Computational Materials System

Design 1st Edition Dongwon Shin

Visit to download the full and correct content document:
https://textbookfull.com/product/computational-materials-system-design-1st-edition-do
ngwon-shin/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Cyber-Physical System Design from an Architecture

Analysis Viewpoint: Communications of NII Shonan
Meetings 1st Edition Shin Nakajima

https://textbookfull.com/product/cyber-physical-system-design-
from-an-architecture-analysis-viewpoint-communications-of-nii-
shonan-meetings-1st-edition-shin-nakajima/

Computational and experimental analysis of functional

materials 1st Edition Reshetnyak

https://textbookfull.com/product/computational-and-experimental-
analysis-of-functional-materials-1st-edition-reshetnyak/

Embedded System Design Introduction to SoC System

Architecture 1st Edition Mohit Arora

https://textbookfull.com/product/embedded-system-design-
introduction-to-soc-system-architecture-1st-edition-mohit-arora/

Computational Materials Science: An Introduction,

Second Edition Lee

https://textbookfull.com/product/computational-materials-science-
an-introduction-second-edition-lee/
Developments in Strategic Materials and Computational
Design V Ceramic Engineering and Science Proceedings
Volume 35 Issue 8 1st Edition Waltraud M. Kriven

https://textbookfull.com/product/developments-in-strategic-
materials-and-computational-design-v-ceramic-engineering-and-
science-proceedings-volume-35-issue-8-1st-edition-waltraud-m-
kriven/

Materials Selection in Mechanical Design Ashby

https://textbookfull.com/product/materials-selection-in-
mechanical-design-ashby/

Computational Chemistry Methodology in Structural

Biology and Materials Sciences 1st Edition Tanmoy
Chakraborty

https://textbookfull.com/product/computational-chemistry-
methodology-in-structural-biology-and-materials-sciences-1st-
edition-tanmoy-chakraborty/

Computational Drug Discovery and Design Mohini Gore

https://textbookfull.com/product/computational-drug-discovery-
and-design-mohini-gore/

Computational Homogenization of Heterogeneous Materials

with Finite Elements Julien Yvonnet

https://textbookfull.com/product/computational-homogenization-of-
heterogeneous-materials-with-finite-elements-julien-yvonnet/
Dongwon Shin · James Saal Editors

Computational
Materials
System Design
Computational Materials System Design
Dongwon Shin • James Saal
Editors

Computational Materials
System Design

123
Editors
Dongwon Shin James Saal
Oak Ridge National Laboratory QuesTek Innovations, LLC
Oak Ridge, TN, USA Evanston, IL, USA

ISBN 978-3-319-68278-5 ISBN 978-3-319-68280-8 (eBook)

https://doi.org/10.1007/978-3-319-68280-8

Library of Congress Control Number: 2017955948

© Springer International Publishing AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

As technological advancement accelerates, critical technology sectors such as

energy, propulsion, and computation are relying on ever more complex materials
systems for improvements in efficiency and capability. As a result, materials
development is increasing in both necessity and difficulty. Over the past two
decades, computational materials modeling and simulation tools have become an
essential component to modern materials design, development, and deployment.
Our understanding of fundamental materials behavior has improved alongside our
computational capabilities, enabling quantitative, predictive modeling of materials
systems.
Two classes of computational tools are critical for materials design: process-
structure and structure-property. Process-structure tools make predictions of the
microstructure of a material (e.g., crystal structure, grain size, defect concentration,
etc.) based on its process history (e.g., cooling rate, annealing time, sintering
temperature, etc.). Process-structure tools consist primarily of thermodynamic
and kinetic models. Thermodynamic computational models (such as CALPHAD)
make predictions by fitting composition-dependent free energy functions to phase-
based thermodynamic properties (such as heat capacity and melting temperature)
and then calculating phase equilibria by minimizing the free energy. Similarly,
kinetic models (such as phase-field or diffusion simulations) add the capability
to predict the time evolution of microstructure for a given processing history by
including atomic mobility and mesoscale-order parameters as an additional layer of
complexity. Structure-property tools make predictions of long-term performance-
critical materials properties (e.g., yield strength, thermal conductivity, ductility, etc.)
from the material’s microstructure. Both process-structure and structure-property
models are only as accurate as the input data on which the principal materials-
dependent parameters are based. Limited experimental data has made theoretically
predicted data, particularly from density functional theory (DFT), critical for the
rapid development and improvement of such tools for materials design.
This book provides a practical overview of the current toolsets available to the
materials designer. Such tools have already demonstrated success for the develop-
ment of novel materials systems, including structural metal alloys and functional

v
vi Preface

materials. Experts in various computational modeling fields provide summaries

of tools, methods, and examples that illustrate how to implement computational
materials design.
This book begins with a chapter by Prof. David McDowell, and provides a broad
overview of the philosophy of materials systems design, with a focus on structure-
property models. Thermodynamics forms the foundation of any materials design
exercise, and this book thusly continues with two chapters by Profs. Zi-Kui Liu and
In-Ho Jung on the computational thermodynamic (widely known as CALPHAD)
approach. The evolution of materials structure with time is then described in the
next two chapters, with summaries of the phase-field approach by Prof. Katsuyo
Thornton and solidification simulation by Prof. Jung. Methods for developing
structure-property relationships are discussed next, with chapters concerning crystal
plasticity modeling by Prof. McDowell and first-principles electronic structures
calculations based on density functional theory (DFT) by Prof. Shyue Ping Ong.
This book concludes with forward-looking chapters on two emerging techniques
that have the potential to greatly affect the development of materials design models:
high-throughput DFT by Prof. Wei Chen and materials informatics by Dr. Bryce
Meredig.

Oak Ridge, TN, USA Dongwon Shin

Evanston, IL, USA James Saal
Contents

1 Microstructure-Sensitive Computational Structure-Property

Relations in Materials Design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
David L. McDowell
2 Design of Materials Processing Using Computational
Thermodynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Cassie Marker, Austin Ross, and Zi-Kui Liu
3 Applications of Thermodynamic Database to the Kinetic
Steelmaking Process Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Marie-Aline Van Ende and In-Ho Jung
4 Phase Field Modeling of Microstructural Evolution . . . . . . . . . . . . . . . . . . . . . 67
Stephen DeWitt and Katsuyo Thornton
5 1D Solidification Model for the Prediction of Microstructural
Evolution in Light Alloys. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Manas Paliwal and In-Ho Jung
6 Multiscale Crystalline Plasticity for Materials Design . . . . . . . . . . . . . . . . . . . 105
David L. McDowell
7 Ab Initio Molecular Dynamics Studies of Fast Ion Conductors . . . . . . . . 147
Zhuoying Zhu, Zhi Deng, Iek-Heng Chu,
Balachandran Radhakrishnan, and Shyue Ping Ong
8 High-Throughput Computing for Accelerated Materials Discovery . . . 169
Wei Chen
9 Materials Data Infrastructure and Materials Informatics . . . . . . . . . . . . . . 193
Joanne Hill, Arun Mannodi-Kanakkithodi, Ramamurthy Ramprasad,
and Bryce Meredig

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

vii
Chapter 1
Microstructure-Sensitive Computational
Structure-Property Relations in Materials
Design

David L. McDowell

1.1 Introduction

The relation between microstructure and properties or responses is a critical element

in design and development of materials to address property and performance
requirements in applications. It is well known that properties of engineering alloys
derive from a hierarchy of length scales of material structure. In fact, the term
“microstructure” is commonly used to designate structures between atomic-scale
lattice arrangement and the scale of applications, with corresponding physical length
scales ranging from the order of several nm (e.g., individual precipitates, interfaces,
short-range ordering) to hundreds of nm (interfaces, defects) and to microns and
above (arrangement of phases or grains).
A transformational early twenty-first-century trend is to incorporate computa-
tional modeling and simulation of material process-structure and structure-property
relations to augment and reduce the number of costly and time-consuming empirical
methods. The integrated computational materials engineering (ICME) initiative [56,
73] has been embraced by industry as a viable path forward to accelerate mate-
rials development and insertion into products by employing more comprehensive
management of data, process monitoring, and integrated computational modeling
and simulation. This has led more recently to the development of the US Materials
Genome Initiative (MGI) [23, 24], as well as companion thrusts in Europe and Asia
[12] which are directed toward the goal of accelerating discovery and development

D.L. McDowell ()

Woodruff School of Mechanical Engineering, School of Materials Science and Engineering,
Georgia Institute of Technology, Atlanta, GA 30332-0620, USA
e-mail: david.mcdowell@me.gatech.edu

© Springer International Publishing AG 2018 1

D. Shin, J. Saal (eds.), Computational Materials System Design,
https://doi.org/10.1007/978-3-319-68280-8_1
2 D.L. McDowell

of new and improved materials via a combined strategy of experiments, theory

and computational simulation, and data science, with emphasis on high-throughput
protocols.
It is most relevant and accurate to consider the role of computational modeling
and simulation as one that principally provides decision support for materials design
and development. In other words, a metric for measuring success in ICME or MGI
is the increase of the fraction of decisions made in the critical path of materials dis-
covery, development, optimization, certification, and deployment that are informed
via modeling and simulation as opposed to experiments. In simulating structure-
property relations for metals and alloys, however, there are tremendous challenges.
One challenge pertains to the quantitative representation of the hierarchical nature of
material structure at various length scales. This challenge has largely dominated the
discussion of ICME over the past decade, focusing attention on schema for materials
characterization and digital material representation at characteristic length scales
ranging from tens of nm to tens of microns. Another challenge that has become
more recently embraced as computational methods and tools develop further is
that of multiscale modeling – primarily based on modeling responses of aggregate
structures at various levels of structure hierarchy.
Both the state-of-the-art and outstanding gaps for computational multiscale
structure-property relations will be presented, focusing on mechanical properties
of single-crystal and polycrystalline metal alloys. We discuss material structure
hierarchy and corresponding demands placed on models at various length and
time scales to provide physically faithful results. We then discuss how materials
design differs from multiscale modeling and how the latter serves the purposes
of the former by assisting to identify the degree of coupling of phenomena at
different length and time scales, as well as providing support for materials design
and development decisions. We close with discussion of some recent inverse design
approaches based on bottom-up multiscale modeling, with the goal of addressing
top-down relations of performance requirements to properties, to structure, and then
to process path.

1.2 Material Structure Hierarchy

Hierarchy of material structure is prevalent in engineering materials and offers

a means by which multiple property requirements can be met simultaneously to
deliver required performance. Responses of metals to applied thermomechani-
cal loads are significantly influenced by generation and motion of dislocations.
Properties of interest in this case involve nonequilibrium evolution of structure.
The most useful material forms are metastable. Figure 1.1 depicts five levels
of structure hierarchy, ranging from the discrete atomic structure of lattices and
interfaces (atomistics) to migration and interaction of dislocation line segments
(discrete dislocations), to collective pattern formation of dislocations (substructure),
to heterogeneous plastic flow within sets of grains/phases (polycrystal plasticity),
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 3

Fig. 1.1 Hierarchy of length scales in metal plasticity ranging from atomic (dislocation cores)
to patterns of dislocations, to multiple grains, and to the macroscopic scale. The primary gap in
modeling and simulation in multiscale modeling exists between the scales of atomistic simulations
and dislocation pattern formation, with discrete dislocation simulations playing an increasing role
(Adapted from McDowell and Dunne [57])

and finally up to the scale of engineering applications where underlying structure

is “smeared” by considering a macroscopic set of properties or responses. The
minimum length scale associated with each of these levels is also shown in Fig.
1.1 and ranges from interatomic spacing to mean free path for dislocations, to
grain size, and onto characteristic dimensions of components or structures. Of
course, the associated issue of relative time scale for processes at each level of
structure hierarchy is implicit in the dynamic to thermodynamic transition indicated
from left to right in Fig. 1.1; dynamic atomistic simulations typically range from
femtoseconds to nanoseconds, while discrete dislocation dynamics simulations
can extend to the ms regime. At the other end of the spectrum, polycrystal and
macroscale plasticity models can address time scales relevant to that of large-scale
laboratory specimens or applications, from seconds to years. Of course, models
of both discrete and continuous type can be cast in quasi-static form, neglecting
dynamics and kinetics in favor of appeal to energy minimization.
Models addressing phenomena corresponding to each of the scales considered
to the right in Fig. 1.1 are typically of increasingly reduced order and reflect
cooperative thermodynamics and kinetics of dislocated crystals. On the other
hand, discrete models that apply to scales at the left in Fig. 1.1 are either fully
dynamical or employ some kind of damped dynamical scheme, tracking locations of
individual particles or defects. Degrees of freedom (DOF) necessary to characterize
the structure of a fixed volume of material decrease as resolution decreases from
left to right in Fig. 1.1; in so doing, information necessary to characterize the
4 D.L. McDowell

dynamical state is shed in favor of a reduced-order continuum thermodynamic

description. Stochasticity is recognized as a hallmark of these intermediate scales
or “mesoscales,” although often deterministic models are applied.
Model constructs that relate structure to responses or properties at various levels
of the length scale hierarchy in Fig. 1.1 differ considerably, both in terms of
their numbers of degrees of freedom and in terms of their fundamental character.
Here the term “model construct” is considered to apply both to simulations and
interpretation of experimental information; as also shown along the top in Fig.
1.1, experimental techniques collect information that is resolved or averaged over
different length and time scales at various levels to measure structure, either in
time-resolved or asynchronous manner. In fact, it is difficult to entirely separate
the issue of the form or structure of a model from the type of experiments that
are used to support and calibrate it. Certain standard models serve as a basis for
interpretation of such measurements of attributes (e.g., time-averaged atomic posi-
tions, explicit dislocation configurations, dislocation densities and lattice curvature,
crystal deformations, and relative orientations of grains or phases). Simulations at
the atomic scale (either based on some form of ground-state density functional
theory or atomistic simulations that employ energy minimization) are useful to
establish understanding of stable or metastable phases, energy functions for defects
or interfaces, dislocation core structures and their effects on material yield strength
and rate sensitivity, constants for bulk elasticity and thermal expansion, diffusion
constants, as well as transition states for unit processes of individual dislocation-
dislocation, dislocation-interface, or dislocation-obstacle interactions. They are
also essential for explicit modeling of point defect interactions with surfaces and
dislocations. Modeling evolution of dislocations in a lattice is complex since they
evolve in a nonequilibrium behavior through sequences of metastable states that can
relax with time at temperature, driven by short- and long-range internal stresses.
Prospects for modeling collective processes at intermediate levels of structure
hierarchy are quite challenging and underdeveloped, owing to spatial heterogeneity
of structure, short-range but persistent dislocation core effects, metastability of
dislocation structures, and kinetics of multiple competing mechanisms in inelastic
deformation. Heterogeneity of dislocation substructure can give rise to nonintuitive
spatiotemporal correlations associated with populations of point and line defects,
as spatial statistics of evolving components of structure must be distinguished
from mean populations. This necessitates introduction of the concept of immobile
and mobile populations of dislocations, which often involves semantics or sub-
jective judgment. For all these reasons, the description of collective, many-body
mesoscopic dislocation evolution processes is extremely challenging, and multiple
strands of associated materials theory and modeling approaches have evolved.
Discussion of these issues and challenges in the context of multiscale crystalline
plasticity is the subject of a later section of this chapter.
Models that pertain to atomistic or discrete dislocation levels can be used in
bottom-up fashion to inform both model form and model parameters at higher levels
of hierarchy, for example, statistical strengthening models or other responses at the
scale of individual grains. Most models (even interatomic potentials) are commonly
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 5

calibrated to information from laboratory experiments obtained at higher length and

time scales. Models that address dislocation patterning and polycrystal plasticity
are often primarily calibrated using experimental information from the top-down
experiments that address length and time scales well above scales of individual
cells (patterns) or grains (polycrystal plasticity). The manner in which models and
experimental information pertaining to different levels of structure hierarchy in Fig.
1.1 are related is manifested in one of two ways, comprising either hierarchical (one-
way, bottom-up) or concurrent (two-way) multiscale schemes. No single multiscale
modeling strategy is superior or even suitable for all cases. Concurrent multiscale
modeling schemes exercise simulations for a range of models of varying fidelity
in length scales over the same time frame. These models can either apply to the
same spatial domain with differing spatial resolution and degrees of freedom or can
be applied in adjacent domains with schemes for communication and consistency
of model responses between coarse-grained and highly resolved regions. The latter
schemes are typically referred to as domain decomposition. Hierarchical multiscale
modeling schemes typically pass responses from modeling results conducted at each
successive scale to the next higher scale, with the intent to instruct model form and
parameters of the latter. Additionally, they may be hierarchical in either length or
time, adding flexibility to the framing of the multiscale modeling problem. More
details and examples will be given later in this chapter.

1.3 Materials Design and Material Structure Hierarchy

It is instructive to consider an example of how this length scale hierarchy spans

not only materials but also devices and structures that are useful in applications,
for example, aerospace propulsion components shown in Fig. 1.2. Most readers
will be familiar with the notion of components or parts, subassemblies (collections
of parts), assemblies, and overall systems, shown to the right of the solid vertical
line in Fig. 1.2. On the other hand, it is equally important to understand that
in reality these scales of hierarchy extend all the way to atomic scale, shown
to the left in this figure. In this case, various scales of atomic arrangements,
interface structures between phases, and distribution of phases or grains confer
properties that are necessary to deliver require part level performance, just as well-
designed and purposed subassemblies serve the needs at higher levels of the overall
system. There are two major differences between hierarchies to the left and to
the right of the solid vertical line in Fig. 1.2. To the right, focus is placed on
systems assemblages of parts, subassemblies, and assemblies. Various interactions
are involved – geometric compatibility, interfaces (mechanical, chemical, electrical,
magnetic, etc.), and functional aspects. These interactions are governed by part
shape, surface characteristics, properties, and contact conformity with neighboring
parts. In contrast, to the left of the vertical black bar, interactions occur between
atoms and phases and are complex and nonlocal, dictated not by simple laws of
local particle and surface interactions but rather by nonlocal material interactions,
6 D.L. McDowell

Fig. 1.2 Extension of systems-based, top-down materials design from part, subassemblies, assem-
blies, and components to hierarchical levels of material structure, treating levels of material
structure and associated responses effectively as subsystem

solid-state thermodynamics, and kinetics. Although design flexibility can be high

for both, degrees of freedom are extremely high for materials with hierarchy,
and controllability of these structures is more complex. The multilevel materials
design contribution to this systems design problem is characterized by very high
uncertainty in process-structure-property relations [58].
We start with the conception of structure hierarchy illustrated in Fig. 1.2 to
provide the context for distinguishing goals and methods of modeling across
levels of structure hierarchy of a metallic material, shown in Fig. 1.1, from
those of systems comprised of an assembly of parts and subsystems. Both are
useful, and in fact vital, from a systems engineering perspective. Historically,
these two classes of hierarchy (materials, product systems) have been addressed
by distinct communities (computational materials science and mechanics on the
one hand and multidisciplinary design optimization on the other) with very weak
communication based largely on tabulated material property sets to represent the
materials component. The solid vertical line in Fig. 1.2 represents the associated
materials selection problem, which demarcates materials design and development
from systems-level engineering design; it involves selection based on tabulated data
from models or experiments and may be approached using Ashby maps [2, 17] and
informatics, e.g., data mining, combinatorics, and so forth [2, 3, 88].
Since the term “materials design” may mean different things to different people,
it is useful to state its definition in the present context. Our use of the term materials
design implies the top-down driven, simulation-assisted, decision-based design
of material hierarchy conducted in a manner that meets ranged sets of product
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 7

performance requirements. Hence, materials design is a decision-making process.

This definition of a simulation-assisted (rather than wholly simulation-based)
strategy aligns closely with that of ICME [73] since the goal is to reduce but not
eliminate empirical routes. In contrast, materials discovery typically has a somewhat
more narrow focus on bottom-up, cyber-enabled search for candidate materials,
typically involving data mining or informatics, with emphasis on simulation-based
science. Materials discovery typically relates to directed searching for bulk phase
and interface structures and properties that hold promise for specific application
requirements based on density functional theory and atomistic simulations. As
explained later, we contend that multiscale modeling is not equivalent to our
conception of materials design. It should be emphasized, of course, that simulation-
assisted design of materials should be model based; models are used as a basis
for designing materials regardless of whether they constructed from theory, sim-
ulation, or empirical evidence. Physically based models that incorporate realistic
mechanisms are of particular value in this regard, as they can reduce uncertainty in
providing decision support for materials design and development.
Objectives for designing metallic systems to tailor the hierarchy of material
structure to deliver required performance requirements related to mechanical prop-
erties/responses include, but are not limited to:
• Control of evolution of structure (e.g., plasticity, phase transformation, diffusion,
etc.)
• Resistance to environmental attack and corrosion
• Resistance to high-temperature creep and coarsening
• Targeted porosity, second phases, inclusions, and process-induced defects or
anomalies
• Resistance to shear localization and formability
• Fatigue and fracture resistance achieved with multiple phases, precipitate
strengthening, crystallographic texture, and grain boundary networks
• Control of phase morphologies of alloy systems for strength, ductility, and
multifunctional applications
• Surface conditions and residual stresses
• Elastic stiffness
Figure 1.3 expresses Olson’s conception [66] of inductive, top-down design of
materials to meet a specified set of performance requirements, targeting levels of
material structure to the left of the vertical line in Fig. 1.2. Deductive, bottom-
up strategies based largely on experiments and increasingly on computational
simulation are typically used to construct process-structure and structure-property
relations. Then, properties can be related to ranged sets of performance requirements
in systems design enterprise via a compromise decision support problem involving
property trade-offs that is framed using goal programming [58].
Empirical exploration of materials process route has been common for alloys and
composites, and structure-property assessments are then carried out via experiments
and computation once interesting structures are realized in the laboratory. These
processes are then scaled up to prototype level to determine if the resulting property
8 D.L. McDowell

Fig. 1.3 Olson’s overlapping

Venn diagram outlining
process-structure-property-
performance relations [66],
with top-down, goals/means
supporting design
distinguished from
bottom-up, deductive
scientific methods for
establishing these relations

sets deliver required or enhanced performance. This is the conventional paradigm.

Note that this requires either a large number of overall iterations or, as more
commonly the case, settling on design solutions that are far from optimal in some
sense. In the new paradigm set forth by Olson in Fig. 1.3, attention shifts early in
the process to top-down, inductive design exploration based on existing bottom-up
experiments coupled with physically based models to accelerate the coupling of
process-structure-property relations with systems-level design requirements and to
reduce the number of iterations in materials design and development.
The process-structure-property-performance diagram shown in Fig. 1.3 should
not be confused with the hierarchy of length scales appearing in Figs. 1.1 and
1.2. In fact, as indicated in Fig. 1.4, fleshing out process-structure and structure-
property relations at or across two successive scales of hierarchy potentially requires
application of sets of models and experiments that span the complete hierarchy
of length scales in Figs. 1.1 and 1.2. Moreover, each level of structure hierarchy
shown in Fig. 1.2 can be addressed by an appropriate model cast at that level of
hierarchy, with degrees of freedom corresponding to the material representation
at that scale (e.g., atoms, dislocations, dislocation patterns, and grains as shown
in Fig. 1.1). The collection of such models to serve the purposes of connecting
information that relates to higher-scale response in Fig. 1.2 constitutes a suite of
so-called hierarchical multiscale models. It is clear that material response at each
level of idealization of material structure in Fig. 1.2 depends on finer-scale structure
and processes, so the goal of hierarchical multiscale modeling is to provide decision
support in tailoring of process-structure and structure-property relations at various
levels of structure hierarchy. Furthermore, it should be evident that modeling and
simulation can provide only partial support for materials design and development
for target applications (simulation-assisted design) – experiments and prototyping
are indispensible in providing decision support.
A practical and compelling goal of this kind of systems strategy for design and
development of materials is to replace an increasing fraction of otherwise empirical
decisions in the materials development and certification cycle with those informed
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 9

Fig. 1.4 Distinction of material structure hierarchy from Olson’s process-structure-property-

performance diagram, clarifying that the latter relations must be established at each of multiple
levels of material structure hierarchy. Physical responses and models at each level may be weakly
or strongly coupled, an issue to be sorted out using multiscale modeling

by computational modeling and simulation or with high-throughput experimental

strategies. For example, if 90% of materials development decisions are made on the
basis of prior experience and experimental evidence, can this be decreased to 80%
or 70% by using modeling and simulation at the same overall level of uncertainty in
material performance? This might represent years shaved off the development and
certification cycle. Of course, the evaluation of effectiveness of this strategy requires
that (1) materials development workflows and corresponding decisions must be
tracked such that reduction of time can be quantified and (2) uncertainty must be
quantified with regard to all relevant information to ensure the designs are feasible
and meet specified ranges with high likelihood.
Ultimately, materials design and development is a decision-making process
that mitigates various sources of uncertainty, with multiscale modeling providing
decision support in this regard, as shown in Fig. 1.5. Moreover, reduction of
the design cycle time necessitates rapid decision support, and this requires the
ability to rapidly explore the design space. So as a vehicle for supporting materials
design, there are two additional requirements of multiscale modeling: (1) it must be
efficient and fast acting, except for cases where high-value information is required
to proceed further with certitude in design, and (2) it must address uncertainty
of models and experiments at each scale, as well as uncertainty propagation
through a chain of models and/or experiments at different levels of hierarchy.
Consideration of uncertainty in multiscale modeling in providing decision support
in materials design has important implications. First, the range of models across
scales shown in Fig. 1.1, when mapped onto the levels of hierarchy in Fig. 1.2,
are most practically approached in hierarchical manner. In so doing, models are
typically calibrated to available information from the bottom-up (lower-scale, high-
fidelity models or experiments) and top-down (e.g., experiments) relative to that
particular scale. These models are almost never of concurrent multiscale type,
10 D.L. McDowell

Fig. 1.5 Modeling across scales of material structure hierarchy (right) is not equivalent to
materials design but rather provides support for decisions in the latter

for several reasons. First, well-known classical models of important mechanisms

are framed at a given level of the hierarchy in Fig. 1.1 and directly address
appropriate, scale-specific measures or attributes of material structure – such
models are inherently microstructure-sensitive. For example, dislocation-obstacle
interactions are typically framed using line tension models for dislocations [29].
Theories of work hardening are framed at the mesoscale due to the many-body
nature of dislocation interactions. Models for ductile-brittle transition [78] and
effects of impurity segregation at grain boundaries are framed in terms of unit
processes occurring at or near the atomic scale [79]. It is understood that each
of these behaviors has an associated scale and structure attributes for which these
correlations or models are most appropriately used. In contrast, it is challenging to
build concurrent multiscale models that embed details of structure specific to each
scale of hierarchy, as they are commonly framed using some kind of multiscale basis
function approach that appeals to self-similarity of structure among multiple scales
to facilitate systematic coarse-graining (cf. [13, 44, 93]) or transferring information
on a common multiresolution domain via evolving boundary conditions with time in
a concurrent FEM computational scheme (cf. [17]). Second, the uncertainty of the
process of linking models in two-way fashion as required by concurrent multiscale
models is extremely difficult to quantify, as it involves much more than simply the
information entropy loss associated with coarse-graining – formal mathematical
approaches for doing this are largely undeveloped. There are both spatial and
temporal contributions to uncertainty, which is also affected by the nonequilibrium
character of material structure evolution. Figure 1.6 summarizes common sources
of uncertainty that arise in models at each scale of structure hierarchy and in
scale linking or transition algorithms. The uncertainty in the coupling of models
across length and time scales can compound other sources of uncertainty related
to material models or material structure at each scale.Quantifying uncertainty in
schemes for linking models at different length and time scales is an immature
field and is affected by the identification of specific scales of hierarchy that may
control properties of interest, approximations made in separating length and time
scales in models used, model form and parameter uncertainty at various scales,
approximations made in various scale transition methods, and lack of complete
characterization of initial conditions and process history effects. Third, to provide
decision support for materials design and development, we are most often interested
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 11

Fig. 1.6 Sources of uncertainty in models at each scale of structure hierarchy (left) and in scale
linking or scale transition algorithms (right)

in representing scale-specific mechanisms and trends of structure-property relations,

rather than upscaling these relations to higher length and time scales of applications.
When uncertainty is considered, input regarding mesoscopic phenomena into
materials design is most effectively provided by models framed at length scales
further to the right in Fig. 1.1 and increasingly makes use of top-down information
to inform model parameters. For example, if variability of system responses of
primary relevance to required properties is chiefly influenced by grain size, shape,
and orientation distributions, then these attributes can serve as design variables,
and application of conventional continuum crystal plasticity with homogenized
description of the behavior within each grain may suffice. As discussed later,
this is the case for fatigue of polycrystalline alloys. Depending on properties,
the structure scale that controls may vary widely. For corrosion or environmental
fracture resistance, the atomic structure of interfaces may play a dominant role
compared to the morphology of grain structure, even if they are coupled [62].
It follows that one of the highest utility applications of multiscale modeling in
providing decision support for materials design is the prediction of sensitivity of
responses or properties to variation of microstructure at each level of structure
hierarchy. This is important for several reasons:
• It is experimentally quite challenging and in many cases impossible to isolate the
sensitivity of responses at specific scales within the material structure hierarchy.
• Sensitivity analyses are keys in any hierarchical system to quantify dominant
design variables among levels of structure hierarchy.
• Sensitivity of process-structure and structure-property relations is central to
concepts of robust design, where the goal is to explore a range of candidate
12 D.L. McDowell

solutions from which the designer can select, incorporating the notion of reduced
sensitivity to process path, material composition, target microstructures, and even
range of use in applications [58].
Most often, the goal of simulated-assisted materials design is not to accurately
predict mean properties at higher scales but rather to understand their sensitivity
to material structure or microstructure and to capture dominant mechanisms and
their transitions that affect material responses or properties. Given the uncertainty
of model form, model parameters, and schemes for linking models pertaining to
various scales in Fig. 1.1, the notion of single-point design optimization using
hierarchical or concurrent multiscale models is not particularly useful in many cases.
Uncertainties associated with process control, initial conditions, and microstructure
randomness are potentially significant as well; by neglecting them, we are ironically
making a decision regarding how to incorporate them – there is no other choice.
For these reasons, extension of systems-based robust design concepts introduced
by Taguchi [92] for process control to multilevel integrated design of materials and
products [52, 53, 58, 67, 73] is logical, focusing on sensitivity of key properties or
responses to variation of structure, which in turn links to variation of composition
and process route [9].
We close this section by noting that the interface between materials design and
development and systems design of products has historically been defined by a focus
on tabulated and certified material properties (i.e., the materials selection “handoff”
designated by the vertical line in Fig. 1.2). When design of the material structure
hierarchy is coupled with that of product applications, more information is available
that compels a shift toward including material structure attributes at various levels,
in addition to properties. In fact, since properties are conferred by structure, the
structure information is more primal in nature and is often in digital format. Design
targets for material structure also serve as a tangible focus on quality control and
inspection in manufacturing and in some cases can be monitored nondestructively or
even online. Moreover, design of the material and product is pursued concurrently,
which, for example, has always been the case with fiber-reinforced composite
materials, and is necessary in certain emergent manufacturing processes such as
additive manufacturing, and it is expected that the material structure will vary
through the part; the resulting heterogeneity renders the notion of “properties”
antiquated and not particularly useful as a means of communication between process
and performance requirements. This drive toward digital representation of material
structure is at the core of both the MGI and ICME initiatives [23, 24, 59, 71, 73].
The following section provides more insight into the demands and capabilities
of multiscale modeling to provide decision support for design and development of
metal alloy systems.
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 13

1.4 Multiscale Modeling in Materials Design

It is evident in Fig. 1.1 that dislocations are of first-order importance in influencing

mechanical properties in metallic systems. Furthermore, point defects (vacancies
and interstitials) are relevant to dislocation mobility, network irreversibility, and
associated damage phenomena. Design of microstructure to achieve enhanced
performance requires microstructure-sensitive model forms that distinguish between
phenomena of nucleation, generation, migration, absorption/desorption, trapping,
and bypass or annihilation of dislocations at various scales of material structure
hierarchy. Most of these phenomena are coupled in across length and/or time scales.
If only a single model is used at a higher scale and calibrated from the top-down,
even if the notion of a characteristic length is introduced to reflect size effects,
it may not have sufficient capability to reflect material structure attributes that
influence properties of interest. There are several compelling reasons to develop
microstructure-sensitive multiscale models of plasticity [50, 52–55], some of which
coincide with the aims of simulation-assisted materials design:
• To support design and development of polycrystalline and/or polyphase
microstructures with tailored properties that relate to plastic deformation, for
example, yield strength, ductility, fatigue, and ductile fracture toughness
• To facilitate accelerated insertion of materials into the product development cycle
by coupling process route with properties in a more predictive way by leveraging
modeling and simulation
• To conduct microstructure-sensitive failure and life prediction
• To quantify influence of environment or complicating contributions of impurities,
manufacturing induced variability, or defects
• To admit competition of distinct mechanical, chemical, and transport phenomena
in multiphysics applications without relying too heavily on intuition to guide
solutions in the case of highly nonlinear interactions
• To build self-consistent estimates for the nucleation and evolution of defects
during plastic flow at various length scales, bridging domains of quantum physics
and chemistry with engineering
A particular challenge to multiscale modeling is to provide support by predicting
the evolution of material structure at levels of hierarchy that are intermediate
to atomistic and polycrystal plasticity levels, as shown in Fig. 1.7. This critical
“mesoscale gap” represents one of the more complex and unsettled horizons in metal
plasticity and arises chiefly due to:
• Difficulties in capturing many-body dislocation interactions of long-range char-
acter
• Identifying microstructure representation at pertinent scales of structure hier-
archy necessary to capture dominant system sensitivities, including effects of
spatial heterogeneity
14 D.L. McDowell

Fig. 1.7 Mesoscale range of length scales in metallic systems involving collective effects of
dislocations, dislocation structures, and interfaces

• The nonequilibrium character of structure evolution associated with dislocations,

including temporal aspects of a spectrum of characteristic transit and relaxation
times
Addressing model form and parameter uncertainty in this mesoscale domain
highlighted in Fig. 1.7 is critical yet has not been extensively addressed in the
literature. While widely recognized as an inherently stochastic regime, surprisingly
few models beyond those addressing discrete dislocation dynamics incorporate
various sources of variability and randomness, including initial conditions. By
their very nature, reduced-order descriptions compromise a precise description
of subscale phenomena in favor of a coarse-grained representation of structure.
Moreover, it seems that a plethora of model forms are employed for ostensibly
the same physical phenomena, yet model form uncertainty is rarely mentioned in
the literature as a scientific topic. Hierarchical multiscale model constructs that
do consider uncertainty typically make use of some sort of statistical or inference
methodology to inform upscale models from fine to coarse scale, corresponding
to different expressions of collective dominant mechanisms that affect responses.
The works of Zabaras and colleagues [6, 15, 37, 80] offer excellent examples of
uncertainty quantification in hierarchical multiscale models, involving both model
parameters and model form.
Recall the distinction between hierarchical and concurrent multiscale models
outlined in a previous section. Hierarchical multiscale models employ a set of
models, each suited to convey degrees of freedom associated with material structure
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 15

and key phenomena at a given scale of hierarchy. As already mentioned, such

models can be referred to as microstructure-sensitive since they couple the structure
attributes at each level of hierarchy (dislocations, dislocation structures, grains,
phases, interfaces, etc.), smearing finer-scale structure by incorporation of its effects
into model form and parameters. These models can be informed using a self-
consistent micromechanics scheme (of Eshelby-Kröner type based on eigenstrain
field mechanics, cf. [16, 40, 64, 75, 89]) for idealized microstructures that employ
Green’s functions or approximate influence functions for long-range interactions or
can be based on direct numerical simulation (DNS) [54, 55, 59, 69] of explicitly
rendered microstructures to incorporate fine-scale phenomena into coarse-grained
constitutive descriptions. The multiresolution continuum theory [44, 45] makes
use of both homogeneous and inhomogeneous contributions to the virtual power
variational principle that underlies the scale transition, allowing for successive
coarse-graining to add additional degrees of freedom to the constitutive description
with upscaling, analogous to the concept of a generalized continuum. Scaling
relations for structure-property relations are another class of hierarchical models
that describe aggregate dislocation behavior and strengthening effects at mesoscopic
length scales and a broad range of time scales. For example, self-organization of
dislocations into periodic low-energy substructures at the mesoscale, leading to
grain subdivision [27, 28, 42], is consistent with the theory of evolution of stress-
screened dislocation structures [38]. Associated scaling laws [28] can be embedded
in continuum crystal plasticity [4, 25] or employed in other formats to provide
decision support for materials design.
Hierarchical multiscale model transitions are almost always framed as relating
information between only two successive levels of hierarchy in Fig. 1.1 to arrive at
effective properties or responses, accounting for microstructure heterogeneity and
dislocation-dislocation and dislocation-structure interactions. For scale transitions
above the realm of discrete dynamical models, regardless of the multiscale modeling
strategy, the concept of a statistically “representative volume element” (RVE)
is frequently invoked in seeking computational estimates of structure-property
relations. The concept of a RVE was introduced by Hill [22] as a volume sufficiently
large to encompass such that the predicted responses or properties do not change
with further increase of size; put another way, it considers all statistical moments of
material structure in structure-property correlations of interest (so-called statistical
homogeneity). The RVE concept is assumed a priori in self-consistent microme-
chanics homogenization approaches, but not in DNS. A caveat is that each response
of interest may have a different RVE size. Certain properties such as ductility or
high-cycle fatigue resistance depend primarily on higher-order spatial statistics of
microstructure attributes, such as largest grains or particles, most severe interactions
of particles or phases, etc., and potentially have much larger RVE sizes than
properties or responses that depend mainly on lower-order moments (e.g., volume
fraction of phases) of microstructure spatial arrangement, such as elastic stiffness
or thermal conductivity [48, 54]. Such properties typically require some form of
DNS to simulate and estimate. Simulations that assume the notion of RVE are
large and computationally demanding using brute-force computing strategies such
16 D.L. McDowell

as a 3D crystal plasticity finite element method (CPFEM) [16, 54]. Moreover, they
do not always address the issue of whether the assumed RVE size is sufficient to
capture each response of interest or how much variability in response might emerge
as a function of the volume element size. To obviate this issue, recent advances
in fast Fourier transform (FFT) CPFEM [5, 41, 74] have dramatically improved
computational efficiency of RVE-level simulations but with certain approximations
regarding interfaces and numerical integration that need to be considered in an
uncertainty analysis. Some authors use the term “statistical volume element” (SVE)
to designate a random sample of microstructure that is too small to satisfy RVE
requirements of statistical homogeneity for a given response function but large
enough to capture the key higher-order (e.g., nearest neighbor, second nearest
neighbor) interactions of microstructure that influence minimum properties or
responses of interest [48, 49, 68]. An ensemble of SVE simulations must be
conducted to build up the collective statistics required to capture suitably high-order
moments of a desired response distribution. Computing with a sufficient number
of SVEs, which are smaller in size and therefore much more computationally
efficient than an associated RVE, facilitates quantification of statistical variability
associated with material size effects with regard to each property or response under
consideration. The nature of applied boundary conditions influences convergence
toward RVE response [68] as a function of number of SVE simulations.
Hierarchical multiscale models often involve statistical description of evolving
microstructure at a given scale and employ some form of “handshaking” methods
for passing information to models framed at the next higher scale of hierarchy. There
are exceptions in which information can be passed far upscale to much different
classes of models; as mentioned earlier, atomistic simulations or DFT can even
provide elastic constants, thermal expansion, and diffusion constants for bulk phases
to support continuum polycrystal plasticity simulations, provided that both the com-
position and lattice structures are adequately treated. For the most part, handshaking
methods can range from intuitive formulation of the form of constitutive models to
estimates of model parameters in coarse-grained models based on high-resolution
simulations. They may also directly inform meta-models or surrogate models for
structure-property relations. Examples of a combined bottom-up homogenization
and handshaking among scales intended to support decision-based design are found
in works of McDowell and colleagues [51, 54, 55, 85] regarding microstructure-
sensitive multiscale models; dislocation density evolution equations are formulated
at the scale of either precipitates or homogenized grains to model cyclic viscoplastic
behavior of Ni-base superalloys and ’“ Ti alloys, and these relations are then
calibrated with experimental elastic stiffness and stress-strain data on single crystals
and polycrystals. A key feature of such models is the incorporation of elements of
microstructure attributes (precipitate or phase volume fractions, sizes, orientation
distribution, etc.) that are sensitive to process route and affect relevant mechanical
properties. Another example of an hierarchical modeling framework is that of a
“multiscale fracture simulator” introduced by Hao et al. [20, 21] to support design of
ultrahigh strength, high-toughness steels; this approach employs information from
atomistics in a scaled interface separation relation for interface fracture between
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 17

metallic matrix and nonmetallic inclusions, and additional models are introduced
at higher scales for coupled dislocation glide-driven void growth, coalescence, and
failure by shear localization and macrofracture. Groh et al. [19] have introduced
a hierarchical multiscale model for plasticity of Al single crystals ranging from
atomic scale through dislocation substructures. Narayanan et al. [65] employed
unit process analysis via molecular dynamics and the nudged elastic band method
via molecular statics to analyze the reaction pathway for coordinated kink pair
formation in shearing of screw dislocation segments in pure Fe, relating this directly
to parameters of and thermally activated flow rule used at the mesoscale (cf. [36]).
Concurrent multiscale modeling schemes involve two-way (bottom-up and top-
down) coupling between models framed at different scales and are especially
useful if fine-scale microstructure evolution (such as cracking or slip localization)
limits failure conditions but cannot be determined a priori at a coarse scale such
as the scale of a structure of interest. In other words, they are more useful for
purposes of multiscale simulation of failure processes, including scales of parts and
subassemblies in Fig. 1.2. As an example, Ghosh et al. [17] introduced a two-way
concurrent scheme to model multiscale damage initiation and growth of fine-scale
damage associated with debonding at fiber-matrix interfaces in composites while
still considering the overall scale of a structure with notches and change of geometry.
In that work, a macroscopic continuum damage model was employed with high-
fidelity computation of stress and strain at local hot spots at the microstructure
(fiber-matrix) level. An intermediate level of modeling was performed at the scale
of a homogenized RVE to monitor the breakdown of coarse-grained continuum
laws. In this way, provisions were made to “zoom in” and “zoom out” in terms
of resolution, enabling a broad area search for critical hot spots in components
along with microstructure-sensitive simulations of failure processes at these hot
spots. Different classes of models were used at each scale, which addresses an
important but challenging requirement for physical consistency at each level of
concurrent multiscale models. Uncertainty and its propagation were not quantified
in this approach.
Prospects for predictive, microstructure-sensitive, concurrent multiscale models
are perhaps most clearly realized in the context of atomistic and coarse-grained,
multiresolution atomistic formulations that move somewhat into the mesoscale
regime (e.g., the quasicontinuum method developed by [86, 87, 90, 91]), since they
are based on an underlying interatomic potential. In contrast, continuum formu-
lations at the mesoscale and above remain underdeveloped and quite challenging
to formulate with quantified uncertainty, particularly for purposes of providing
decision support for materials design and development. Both concurrent and
hierarchical multiscale modelings typically employ models that are increasingly
coarse-grained with increase of the scale of material structure hierarchy, with
reduction of degrees of freedom at longer length and time scales. From a practical
perspective, however, the formulation of concurrent multiscale modeling requires
intimate, direct mathematical and computational coupling between models framed
at successive scales. To facilitate scale bridging, the same mathematical model
structure is often assumed at each scale, albeit successively coarse-grained. So it
18 D.L. McDowell

is difficult to address the transition from discrete to continuous models or from

dynamic to thermodynamic descriptions in the context of two-way, concurrent
coupling. Models that attempt to transition between atomistic and continuum
descriptions are therefore limited in the degree of upscaling that can be achieved.
Moreover, their ability to convey information regarding details of material structure
hierarchy is typically limited.
This discussion has been primarily focused on mechanical properties and
responses. A multifunctional materials design must address multiple property or
response requirements. Often these properties conflict in terms of their demands
on material structure; a good example is the classical trade-off of strength and
ductility. In some cases, multifunctional materials design must address property
requirements in different physical domains, for example, conductivity, oxidation
resistance, tensile strength, elastic stiffness, and creep and fatigue resistance in
high-temperature gas turbine engine disk or blade materials (refer to Fig. 1.2).
Multiple property goals cannot be met by optimizing individual models at different
levels of hierarchy in Fig. 1.2, for example, but only by considering the entire
hierarchy of scales. A multilevel systems approach is essential. In the case of
coupled multiphysics material responses (e.g., chemo-mechanical, electromechan-
ical, magnetomechanical) of energy materials or smart materials, for example,
concurrency in time is necessitated by the coupling of responses related to each
physical process unless time scale separation can be established. Typically, such
models have been pursued mainly at distinct levels of spatial structure hierarchy
and associated idealization, with few attempts to bridge both spatial and temporal
scales.
As illustrated in Figs. 1.4 and 1.5, Olson’s hierarchy in Fig. 1.3 [66] should not
be confused with a multiscale modeling strategy. Multiscale modeling, whether
hierarchical or concurrent, is not equivalent to materials design. The goal of
linking process-structure-property relations is not focused on multiscale modeling
per se. Materials design is effectively a multilevel, multiobjective optimization
problem in which ranged sets of solutions are sought that satisfy ranged sets of
performance requirements [7, 52, 58, 70, 72, 81, 82]. It does not rely on the
premise of explicit linkage of responses at multiple length scales. In fact, it is
often preferable to introduce rather more elementary, validated model concepts
at each scale than accept the uncertainty of complex, coupled multiscale models
for which parameter identification and validation are difficult [7]. McDowell et al.
[58] have summarized this body of work and have written extensively on systems
engineering principles in materials design and how hierarchical multiscale modeling
can provide useful information regarding process-structure and structure-property
relations. Ultimately, design is an inductive, top-down exercise and requires that one
pursues top-down, inverse strategies to explore property-structure-process relations.
Lack of ability to invert process-structure and structure-property relations is the rule,
both for computational modeling and simulation and experimental correlations, due
in large part to a number of issues:
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 19

• Nonlinear, nonequilibrium path-dependent behavior, limiting parametric study

and imparting dependence upon initial conditions
• Dynamic to thermodynamic model transitions in multiscale modeling, with
nonuniqueness arising from reduction of model degrees of freedom
• Approximations made in digital microstructure representation and reconstruction
of material structure
• Dependence of certain properties such as ductility, fracture toughness, fatigue
strength, etc. on extreme value distributions of microstructure, rather than on
simple means
• Microstructure metastability and long-term evolution
• Uncertainty and nonuniqueness of representation of material structure, model
forms, and model parameters
• Lack of experimental data
• Variability and uncertainty of experimental data
To facilitate top-down exploration for robust design solutions, an iterative
approach is essential for bottom-up information flow (simulations, experiments),
guided from the top-down by performance requirements for applications. One
such approach to pursue inverse design that employs multiscale modeling and
experiments in early-state design exploration is the so-called inductive design
exploration method (IDEM), introduced by Choi [7] and Choi et al. [8] and applied
to a several exemplary case studies (cf. [58]) and more recently to the design of
ultrahigh-performance concrete [9]. Simulations are typically conducted in bottom-
up manner and cannot strictly be inverted from right to left in Fig. 1.2. IDEM has
two major objectives: (i) to guide bottom-up modeling and simulation to explore
and conduct top-down, requirement-driven design and (ii) to manage uncertainty
propagation in chains of process-structure-property relations. The full hierarchy of
material structure is admitted, constituting a multilevel design approach. Effectively,
bottom-up computation or experiments can be conducted in parallel using design of
experiment strategies over much of the material structure design space, establishing
projections among spaces of process-structure-properties for each level of structure
hierarchy, followed by formulation of meta-models or data science correlations to
invert these relations subject to certain feasibility and uncertainty constraints. We
can carry the analogy of parts, subassemblies, and assemblies of product design
to the material structure hierarchy. Identification of subsystems in the material
hierarchy with weak coupling to responses of subsystems at other levels of hierarchy
is an important step [72], as these subsystems can often be analyzed indepen-
dently with regard to their process-structure and/or structure-property relations.
Multiscale modeling provides the decision support to evaluate the strength of these
couplings (see Fig. 1.4), along with modern data science methods for materials and
microstructure informatics as appropriate [34, 60, 61, 76, 77, 88]. Seepersad [83]
has advanced the treatment of uncertainty propagation in multilevel materials design
by considering Bayesian network classifiers in design of materials with hierarchy
[47, 84]. Mahadevan and coworkers [43, 63] have addressed multilevel uncertainty
integration, relevance, calibration, and validation in hierarchical materials design
20 D.L. McDowell

and analysis problems. The field is rapidly advancing, and there is a pressing
need for integration of systems approaches to uncertainty analysis with common
materials design and development workflows, both experimental and computational.
The materials community must facilitate this integration, both in industry practice
and academic educational programs.
There are other approaches to inverse design as well, often specific to certain
classes of materials design problems. Olson’s strategy [53] of limited iteration
is a practical and fairly general extension of traditional design approaches but is
somewhat less amenable to parallelization and design exploration – it presumes
considerable insight and knowledge of dominant mechanisms, levels of material
structure hierarchy, and candidate solutions entering into the design process and
therefore is more suited toward detailed design than design exploration. A range of
strategies too numerous to list have emerged to pursue materials design (cf. [15, 35,
80, 94]), some of which target microstructure-sensitive design problems for which
direct analytical or computational inverse problems can be effectively pursued,
such as texture control of elastic properties [1, 14, 30, 31, 46]. Advances in rapid
inverse design have been afforded by the Materials Knowledge Systems approach of
Kalidindi and coworkers [10, 11, 32, 33, 39], which effectively combines Eshelby-
Kröner micromechanics-type relations based on eigenstrain fields with DNS to
calibrate approximate representations of Green’s functions to facilitate efficient
inverse estimation of local responses (e.g., grain level plastic strain and stress
distributions) for random microstructures at the RVE level [10, 34, 59, 60].

1.5 Summary and Conclusions

This paper has attempted to clarify goals and methods of materials design, as well
as the nature of decision support rendered by multiscale models executed at various
levels of material structure hierarchy, with a focus on the modeling of dislocations in
metallic systems. We have clearly distinguished multiscale modeling from materials
design, with the former serving the purposes of the latter. Multiscale modeling in
support of materials design is a relatively new endeavor, which holds promise to
intensify in degree with advances in high-performance computing and data science.
The issue of multiple distinct levels of material structure hierarchy has been outlined
as a key issue in defining and applying multiscale modeling approaches.
With the confluence of continued development of computational modeling and
simulation of material structure-response relations at various levels of material
structure hierarchy, application of data science and associated big data tools,
high-performance computing into the exascale regime and beyond, in situ and
ex situ multiresolution experimental measurements, and both computational and
experimental high-throughput screening methods, the worldwide community of
materials, data, computation, and systems engineering researchers is witnessing an
explosion of capabilities, developing at such a rapid pace that single investigators
or laboratories find difficult to integrate. Pursuit of the wide range of method
1 Microstructure-Sensitive Computational Structure-Property Relations. . . 21

and approaches discussed in this chapter extends well beyond a single researcher,
research group, or even entire university, industry, or government research laborato-
ries. To take advantage of these advances and to incorporate multiscale modeling
into the stream of decision support for materials design will require integrated
infrastructure – of regional, national, and international character – to integrate the
collaborative computational, experimental, and data science aspects. To this end,
aspirations of the MGI and ICME will carry forward into the future by virtue of
close integration of formerly disparate disciplines. More specifically, the materials
innovation infrastructure described by McDowell and Kalidindi [60] might serve
as a model architecture for these distributed interactions, involving coupling of
experimental, computational, and data sciences infrastructure via distributed e-
collaboration. The interested reader may wish to consult the recent ICME-directed
monograph by Horstemeyer [26] to provide an additional and, in some respects,
alternative perspective regarding the role of multiscale modeling in materials design.

Acknowledgments The author is grateful for the support of the Carter N. Paden, Jr. Distinguished
Chair in Metals Processing at Georgia Tech, as well as prior support of AFOSR, ONR D3D, Eglin
AFB, DARPA, NAVAIR, QuesTek, the NSF-funded PSU-GT Center for Computational Materials
Design, SIMULIA, NSF CMMI-1232878, NSF CMMI-0758265, and NSF CMMI-1030103.

References

1. Adams, B.L., Lyon, M., Henrie, B.: Microstructures by design: linear problems in elastic-
plastic design. Int. J. Plast. 20(8–9), 1577–1602 (2004)
2. Ashby, M.F.: Materials Selection in Mechanical Design, 2nd edn. Butterworth-Heinemann,
Oxford (1999)
3. Billinge, S.J.E., Rajan, K., Sinnott, S.B.: From cyberinfrastructure to cyberdiscovery in
materials science: enhancing outcomes in materials research, education and outreach. Report
from NSF-sponsored workshop held in Arlington, Virginia, 3–5 Aug (2006)
4. Butler, G.C., McDowell, D.L.: Polycrystal constraint and grain subdivision. Int. J. Plast. 14(8),
703–717 (1998)
5. Chen, L., Chen, J., Lebensohn, R., Chen, L.-Q.: An integrated fast Fourier transform-based
phase-field and crystal plasticity approach to model recrystallization of three dimensional
polycrystals. Comp. Meth. Appl. Mech. Eng. 285, 829–848 (2014)
6. Chen, P., Zabaras, N.: Uncertainty quantification for multiscale disk forging of polycrystal
materials using probabilistic graphical model techniques. Comput. Mater. Sci. 84, 278–292
(2014)
7. Choi, H.-J.: A robust design method for model and propagated uncertainty. PhD Dissertation,
G.W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta
(2005)
8. Choi, H.-J., McDowell, D.L., Allen, J.K., Rosen, D., Mistree, F.: An inductive design
exploration method for the integrated design of multi-scale materials and products. J. Mech.
Des. 130(3), 031402 (2008)
9. Ellis, B.D.: Multiscale modeling and design of ultra-high-performance concrete. PhD Disser-
tation, Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta
(2013)
10. Fast, T., Kalidindi, S.R.: Formulation and calibration of higher-order elastic localization
relationships using the MKS approach. Acta Mater. 59, 4595–4605 (2011)
22 D.L. McDowell

11. Fast, T., Niezgoda, S.R., Kalidindi, S.R.: A new framework for computationally efficient
structure-structure evolution linkages to facilitate high-fidelity scale bridging in multi-scale
materials models. Acta Mater. 59(2), 699–707 (2011)
12. Featherston, C., O’Sullivan, E.: A review of international public sector strategies
and roadmaps: a case study in advanced materials. Centre for Science Technology
and Innovation, Institute for Manufacturing. University of Cambridge, UK.
http://www.ifm.eng.cam.ac.uk/uploads/Resources/Featherston__OSullivan_2014_-
_A_review_of_international_public_sector_roadmaps-_advanced_materials_full_report.pdf.
(2014). Accessed 24 Sept 2017
13. Fish, J.: Multiscale Methods: Bridging the Scales in Science and Engineering. Oxford
University Press, 1st edn. ISBN 978–0–19-923385-4 (2009)
14. Fullwood, D.T., Niezgoda, S.R., Adams, B.L., Kalidindi, S.R.: Microstructure sensitive design
for performance optimization. Prog. Mater. Sci. 55(6), 477–562 (2010)
15. Ganapathysubramanian, S., Zabaras, N.: Design across length scales: a reduced-order model of
polycrystal plasticity for the control of microstructure-sensitive material properties. Comput.
Methods Appl. Mech. Eng. 193(45–47), 5017–5034 (2004)
16. Geers, M.G.D., Kouznetsova, V.G., Brekelmans, W.A.M.: Multi-scale computational homoge-
nization: trends and challenges. J. Comput. Appl. Math. 234(7), 2175–2182 (2010)
17. Ghosh, S., Bai, J., Raghavan, P.: Concurrent multi-level model for damage evolution in
microstructurally debonding composites. Mech. Mater. 39(3), 241–266 (2007)
18. Granta Design, Granta CES Selector. https://www.grantadesign.com/products/ces/ 2016.
(Accessed 21 June 2016)
19. Groh, S., Marin, E.B., Horstemeyer, M.F., Zbib, H.M.: Multiscale modeling of the plasticity in
an aluminum single crystal. Int. J. Plast. 25(8), 1456–1473 (2009)
20. Hao, S., Moran, B., Liu, W.K., Olson, G.B.: A hierarchical multi-physics model for design of
high toughness steels. J. Computer-Aided Mater. Des. 10, 99–142 (2003)
21. Hao, S., Liu, W.K., Moran, B., Vernerey, F., Olson, G.B.: Multi-scale constitutive model and
computational framework for the design of ultra-high strength, high toughness steels. Comput.
Methods Appl. Mech. Eng. 193, 1865–1908 (2004)
22. Hill, R.: Elastic properties of reinforced solids: some theoretical principles. J. Mech. Phys.
Solids. 11, 357–372 (1963)
23. Holdren, J.P.: National Science and Technology Council, Materials Genome Initiative
for Global Competitiveness. http://www.whitehouse.gov/sites/default/files/microsites/ostp/
materials_genome_initiative-final.pdf, (2011). (Accessed 21 June 2016)
24. Holdren, J.P.: National Science and Technology Council, Committee on Technol-
ogy, Subcommittee on the Materials Genome Initiative, Materials Genome Initia-
tive Strategic Plan. https://www.whitehouse.gov/sites/default/files/microsites/ostp/NSTC/
mgi_strategic_plan_-_dec_2014.pdf (2014). (Accessed 21 June 2016)
25. Horstemeyer, M.F., McDowell, D.L.: Modeling effects of dislocation substructure in polycrys-
tal elastoviscoplasticity. Mech. Mater. 27, 145–163 (1998)
26. Horstemeyer, M.F.: Integrated Computational Materials Engineering (ICME) for Metals: Using
Multiscale Modeling to Invigorate Engineering Design with Science, 1st edn. Wiley, Hoboken
(2012)
27. Hughes, D.A., Hansen, N.: High angle boundaries and orientation distributions at large strains.
Scripta Metall. Mater. 33(2), 315–321 (1995)
28. Hughes, D.A., Liu, Q., Chrzan, D.C., Hansen, N.: Scaling of microstructural parameters:
misorientations of deformation induced boundaries. Acta Mater. 45(1), 105–112 (1997)
29. Hull, D., Bacon, D.J.: Introduction to Dislocations, 5th edn. ButterworthHeinemann, Oxford
(2011)
30. Kalidindi, S.R., Houskamp, J.R., Lyon, M., Adams, B.L.: Microstructure sensitive design of
an orthotropic plate subjected to tensile load. Int. J. Plast. 20(8–9), 1561–1575 (2004)
31. Kalidindi, S.R., Houskamp, J., Proust, G., Duvvuru, H.: Microstructure sensitive design with
first order homogenization theories and finite element codes. Mater. Sci. Forum, v 495–497, n
PART 1, Textures of Materials, ICOTOM 14 – Proc. 14th Int. Conf. on Textures of Materials,
23–30 (2005)
Another random document with
no related content on Scribd:
Hij ging de knip van de deur doen en de jonge man, die hem op
zoovele zijner gevaarlijke avonturen had vergezeld, trad het vertrek
binnen.

—Heb je haar gesproken? vroeg hij.

—Ja. Alles is in orde. Als zij zich niet buiten vertoont, bestaat er geen
vrees dat de politie haar zal vinden, noch de handlangers van Dr.
Fox, den chef van het Genootschap van den Gouden Sleutel, die het
op haar leven voorzien heeft, zoowel als op dat van haren minnaar,
Raoul Beaupré.

—Je hebt haar natuurlijk alles medegedeeld omtrent [7]den

gevaarlijken toestand, waarin zich die Fransche bandiet bevindt?

—Ja—daarom ging ik immers daarheen! Zij schijnt dien man zeer lief
te hebben, want zij siddert voor zijn leven, en zelfs voor zijn vrijheid.
Ik acht het dan ook mijn plicht dien man te redden!

Charly keek Raffles aan, alsof hij hem niet goed verstaan had.

—Dat kan je toch geen ernst zijn? vroeg hij.

—Volle ernst. Beaupré kan mij later diensten bewijzen, als ik den
strijd met Fox weder aanbind, en al was dat niet zoo—belofte maakt
schuld! Zijn minnares is door mijn toedoen in den toestand geraakt
waarin zij zich nu bevindt—dat wil zeggen, van een stuk wild dat door
de politie, zoowel als door de mannen van Fox wordt opgejaagd. Zij
heeft de plannen van de bende aan mij verraden, en is dus volgens
de bepalingen van het Genootschap van den Gouden Sleutel des
doods schuldig!

—Maar—als zij jou niet ontmoet had, dan zou zij eenvoudig een
ander in den arm hebben genomen, om Fox onschadelijk te maken—
de politie bijvoorbeeld! riep Charly uit.
—Dat is best mogelijk! hernam Raffles kalm. Maar dat verandert voor
mij de zaak niet. Ik ben nu eenmaal geen politie, maar een inbreker—
een soort collega dus, al verschillen onze werkmethoden en onze
doeleinden wellicht een weinig! Hoe het ook zij—Beaupré moet gered
worden, omdat ik zijn minnares in een gevaarlijken toestand heb
gebracht en daardoor ook alle daarop volgende gebeurtenissen in het
Ziekenhuis heb teweeggebracht!

Charly slaakte een zucht, maar hij sprak niet verder tegen, daar hij
wel wist, dat hem dit bitter weinig zou baten.

Raffles had eenmaal een plan opgevat, een plan, dat zijn gevoel voor
recht en billijkheid hem ingaf, en vooral zijn begrip van wat de eer
eischt—en hij zou het ten uitvoer brengen, ook al zou het hem het
leven kosten.

En zoo duurde het niet lang, of de twee mannen zaten tegenover

elkander in de kleine rookkamer, welk vertrek bij voorkeur werd
uitverkoren voor dergelijke krijgsraden en overleggen.

Zij begrepen maar al te goed, dat hun voornemen op de grootste

moeilijkheden zou stuiten, want de bewaking van den zoogenaamden
„Dubois” die zich had ontpopt als een lang gezocht Parijsch bendelid,
zou zeker streng zijn, en nog veel strenger worden, zoodra de
toestand van den gewonde toeliet, dat er vrees voor een ontsnapping
zou bestaan.

Er moest dus zoo snel mogelijk gehandeld worden en met list, daar
hier met geweld niets te bereiken viel.

Een groot deel van den morgen was reeds verstreken en nog hadden
de beide vrienden niets gevonden.

Om half twaalf stond Raffles ongeduldig op en zeide:

—Neem je hoed en je stok en ga mede! Op straat zullen wij
misschien een weinig beter geïnspireerd worden—vooral den kant op
van het ziekenhuis! Misschien geeft het zien van dat gebouw wel een
goed plan aan de hand!

En zoo stonden de twee mannen een oogenblik later op straat, thans

als William Aberdeen en zijn secretaris, zooals vele Londenaren hen
beiden goed kenden.

Een huurauto bracht hen tot op een honderdtal meters van het
ziekenhuis, hetwelk voor hen het tooneel van zoovele schokkende
voorvallen was geweest.

Zij liepen er eens om heen, in den gewonen wandelpas, om geen

argwaan te wekken, maar aan den achterkant van het groote gebouw
gekomen, dat aan alle kanten met een grooten tuin omringd was,
waar hier en daar paviljoens voor besmettelijke ziekten stonden, hield
Raffles eensklaps zijn schreden in en keek naar boven.

—Wat is er? vroeg Charly nieuwsgierig, daar hij een schittering in de

schrandere grijze oogen van zijn vriend had gezien, die heel wat
verried.

—Kijk eens naar het dak! beval Raffles.

Charly gehoorzaamde en zag eerst een klein rookwolkje ten hemel

stijgen, en daarop de gedaanten van een tweetal mannen, die
blijkbaar bezig waren met iets aan het zinken dak van het ziekenhuis
te repareeren.

Charly keek Raffles glimlachend aan en zeide veelbeteekenend:

—Dat zou misschien wel iets voor ons zijn, dunkt je niet?
—Het hangt er van af, hoever die menschen met hun werk staan! Ik
hoop maar, dat zij nog vandaag gereed komen! [8]

—Waarom?

—Omdat wij dan morgen daar naar boven kunnen gaan en zeggen,
dat er nog iets verricht moet worden, wat heden vergeten werd!

—Maar als wij ons eens gingen aanbieden als knechts!

—Ik vrees, dat dit in het onderhavige geval niet veel zou uitwerken,
mijn jongen! Die mannen zijn maar met hun beiden, en de reparatie is
dus blijkbaar niet zeer omvangrijk. Zij schijnen het gemakkelijk met
zijn beiden af te kunnen en zullen er dus niet twee mannen bij
nemen!

—Maar één toch misschien wel!

—Het zou te probeeren zijn, maar eerst willen wij eens afwachten of
het werk reeds is afgeloopen!

De twee mannen gingen een weinig verder en traden een fijne bar
binnen, waar, wegens het vroege uur, plaats in overvloed was bij een
der ramen.

Vandaar hadden zij een uitstekend gezicht op het dak van het
ziekenhuis en zij konden de twee loodgieters duidelijk aan het werk
zien.

Het werd tijd om te lunchen en Raffles bestelde een maaltijd.

En juist toen hij en Charly aanstalten maakten toe te tasten, zagen zij
hoe de twee arbeiders daar boven op het platte dak hun
gereedschappen bijeen pakten, het vuurtje onder een kleine kachel
doofden, deze afbraken en in een grooten zak deden.
—Wij treffen het! mompelde Raffles zachtjes. Het werk is gedaan,
anders zouden die arbeiders niet midden op den dag hun boeltje
pakken en verdwijnen!

—Misschien gaan zij wel lunchen! meende Charly.

—Met hun gereedschappen en hun kachel? vroeg Raffles spottend.

Dat klinkt niet erg waarschijnlijk! Kijk, zij laten alles al aan een touw
zakken en beneden wordt alles in ontvangst genomen door den
derden man, die zoo even met een kar is komen aanrijden! Neen, het
werk is wel degelijk afgeloopen!

—Maar die lieden zullen aan de directie gaan mededeelen, dat zij
gereed zijn, riep Charly uit.

—Welnu—dan komen wij haar morgenochtend zeggen, dat zij zich

vergist hebben!

—Dan zullen wij moeten weten van welke firma zij zijn!

—Dat is tenminste veiliger, voor het geval er soms naar mocht

worden gevraagd, en daarom zullen wij snel moeten eten en hen
volgen!

—Maar geloof je, dat het ons van nut kan zijn, als wij daar op het dak
zijn, vroeg Charly.

—Van zeer groot nut! Ik heb niet meer dan vijf minuten noodig! Ik
weet waar het vertrek gelegen is, waar men gewonden en zieken,
waarvan tijdens hun verblijf gebleken is, dat zij nog een appeltje met
de justitie te schillen hebben, heen brengt. Het is op de vijfde, dat wil
zeggen, op de bovenste verdieping, en aan de zijde van de
binnenplaats. Je kunt het hier natuurlijk niet zien, maar het platte dak
loopt om die binnenplaats heen, en het zal zeker niet moeilijk vallen,
ons met den gewonde in verbinding te stellen, en hem de noodige
instructies te geven. En daar valt mij nog een beter plan in, waardoor
ons verblijf op het dak niet noodeloos gerekt zal worden—ik ben een
werkbaas van de loodgieterszaak, die komt zien, of het karweitje wel
behoorlijk verricht is!

—En als het uitkomt, dat je de directie wat op de mouw hebt

gespeld? Als bijvoorbeeld later eens een échte werkbaas komt
kijken?

—Wel, dan heeft Beaupré in ieder geval zijn instructies en de rest! En

wij moeten dan maar hopen, dat die werkbaas n i e t komt! En laten
wij ons nu wat haasten, want ik zie de arbeiders niet meer op het dak,
en zij zullen wel aanstonds met hun kar vertrekken.

De twee vrienden beëindigden spoedig hun maaltijd, en betaalden

den kellner.

Zij traden juist bijtijds weder naar buiten, om er getuige van te zijn,
dat de drie loodgieters met hun handkar vertrokken.

Zij liepen hen na, daar zij begrepen, dat de loodgieterszaak

onmogelijk ver uit de buurt kon zijn, anders hadden de arbeiders wel
paard en wagen, of een auto gebruikt.

Dit hadden zij blijkbaar goed ingezien, want nog geen half uur later
reed de kar een winkelstraat in, en reeds van verre zagen de beide
mannen het uithangbord van de zaak, welke de loodgieters had
uitgezonden: „Dribbel and Jones, Aanleg van gas en waterleiding”.

Zij bleven voor de zekerheid nog even staan en zagen, hoe de kar
een smalle gang naast den winkel werd binnen gereden. [9]

Toen sprak Raffles glimlachend:

—Het was een goede digestie-wandeling, Charly—wij weten nu
tenminste, wat wij weten wilden! En nu snel naar ons huis in de
nieuwe straat, die nog zelfs geen behoorlijken naam heeft—om
Marthe Debussy in veiligheid te brengen, en dan de handen vrij te
hebben!

[Inhoud]
HOOFDSTUK III.
 De vlucht.

Den volgenden morgen om negen uur kwamen zich twee mannen

aan het ziekenhuis in de Sloane Street aanmelden, die verklaarden,
te moeten nazien of de reparatie op het dak wel goed verricht was.

Zij noemden den naam van de firma Dribbel and Jonas en daar zij er
juist uitzagen als de personen, die zij voorgaven te zijn, werden
Raffles en Charly zonder eenig bezwaar binnen gelaten, en wees
men hun den weg naar het groote dak.

Daar gekomen, nam Raffles de reparatie schijnbaar met de grootste

zorg op, schudde nu en dan het hoofd, en zeide eindelijk tegen den
surveillant, die hem en Charly hierheen had vergezeld:

—Er zijn nog een paar plekken, die nog wel wat afgewerkt mochten
worden, mijnheer! Het werk van een uur! Laat u niet ophouden—wij
vinden den weg wel weer!

De zoogenaamde werkbaas wierp zijn tasch met gereedschap neer

en bekommerde zich niet langer om den surveillant, die vond, dat hij
zijn plicht had gedaan en er bitter weinig voor gevoelde in het
snerpend koude Novemberweder langer op het dak te blijven dan
volstrekt noodzakelijk was.

—Meldt u dan even aan als het werk gedaan is, zeide hij in de
handen wrijvend.

—Tot uw dienst, mijnheer! kwam Charly, die reeds de tasch geopend

had en er een klein komfoor had uitgehaald.

De surveillant verdween en Raffles en Charly hadden nu het rijk

alleen, en konden ongestoord de omgeving opnemen.
Diep onder hen gaapte het groote, vierkante gat van de binnenplaats.

Recht tegenover de plek waar zij stonden verrees de achtergevel van

het voorgebouw, aansluitend bij de zijgevels, die op hun beurt weder
aansluiting hadden bij het achtergebouw.

Het dak liep dus inderdaad onafgebroken om de binnenplaats heen,

en dat was een groot voordeel voor de twee gewaande loodgieters.

Want vlak over hen, aan de andere zijde van die plaats, op bijna
zestig meter afstand, was het getraliede raam van de kamer, waar
men de misdadige zieken en gewonden verpleegde, die slechts
wachten op het oogenblik dat zij deze kamercel zouden verruilen
tegen de cel eener gevangenis. [10]

De bovenkant van het raam raakte bijna de daklijst, die hier slechts
een smalle goot droeg.

Het raam was gesloten, maar de gordijnen waren terzijde geschoven,

en vaag was daarbinnen de omtrek van het bed zichtbaar.

Raffles trok zich zoover terug, dat men van den anderen kant niets
zou kunnen zien, tenzij uit de ramen, die op gelijke hoogte met dat
van de cel gelegen waren, en haalde een sterk vergrootenden
verrekijker uit zijn zak, in den vorm van een gewonen tooneelkijker,
maar met dit verschil dat men door eenvoudig een scharnier te doen
draaien, met behulp van een spiegelinrichting ook terzijde, en zelfs
achter zich kon zien.

Hij bracht het toestel voor het oog, zorg dragende dat hij half achter
een schoorsteen verborgen was, en keek naar het raam.

Een seconde later liet hij den kijker weder zakken en zeide:
—Hij ligt alleen! Het is tijd, om hem onze boodschap te doen
toekomen!

—Maar dat zal niet mogelijk zijn! kwam Charly. Het raam is immers
gesloten!

—Dan moet het geopend worden, hernam Raffles kortaf. Jij blijft hier
en ik ga naar dien overkant, tot boven het venster. Doe alsof je aan ’t
werk waart, maar let goed op mij, en houd den kijker bij je, zoodat je
in het vertrek kunt zien. Er zal zoo dadelijk wel een verpleger
binnenkomen, en het raam openen. Zoodra je ziet, dat hij zich weder
verwijderd heeft, ga je op een schoorsteenrand zitten. Dat zal voor
mij het teeken zijn, dat alles veilig is en ik kan handelen! Mocht er
echter, juist terwijl ik aan den overkant kom, iemand de cel
binnentreden, wat ik natuurlijk niet zou kunnen zien, zonder mijn
tegenwoordigheid te verraden, waarschuw mij dan door je pet af te
nemen en over je hoofd te strijken met je zakdoek.

—In orde! zeide Charly. Ik zal goed acht geven.

Raffles nam een dunne koperen staaf uit de gereedschapstasch en

begon zijn tocht over het dak, na Charly te hebben toegeknikt.

Nu en dan stond hij stil en deed alsof hij de looden dakbedekking aan
een inspectie onderwierp.

Zoo bereikte hij de overzijde, juist boven het venster van de cel.

Charly zag hoe hij een stuk papier uit zijn zak haalde en er blijkbaar
met groote letters iets op schreef.

Daarop bond Raffles het papier aan het uiteinde van zijn dunne
koperen staaf en keek naar de overzijde.
Charly had zijn kijker gebruikt en gezien, dat de kamer nog altijd leeg
was op den zieke na.

Het moest dus nu gewaagd worden.

Charly verroerde zich niet en Raffles begreep hieruit, dat alles veilig
was.

Toen trok Raffles de stang weder naar zich toe, en wachtte.

Zou Beaupré wakker zijn geweest?

Zou hij het papier wel hebben gezien, en zoo ja—zou hij dan hebben
gezien wat er op te lezen stond?

Binnen enkele minuten moest dit beslist worden.

Charly hield zijn blikken door den kijker onafgewend op het

ziekenvertrek gericht, ten einde Raffles op tijd te kunnen
waarschuwen.

En nu zag hij hoe de deur open ging en er een verpleger binnen trad.

Blijkbaar had Beaupré de boodschap kunnen lezen, en op een bel

naast zijn bed gedrukt.

De verpleger scheen eenige woorden met den gewonde te wisselen

—en trad toen naar het raam toe.

Charly bukte zich dadelijk, en deed alsof hij druk bezig was met de
dakbedekking.

Toen hij even later weder keek, zag hij dat het raam geopend
was.…..
Weer sprak de verpleger even met Beaupré, en daarop verwijderde
hij zich weder.

Charly zette zich dadelijk op den rand van een schoorsteen, zooals
hij met Raffles had afgesproken.

Deze ging plat op zijn buik op het platte dak liggen, zoodat zijn hoofd
over de goot heen stak, en deed alsof hij deze onderzocht.

Maar inderdaad wierp hij met de hand, die de sterkte van de goot
scheen te beproeven, een klein pakje naar binnen, dat juist op het
bed van den gewonde terecht kwam.

Hij kwam weer overeind, liep voor de leus nog eenige passen verder,
trok hier en daar aan een steunstang van een schoorsteen, en
keerde toen weder kalm op zijn schreden terug.

Spoedig had hij zich weder bij Charly gevoegd, die vol ongeduld zijn
terugkomst afwachtte. [11]

Op zachten toon, alsof men hem hier had kunnen hooren, vroeg de
jonge man:

—Wat heb je nu eigenlijk gedaan?

—Ik heb hem eerst gevraagd, een verpleger te roepen en dien te

verzoeken het venster onder een of ander voorwendsel te openen.
Daarop heb ik een pakje op zijn bed geworpen, hetwelk in de eerste
plaats een middel van mijn vinding bevat, bij uitstek geschikt om de
uitgeputte levenskrachten zeer snel te herstellen, veel sneller dan
waarop de geneesheeren kunnen rekenen, vervolgens een sterke vijl,
om de tralies door te vijlen, en een dun maar zeer stevig
knoopentouw, aan het einde van een stalen haak voorzien, waaraan
hij zich zal kunnen laten zakken, en eindelijk een kort briefje met de
noodige aanwijzingen.
—Wat behelst dat briefje als ik vragen mag?

—Ik zeg hem daarin, dat hij zich bij voortduring zeer zwak en ziek
moet houden, om de verplegers om den tuin te leiden, zoodat zij niet
al te waakzaam zijn, dat hij over drie dagen sterk genoeg zal zijn om
te ontvluchten, al zijn zijn wonden dan zeker nog lang niet genezen,
en dat er precies om drie in den nacht een auto zal gereed staan op
den hoek van de Sloane Street en Beggar Road, om hem op te
nemen.

—En wij?

—Wij wachten op hem!

—Maar zal hij dat pakje wel kunnen verbergen?

—Als hij dat niet kan, is hij geen knip voor den neus waard, en zeker
niet geschikt om aan het hoofd eener Parijsche dievenbende te
staan! antwoordde Raffles verachtelijk. Hij kan het onder aan zijn
matras binden, met behulp van het touw dat er omheen gebonden
was, maar hij zal nu wel niet verbed worden, en daarom kan hij het
eenvoudig onder zijn peluw schuiven. En nu hebben wij hier verder
niets meer te doen, mijn jongen! Ik zie, dat de reparatie toch wel heel
deugdelijk is uitgevoerd! Op weg maar!

De twee mannen namen hun zak met gereedschap weder op en

klommen door het dakluik naar den zolder, waarop zij de trappen
afdaalden en beneden bij den portier hun bevindingen omtrent den
staat van het dak gingen mededeelen.

Nog geen vol half uur nadat zij gekomen waren, stonden de
gewaande loodgieters weer op straat.

Drie dagen waren verloopen.

Het was omstreeks een uur in den nacht.

De surveillanten hadden zooeven hun laatste ronde door de

particuliere ziekenkamers gedaan, en ook Beaupré bezocht, die
schijnbaar koortsig op zijn leger heen en weer woelde.

En daarop was de ijzeren deur in het slot gedraaid, en de gewonde

hoorde, hoe een der surveillanten een stoel naderbij trok, en hoe het
meubel onder zijn gewicht kraakte, toen hij er op ging zitten, om
naast de deur te waken.

Beaupré was verre van koortsig—of het moest zijn van spanning en
opwinding.

Integendeel—hij gevoelde zich sterk en krachtig en alleen zijn

wonden deden hem pijn.

Het geheimzinnige middel, hem door Raffles verstrekt, had wonderen

uitgewerkt!

Hij had zijn kracht als het ware met ieder uur voelen toenemen, en nu
was hij vastbesloten een poging te wagen om zijn vrijheid te
herkrijgen.

Hij wist wel dat de bewaker meestal na een half uur insluimerde,
oordeelende, dat iedere poging om te ontvluchten van den zoo
verzwakten en uitgeputten gewonde tot de onmogelijkheden
behoorde.

En zoo luisterde hij ingespannen naar alle geluiden welke door de

ijzeren deur tot hem doordrongen.

Hij wist precies hoe laat het was, want alles was in dit ziekenhuis met
de regelmaat van een klok ingericht.
Het oogenblik ging aanbreken waarop hij aanstalten zou moeten
maken om te ontvluchten.

Hij zou zeker wel een uur noodig hebben om de tralies van zijn
venster door te vijlen.

Hij voelde zijn slapen kloppen en een zenuwachtige rilling liep nu en

dan over zijn lichaam, want het volgende uur moest hem de vrijheid
of een langdurige, misschien wel levenslange gevangenisstraf
brengen.

Eindelijk vernam zijn oor een zacht gerucht, dat hem wel bekend was
—het was het gesnurk van den surveillant aan de andere zijde van
de deur, dat tot hem doordrong.

Dadelijk wierp Beaupré het dek van zich af en liet zich behoedzaam
uit het bed glijden.

Tot zijn geluk droeg hij een warmen, flanellen pyjama en hij wist dat
zijn overjas in een kast was opgehangen. [12]

Hij sloop daar naar toe, nam er het kleedingstuk uit en trok het aan.

Vervolgens sloeg hij nog een deken om, trad op het venster toe en
opende het zonder geraas te maken.

Hij nam nu de vijl ter hand, besmeerde een der staven met een
weinig vet ten einde het gekras van het werk zooveel mogelijk te
dempen en begon met zijn arbeid.

Het ging vlugger dan hij vermoed had, want de vijl was van
uitmuntende kwaliteit en tastte het ijzer der tralies snel aan.

Toch duurde het een vol uur alvorens Beaupré een paar der spijlen
geheel had doorgevijld, en had kunnen ombuigen, zoodat er een gat
ontstond, groot genoeg om hem doorgang te geven.

Van tijd tot tijd was hij aan de deur gaan luisteren, waarachter nog
steeds het regelmatig gesnurk van den surveillant klonk.

Nu zou hij zich echter moeten haasten, want over een kwartier zou
de hoofdverpleger van deze verdieping zijn ronde doen om zich te
vergewissen of alles in orde was en dan kon hij het wel in zijn hoofd
krijgen een blik in de celkamer te werpen.

Het oogenblik om te handelen was dus aangebroken.

Beaupré sneed een reep van de wollen deken, met behulp van het
kleine mes hetwelk hij eveneens in het pakje had gevonden, dat
Raffles hem zoo behendig in handen had weten te spelen en knoopte
deze bij wijze van een bouffante om zijn hals, ten einde aldus
eenigszins beschermd te zijn tegen de koude van den
Novembernacht.

Hij was blootsvoets, maar daaraan was niets te doen, en in ieder

geval zou het hem bij het klimmen van nut zijn.

Hij stak dus eerst het hoofd door de opening welke hij zooeven
gemaakt had, en vervolgens het bovenlichaam, daarop trok hij zich,
terwijl hij zich met beide handen aan de nog vast zittende tralies vast
hield, geheel naar buiten en plaatste de voeten op de smalle
vensterbank.

Het knoopentouw met den stalen haak aan het eind had hij om zijn
middel gewikkeld zoodat het hem bij zijn pogingen niet kon hinderen.

Hij kon nu met zijn handen juist den rand van de dakgoot bereiken en
trok zich nu op, met de voeten steun zoekend op de dwarsloopende
tralies.
Het was zwaar werk voor een man, wiens kracht op kunstmatige
wijze was opgewekt, en die bijna een week op het ziekbed had
gelegen, maar de zucht tot vrijheid staalde zijn spieren en deed hem
alle moeilijkheden overwinnen.

Hij wist een been over den gootrand te slingeren, zette zich nu met
het andere been met kracht af—en lag een oogenblik later hijgend in
de smalle goot.

De nachtvorst had het dunne laagje water in ijs doen verkeeren, en

Beaupré voelde de koude door zijn ledematen dringen.

Een soort verdooving maakte zich van hem meester, waartegen hij uit
alle macht streed, want als hij daaraan toegaf, dat begreep hij wel,
dan was hij verloren.

Hij richtte zich dus op, kroop uit de goot op het dak, en liep zoo vlug
hij kon naar den kant waar zich de rechterzijgevel bevond.

Hij was het ziekenhuis vaak genoeg voorbij gekomen, om te weten

dat zich daar de brandtrap bevond, die ongeveer drie meter lager, bij
de bovenste verdieping begon.

Er stond een straatlantaarn tegenover de trap en hij kon haar vrij

duidelijk onderscheiden, ondanks de duisternis.

Beaupré maakte het eene uiteinde van het touw met behulp van den
stalen haak aan een steunijzer van een der kolossale schoorsteenen
vast en wierp het uiteinde over den rand van het dak.

Hij beproefde de sterkte, door er uit alle macht aan te trekken en

bevond, dat het zijn gewicht gemakkelijk zou kunnen dragen.

Het koord bleek uit zijde gevlochten te zijn en zeker even sterk als
staaldraad van dezelfde dikte.
Hij had nog slechts enkele minuten den tijd en hij begreep, dat hij niet
mocht aarzelen, want dadelijk na zijn ontsnappen zou er alarm
gemaakt worden en men zou natuurlijk allereerst aan de brandtrap
denken.

Hij vatte het touw dus boven een der knoopen vast, slingerde zich
moedig over den rand van het dak, en liet zich behoedzaam zakken,
telkens met zijn bloote voeten steun zoekend op een der knoopen,
daar hij niet krachtig genoeg was om alleen op zijn armen te kunnen
vertrouwen. [13]

Na eenige minuten voelde hij met den rechter voet iets kouds—de
ijzeren leuning van het kleine portaal, waarmede de brandtrap
eindigde.

Nog iets lager—en hij stond veilig en wel op de brandtrap.

Hij duizelde van de inspanning, maar hij begreep dat hij hier niet
mocht blijven staan, niet alleen omdat hij nog lang niet veilig was,
maar ook omdat de koude nachtlucht hem in zijn dun gewaad zeker
ernstig zou kunnen schaden.

Hij begon dus zoo snel hij kon de brandtrap af te loopen.

Deze eindigde een paar meter boven de oppervlakte van de straat,

en Beaupré liet zich vallen.

Zijn wonden deden hem pijn, en een oogenblik bleef hij versuft staan.

Maar een snerpend gefluit boven zijn hoofd, dat onmiddellijk in het
ziekenhuis en op de binnenplaats herhaald werd, joeg hem
eensklaps weder op!

Zijn vlucht was ontdekt en de surveillant had alarm gemaakt.

Beaupré bezon zich geen seconde, maar zette het aanstonds op een
loopen in de richting van de Beggar Street.

Juist toen hij den hoek bereikte, kwam er een agent van politie
aanstormen, die door het fluitsignaal was aangelokt, waarvan hij de
beteekenis blijkbaar kende.

Er had een botsing plaats tusschen den man in blauw laken en den
vluchteling in zijn grijs flanellen pyjama, en bijna was de Fransche
markies ter aarde geslingerd.

Hij wist echter op de been te blijven, bracht den agent een hevigen
vuistslag tegen de kaak toe, die hem deed duizelen, en snelde zoo
vlug hij kon voort.

Voorbijgangers waren er gelukkig op dit uur niet te zien, en op

honderd pas afstand schitterde eensklaps het krachtige licht van een
autolantaarn.

Nog een laatste inspanning en Beaupré voelde zich door krachtige

handen in de wachtende auto getrokken, die dadelijk met de grootste
snelheid wegreed.

Charly Brand, die het avontuur niet had willen missen, legde dadelijk
een dikken plaid over den vluchteling, die nu met gesloten oogen en
spierwit gelaat achterover in de kussens leunde, terwijl Raffles hem
den inhoud van een klein fleschje, dat hij uit zijn vestzak had gehaald,
tusschen de lippen goot.

Wat de man achter het stuurwiel betreft, dat was de trouwe

Henderson, de reusachtige chauffeur, die eveneens zijn rol bij de
ontvoering had gespeeld.

De tocht duurde ongeveer drie kwartier, en toen stond de auto stil

vóór het huis, waarin ook de minnares van Beaupré zich bevond.