantify the influence of uncertainty of the individual parameters in the constitutive equations on the uncertainty in damage. In metallurgical studies, it is difficult to independently quantify the effects of microstructural parameters when complex interactions are inherent. The quantitative predictions of the numerical study are presented in light of metallurgical findings to assure that the numerical results provide insights that are consistent with physical observations.

This paper is organized as follows. A brief description of the microstructure -property relationship model of Horstemeyer et al. [3] is given in the next section. Section 3 discusses material microstructure characterizations and their uncertainties. Section 4 presents the calculation of statistical moments of damage using univariate dimension reduction. Section 5 gives a brief description of probability distribution fitting by using the extended generalized lambda distribution. Section 6 presents the results of the uncertainty analysis for damage, where sensitivities of various factors are also compared. Finally, concluding remarks appear in Section 7.

## Microstructure-property relationships

An effective method to capture the microstructure-property relationships is by use of internal state variable (ISV) evolution equations, which are formulated at the macroscale level. The ISVs reflect lower length scale microstructural rearrangements so that history effects can be modeled. With the help of such a material model, it would be possible to relate structural responses of interest, such as stress, strain, and toughness, to key material parameters such as particle size, interfacial strength, and spacing at the lower length scale.

The microstructure-property relationship modeling framework used here is that developed by Bammann et al. [1] [2] and extended by Horstemeyer [3] to account for stress-state-dependent damage evolution. The pertinent equations in this model are denoted by the rate of change of the observable and the internal state variables. For the sake of clarity and completeness, a listing of these equations and their relation to the material microstructure is briefly given here with additional details provided in [3]. The first equation is the modified Hooke's law that includes damage, and is given as

 $\underline{\overset{\circ}{\sigma}} = \lambda'(1-\phi)\operatorname{tr}(\underline{D^e})\underline{I}+2\mu'(1-\phi)\underline{D^e}-{{\dot{\phi}}\over{1-\phi}}\underline{\sigma}$ (1)

where $\mu'$ and $\lambda'$ are the temperature dependent shear modulus and the Lamé constant given as follows:

 $\mu'=\mu \left[ 1 - \frac{T}{T_{melt}}\exp \left(aa \left(1-\frac{T}{T_{melt}} \right) \right)\right],$ $\lambda'=K-bb\frac{T}{T_{melt}} - \frac{2}{3}\mu'$ (1a)

where aa is the shear modulus temperature dependent coefficient, bb is the Lamé constant temperature dependent coefficient, $T_{melt}$ is the melting temperature in Kelvin, T is the current temperature in Kelvin, $\mu$ and $K$ are shear and bulk moduli of base material, $\underline{\sigma}$ and $\underline{\overset{\circ}{\sigma}}$ are the Cauchy stress and the co-rotational rate of the Cauchy stress, respectively, $\phi$ is an ISV that represents the damage fraction or state of material within a continuum element in the context of FEA with $\dot{\phi}$ representing its material time derivative, $\underline{D^e}$ is the elastic deformation tensor, and $\underline{I}$ is the second-order identity tensor. The underscore symbol indicates a second-rank tensor. Recognizing that $\underline{D^e}=\underline{D}-\underline{D^p}$, the ISV representing the plastic deformation tensor or inelastic flow rule, $\underline{D^p}$, is given by the relationship

 $\underline{D^p}=f(T)\operatorname{sinh}\left[ \frac{\left\Vert\underline{\sigma'}-\underline{\alpha}\right\Vert-\{R+Y(T)\}(1-\phi)} {V(T)(1-\phi)} \right] \frac{\underline{\sigma'}-\underline{\alpha}}{\left\Vert \underline{\sigma'}-\underline{\alpha}\right\Vert}$ (2)

where $\underline{\sigma'}$ is the deviatoric part of stress tensor, $T$ is the temperature in Kelvin, $\underline{\alpha}$ is the kinematic hardening (an ISV reflecting the effect of geometrically necessary dislocation) and R is the isotropic hardening (an ISV reflecting the effect of statistically stored dislocation density). The function $V(T)$ determines the magnitude of rate-dependence on yielding, $Y(T)$ is the rate-independent yield stress, and $f(T)$ determines when the rate-dependence affects initial yielding. The functions $f(T), V(T)$, and $Y(T)$, which are related to yielding with Arrhenius-type temperature dependence, are given as

 $V(T)=C_1 e^{(-C_2/T)}$ (3a) $Y(T)=C_3 e^{(-C_4/T)}$ (3b) $f(T)=C_5 e^{(-C_6/T)}$ (3c)

where $C_1$ through $C_6$ are the yield-stress-related material parameters obtained through different monotonic stress-state tests (tension, compression, and torsion) at different temperatures and strain rates. The evaluation of $\underline{D^p}$ in Eq. (2) also requires the corotational rate of the kinematic hardening, $\overset{\circ}{\underline{\alpha}}$ , and the material time derivative of isotropic hardening, $\dot{R}$, expressed in a hardening-recovery format as

 $\underline{\overset{\circ}{\alpha}} = \left\{ h(T)\underline{D^p}- \left[ \sqrt{\frac{2}{3}}r_d(T)\left\Vert\underline{D^p}\right\Vert+r_s(T) \right] \left\Vert\underline{\alpha}\right\Vert\underline{\alpha} \right\} \left[ \frac{DCS_0}{DCS} \right]^z$ (4) $\dot{R} = \left\{ H(T)\underline{D^p}- \left[ \sqrt{\frac{2}{3}}R_d(T)\left\Vert\underline{D^p}\right\Vert+R_s(T) \right] R^2 \right\} \left[ \frac{DCS_0}{DCS} \right]^z$ (5)

where DCS0, DCS and z capture the microstructure effect of grain size. In Eqs. (4) and (5), $r_d(T)$ and $R_d(T)$ are scalar functions that describe dynamic recovery whereas $r_s(T)$ and $R_s(T)$ are scalar functions that describe thermal (static) recovery with $h(T)$ and $H(T)$ representing the anisotropic and isotropic hardening modulus, respectively. Hence, the two main types of recovery that are exhibited by populations of dislocations within crystallographic materials are captured in the ISVs. The temperature-dependent functions in Eqs. (4) and (5) are found as

 $r_s(T)=C_{11}e^{(-C_{12}/T)}$ (6) $R_s(T)=C_{17}e^{(-C_{18}/T)}$ (7) $r_d(T)=C_7+\left[ 1 +C_a\left(\frac{4}{27}-\frac{{J'}^2_3}{{J'}^3_2}\right) -C_b\left(\frac{J'_3}{J'_2}\right)^{(3/2)} \right] e^{(-C_8/T)}$ (8) $R_d(T)=C_{13}+\left[ 1 +C_a\left(\frac{4}{27}-\frac{{J'}^2_3}{{J'}^3_2}\right) -C_b\left(\frac{J'_3}{J'_2}\right)^{(3/2)} \right] e^{(-C_{14}/T)}$ (9) $h(T)=C_{9}+\left[ 1 +C_a\left(\frac{4}{27}-\frac{{J'}^2_3}{{J'}^3_2}\right) -C_b\left(\frac{J'_3}{J'_2}\right)^{(3/2)} \right] e^{(-C_{8}/T)}-C_{10}T$ [/itex] (10) $H(T)=C_{15}+\left[ 1 +C_a\left(\frac{4}{27}-\frac{{J'}^2_3}{{J'}^3_2}\right) -C_b\left(\frac{J'_3}{J'_2}\right)^{(3/2)} \right] e^{(-C_{8}/T)}-C_{16}T$ (11)

where $J'_2=\frac{1}{2}(\underline{\sigma'}-\underline{\alpha})^2, J'_3 = \frac{1}{3}(\underline{\sigma'}-\underline{\alpha})^3$ $C_7$ through $C_{12}$ are the material plasticity parameters related to kinematic hardening and recovery terms, C13 through C18 are the material plasticity parameters related to isotropic hardening and recovery terms, and Ca and Cb are the material plasticity parameters related to dynamic recovery and anisotropic hardening terms. Constants C1 through C18 are found from tension, compression and shear tests at different temperatures and strain rates.

The mechanical properties of a material depend upon the microdefects within its structure that can change as a result of deformation. When the number of microdefects accumulates, damage is said to have grown. The three components of damage progression mechanism are void nucleation, growth and coalescence from second-phase particles and pores. In this regard, the material time derivative of damage, $\dot{\phi}$, is expressed as

 $\dot{\phi}=\left(\dot{\phi}_{particles}+\dot{\phi}_{pores}\right)C + \left(\phi_{particles}+\phi_{pores}\right)\dot{C}$ (12)

### Insert Here

TODO: Finish section

## Material microstructure features and associated uncertainties

The microstructure of a typical metallic material contains a large number of microdefects such as microcracks, dislocations, pores, and decohesions. Some of these defects are induced during the manufacturing process and are present before the material is subjected to mechanical loads and thermal fields. In general, these defects are small and distributed throughout most of the volume.

TODO: Finish section

## Calculation of statistical moments of damage using univariate dimension reduction

As noted earlier, uncertainty analysis requires the calculation of the statistical moments of the response function. Note that, in this case, the response function of interest is damage. The easiest way to estimate the first two statistical moments of a response function Y.X/ is to use a first-order Taylor series (FOTS) approximation, which gives

TODO: Finish section

## Estimating the probability distribution of damage using the EGLD

Approximating the distribution of a random variable using a few of its statistical moments has been of interest to many researchers. Usually, the first four moments are used for this purpose. Popular distribution fitting techniques include the Johnson distribution, Pearson distribution, saddlepoint approximations, and generalized lambda distributions. The main disadvantage of the Johnson distribution is that it is not very easy to determine the four parameters from the moments of the sample data. The major drawback of the Pearson family of distributions can lead to unstable results near the boundaries of families in the skewness�kurtosis plane [18], and for the saddlepoint approximation, the main shortcoming is that

TODO: Finish section

## Uncertainty analysis of damage evolution

### Uncertainty characterization of damage at different strain values

An un-notched tension specimen made of castAA356-T6 is loaded at a strain rate of 10􀀀3 1=s until the point of failure. The plot of damage growth as a function of strain, ", is shown in Fig. 4. The plots of damage probability distributions at different points along the damage�strain curve (Fig. 4) are shown in Fig. 5. The scatter (COV) in damage (C�) and mean value of damage appear to increase as the strain value increases. The comparison of UDR+EGLDbased PDF curves with those found using MCS (of sample size 10,000) indicates a good agreement between the two distributions. In addition, normal distribution fits to damage are also shown in Fig. 5. Since the number of random variables is large (38 random variables), in view of the central limit theorem, the probability distribution of damage can also be represented reasonably well with a normal distribution.

TODO: Finish section

### Sensitivity of damage uncertainty at different strain values

The sensitivity of damage with respect to an input random variable Xi is measured using the formula

TODO: Finish section

## Concluding remarks

This paper has presented a new approach for stochastic uncertainty analysis using the combination of univariate dimension reduction (UDR) technique and the extended generalized lambda distribution (EGLD). This approach was used to perform an uncertainty analysis on an un-notched cast AA356-T6 specimen using an internal state variable plasticity�damage model. The effects of uncertainty in material microstructural features (i.e., voids, cracks, inclusions) on damage initiation and evolution (or accumulation) were investigated. The uncertainty analysis results based on UDR+EGLD were compared with those obtained from first-order Taylor series (FOTS) expansion and Monte Carlo simulation. The proposed approach is found to be computationally efficient and provides more accurate estimates of parametric uncertainty than the FOTS. The influence of uncertainty in individual parameters appearing in the constitutive equations on damage uncertainty was also studied. From the results obtained in this study, we can draw the following conclusions: (i) the scatter in damage, as measured by the coefficient of variation of damage, can increase with strain even though the uncertainties in the input variables are kept fixed; (ii) the sensitivities of damage uncertainty to the uncertainties in the input random variables depend on the strain values. As the strain value changes (i.e., as damage evolves), the importance of the random variables changes; (iii) the sensitivities are found to be consistent with the physics of the damage progression. At the very beginning, the void properties (initial size and growth parameters) are found to drive the damage evolution. Then, void coalescence becomes the main driver, and finally, near the failure condition, macroscopic properties such as fracture toughness dominate the damage evolution process.

## Acknowledgements

This material is based upon work supported by the Department of Energy under Award Number DE-FC26-06NT42755. This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed therein do not necessarily state or reflect those of the United States Government or any agency thereof.

## References

1. 1.0 1.1 Bammann DJ, Chiesa ML, Horstemeyer MF, Weingarten LI. Failure in ductile materials using finite element methods, structural crashworthiness and failure. In: Wierzbicki T, Jones N, editors. Elsevier applied science, The Universities Press (Belfast) Ltd.; 1993.
2. 2.0 2.1 Bammann DJ, Chiesa ML, Johnson GC. Modeling large deformation and failure in manufacturing processes. In: Tatsumi T, Wannabe E, Kambe T, editors. Theoretical and applied mechanics, Elsevier Science; 1996. p. 359-76.
3. 3.0 3.1 3.2 3.3 Horstemeyer MF. From atoms to autos a new design paradigm using microstructure-property modeling, Part 1: Monotonic loading conditions. Sandia National Laboratories, Sand2000-8662, March 2001.
4. Ganapathysubramanian S, Zabaras N. Design across length scales: A reducedorder model of polycrystal plasticity for the control of microstructuresensitive material properties. Computer Methods in Applied Mechanics and Engineering 2004;193:5017-34.
5. Shilkrot LE, Miller RE, Curtin WE. Multiscale plasticity modeling: Coupled atomistics and discrete dislocation mechanics. Journal of the Mechanics and Physics of Solids 2004;52(4):755-87.
6. Graham-Brady LL, Arwade SR, Corr DJ, Gutierrez MA, Breysse D, Grigoriu M, Zabaras N. Probability and materials: From nano- to macro-scale: A summary. Probabilistic Engineering Mechanics 2006;21:193-9.
7. 7.0 7.1 Horstemeyer MF, Solanki K, Steele WG. Uncertainty Methodologies to Characterize a Damage Evolution Model. In: Proceeding of international journal of plasticity conference. 2005.
8. Solanki KN. Physically motivated internal state variable form of a higher order damage model for engineering materials with uncertainty. Ph.D. dissertation. Mississippi State University (Eds.); 2008.
9. Rahman S, Xu H. A univariate dimension-reduction method for multidimensional integration in stochastic mechanics. Probabilistic Engineering Mechanics 2004;19(4):393-408.
10. Karian ZE, Dudewicz EJ, McDonald P. The extended generalized lambda distribution system for fitting distributions to data: History, completion of theory, tables, applications, the final word on moment fits. Communications in Statistics-Computation and Simulation 1996;25(3):611-42.
11. Karian Z, Dudewicz E. Fitting statistical distributions to data: The generalised lambda distribution and the generalised bootstrap methods. Boca Raton (Florida): CRC Press; 2000.
12. Lampasi DA, Nicola FD, Podesta L. The generalized lambda distribution for the expression of measurement uncertainty. IEEE instrumentation and measurement conference. 2005.
13. King RAR, MacGillivray HL. Fitting the generalized lambda distribution with location and scale-free shape functionals. In: Proceedings of the symposium on fitting statistical distributions to data. 2006.
14. King RAR, MacGillivray HL. Fitting the generalized lambda distribution with location and scale-free shape functionals. In: Proceedings of the symposium on fitting statistical distributions to data. 2006.