Professional Documents
Culture Documents
Computer Simulations Advances Research Applications 8
Computer Simulations Advances Research Applications 8
COMPUTER SIMULATIONS
ADVANCES IN RESEARCH
AND APPLICATIONS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
COMPUTER SCIENCE, TECHNOLOGY
AND APPLICATIONS
COMPUTER SIMULATIONS
ADVANCES IN RESEARCH
AND APPLICATIONS
MICHAEL D. PFEFFER
AND
ERIK BACHMEIER
EDITORS
Copyright © 2018 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted
in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying,
recording or otherwise without the written permission of the Publisher.
We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to
reuse content from this publication. Simply navigate to this publication’s page on Nova’s website and
locate the “Get Permission” button below the title description. This button is linked directly to the
title’s permission page on copyright.com. Alternatively, you can visit copyright.com and search by
title, ISBN, or ISSN.
For further questions about using the service on copyright.com, please contact:
Copyright Clearance Center
Phone: +1-(978) 750-8400 Fax: +1-(978) 750-4470 E-mail: info@copyright.com.
Independent verification should be sought for any data, advice or recommendations contained in this
book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to
persons or property arising from any methods, products, instructions, ideas or otherwise contained in
this publication.
This publication is designed to provide accurate and authoritative information with regard to the subject
matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in
rendering legal or any other professional services. If legal or any other expert assistance is required, the
services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS
JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A
COMMITTEE OF PUBLISHERS.
Additional color graphics may be available in the e-book version of this book.
Preface vii
Chapter 1 Computer Simulations of Passenger Car Radial
Tires Using the Finite Element Method 1
Mir Hamid Reza Ghoreishy
Chapter 2 Computer Simulations as a Solution Method
for Differential Games 63
Guennady A. Ougolnitsky and Anatoly B. Usov
Chapter 3 The Computer Simulation of the Talbot Effect and
Carpet via the Iterative Fresnel Integrals Method 107
Kazi Monowar Abedin, Aamna Al-Saedi
and S. M. Mujibur Rahman
Chapter 4 Modeling of the Light Scattering by a Yarn 129
Zeguang Pei
Chapter 5 The Unsteady Analysis of a Solar Chimney
Power Plant 149
Ahmed Ayadi, Zied Driss, Haythem Nasraoui,
Abdallah Bouabidi and Mohamed Salah Abid
vi Contents
scattering model is adopted to simulate the light scattering from the yarn
surface and to analyze the relation between the scattering pattern and the
surface twist angle. The solution of the scattering intensity distribution for
the yarn surface profile is derived and numerical results indicate that the
highest light intensity on the backward light scattering pattern lies along
the direction perpendicular to the surface fibers regardless of the fiber or
yarn parameters. The result of this chapter can serve as the theoretical basis
for the measurement of yarn surface twist angle based on backward light
scattering by the yarn. The authors go on to present a numerical study of
the unsteady airflow characteristics inside a solar chimney power plant.
Ansys Fluent 17.0 is used to simulate the air flow within the solar setup.
To validate the authors’ study, a solar prototype is built in the National
School of Engineers of Sfax, University of Sfax, Tunisia in the North
Africa. Subsequently, the authors also present a distributed boundary value
problem of thermal desorption with nonlinear dynamical boundary
conditions. The problem is reduced to the nonlinear functional differential
equation of neutral type for surface concentration. A numerical method is
developed for TDS spectrum simulation, where only integration of a
nonlinear system of low order (compared with, e.g., the method of lines)
ordinary differential equations (ODE) is required. Lastly, the book presents
an aggregation procedure for hydrogen permeability experiments
precluding depressurization and (or) change of the samples of investigated
material. Additionally, the corresponding mathematical software for
correct processing of measurements along with appropriate assembly of the
experimental unit is described.
Chapter 1 - This monograph is intended to briefly describe the methods
and techniques developed so far for the computer simulation of steel-belted
radial tires under different mechanical and thermal loads. The chapter
begins with a concise overview and background of the topic during the past
decade. Then, the finite element method and the commercial available
codes are introduced with an emphasis on Abaqus as the mostly used
software in this field. The next part deals with the tire geometry and
creation of a consistent and computationally effective finite element mesh
from 2D and 3D drawings. Following that, the material models used for the
Preface ix
studies into the diffusion limited regime (DLR) and the surface limited
regime (SLR)).
Chapter 7 - High-purity hydrogen is required for clean energy and a
variety of chemical technology processes. The various alloys potentially
well-suited for use in gas-separation plants for the membrane technology
of high- purity hydrogen production were investigated by measuring
specific hydrogen permeability. For structural materials one had to
estimate the parameters of diffusion and sorption to numerically model the
different scenarios and experimental conditions of the material usage
(including extreme ones), and identify the limiting factors. The proposed
mathematical model matches the experimental method of hydrogen
permeability and takes into account only the main limiting factors for the
applied membrane filtering problem and the informative capabilities of the
experiment modifications. The penetration method allows determining the
diffusion coefficient by so-called lag time. The accuracy of the estimation
depends on the degree of proximity to the DLR (diffusion limited regime)
mode. The method of ‘communicating vessels’ is more sensitive to surface
processes. ‘Separate’ application of these methods leads to a situation
where the materials studied are in fact somewhat different (for example,
due to different impacts on the surface), and there appear significant
differences in hydrogen permeability parameter estimates. The main idea
of this chapter is to develop an aggregation procedure for hydrogen
permeability experiments precluding depressurization and (or) change of
the samples of the investigated material. The authors also present the
corresponding mathematical software for correct processing of
measurements along with appropriate assembly of the experimental unit.
This chapter describes the proposed cascade experiment technique and the
mathematical software: model, computer simulation and parametric
identification algorithms. This allows increasing the informative capacity
of the experimental studies and the accuracy of the estimation of hydrogen
permeability parameters (diffusion, absorption, desorption).
In: Computer Simulations ISBN: 978-1-53613-095-9
Editors: Michael D. Pfeffer et al. © 2018 Nova Science Publishers, Inc.
Chapter 1
ABSTRACT
Corresponding Author Email: M.H.R.Ghoreishy@ippi.ac.ir.
2 Mir Hamid Reza Ghoreishy
complex frictional behavior between tire tread surface and road are also
taken into consideration for the analyses in which the non-isothermal and
dissipations effects become prominent. The latter case is mainly
associated with prediction of rolling resistance and fuel economy.
Different methods for tire analysis are then explained. These can be
categorized into static loadings (rim mounting, inflation and footprint),
steady state rolling (force and moment, slip and camber angles, material
history for prediction of rolling resistance and wear) and finally transient
loading (rolling over road disturbance and curb, accelerating and
braking). The results of the analyses of some radial tires under the
mentioned loading conditions are presented and discussed. These results,
wherever possible, are also compared with experimental data to verify the
applicability of the developed models and simulation methods.
1. INTRODUCTION
2. BACKGROUND
Due to the large deformations arise in tire and also nonlinear behavior
of rubberlike materials as well as the flexible contact area between tire and
road, the use of nonlinear finite element modeling is inevitable.
Consequently, nonlinear finite element code such as Abaqus [24] must be
used to develop a proper model for tires. As it was pointed out in the
introduction, a tire undergoes very widespread types of loads during its
service life. Therefore, various finite element models have been developed
so far for the prediction of tire response to individual or combined
loadings. These generally include 1) axisymmetric two-dimensional
models for rim mounting and applying of inflation pressure 2) three-
dimensional models under static conditions such as footprint (normal) and
shear loads 3) steady state rolling analysis for straight line motion and
cornering and 4) transient analysis for the modeling of tire under time-
dependent conditions. All the models and procedures given in the
following sections are based on the use of Abaqus code [24]. This is a very
well-known computer program which is widely used for the analysis of
tires by many investigators. Like any other commercial codes, the use of
this program needs carrying out three consecutive stages, known as pre-
processing, processing and post processing. The purpose of the pre-
processing step is to develop finite element mesh including elements and
nodes, selection of proper material models and assigning them to elements,
applying boundary and initial conditions as well as necessary loads and
finally providing supplementary information such as defining time period,
time step and function, increment size and the extra geometrical data like
Computer Simulations of Passenger Car Radial Tires … 9
thickness. This step is very time consuming and requires spending much
time by model developer. Abaqus/CAE is the pre-processing section of the
Abaqus software. Processing stage involves the solution of the equations
and obtaining the results. For tire analysis, depending on the type of model
(2D vs. 3D or steady state vs. transient) and computer hardware, it may
take a few seconds up to several hours to complete a job on the system.
Abaqus/Standard and Abaqus/explicit are the main processing modules
which are used for the tire analysis. The post-processing stage deals with
the representation of the results in such way that they can be best used for
interpretation of the outcomes and optimize of the initial designs. This step
in Abaqus is performed by Abaqus/Viewer. The details of each of these
steps for different tire model are given in next sections.
Figure 5. Finite element mesh of a 205/60R15 tire with total number of elements and
nodes equal to 1168 and 940, respectively.
Figure 6. Finite element mesh of a 185/65R14 tire with total number of elements and
nodes equal to 850 and 687, respectively.
Computer Simulations of Passenger Car Radial Tires … 13
Figures 5 and 6 show the finite element meshes of two 205/60R15 and
185/65R14 radial tires with axisymmetric solid and surface elements. As it
can be seen, due to the complex shape of the tire cross section, both
quadrilateral and triangular elements are used to create a convergent and
fully verified mesh. It is worth mentioning that although tire lay-out in 2D
view represents symmetry, due to the requiring of a full model for rolling
analysis it is preferred not to exploit symmetry to construct the mesh.
𝐁 = 𝐅. 𝐅 𝐓 (1)
𝑘
3( 0 )−2
𝜇0
𝜈= 𝑘 (4)
6( 0 )+2
𝜇0
where 𝑘0 , 𝜇0 and 𝜈 are the initial bulk modulus, initial shear modulus and
Poisson's ratio, respectively. In a fully incompressible material 𝑘0 → ∞
and thus 𝜈 = 0.5. Increasing the compressibility gives rise to lowering 𝑘0 ,
hence Poison's ratio becomes lower than 0.5. These types of materials are
called as nearly incompressible. For better description of the mechanical
behavior of these materials, the deviatoric part of the deformation is
separated from volumetric changes that may take due to compressibility or
temperature changes. In order to develop the corresponding mathematical
relationship, a new deformation gradient tensor is defined as:
1
𝐅̅ = 𝐽−3 𝐅 (5)
2
𝐼1̅ = 𝐽−3 𝐼1 = 𝜆̅12 + 𝜆̅22 + 𝜆̅23
4
𝐼2̅ = 𝐽−3 𝐼2 = 𝜆̅12 𝜆2̅2 + 𝜆̅12 𝜆2̅3 + 𝜆̅22 𝜆2̅3 (6)
𝐼3̅ = 1
1
𝜆̅𝑖 = 𝐽−3 𝜆𝑖 (7)
The new form of the left Cauchy-Green tensor is also given by:
2
̅ = 𝐅̅. 𝐅̅ 𝐓 = 𝐽−3 𝐁
𝐁 (8)
where 𝑊𝑑𝑒𝑣 and 𝑊𝑣𝑜𝑙 are the deviatoric and volumetric components,
respectively. 𝐽𝑒𝑙 is the elastic volume ratio defined as:
𝐽
𝐽𝑒𝑙 = (10)
𝐽𝑡ℎ
𝐽𝑡ℎ = (1 + 𝜖 𝑡ℎ ) (11)
Neo-Hookean
This is the first and simplest mechanistic model based on the statistical
mechanics and thermodynamics given as [31]:
1 2
𝑊 = 𝐶10 (𝐼1 − 3) +
𝐷1
(𝐽𝑒𝑙 − 1) (12)
where 𝐶10 and 𝐷1 are the material parameters which can be generally
dependent on the temperature. This model can precisely predict the real
behavior of rubber up to 40% and 90% in tension and shear modes,
respectively. The initial shear and bulk moduli are also expressed by:
𝜇0 = 2𝐶10
2 (13)
𝑘0 = 𝐷
1
Mooney-Rivlin
This model is based on the nonlinear elastic theory and has been first
proposed by Mooney as:
1 2
𝑊 = 𝐶10 (𝐼1̅ − 3) + 𝐶01 (𝐼2̅ − 3) + 𝐷 (𝐽𝑒𝑙 − 1) (14)
1
Polynomial Models
The neo-Hookean and Mooney-Rivlin models are special forms of a
general equation expressed by:
Computer Simulations of Passenger Car Radial Tires … 17
𝑖 1 2𝑖
𝑊 = ∑𝑛𝑖=0 ∑𝑚 ̅ ̅ 𝑗 𝑛 𝑒𝑙
𝑗=0 𝐶𝑖𝑗 ( 𝐼1 − 3) (𝐼2 − 3) + ∑𝑖=1 𝐷 (𝐽 − 1) (15)
𝑖
Reduced Polynomials
These models are like polynomial models with the exception that the
deviatoric part of the model is only dependent to first invariant of the left
Cauchy-Green tensor (𝐁 ̅ ), i.e.:
1 2𝑖
𝑊 = ∑𝑛𝑖=1 𝐶𝑖0 (𝐼1̅ − 3)𝑖 + ∑𝑛𝑖=1 𝐷 (𝐽𝑒𝑙 − 1) (16)
𝑖
1 2
𝑊 = 𝐶10 (𝐼1̅ − 3) + 𝐶20 (𝐼1̅ − 3)2 + 𝐶30 (𝐼1̅ − 3)3 + (𝐽𝑒𝑙 − 3) +
𝐷1
1 4 1 6
𝐷2
(𝐽𝑒𝑙 − 3) +
𝐷3
(𝐽𝑒𝑙 − 3) (17)
Arruda-Boyce Model
In this model the molecular structure of an elastomeric material is
considered to be described by a non-Gaussian behavior [35]. It is a
mechanistic model expressed by the following equation:
18 Mir Hamid Reza Ghoreishy
2
𝐶𝑖 𝑖̅ 1 (𝐽𝑒𝑙 ) −1
𝑊 = 𝜇 ∑5𝑖=1 𝑖
2𝑖−2 (𝐼1 − 3 ) + 𝐷 ( 2
− ln(𝐽𝑒𝑙 )) (18)
𝜆𝑚
where
1 1 11 19 519
𝐶1 = , 𝐶2 = , 𝐶3 = , 𝐶4 = , 𝐶5 =
2 20 1050 7000 673750
Ogden Model
It is also a phenomenological model which directly relates the
deviatoric part of the hyperelastic model to the principle extension ratios as
[36]:
2𝜇 𝛼 𝛼 𝛼 1 2𝑖
𝑊 = ∑𝑛𝑖=1 𝛼 2𝑖 (𝜆̅1 𝑖 + 𝜆̅2 𝑖 + 𝜆̅3 𝑖 − 3) + ∑𝑛𝑖=1 𝐷 (𝐽𝑒𝑙 − 1) (19)
𝑖 𝑖
3 2
2 𝐼̃−3 2 1 𝐽𝑒𝑙 −1
𝑊= 𝜇 {−(𝜆2𝑚 − 3)[ln(1 − 𝜂) + 𝜂] − 𝑎( ) }+ ( −
3 2 𝐷 2
ln 𝐽𝑒𝑙 ) (20)
where
𝐼̃ − 3
𝐼̃ = (1 − 𝛽)𝐼1̅ + 𝛽𝐼2̅ , 𝜂 = √ 2
𝜆𝑚 − 3
Marlow Model
The Marlow model does not contain any explicit relation between
strain energy density and invariants or stretch ratios [38]. It is assumed that
the strain energy density is only a function of the first invariant of the
strain tensor, i.e.,
1
𝐼1̅ = 𝜆̅12 + 𝜆̅22 + ̅2 ̅2 (22)
𝜆1 +𝜆 2
It can be shown that the minimum and maximum values of the 𝐼1̅ are
between 3 and +, respectively. The uniaxial form of 𝐼1̅ (𝜆̅2 = 1⁄ ) can
√𝜆̅1
be written as:
2
𝐼1̅ = 𝜆̅2 + ̅ (23)
𝜆
̅
𝜆 (𝐼)−1̅
𝑊𝑑𝑒𝑣 (𝐼 )̅ = ∫0 𝑇 𝜎(𝜖)𝑑𝜖 (24)
where and are the nominal strain and stress in uniaxial tension test,
respectively, that are directly substituted in equation (24).
20 Mir Hamid Reza Ghoreishy
Uniaxial Test
Uniaxial experiment in tension mode is generally run using a universal
machine on dumbbell test specimen according to ASTM D-412. The
corresponding uniaxial compression test is performed on a cylindrical
specimen using the same machine according to ISO-7743 standard. Figures
7 and 8 show two fixtures used for tensile and compression tests installed
on a universal testing machine with their corresponding samples,
respectively.
Computer Simulations of Passenger Car Radial Tires … 21
Figure 7. Rubber sample with fixtures and extensometer used for uniaxial tensile test.
Figure 8. Rubber sample with supporting and forcing plates used for uniaxial
compression test.
Equibiaxial Test
In an equibiaxial test the rubber sample undergoes simultaneous
tensions in two perpendicular directions. To carry out such an experiment,
the testing machine needs to have two axes in which each axis has its own
independent moving mechanism. Therefore, several mechanisms have
been developed to do such test. One of the simplest methods is to use a
universal testing machine in conjunction with a special fixture that
emulates the biaxial tension in two directions. For example, the fixture
shown in Figure 9 generates tension in two directions.
22 Mir Hamid Reza Ghoreishy
Planar Test
In a planar tension or pure shear test the plane strain condition is
enforced on the test specimen by preventing the edges of the specimen
from contracting. This is achieved through the use of test samples with
high aspect ratios (large width to length). Figure 10 displays a typical
rubber sample fixed between the grips of a specially designed fixture to
carry out this using any standard universal testing machine.
Figure 10. Rubber sample with fixture for planar tensile (pure shear) test.
Computer Simulations of Passenger Car Radial Tires … 23
𝑑
𝜖 = 2𝑡 (25)
𝐹
𝜎 = 2𝐿𝑊 (26)
where 𝐿, 𝑊, and 𝑡 are the length, width and thickness of each sample,
respectively. It is worth noting that some software such as Abaqus does not
accept shear test data. In this case the previously mentioned test procedures
(uniaxial, equibiaxial and planar) can only be used.
Figure 11. Rubber samples with fixtures for quadruple shear test.
Volumetric Test
The last experiment from which the parameters of the volumetric part
of a hyperelastic model are computed is the volumetric test. The
24 Mir Hamid Reza Ghoreishy
Figure 12. A volumetric test fixture (left) mounted in a universal test machine (right).
The applied force and the displacement of the pin are recorded. These
values are then converted to the pressure and volume ratio using the
following equations:
𝐹
𝑃𝑟𝑒𝑠𝑠𝑢𝑟𝑒 = 𝐴 (27)
where 𝐹 is the force, 𝐴 is the area of the cross section of the test sample
and 𝜖 is the nominal strain.
The minimum experiment required to calibrate a hyperelastic model is
a uniaxial test on a rubber dumbbell sample. However, the accuracy of the
predicted values of the model parameters and goodness of fitting depend
Computer Simulations of Passenger Car Radial Tires … 25
on smoothing of the measured data, stability and types of the selected tests.
Smoothing is the selection of a mathematical procedure for removing of
the noise in experimental data so that the quality of the fitting is enhanced.
Stability is taken by Drucker stability condition [42] in which the
following inequality should be satisfied:
planar test, the deviation from experimental data for parameters calculated
from uniaxial results are much higher than when full data sets are utilized.
This clearly indicates that increasing the accuracy of the model parameters
in different deformational modes requires more experimental data to be
used during curve fitting. The same approaches were adopted for the
polynomial model with N = 2 as shown in Figures 15 and 16. The results
are similar to those reported for the Arruda-Boyce model. As can be in
Figure 15, for uniaxial mode, the data from both uniaxial only and full set
experiments gives a very good fitting. However, the deviation from
experimental data for planar test using only uniaxial test data, as shown in
Figure 16, is much greater than the similar case when the Arruda-Boyce
model is employed (Figure 14). This is because that the Arruda-Boyce
equation is a mechanistic model in which using only the uniaxial test data
can give rise to relatively good prediction in all deformational modes.
Nevertheless, the polynomial model is a phenomenological model which
requires data from multiple experimental tests becomes available to give a
high accurate curve fitting.
Figure 13. Experimental and computed stress-strain curve using Arruda-Boyce model
in uniaxial mode using uniaxial only and full set data.
Computer Simulations of Passenger Car Radial Tires … 27
Figure 14. Experimental and computed stress-strain curve using Arruda-Boyce model
in planar mode using uniaxial only and full set data.
𝑡
− 𝐺
𝑔𝑅 (𝑡) = 1 − ∑𝑁 𝑃
𝑖=1 𝑔̅𝑖 (1 − 𝑒 𝜏
𝑖 ) (31)
Computer Simulations of Passenger Car Radial Tires … 29
𝑡
𝑝 −
𝐶𝑖𝑗𝑅 (𝑡) = 𝐶𝑖𝑗0 (1 − ∑𝑁
𝑘=1 𝑔̅𝑘 (1 − 𝑒
𝜏𝑘
)) (32)
where 𝐶𝑖𝑗𝑅 (𝑡) and 𝐶𝑖𝑗0 are the reduced and instantaneous material constants,
respectively. Similar relations can be written for parameters in other
material models.
1
𝜖̅̇𝑐𝑟 = (𝐴𝑞 𝑛 [(𝑚 + 1)𝜖̅𝑐𝑟 ]𝑚 )𝑚+1 (33)
- Hyperbolic-sine law
belts, polyester or rayon for main body plies and nylon for cap plies. Two
material constants including elastic or Young’s modulus (𝐸) and Poisson’s
ratio (𝜈) are required to be defined. Young’s modulus is generally
determined from a uniaxial test on the preconditioned cord and for the
Poisson’s ratio a value of 0.3 is normally assumed. It should be noted that
due to the very stiff materials used for tire beads, the linear elastic model is
also used for this part of the tire.
Figure 18. Finite element mesh of a 205/60R15 tire with applied pressure and
boundary conditions.
32 Mir Hamid Reza Ghoreishy
tread pattern into consideration. In fact, like 2D meshes, the tread patterns
are modeled using simple ribs.
Figure 20. Finite element mesh of Figure 21. Finite element mesh of
the repeating unit (symmetric part the repeating unit (tread pattern)
without tread pattern) [45]. [45].
axis only for a small angle which can be the rotational angle of a repeating
unit of the tire tread blocks. The geometry of the repeating section was
then created by a 3D modeling software and meshed using 3D brick
elements. Either a uniform or variable pitch for the geometry of the tread
blocks can be assumed. Figures 20 and 21 show the finite element meshes
of the symmetric part (tire without tread) and tread blocks for a 175/70R14
steel-belted radial tire, respectively. These two meshes are then assembled
and a unique mesh is created as shown in Figure 22.
Figure 22. Finite element mesh of the repeating unit (symmetric part + tread pattern)
[45].
Figure 23. Finite element mesh of the full 3D model with tread blocks.
Computer Simulations of Passenger Car Radial Tires … 35
4.1. Two-Dimensional
𝐿 𝑑𝐿 𝐿
𝜖𝑡 = ∫𝐿 = 𝐿𝑛 𝐿 (36)
0 𝐿 0
Figure 24. Distribution of the belt-edge out-of-plane true shear strain (LE13) for belt
angle = 18o.
Figure 25. Distribution of the belt-edge out-of-plane true shear strain (LE13) for belt
angle = 20o.
Figure 26. Distribution of the belt-edge out-of-plane true shear strain (LE13) for belt
angle = 22o.
38 Mir Hamid Reza Ghoreishy
As it can be seen in these figures, decreasing the belt angle will also
decrease the maximum shear strain which is due to the decrease in tire
shape growth. This may have direct influence on the tire durability. To
confirm this, the stored strain energy in tires have also be computed which
are 111.6, 114.12 and 117.7 Joule for tires with belt angles of 18o, 20o and
22o, respectively. A positive correlation is found between increasing the
out-of-plane shear strain in belt-edge and stored strain energy in tires.
However, it should be noted that varying the belt angle would also results
in changing of the other variables especially in contact area which must be
considered separately.
Figure 27. Load-deflection data for a 185/65R15 tire. Experimentally measured vs.
FEM prediction.
Figure 28. Footprint area of the tire (185/65R14) predicted at 70% of the rated load
(L = 3840N).
Figure 29. Contact pressure of the tire (185/65R14) predicted at 70% of the rated load
(L = 3840N).
Computer Simulations of Passenger Car Radial Tires … 41
Figure 30. Predicted load-deflection curve for a 175/70R14 tire with and w/o tread
pattern detail.
Figure 31. Predicted contact pressure distribution for a 175/70R14 tire w/o tread
pattern detail.
42 Mir Hamid Reza Ghoreishy
Figure 32. Predicted contact pressure distribution for a 175/70R14 tire with tread
pattern detail.
Figure 33. Direction of applied loads in radial, longitudinal, lateral and torsional modes
on tire [18].
Once the footprint analysis has been completed, the tire can further be
analyzed and studied by applying longitudinal, lateral and torsional loads
to the contact surface. Figure 33 shows a typical radial tire presenting the
directions of the above-mentioned loads. Having analyzed the tire under
these loads, the corresponding force and deformation in tangential and
lateral directions and also moment and angular rotation for torsional mode
are determined. The tangential, lateral and torsional stiffness of the tire are
then calculated as the initial gradient or slope of the load and deflection
(moment and angular rotation for torsional stiffness) [18]. These
Computer Simulations of Passenger Car Radial Tires … 43
parameters are primarily utilized for the evaluation of the ride and handling
of tires under low speed conditions especially in urban applications.
Longitudinal stiffness is mainly used for the study of the enveloping
property of the tire encountering the road disturbances. The lateral stiffness
of a non-rolling tire is used in conjunction with cornering stiffness for the
evaluation of the relaxation property of a rolling tire and finally torsional
stiffness is useful for the assessment of steering property of the tire at
parking position and nearly zero speed condition [51].
The predicted stiffness for a 185/65R15 radial tire using above
described methods are given in Table 2 and compared with experimental
data. As can be seen, the finite element analysis were carried out using
three different hyperelastic models to compare the ability of the chosen
constitutive equations on the accuracy of the results.
selected for the description of the rubber is the dominant part for the
mechanical behavior of the cord-rubber composite in shear mode.
Therefore, the use of Ogden, Yeoh and Marlow (or any other hyperelastic
equations) determines the accuracy of the predicted stiffness in shear
modes [18].
𝑉
𝜔𝑚𝑖𝑛 = 𝑅 (37)
𝑚𝑎𝑥
𝑉
𝜔𝑚𝑎𝑥 = (38)
𝑅𝑚𝑖𝑛
where 𝑅𝑚𝑖𝑛 and 𝑅𝑚𝑎𝑥 are the minimum and maximum radii of the loaded
tire located at the center of the contact zone and the opposite side,
respectively. Figure 34 shows the reaction force parallel to the ground on
tire axle (applied at the center of the tire) at different spinning velocities
for a 185/65R15 tire. The ground velocity is assumed to be 80 km/h and
the spinning velocities are changed from 70 to 80 rad/s. As it can be seen,
the free rolling for this tire (RF = 0) occurs at spinning velocity of 73.637
rad/s. Partial braking occurs at spinning velocities smaller than this value
where the contact points between tire and road are slipping in backward
direction.
Figure 34. Longitudinal force vs. spinning velocity at constant ground velocity.
On the other hand, for those spinning velocities which are greater than
free rolling speed (i.e., >73.637 rad/s), partial traction occurs which means
that the mentioned contact points are slipping in forward direction. For a
46 Mir Hamid Reza Ghoreishy
𝑉
𝑅𝑑 = (39)
𝜔𝑓
106
𝑅𝑘 = 2𝜋𝑅 (40)
𝑑
𝑉𝑥 = 𝑉𝑐𝑜𝑠(𝛼) (41)
Computer Simulations of Passenger Car Radial Tires … 47
𝑉𝑦 = 𝑉𝑠𝑖𝑛(𝛼) (42)
where 𝛼 is the slip angle and 𝑉𝑥 and 𝑉𝑦 are the components of velocity
vector in 𝑥 and 𝑦 directions, respectively. The 𝑥, 𝑦 and 𝑧 coordinates refer
to longitudinal (traveling), lateral and vertical directions, respectively.
Figure 36. Lateral force vs. slip angle at different vertical loads.
However, reducing the vertical load especially below the rated load
makes that the nonlinearity to be occurred at lower slip angles. To further
study, the cornering stiffness (CS) is defined as the absolute value of the
slope of the lateral force curve at the zero slip angle. The variation of the
cornering stiffness with vertical load is shown in Figure 37. As seen, it
Computer Simulations of Passenger Car Radial Tires … 49
initially increases with vertical load and then tends to reach a maximum
level and finally slumps to lower values with further increase of load. It
was previously shown that [52] for a better handling it is desirable to shift
the peak value in CS vs. load curve to higher values of the vertical load.
Figure 39. Predicted camber force vs. camber angle for a 205/60R15 tire.
Figure 40. Longitudinal force vs. spinning velocity for the steady rolling of the tire
with and without viscoelasticity [13].
𝑘
𝑞̇ = 𝐻 𝑃𝐴𝛾̇ (43)
where 𝑞̇ is the volumetric rate of material loss due to wear of the surface, 𝑘
is a constant known as wear coefficient, 𝐻 is the hardness or elastic
52 Mir Hamid Reza Ghoreishy
𝑘
Figure 41. Element set of the tire tread before and after wearing ( = 1𝑥10−14 ).
𝐻
Computer Simulations of Passenger Car Radial Tires … 53
Figure 50. Variations of the longitudinal force vs. time for tire moving over curb.
Computer Simulations of Passenger Car Radial Tires … 57
CONCLUSION
REFERENCES
[17] Dorfmann, A.; Ogden, R. W. Int. J. Solids Struct. 2004, 41, 1855-
1878.
[18] Ghoreishy, M. H. R. Tire Technology International Annual Review
2014, 52-56.
[19] Rafei, M.; Ghoreishy, M. H. R.; Naderi, G. Math. Comput. Simulat.
2017 (in press).
[20] Ghoreishy, M. H. R.; Naderi, G.; Rafei, M. Iran. J. Polym. Sci.
Technol. (Persian) 2016, 29, 139-155.
[21] Ghoreishy, M. H. R. In Computational Materials; Oster, W. U.; Ed.;
Nova Science Publisher: New York, 2009; pp 445-478.
[22] Ghoreishy, M. H. R. Iran Polym. J. 2016, 25, 89-109.
[23] Ghoreishy, M. H. R.; Naderi, G. Iran Polym. J. 2005, 14, 735-743.
[24] Abaqus Analysis User's Guide. 2016, Simulia Inc. p. Abaqus
Documentation.
[25] Ghoreishy, M. H. R. P. I. Mech. Eng. D-J Aut. 2006, 220, 713-721.
[26] Ghoreishy, M. H. R. Plast. Rubber Compos. 2006, 35, 83-90.
[27] Ghoreishy, M. H. R.; Firouzbakht, M.; Naderi, G. Mater. Des. 2014,
53, 457-465.
[28] Ghoreishy, M. H. R. Mater. Des. 2012, 35, 791-797.
[29] Muhr, A. Rubber Chem. Technol. 2005, 78, 391-425.
[30] Ghoreishy, M. H. R.; Naderi, G.; Roohandeh, B. Iran Polym. J. 2015,
24, 1015-1024.
[31] Gent, A. N. In Engineering with Rubber; Gent A. N.; Ed.; Third
Edition; Carl Hanser Verlag: Munich, 2011; pp 37-88.
[32] Yeoh, O. H., In Engineering with Rubber; Gent A. N.; Ed.; Third
Edition; Carl Hanser Verlag: Munich, 2011; pp 345-364.
[33] Yeoh, O. H. Rubber Chem. Technol., 1993, 66, 754-771.
[34] Yeoh, O. H. Rubber Chem. Technol., 1990, 63, 792-805.
[35] Arruda, E. M.; Boyce, M. C. J. Mech. Phys. Solids 1993, 41, 389-
412.
[36] Ogden, R. W. Proc. R. Soc. Lond. A. 1972, 326, 565-584.
[37] Seibert, D. J.; Schöche, N. Rubber Chem. Technol. 2000, 73, 366-
384.
60 Mir Hamid Reza Ghoreishy
Chapter 2
ABSTRACT
This work was supported by the Russian Science Foundation, project no. 17-19-01038.
†
Corresponding Author Email: gaugolnickiy@sfedu.ru.
64 Guennady A. Ougolnitsky and Anatoly B. Usov
1. INTRODUCTION
within the game-theoretic approach (Basar and Olsder, 1999). The control
system includes one agent at the upper control level and several agents at
the middle and lower levels. The relations between the agents have
hierarchical organization. The upper-level agent exerts an influence on the
middle-level agents while the latter, in turn, have an impact on the lower-
level agents. The lower-level agents exert a direct influence on the state of
a controlled dynamical system. As a rule, the upper-level agent has no
direct impact on the lower-level agents and on the state of the controlled
system.
Such control systems are studied using the “principal - supervisor -
agent - object” hierarchical chain in different modifications within the
framework of optimal control theory and the theory of differential games.
At the beginning, the principal chooses his strategy (moves first) and
reports it to the supervisor and the agents. He maximizes his payoff
functional taking into consideration their best responses on the set of
strategies that support some conditions of sustainable development
(Ougolnitsky, 2011, 2014, 2015). The supervisor chooses her strategy
when the principal’s choice is already known but the agents’ strategies are
not fixed. She seeks to maximize her payoff functional with no regard to
the conditions of sustainable development. Any agent chooses his strategy
when the strategies at the upper levels are fixed, seeking to maximize his
payoff functional only.
In the special case of an indifferent principal (with a constant payoff
functional), we consider a simplified two-level hierarchical control system
containing one supervisor and several agents.
Suppose there are n agents. Denote by vi (t ), ui (t ) the controls of the
supervisor and agents, respectively (i 1,2,..., n) ; v(t ) (v1(t ), v2 (t )..., vn (t )) ,
u(t ) (u1(t ), u2 (t ),..., un (t )) are the control variable vectors of the players.
The payoff functionals have the following form:
e T S0 ( x(T )) max ;
T
J 0 (v(), u ())
0
e t G0 (v(t ), u (t ), x(t ))dt (1)
e T Si ( x(T )) max .
J i (v(), u ())
0
e t Gi (v(t ), ui (t ), x(t ))dt (2)
dx
= f ( x(t ), u (t )), x( 0 ) = x0 . (4)
dt
(v, u)(k ) (v1(k ) , v2(k ) ,..., vn(k ) , u1(k ) , u2(k ) ,..., un(k ) ); vi(k ) ViQRS; ui(k ) UiQRS ; i 1,2,.., n.
(a) for any two elements of the set (v, u)(i ) , (v, u)( j ) QRS ,
| J 0(i ) J 0( j ) | / J 0max ;
(b) for any other element (v, u )(l ) QRS , there exists an element
U1QRS ... U nQRS QRS gives a QRS-set of the game, which includes
n
m | QRS | U iQRS K n elements. In this case, each representative
i 1
UiQRS ui(1) , ui(2) ,..., ui( K ) ; | U iQRS | K ; i 1, 2, .., n .
n
For a given supervisor’s scenario, there are m | QRS | U iQRS K n
i 1
qualitatively different scenarios in the game.
We propose the following algorithm to design the set of Nash
equilibria in pure strategies on the QRS -set (denoted by NE (QRS ) ).
All input functions and parameters of model (2)–(4) are given. Set
i 1.
For each agent i , construct the set of his individually non-improvable
strategies Nei by
Nei (v(0) ) (v1(0) , v2(0) ,..., vn(0) , u1 , u2 ,..., ui 1, ui(*) , ui 1,..., un ) :
n
Ui
QRS
K n 1 operations are required for checking of all QRSs of
i 1
j i
the game under a fixed strategy of player i . If the number of players is not
too large (for example, n 10 , K 3 or 4 ), then the calculations are
technically feasible, which gives a base for the proposed method. If there
exists a Nash equilibrium in pure strategies in model (2)-(4) but is not
yielded by the proposed algorithm (i.e., NE (QRS ) ), then it is
necessary to update the QRS-set of the game by adding new scenarios for
each player and return to Step 2 of the algorithm for i 1 . For a successful
choice of the QRS-set of the game, the Nash equilibrium will be calculated
by a finite number of iterations.
Now, let us study system (1)–(4) from the supervisor’s viewpoint. The
informational structure corresponds to a Stackelberg game, and the
supervisor uses impulsion (Ougolnitsky, 2011, 2014). In this case, the
supervisor’s controls are unknown and have to be found.
Again, for each agent and for the supervisor with respect to each agent,
let a QRS set contain K elements, i.e.,
U iQRS ui(1) , ui(2) ,..., ui( K ) ; ViQRS vi(1) , vi(2) ,..., vi( K ) ;
| UiQRS || ViQRS | K ; i 1, 2, .., n .
n n
Then there are m | QRS | ViQRS U iQRS K 2n different
i 1 i 1
scenarios (sets of control variables) of the game. The supervisor
n
has Vi
QRS
K n different strategies of the form
i 1
74 Guennady A. Ougolnitsky and Anatoly B. Usov
v( k ) (v1( k ) , v2( k ) ,..., vn( k ) ), vi( k ) Vi QRS , i 1, 2,..., n, k 1, 2,..., K N ; each
agent has U iQRS K strategies of the form ui(1) , ui(2) ,..., ui( K )
( i 1, 2, .., n ).
We propose the following algorithm to design the set of Stackelberg
equilibria on a QRS-set.
1. All input functions and parameters of model (1)–(4) are given. Set
k 1.
2. A current ( k th) supervisor’s strategy is fixed, i.e.,
v(k ) (v1(k ) , v2(k ) ,..., vn(k ) ); vi(k ) ViQRS; i 1,2,..., n .
3. For the given supervisor’s strategy, calculate the Nash equilibria
n
NE (QRS ) Nei (v( k ) ) by the complete enumeration according
i 1
to the algorithm described in section 2.1.
maximum of the supervisor’s payoff functional (1). Save the greater value
of (1) and the corresponding set of controls.
If there exist unexamined strategies of the supervisor, then go to the
next strategy (k : k 1 ) and return to Step 2.
n
When all feasible strategies of the supervisor ( K strategies) are
examined, the strategies of the supervisor and agents that maximize (1) or
form the set arg max max J 0 (v ( k ) , u ) , constitute the
v (k )
;vi( k ) ViQRS n Ne ( v ( k ) )
i
i 1
3. APPLICATIONS
dP P
dP(1 ) w(t ) P; P(0) p0 . (5)
dt R
76 Guennady A. Ougolnitsky and Anatoly B. Usov
J L (q(), w())
0
e t [ H ( P(t ) P 0 (t )) 2 C (q(t ), w(t ))]dt min ; (6)
J F (q(), w())
0
e t [a(1 s(t ))w(t ) 0.5bw2 (t )]P(t )dt max . (7)
0 q(t ) 1; 0 t T ; (8)
b1 , if 0 t t1 ,
b , if t1 t t 2 ,
b(t ) 2 (10)
...
bM , if t M 1 t T ,
where bi are constant controls of the supervisor or agents for t [ti 1, ti ) ;
ti i t; t T / M ; M is the number of time intervals with constant
controls. Thus, the control of each agent becomes a grid function, for
example, w(t ) wi iM1 , where wi const is the agent’s control for
78 Guennady A. Ougolnitsky and Anatoly B. Usov
t [ti 1, ti ) . Then, using (6), (7) and (10), we transform the payoff
functionals into the payoff functions as follows:
M tk
J L (wi iM1, qi iM1) ( e t H ( P(t ) P0 (t ))2dt
k 1 t k 1
1
C (qk , wk ) (e t k 1 e t k )) ; (11)
tk
M
J F (wi i 1 , qi i 1 )
(a(1 sk ) wk 0.5bwk2 ) e t P(t )dt .
M M
(12)
k 1 tk 1
0 qk 1; k 1,2,..., M ; (13)
0 wk qk ; k 1,2,..., M . (14)
solved and the grid functions w wi iM1 and q qi iM1 can be
calculated according to the Stackelberg informational structure.
Computer Simulations as a Solution Method … 79
a function of the supervisor’s control, i.e., wi* (qi ) i 1 .
M
M
2. Substitute the agent’s controls wi* (qi ) i 1 into (5), (11). Solve
problem (11), (13), (5) and denote the resulting solution by qi* i 1
M
grid functions w wi iM1 and q qi iM1 . Introduce the sets
1 2 K 1
QiQRS 0; ; ; ...; ;1 and
K K K
q 2q K 1
Wi QRS 0; i ; i ;...; qi ; qi , i 1, 2,..., M . (15)
K K K
80 Guennady A. Ougolnitsky and Anatoly B. Usov
scenarios, i.e., | QRSL || QRSF | ( K 1)M . The game involves two
players and hence the QRS-set of the game includes ( K 1)2 M elements,
or | QRS | ( K 1)2M .
In real-world systems, the numbers K and M are not large (not
exceeding 3 or 4), which allows a numerical implementation of the
approach.
We propose the following numerical algorithm of Stackelberg
equilibrium design in model (11)–(14), (5) on the QRS-set is proposed.
qi 0 , then WiQRS 0 ; if qi
1 1
Q1QRS Q2QRS 0; ; 1 . If , then
2 2
1 1 1
WiQRS 0; ; ; finally, if qi 1 , then WiQRS 0; ;1 ( i 1,2 ).
4 2 2
The subscript i denotes an instant of time when the player chooses a
corresponding strategy (for M 2 , i 1,2 ).
Considering all possible combinations of elements from the sets
Q1QRS and W1QRS and W2QRS and Q2QRS as well, we found 7 scenarios for
the first instant of time when the players chose their strategies, i.e., for
i 1,
Q1QRS W1QRS (q1 , w1 ) k k 1 (0,0); ( ,0); ( , ); ( , ); (1,0); (1, ); (1,1) ,
7 1 1 1 1 1 1
2 2 4 2 2 2
Q2QRS W2QRS (q2 , w2 ) k k 1 (0,0); ( ,0); ( , ); ( , ); (1,0); (1, ); (1,1) .
7 1 1 1 1 1 1
2 2 4 2 2 2
Table 3.1.1.
min | J 0 J 0 | / J 0
(k ) ( j) min
Strategies (q, w)( j ) J L( j ) (q0 , q1 , w0 , w1 )
1 k 49
(c.u.) k j
q0 q1 w0 w1
0 0 0 0 22.0 0.1
0.5 0.5 0.5 0.5 17.7 0.03
0.5 0.5 0.25 0.25 19.8 0.1
0.5 0.5 0 0 31 0.32
1 1 0 0 58 1
1 1 0.5 0.5 17.87 0.05
1 1 1 1 14 0.3
0 1 0 1 21.8 0.1
0 1 0 0.5 21.9 0.1
0 1 0 0 40.1 0
1 0 1 0 21.7 0.1
1 0 0 0
1 0 0.5 0 21.6 0.1
0.5 1 0 0 44.7 0
0.5 1 0.5 0.5 17.83 0.05
0.5 1 0.5 1 17.78 0.05
1 0.5 1 0.5 17.73 0.05
1 0.5 0.5 0.5 17.67 0.06
1 0.5 0 0.5 40.2 0.1
1 0.5 0.5 0 26.5 0.05
1 0.5 1 0 26.4 0.1
1 1 0.5 0 40.3 0.3
1 1 0 0.5 40 0.3
1 1 0.5 1 17.62 0.05
1 1 1 0.5 17.65 0.03
Computer Simulations as a Solution Method … 83
Here J 0min is the global minimum of the payoff functional (1) (payoff
function (11)), as the minimization problem was solved. The minimum was
searched in the control variables of the supervisor and agent
simultaneously (for (1) in the two functions q(t ) and w(t ) ; for (11), in the
two grid functions w wi iM1 and q qi iM1 ).
Table 3.1.2 illustrates part of the numerical calculations to check the
second condition from the definition of a QRS-set with precision for the
hierarchical game with the input data of Example 3.1.1 and M K 2 .
According to the results in Tables 3.1.1 and 3.1.2, the precision is
0.03 . Thus, the resulting QRS-set satisfies the definition in the case
M K 2 . Recall that the value K determines the number of QRSs for
each player at a fixed instant of time.
Table 3.1.2.
(q, w) ( j)
QRS (c. u.) (q, w) (k )
QRS (c.u.) / J 0min
q0( j ) ( j)
q1( j ) w0 w1( j )
(k )
q0( k ) q1 w0( k ) w1( k )
0.6 0.6 0.3 0.1 26 0.5 0.5 0.25 0 25.8 0.014
0.6 0.6 0.6 0.1 20.3 0.5 0.5 0.5 0 20.2 0.007
0.6 0.6 0.1 0.6 20.4 0.5 0.5 0 0.5 20.45 0.004
0.6 0.6 0 0 31.3 0.5 0.5 0 0 31 0.021
0.6 0.6 0.3 0.3 19.84 0.5 0.5 0.25 0.25 19.8 0.003
0.6 0.6 0.6 0.6 17.95 1 1 0.5 0.5 17.87 0.006
0.6 0.6 0.4 0.4 17.65 1 1 1 0.5 17.65 0
0.3 0.3 0.3 0.3 19.7 0.5 0.5 0.25 0.25 19.8 0.007
0.3 0.3 0.1 0.1 20.1 0.5 0.5 0.5 0 20.2 0.007
0.3 0.3 0.3 0.1 20.1 0.5 0.5 0.5 0 20.2 0.007
0.3 0.3 0.1 0.3 20.3 0.5 0.5 0 0.5 20.5 0.014
0.3 0.1 0.3 0.1 22.7 0.5 0 0.5 0 23 0.021
0.3 0.1 0.1 0.1 43.8 0.5 0 0 0 44 0.014
0.3 0.1 0 0 43.8 0.5 0 0 0 44 0.014
0.8 0.8 0.8 0.8 13.9 1 1 1 1 14 0.007
0.8 0.8 0.8 0.1 19.0 1 1 1 0 19.3 0.021
0.8 0.8 0.8 0.3 17.64 1 1 1 0.5 17.65 0.007
0.8 0.8 0.8 0.6 17.66 1 1 1 0.5 17.65 0.007
0.8 0.8 0.6 0.8 17.63 1 1 0.5 1 17.62 0.007
0.8 0.8 0.3 0.8 17.62 1 1 0.5 1 17.62 0
84 Guennady A. Ougolnitsky and Anatoly B. Usov
where J L(0) is the value of the payoff function (11) for K 32 . The value
shows the difference (in percentage terms) between the supervisor’s
payoffs in the Stackelberg equilibrium for K = 32 and for a current value
K.
Table 3.1.3.
K
J L* (c.u.) (%)
2 21.9 2.8
4 21.5 0.9
8 21.5 0.9
16 21.4 0.46
32 21.3 0
Table 3.1.4.
K
J L* (c.u.) (%)
2 788 0,63
4 785 0.25
8 785 0.25
16 783 0
32 783 0
Table 3.1.5.
K
J L* (c.u.) (%)
2 3.65 2.2
4 3.59 0,56
8 3.59 0.56
16 3.57 0
32 3.57 0
Example 3.1.5. For the input data of Example 3.1.1 and s(t ) 0.1 , the
results of calculations are given in Table 3.1.6.
Table 3.1.6.
K
J L* (c.u.) (%)
2 21.9 5,2
4 21.0 1,0
8 20.8 0
16 20.8 0
32 20.8 0
Example 3.1.6. For the input data of Example 3.1.1 and R 1 , the
results of calculations are given in Table 3.1.7.
Table 3.1.7.
K
J L* (c.u.) (%)
2 181 312 0.55
4 180 815 0.27
8 180 567 0.14
16 180 315 0
32 180 315 0
86 Guennady A. Ougolnitsky and Anatoly B. Usov
Example 3.1.7. For the input data of Example 3.1.1 and d 5 , the
results of calculations are given in Table 3.1.8.
Table 3.1.8.
K
J L* (c.u.) (%)
2 22 49.6
4 17.4 21
8 17 15.6
16 16.9 15.5
32 14.7 0
T
J i e t [ gi (ri ui (t )) si (t )c( x(t ))]dt e T si (T )c( x(T )) max, (16)
0
Computer Simulations as a Solution Method … 87
0 ui (t ) ri , i B, P, C. (17)
(19)
J i (u ()) e t [ki (ri ui (t )) pi si (t )cx (t )]dt e T si (T )cx (T ) max,
0
0 ui (t ) ri , i N , (20)
M T
J i (uik kM1) (ki (rik uik ) pi (e tk 1 e tk ) / e t si (t )cx(t )dt
k 1 0
, u , u
functions uВk
Ne M
k 1
Ne M
Рk k 1
Ne M
Сk k 1
such that none of the players
Ne
benefits by a unilateral deviation from the strategy uik M
k 1 (i N ) .
We will calculate Nash equilibria in game (21)–(23) using the QRS
method. Introduce a QRS-set of player i at the instant of time t kt ,
assuming that it contains L+1 elements:
QRS
r 2r ( L 1)rik
U ik 0; ik ; ik ; ... ; ; rik ; i B, P, C; k 1,2,..., M .
L L L
The game involves three players and hence the QRS-set of the game
contains ( K 1)3M scenarios, i.e., QRS QRSB QRSP QRSS ; | QRS | ( K 1)3M .
In real-world systems, the numbers K and M are not large (not
exceeding 3 or 4), which allows a numerical implementation of the
approach.
We propose the following numerical algorithm to design the set of
Nash equilibria on the QRS-setin game (21)–(23).
90 Guennady A. Ougolnitsky and Anatoly B. Usov
1. All input functions and parameters of the model are given. Set
j 1.
2. Fix scenario j of the game, i.e.,
, u , u
(uВ( j ) , uР( j ) , uС( j ) ) uВk
( j) M
k 1
( j) M
Рk k 1
( j) M
Сk k 1
.
r 3r 5r
QikQRS 0; ik ; ik ; ik ;...; rik ; i B, P, C; k 1,2,..., M .
L L L
rC rC
uPNE (rP , rP ), uCNE ( , ) on the QRS-set for L 4 . That is, the
2 2
strategy of the third player (C) was reduced by half in comparison with
Examples 3.2.2 and 3.2.3.
Example 3.2.5. For the input data of Example 3.2.4, let k В 1.3 c.u.
rВ rВ
In this case, the algorithm yielded the Nash equilibrium uBNE ( , ),
2 2
rC rC
uPNE (rP , rP ), uCNE ( , ) on the QRS-set for L 4 . That is, the
2 2
strategy of the first player (B) was reduced by half.
Example 3.2.6. For the input data of Example 3.2.5, let k Р 0.9 c.u.
rВ rВ
In this case, the algorithm yielded the Nash equilibrium uBNE ( , ),
2 2
rР rР r r
uPNE ( , ), uCNE ( C , C ) on the QRS-set for L 8 . That is, the
2 2 2 2
strategy of the second player (P) was reduced by half.
According to the results of numerical calculations for a wide class of
the input functions and parameters, it is possible to find a Nash equilibrium
in pure strategies on small QRS-sets that contain 2–4 elements. Further
extension of this set doesn't affect the results.
Now, consider the following hierarchical setup of the social
partnership control problem for the continuous education system. Suppose
the players from the set N {B, P, C} (agents) with the payoff functionals
(16) and constraints (17) are located at the lower level while the upper
level is represented by a governmental body (supervisor) that maximizes
the total payoff functional (social welfare):
T
J 0 ( s(), u ()) e t [ gi (ri ui (t )) si (t )c( x(t ))]dt e T si (T )c( x(T )) max.
(24)
iN 0
Computer Simulations as a Solution Method … 93
The supervisor’s control variables are the shares of agents in the social
utility determined by the students’ professional development. Her control
constraints have the form
As all the strategies are constant on the same time intervals (10), the
and her payoff functional (24) can be written as the payoff function
M T
J 0 (uik kM1 , sik kM1 , ) (k i (rik uik ) pi (e tk 1 e tk ) / e t si (t )cx (t )dt
iN iN iN k 1 0
iN
e T cx (T ) max . (26)
sik (t ) 0,
iN
sik (t ) 1; k 1,2,..., M ; i N . (27)
r 2r ( L 1)rik
QRS
U ik 0; ik ; ik ; ...; ; rik ; i B, P, C; k 1,2,..., M .
L L L
i 1 i 1 i 1
1 sik 2(1 sik ) ( L 1)(1 sik )
i 1
j 1 j 1 j 1
S ikQRS 0; ; ; ... ; ; (1 sik ) ; i B, P, C; k 1,2,..., M .
L L L j 1
QRSi U iQRS
1 ... U iM
QRS
; | QRSi | ( K 1) M i B, P, C,
while the supervisor has not greater than ( K 1)3M scenarios of the form
QRS 0 S ВQRS
1 ... S ВM S Р1 ... S РM S С1 ... S СM
QRS QRS QRS QRS QRS
;
| QRS 0 | ( K 1) 3M
.
Computer Simulations as a Solution Method … 95
r
U ikQRS 0; ik ; rik ;
2
i 1
1 sik
j 1 i 1
S ikQRS 0; ; (1 sik ) ; i B, P, C; k 1,2,..., M .
2 j 1
Then there are nine QRSs for each agent ( i B, P, C ) of the form
QRSi U iQRS
1 U iQRS
2
(ui1 , ui 2 ) 6k 1
ri 2 ri1 ri1 ri 2 ri1 ri 2
(0,0); (0, ); (0, ri 2 ); ( ,0); ( , ); ( , ri 2 ); (ri1 ,0); (ri1 , ); (ri1 , ri 2 ) .
2 2 2 2 2 2
For the supervisor, there are six QRSs at each instant of time ( k 1,2 )
as follows:
96 Guennady A. Ougolnitsky and Anatoly B. Usov
S kQRS S Вk
QRS
S Рk
QRS
SСk
QRS
(s Вk , s Рk , sСk ) 6j 1
1 1 1 1 1 1
(0,0,1); (0,1,0); (1,0,0); (0, , ); ( ,0, ); ( , ,0) .
2 2 2 2 2 2
For all 26244 scenarios of the game, we numerically checked the first
condition in the definition of a QRS-set. These calculations took about 320
seconds on a PC with Pentium (R) Dual Core E5400 CPU. First, we found
the global maximum J 0max of the payoff functional (24) (payoff function
(26)). The maximum J 0max was searched in the control variables of the
supervisor and all three agents (for (24), in the six functions
s В (t ), s Р (t ), sC (t ) and u В (t ), u Р (t ), uC (t ) ; for (26), in the six
uВ uВk k 1 , uР uРk k 1 , uС uСk k 1
M M M
grid functions and
1 1 41 860 0.023
( , ,0,0,0,1,50,50,15,50,25;30)
2 2
1 1 1 1 42 076 0.025
( , ,0,0, , ,50,50,15,0,25;30)
2 2 2 2
1 1 26 282 0.003
( , ,0,0,0,1,100,50,30,100,50;30)
2 2
1 1 42 203 0.016
( , ,0,0,0,1,50,50,30,50,25;0)
2 2
1 1 37 588 0.04
( , ,0,0,0,1,100,0,30,50,50;15)
2 2
Table 3.2.1. (Continued)
1 1 42 268 0.009
( ,0, ,0,0,1,50,50,30,50,0;0)
2 2
1 1 37 713 0.011
( ,0, ,0,0,1,100,50,15,50,25;30)
2 2
1 1 34 681 0.005
( ,0, ,0,0,1,50,50,30,50,50;30)
2 2
1 1 37 433 0.012
( ,0, ,0,0,1,0,50,30,0,25;30)
2 2
1 1 44 435 0
( ,0, ,0,0,1,50,0,30,50,25;0)
2 2
1 1 1 1 34 906 0.004
( ,0, , ,0, ,50,50,30,0,50;30)
2 2 2 2
1 1 1 1 42 017 0.016
( ,0, , ,0, ,50,0,0,50,50;30)
2 2 2 2
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , J 0( j ) (( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , min | J 0(k ) J 0( j ) |
(c.u.) 1 k 26244
Strategies k j
u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) ) / J 0max
1 1 1 1 30 526 0.05
( ,0, , ,0, ,100,50,30,50,50;30)
2 2 2 2
1 1 39 456 0.007
( ,0, ,0,0,1,50,50,30,50,50;15)
2 2
1 1 1 1 41 923 0.03
( ,0, , ,0, ,50,50,15,50,25;30)
2 2 2 2
1 1 1 1 39 515 0.008
( ,0, , ,0, ,50,50,15,50,50;30)
2 2 2 2
1 1 41 803 0.024
( ,0, ,0,0,1,50,50,15,50,0;30)
2 2
1 1 47 030 0.012
( ,0, ,0,0,1,50,50,15,50,0;0)
2 2
1 1 44 177 0.0087
( ,0, ,0,0,1,0,0,15,50,25;30)
2 2
1 1 42 397 0.0078
( ,0, ,0,0,1,50,50,0,0,25;30)
2 2
1 1 47 297 0.009
( ,0, ,0,0,1,0,50,15,0,25;0)
2 2
Table 3.2.2.
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( j ) (c.u.) ( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( k ) (c.u.) | J 0(k ) J 0( j ) |
Strategies “Close” strategy
sС 2 , u В1 , u Р1 , uC1 , u В 2 , sС 2 , u В1 , u Р1 , uC1 , u В 2 ,
/ J 0min
u Р 2 , uC 2 ) ( j)
QRS u Р 2 , uC 2 ) (k )
QRS
(0.2,0.2,0.6,0.2,0.2,0.6, 49 238 (0,0,1,0,0,1,50,0,15,50,0;0) 49 234 0.00008
0,20,0,40,20,20)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 453 (0,0,1,0,0,1,50,25,15,0,0,15) 49 466 0.0002
10,10,10,10,10,10)
(0.2,0.2,0.6,0.2,0.2,0.6, 48 807 (0,0,1,0,0,1,0,30,15,0,30,15) 48 810 0.00006
20,20,20,20,20,20)
(0.2,0.2,0.6,0.2,0.2,0.6, 39 531 (0,0,1,0,0,1,50,25,30,50, 39 532 0.00002
30,30,30,30,30,30) 25,30)
(0.3,0.3,0.4,0.2,0.2,0.6, 49 453 (0,0,1,0,0,1,0,0,15,0,0,15) 49 402 0.00008
10,10,10,10,10,10)
(0.2,0.2,0.6,0.3,0.3,0.4, 48 807 (0,0,1,0,0,1,0,30,15,0,30,15) 48 810 0.00006
20,20,20,20,20,20)
(0.3,0.3,0.4,0.3,0.3,0.4, 39 531 (0,0,1,0,0,1,50,25,30,50, 39 532 0.00002
30,30,30,30,30,30) 25,30)
(0.4,0.4,0.2,0.2,0.2,0.6, 49 762 (0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0
0,0,0,0,0,0)
Computer Simulations as a Solution Method … 103
A direct analysis of Tables 3.2.1 and 3.2.2 implies that the precision is
0.0012 . Thus, the resulting set of scenarios can be taken as a QRS-set
for the input data of Example 3.2.1 in the case M K 2 .
CONCLUSION
REFERENCES
Reviewed by:
Professor Dmitry A. Novikov
Trapeznikov Institute of Control Sciences
Russian Academy of Sciences, Russia
In: Computer Simulations ISBN: 978-1-53613-095-9
Editors: Michael D. Pfeffer et al. © 2018 Nova Science Publishers, Inc.
Chapter 3
ABSTRACT
Corresponding Author Email: abedin@squ.edu.om.
108 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
1. INTRODUCTION
The Talbot effect was first discovered by the English photographer and
inventor Henry Talbot in 1836. He observed that the light diffracted behind
a diffraction grating has a regular and repetitive pattern in the axial
direction [1]. A periodicity is also observed in the lateral directions due to
the periodicity of the grating. The regular distance in the axial direction is
called the Talbot distance. After its discovery, the Talbot effect was
forgotten for nearly half a century, until it was rediscovered by Lord
Rayleigh. Lord Rayleigh [2] explained this repetition of images within the
paraxial approximation in 1881, and showed that the Talbot distance was
given by D2/, where D is the slit spacing of the grating and is the
wavelength of the light used.
The Talbot effect has been extensively studied both theoretically and
experimentally in the intervening period. The theoretical investigations are
often concerned with the origin of the Talbot effect, and the associated
wave and coherence phenomena. The relevant theoretical research has
been done within the framework of Fresnel diffraction, Fresnel images,
Fourier images, and the theory of image formation [3-6]. In addition,
rigorous theoretical analyses were done using, e.g., the finite-difference
The Computer Simulation of the Talbot Effect … 109
time domain (FDTD) method [7], electromagnetic theory [8], and the
Rayleigh-Sommerfeld diffraction formula [9], for the effects of aberrations
on the self-imaging phenomena. Other authors have pointed out the
resemblance of the Talbot effect to fractals [10].
On the experimental side, many applications have been proposed for
the Talbot effect. It has been proposed, for example, for imaging [11],
sensing a distance and a displacement [12], measuring the refractive index
of a medium [13], array illumination [14], lithography [15], sub-
wavelength focusing of light [16], and laser beam combination [17].
The Talbot effect is an experimental observation in the first place. One
needs some elaborate equipment, such as light sources or lasers, diffraction
grating, translation stages, a suitable camera and other associated
equipment to observe the effect. To generate the repetitive but beautiful
pattern called the Talbot carpet is also a nontrivial task, since the image is
composed of many discrete images captured at different known axial
distances from the grating.
Previously, we used the Iterative Fresnel Integrals Method (IFIM) to
simulate the diffraction effects from rectangular apertures in the near field,
and later, from N-slits or gratings in the near-field (Fresnel) regime [18-
19]. We have used the powerful, general purpose software MATLAB,
running on an ordinary PC to perform these simulations. The method was
found to be powerful, accurate and general, being able to generate many
features of the diffraction, namely, the missing order, principal and
secondary maxima, Fresnel-Fraunhofer transitions, etc. Therefore, our idea
was to use this IFIM method to try to simulate the Talbot effect and the
Talbot carpet, entirely by computer. We were successful in our efforts.
In this Chapter, we describe how the IFIM method can be used to
simulate the Talbot effect from a periodic grating structure. In addition, we
were able to synthesize the complete Talbot carpet with a certain
resolution. The only equipment needed for this, is, of course, a PC (or a
laptop) and the well-known software MATLAB. A photo-processing
software is also required to synthesize the Talbot carpet.
The Chapter is divided into the following sections: First, we discuss
the theoretical basis of the method as it applies to the generation of near-
110 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
field images and the Talbot effect. In the next section, methods of
generating the Talbot carpet are discussed. That is followed by a section, in
which we present examples of Talbot carpet generated by the method.
Conclusions are drawn in the final section.
2. THEORETICAL BACKGROUND
where Eu is the unobstructed electric field at P (i.e., the electric field that
would be observed if the aperture were removed). Here, C(u) and S(u) are
known as the Fresnel cosine and sine integrals, being defined by,
w w
C ( w) cos( w' 2 / 2)dw' and S ( w) sin ( w' 2 / 2)dw'.
0 0 (2)
112 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
u y 2( p0 q0 ) / p0 q0 , z 2( p0 q0 ) / p0 q0 . (3)
The limits u1 and u1’ are the values of the dimensionless variable u
corresponding to the two edges of aperture 1, i.e., for y1 = a/2+b and
y1’ = 3a/2+b, respectively. Similarly, the limits v and v’ are the values of
the dimensionless variable v corresponding to the lower and the upper
edges of this aperture, i.e., for z = -c and z’ = +c, respectively.
The electric field contributed by the aperture n to the right is,
where un and un’ are the values of u corresponding to the two vertical edges
of aperture n, i.e., for yn= (2n-1)a/2+nb and yn’ = (2n+1)a/2+nb,
respectively. The values v and v’ are the same as in equation (1).
The electric field contributed by the central aperture 0 is,
where u0 and u0’ are the values of u corresponding to the two vertical edges
of the central aperture, i.e., for y0= -a/2 and y0’ = a/2, respectively. The
values v and v’ are the same as in equation (1) or equation (4). The electric
field contribution by the -1 aperture to the left is similarly obtained from,
where u-1 and u-1’ are the values of u corresponding to the two edges of
the -1 aperture, i.e., for y-1’ = -a/2-b and y-1= -3a/2-b.The electric field
contribution by the last aperture -n is given by
The Computer Simulation of the Talbot Effect … 113
E Pn E2u C (u) jS (u) uunn ' C (v) jS (v) vv ' (7)
where un and un’ are the values of u corresponding to the two vertical edges
of aperture -n, i.e., for y-n’= -(2n-1)a/2-nb and y-n= -(2n+1)a/2-nb,
respectively. The values v and v’ are as before.
The total complex electric field at P contributed by the (N = 2n+1)
apertures is simply the sum of all the complex amplitudes EPn, EP1, EP0,
EP-1,.EP-n,
E N E Pn .. E P1 E P 0 E P1 .... E P n
{C (u ) jS (u ) un ' ... C (u ) jS (u ) uu1' C (u ) jS (u ) u0 ' C (u ) jS (u ) uu1' ... C (u ) jS (u ) u n ' }
Eu
u u u
2 n 1 0 1 n
(8)
Separating the cosine and the sine integrals, and using the summation
notation for the Fresnel sines and the cosines, we derive the total
complex electric field at P as,
The summation on i from –n to +n for both the Fresnel cosine and sine
integrals (for the u variable only), in effect, carries out the summation of
the complex electric field contributions from aperture –n to aperture n, i.e.,
for a total of N = (2n+1) apertures. No such summation is needed for the v
variable, since only the two edges (upper and lower) of the apertures are
involved in this z-direction for all the (2n+1) apertures.
The net intensity at P is proportional to the square of the net electric
field, i.e.:
I P 40 ( EN E N ),
I *
(10)
Where I0 denotes the intensity of the unobstructed wave, i.e., I0= Eu2.
114 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
From equation (9), it is clear that the calculation of the electric field or
intensity at a point P requires, in general, the calculation of 2N Fresnel
cosine and 2N Fresnel sine integrals, corresponding to the 2N edges of the
N aperture system for the u(y) variable. In addition, two pairs of Fresnel
cosine and sine integrals are required for the v(z) variable. The cosine
integrals form the real parts of the electric field, and the sine integrals form
the imaginary parts. After calculating the complex electric field, the
intensity at P is calculated from equation (10) by multiplying the field by
its complex conjugate. All these mathematical operations can be efficiently
performed by the MATLAB software. These equations are the basis of
calculation of the complete intensity distribution of the Fresnel diffraction
pattern from an N-aperture system, as will be explained in Sec. IIIA.
Equation (9) was used to calculate first the complex electric field, and then
the intensity was calculated by taking the square of it. Equation (9) has two
factors, the first factor involves only the u (or y) coordinate, and expresses
the dependence of the electric field or intensity in the y direction. The
second factor involves only the v (or z) coordinate, and it expresses the
dependence of the electric field or intensity in the z direction.
In the rest of this Chapter, for the sake of simplicity, we assume that
the source S is located at infinity. This can be achieved experimentally by
placing a convex lens between S and the aperture, and placing the source S
at the back focus of the lens. In this case, Eq. (3) simplifies to,
u y 2 / q0 , z 2 / q0 . (11)
(11)
The Computer Simulation of the Talbot Effect … 115
Equations (9) and (10) describe the electric field and the intensity at
P (the center of the observation screen), respectively. In order to apply
these for an off-axis arbitrary point P', we use the following technique
described in Ref. [18] and [19]: The observation screen and the SOP line
were held fixed. Then instead of moving P, the entire aperture in the yz
plane was moved in the opposite direction, so that the relative position of
116 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
the aperture in this new position and point P remains unaltered. We then
calculate the electric field at P instead of at P'. The point P' will observe a
new set of values for the y’s and z’s (and hence for the u’s and v’s). For
example, to find the intensity at point P', 1mm to the right of P, the screen
was kept undisturbed and the aperture system was moved 1mm leftwards,
and the intensity at P in this configuration is calculated (instead of at P').
As a result, the point P will now observe a new set of values for y’ (and
therefore, for the values of u) in equations (8) and (9). The electric field
(and the corresponding intensity) at any arbitrary point P' on the image
plane (screen) can be found in this way by making appropriate virtual
movements of the aperture in the y and z directions, and in equation (8) or
equation (9), using correspondingly a new set of values for u and v.
The flow chart of the algorithm is given in Figure 2. We can realize
that for the calculation of the electric field distribution using equation (9) at
all the points (or, pixels) on the image plane, a large number of the Fresnel
cosine and sine integrals will need to be evaluated quickly, and this is
efficiently done by the special functions mfun (‘FresnelC', i:s:f) and mfun
(‘FresnelS', i:s:f)in MATLAB.
As shown in Figure 3 for N= 3 (three apertures again for simplicity),
the whole aperture system was displaced (virtually) by a distance W, and
the observed positions of the 2N edges of the displaced aperture was
determined and compared to the corresponding positions for the aperture
before the displacement, to find the required range of the u and v values.
These positions are marked in Figure 3. In a more general N-aperture
system, when the aperture is moved by W to the left, the ranges for the 2N
vertical edges of the N= (2n+1) apertures will be given by:
Figure 3. Apparent limits of the virtually displaced aperture seen from the observation
plane for N = 3. (Reprinted from Abedin etal. Optik Vol.126. pp. 3743-3751,
Copyright (2015) with permission from Elsevier).
inverting the D matrix twice (lines 27-35). Here, we use the inversion
symmetry of the image around the u and v-axes. At the end of the program,
the generated image is displayed as a grayscale (black and white) image by
the images c statement with the appropriate scale axes (line 37).
We used the program to first simulate near field images for a particular
diffraction grating (with a= 0.04mm, b= 0.16mm, c= 8mm and N= 15). In
other words, the grating linewidth is D= (a+b) = 0.20 mm, and the grating
constant is 50 lines/cm. The wavelength of the light is λ = 500nm, the
image area is W= 3 mm, and the resolution of the simulation is
s= 0.01 mm. The image at an aperture-screen distance, q0= 10 mm, is
shown in the Figure 4(a). It shows the closely-spaced, characteristic pattern
of Fresnel diffraction images in the near field.
To examine the near-field images at regular intervals, and to see if they
are repeated at intervals of the Talbot lengths (self-imaging), we simulated
Fresnel images at q0= 10mm, 20mm, 80mm, 90mm, 100mm, 160mm. The
grating constants, the wavelength and all the other calculation parameters
were left unchanged. The simulated images are all shown in Figure 4.
Remembering that the Talbot distance in our case is:
Figure 4. Computer simulation for the Talbot effect for a 15 slit system with the
simulation parameters: a = 0.04 mm, b = 0.16 mm (D = 0.2 mm), c = 8 mm, w = 3 mm,
s = 0.01 mm and λ = 500 nm. q0 is increased step by step in the near field region.
Self-repeating of images at intervals of 80 mm is clearly observed.
A horizontal strip across the center of each image is cut using the
software Photoshop, and the strip is laid out on a scaled map, the scale
running from 1mm to 160 mm. The scale represents the axial distance q0 of
the image behind the grating. The process is repeated for each of the
simulated images. The result is a mosaic of images, consisting of about 160
horizontal strips, each strip being placed at the appropriate distance, where
they would be observed in reality. The complete image is shown in
Figure 5.
122 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
4. DISCUSSION
Figure 6. Synthesized Talbot carpet at an axial resolution of 0.1 mm, from q0= 1 mm to
70 mm. Other parameters of the simulations were not changed.
matter waves associated with trapped atoms [24]. For light waves in the
visible region, the highest resolution carpet that can be produced is limited
by the spatial or pixel resolution of the imaging camera. But in our case,
we reproduce it completely on computer, where all the essential features
can be simulated without any experimental equipment. All the parameters,
such as the wavelength of light, the slit width, and the separation can be
arbitrary changed in the simulations, and the corresponding Talbot carpet
can be generated. The ability to generate a Talbot carpet at any wavelength
is particularly noteworthy, since it enables one to generate the Talbot
carpet for any wavelength of the illuminating light.
It takes a huge amount of computational effort and a considerable
amount of time to generate a full Talbot carpet. For example, using a
Pentium core-i7 PC with 8 MB of RAM, it took approximately 4 hours to
synthesize a Talbot carpet at 1 mm resolution, and more than 12 hrs to
synthesize it at 0.1 mm resolution to a length of 70 cm. To generate it at
higher resolutions will take such a long time as to render the process
impractical.
It may be possible to automate the whole process by devising suitable
software that can perform all the steps, i.e., generating all the simulations
and performing the synthesis of the carpet automatically. Then the whole
program should be able to automatically generate the carpet, once the
necessary input parameters, such as, the wavelength of light, the slit width,
the separation, and the axial resolution are specified. In principle, there is
no upper limit to the resolution that can be achieved; this is limited only by
the computational resources available. Viewing and printing of a such high
resolution carpet in its entirely will require a large, high-resolution
computer screen, and a correspondingly large-format printer.
CONCLUSION
features of the Talbot effect and the Talbot carpet can be readily observed
in the simulated images. Our method does not require any experimental
apparatus, and moreover, it can be extended to any arbitrary axial
resolution. It also shows that the Iterative Fresnel Integrals Method is a
powerful technique to simulate the various aspects of near-field diffraction.
The only limitation of our method is the large amount of computational
effort and time involved, especially, if the Talbot carpet is to be
synthesized at a higher resolution. But even at these high resolutions, an
ordinary PC or laptop is sufficient to generate the desired Talbot carpet.
APPENDIX
REFERENCES
Chapter 4
Zeguang Pei
College of Mechanical Engineering,
Donghua University, Shanghai, China
ABSTRACT
Corresponding Author E-mail: zgpei@dhu.edu.cn.
130 Zeguang Pei
surface and to analyze the relation between the scattering pattern and the
surface twist angle. The solution of the scattering intensity distribution
for the yarn surface profile is derived and numerical results indicate that
the highest light intensity on the backward light scattering pattern lies
along the direction perpendicular to the surface fibers regardless of the
fiber or yarn parameters. The result of this chapter can serve as the
theoretical basis for the measurement of yarn surface twist angle based on
backward light scattering by the yarn.
1. INTRODUCTION
tan
T (1)
πD
where T is the yarn twist in turns per unit length, β is the surface twist
angle, and D is the yarn diameter. In order to control and evaluate the
quality of a yarn, measurement of the yarn twist is a critical and
indispensable process. Using Eq. (1) to determine the yarn twist requires
the measurement of the surface twist angle β and yarn diameter D. For the
past decades, various methods and devices have been proposed for the
measurement of the twist angle of a yarn. In addition to mechanical
methods [2], optical methods have also played an important role in these
measurement methods. Plöckl [3] and Hensel et al., [4] developed
Modeling of the Light Scattering by a Yarn 131
the solution of the scattering intensity distribution for the yarn surface
profile is derived. The feasibility of determining yarn twist angle based on
the backward light scattering by the yarn surface is determined by
analyzing the relation between the scattering pattern and the surface twist
angle. Numerical results are finally confirmed by comparing it with the
experimental results.
surface which has a rough and periodic texture. Here, the yarn is assumed
to be of a cylindrical shape and lies in the horizontal direction. At a point
on the yarn surface, the yarn surface is approximated by a tangent plane S
to the cylinder. The normal of S at this point inclines an angle α relative to
the Z axis pointing vertically upward, as shown in Figure 2. A local
Cartesian coordinate system, x0y0z0 which obeys the right-hand rule, is set
on the yarn with y0 parallel to the yarn axis, z0 superposed with the normal
of S, and x0y0 lie in S. The scattering model is applied to this tangent plane
S as an incident wave lying in the x0z0 plane and propagating
perpendicularly to both the y0 axis and Z axis strikes the yarn surface with
π
an angle of incidence θ1 α . A screen is placed at a distance DS from
2
the yarn to receive the scattered waves. A screen coordinate system XSYSZS
is set on the screen. Let k1 and k2 denote the propagation vectors of the
incident wave and scattered wave, respectively, as shown in Figure 3, we
have:
2π
where k is the incident wave number, λ is the incident wavelength,
λ
θ2 and θ3 are the scattering angles, x0, y0, z0 are unit vectors for the local
Cartesian coordinate system x0y0z0.
Let P be the point of observation in the Fraunhofer zone of diffraction,
i.e., on the screen. The scattered field E2 at point P can be expressed by the
Helmholtz-Kirchhoff integral:
1 E
E2 P E n n dS (4)
4 S
134 Zeguang Pei
e
i k2 R0 k 2 r
with , where R0 is the distance from point P to the origin of
R0
the coordinate system x0y0z0, r is the vector pointing from the origin to the
considered point on the yarn surface and is expressed as:
r xx0 yy 0 x, y z 0 (5)
Figure 2. Side view of the schematic diagram of the light scattering by a yarn.
E
E S 1 R E1 , 1 R E1k 1 n (6)
n S
Modeling of the Light Scattering by a Yarn 135
ik 1 r
where E1 e is the incident plane wave of unit amplitude, R is the
Fresnel reflection coefficient, n is the normal to the surface at the
considered point.
Figure 3. Perspective view of the schematic diagram of the light scattering by the yarn
surface.
For a surface extending from x0 = -X0 to X0, and y0 = -Y0 to Y0 with its
area A = 4X0Y0 ≫ λ2 and a constant R, a scattering coefficient representing
the normalized scattered field E2 can be introduced as:
F3
θ1 ,θ 2 ,θ3 e
ivr
dxdy (7)
A A
where
λ
sin θ 2 cosθ3 sin θ1 m m 0, 1, 2,... (10)
x
λ
sin θ 2 sin θ3 n n 0, 1, 2,... (11)
y
π π
in which 0 θ1 , 0 θ2 .
2 2
The surface profile function x, y of the yarn is determined by the
alignment characteristic of the fibers on the yarn surface and the shape of
the fiber cross section. Circular cross sections are adopted by some man-
made fibers such as polyester and nylon, while some fibers used in the
textile industry may also have non-circular cross sections, such as cotton
and ramie fibers which are both natural fibers. Here for the simplification
of modeling, the yarn is supposed to have a simple helix structure and the
fibers in the yarn have circular cross sections. For calculation convenience,
the origin of the local coordinate system x0y0z0 is set at a point on the axis
of a surface fiber. Another local coordinate system x1y1z1 is also applied to
the tangential plane S. This local coordinate system shares the same origin
Modeling of the Light Scattering by a Yarn 137
with x0y0z0 with x1 along the surface fiber axis and z1 superposed with z0, as
shown in Figure 4. According to geometrical relation,
y0 rC cos (12)
x0 rC sin (13)
y1 rC cos rC sin rC sin cos cos sin
2
y0 sin x0 cos
(14)
(15)
namely,
F3 X0 Y2
i vx x v y y vz
2X y
X 0 Y1
e dxdy
F3
exp ik sin θ1 sin θ 2 cos θ3 x sin θ 2 sin θ3 y (17)
X0 Y2
2X y
X 0 Y1
2 x cos d 2 x cos d
where Y1 , Y2 .
2sin 2sin
138 Zeguang Pei
Figure 4. Schematic diagram of the geometrical relations for expressing yarn surface
profile: (a) x0y0 plane; (b) y1z1 plane (A-A section).
1 tan r sin r
2 2
R
2
1 2
R R
2
2 tan 2 r sin 2 r
(18)
sin
nf (19)
sin r
π
cos sin σ-atan cotαcos sin α cos σ-atan cotαcos cos α cos , 0,
2
(20)
where σ 0, π is the angle between the normal to the fiber surface and
the y0 axis.
The average value of R is calculated by taking its integral over the
illuminated part of the fiber surface:
1 π
π 0
R Rd (21)
D
yS sin DS sec tan θ3 DS sec tan θ3 (22)
2
D D
zS sin DS cot δ2 cos DS cot δ2 (23)
2 2
where δ2 is determined by
D π
yS DS sin csc δ2 cos δ2 tan θ3 DS csc δ2 sin δ2 tan θ3
2 2
(26)
D D
zS DS sin cot δ2 cos DS cot δ2 (27)
2 2
π
tan δ1 tanθ3 cos δ2 tanθ3 sin δ2 (28)
2
3. NUMERICAL IMPLEMENTATION
R (Calculated by
0.3509 0.3348 0.3401 0.3529 0.3398
Eq.(21))
Figure 6. Modeled light scattering process of an incident ray by a yarn (α = 30°): (a)
Modes of m <0; (b) Modes of m >0. Note that 1 mode in every 8 modes is shown in
(a).
Figure 8. Calculated scattering patterns of (a) case 2 (b) case 3 (c) case 4 (d) case 5.
Modeling of the Light Scattering by a Yarn 145
Figure 9. The backward scattering image of a Ne 32 cotton yarn with Z twist captured
by the CCD camera.
Figure 10. Comparison of twist angles measured based on the backward light
scattering and microscopic images of various yarn samples.
146 Zeguang Pei
CONCLUSION
ACKNOWLEDGMENTS
REFERENCES
Chapter 5
ABSTRACT
*
Corresponding Author Email: ahmed.ayadi.gem@gmail.com.
150 Ahmed Ayadi, Zied Driss, Haythem Nasraoui et al.
NOMENCLATURE
1. INTRODUCTION
The solar chimney power plants (SCPP) are devices which are
designed to generate electrical energy using the solar radiation. Generally,
The Unsteady Analysis of a Solar Chimney Power Plant 151
2. MATERIAL
3. NUMERICAL METHOD
The CFD code, Ansys Fluent 17.0, is used to simulate the air flow
within the solar chimney power plant. The air flow is governed by the
continuity and the motion equations which are presented in the cylindrical
coordinate. The continuity equation is written as follows:
r
V
z (2)
1 p 1 r 2
1 2 r
2
g
r r r r 2
2 2
z
2
r r (4)
The Unsteady Analysis of a Solar Chimney Power Plant 155
3.3. Meshing
simulations since this meshing gives a good agreement compared with the
experimental data.
From figure 6, it has been noted a difference in values of the static pressure
from one case to another. Over the considered instances, the values of the
static pressure vary. For example, at t = 10:00a.m., t = 12:00a.m., t =
02:00p.m. and t = 04:00p.m., the maximum values of the static pressure are
equal respectively to p = 101326.41 Pa, p = 101325.64 Pa, p = 101325.78
Pa and p = 101325.45 Pa.
CONCLUSION
inside the SCPP. The generated power has an obvious hourly and daily
variation. In fact, the local characteristics vary with the variation of the
environmental conditions. In the future, we are going to propose a new
model of a solar chimney power plant with a heat storage system.
REFERENCES
Chapter 6
Abstract
One of the technological challenges for hydrogen materials science is
the currently active search for structural materials with important appli-
cations (including nuclear power industry and the ITER project) that will
have predetermined limits of hydrogen permeability. One of the experi-
mental methods is thermal desorption spectrometry (TDS). A hydrogen-
saturated sample is degassed under vacuum and monotone heating. The
desorption flux is measured by mass spectrometer to determine the char-
acter of interactions of hydrogen isotopes with the solid.
The chapter presents a distributed boundary value problem of ther-
mal desorption with nonlinear dynamical boundary conditions. The prob-
lem is reduced to the nonlinear functional differential equation of neu-
tral type for surface concentration. A numerical method is developed for
∗
Corresponding Author Email: zaika@krc.karelia.ru
168 Yury V. Zaika and Ekaterina K. Kostikova
1. Introduction
The interaction of hydrogen isotopes with solids has many applications [1–11].
It is sufficient to mention outlooks for the energy industry, protection of metals
from hydrogen corrosion, design of chemical reactors and rocket engineering.
Some problems of hydrogen materials science related to this paper were inves-
tigated in [12–20]. This work is a continuation of [21–24]. Practice has shown
that the limitations are not only diffusion processes but also physical-chemical
effects on the surface [1, 2]. Transfer parameters depend on the process charac-
teristics of producing the material batch, and one needs effective algorithms for
modeling instead of focusing on “tabular data”. We shall focus on the thermal
desorption method (TDS) and take into account only the limiting factors and the
information capacities of TDS experiments. Papers [16–18, 24] were dedicated
to interpretation of TDS peaks. Analysis of the causes of various TDS surges
is crucial for the selection of structural materials contacting with hydrogen iso-
topes. A sufficiently detailed review is presented in [17, 18]. Where modeling
does not take dynamic surface processes into account, TDS peaks are inevitably
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 169
where Qt∗ = (0, t∗ ) × (0, `), c(t, x) is the diffusing H concentration. The
diffusion coefficient D is a function of the temperature T = T (t) given by the
Arrhenius equation: D(T ) = D0 exp −ED /[RT ] . We can take into account
different channels of diffusion, but the information content of the TDS experi-
ment is limited. Therefore, the coefficient D in Equation (1) is assumed to be
an integral effective index.
Seeking a write-up of the TDS peaks set it is handier to use the following
model
m
∂c ∂ 2c X h − i
aν 1 − Zν c(t, x) − a+
=D 2 − ν zν (t, x) ,
∂t ∂x ν=1
∂zν +
= a−
ν (T ) 1 − Zν c(t, x) − aν (T )zν (t, x), (2)
∂t
where zν (t, x) is the concentration of hydrogen atoms captured by defects of
different types; a∓ν are the coefficients of H capture and release by traps; Zν ≡
ν ν
zν (t, x)/zmax is the defects saturation degree zmax = max zν . Capture is
taken into account at its integral level for practical purposes. A more precise
definition of the defects’ geometry and distribution would add complexity to
the model. If, for instance, the defect is not a micro-cavity but hydride phase
inclusions, then at the degassing stage the corresponding coefficient a− j (T ) is
+
identically zero and aj (T ) value is positive only if the critical temperature is
reached:T (t) ≥ Tcrit.
It is easy to simulate the required number of TDS-peaks using different
binding energies (coefficients Ea). Algorithms based on difference schemes
and modeling results were presented in [21], [23]. In this paper we use only
Equation (1) (D = Deff ). For thin membranes of quickly permeable material
used in experiments this approximation is usually sufficient.
Here, c0 (t) ≡ c(t, 0), c` (t) ≡ c(t, `) (identity sign is used in terms of equal-
ity by definition); q0,` (t) are the surface concentrations (x = 0, `, q̇ ≡ dq/dt);
g(T ) is the parameter of local equilibrium between concentrations on the surface
and in the subsurface bulk (coefficient of quick dissolution); µ is the gas-kinetic
coefficient; p0,` (t) are the gaseous (H2 ) pressures; b(T ) is the desorption coef-
ficient. The well-posedness of these boundary conditions was proved in [26].
Hydrogen occurs in the model both as molecules and in monoatomic form.
For consistency, the count is based on atoms: [c] = 1/cm3 , [q] = 1/cm2 ,
[Dcx] = [bq 2 ] = 1/(cm2 s). The factor µp determines the number of parti-
cles (H2 molecules) encountered per unit time and area. If the coefficient s
is changed to 2s, then the value of s denotes the share of adsorbed H atoms. √
Let us fix [p] = Torr. Then µ(T ) = (2πmkT)−1/2 ≈ 2.474 × 1022 / T ,
[µ] = 1H2 /(Torr cm2 s), [T ] = K (temperature numerical value under the rad-
ical), k is the Boltzmann constant, m is the hydrogen √ molecular mass. The
kinetic constant’s dependence on temperature ( µ ∝ 1/ T ) is usually neglected
on the scale of the exponential s(T ).
surface is isotropic (in terms of Ek− ≈ Ek+ ), then the parameter g(T ) is little
dependent on temperature. The density of the H atoms adsorption flux can be
modeled by the term µsp[1 − θ]2 for balance Equations (5), (6). For the ranges
of weak concentrations and sufficiently high working temperatures the degree
of surface coverage satisfies θ(t) = q(t)/qmax 1. These simplifications are
in agreement with the limited information capacities of the TDS experiments.
Experimental data are more easily approximated for a large number of param-
eters. But the uniqueness of the estimations is then failed. And thus essential
errors may occur at extrapolation of the results “from thin plate to wall”. Such
extrapolation is supposed by default.
The information capacities of the initial and final stages of the TDS experiment
are low. So it is sufficient to limit ourselves to t ≤ t∗ , when the flux from the
sample has decreased by an order of magnitude compared to maximum. The
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 173
experimental data are the desorption density curves J(t) or TDS spectra J(T )
(T (t) ↔ t) for different saturation conditions and heating rates. The conversion
p(·) 7→ J(·) is made more specific by taking into account the parameters of the
experimental apparatus. Modern equipment provides a means of creating deep
vacuum (10−9 − 10−7 Torr). For this reason, the component P ≡ µsp (P 1)
is the key control for the saturation stage, but resorption is neglected for the
degassing stage (P 1).
Note that only one saturation method is presented. Cryogenic technology
and the electrochemical method can also be used. There are good reasons to
bring the saturation process into agreement with the operating conditions of the
structural material. The corresponding technical adjustments are nonessential.
γθ1
k1 θ1 [1 − θ2 ] = k2 θ2 [1 − θ1 ] ⇒ θ2 = .
1 + [γ − 1]θ1
3. For definiteness, we take into account only one type of traps (m = 1).
We use Equation (2). For symmetry, we add the saturation factor [1 − θ2 ] for
174 Yury V. Zaika and Ekaterina K. Kostikova
√
The curves Θ = f ( p) have the “growing wave” form. It is noticeable
in a wide pressure range only. For the given parameters, the inflection point
is the most vivid on the curves marked with pentagrams. The analysis of each
additive component in the formula (9) for the total concentration shows that the
position of the inflection point is determined by the moment when the first and
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 175
third addends have turned to the saturation mode with the following prevalent
rise of θ2 . Note that the curves for a narrower pressure range are practically
linear, in line with ranges of the Sieverts’ law. In a wide range of pressures, the
wave-like character of the graphs is observed experimentally [12].
A warning has to be made. When estimating the hydrogen permeability of
materials in experimental practices, Richardson approximation is often used for
the penetration flux density:
p p
J(t) = Φ pin (t) − pout(t) `−1 , pout < pin .
(10)
Membranes are usually very thin and their permeability is relatively high. We
can accept for the concentration gradient cx = [cout − cin ]/`. In accordance
with the Fick’s law, the formula J = −D[cout − cin ]/` is “exact”. Here, the
function J(t) is the monoatomic hydrogen permeation flux density. The ac-
tual boundary volume concentrations cannot however be measured, and thus
a quasi-equilibrium approximation is used, where the concentration c is sub-
√
stituted with the equilibrium concentration c̄ = Γ p in accordance with the
Sieverts’ law (Γ is the solubility coefficient, or solubility for short). On account
of the inequalities c̄in > cin , c̄out < cout, this substitution overrates the second
factor in the formula J = D[cin − cout]/`: c̄in − c̄out > cin − cout. Hence, the
value of D needs to be formally marked down to retain the equality. So, if the
value of Φ (permeability) is found from experimental data by fitting as indicated
in formula (10), then we obtain Φ < DΓ. The Φ and Γ values from the (typi-
cally used) formula Φ = DΓ determine the lower bound of the diffusion coeffi-
cient D. Furthermore, dissolved diffusing atomic hydrogen is mainly involved
in the steady-state (quasi-steady-state) permeability mode. In the context of the
√ p
model (4)–(6), we obtain c̄ = Γ p, Γ = g µs/b. The “saturation-degassing”
experiment yields the total concentration value c̃ > c̄ and an overstated (for the
√
permeability problem) value Γmax : c̃ = Γmax p.
Hence, we can estimate the Φ value (permeability) by fitting using for-
mula (10). This information has practical value as a convenient coefficient for
translation from pressures to flux. If, however, the D value is taken from one ex-
periment (or paper) and the Γ value is taken from another source, then, strictly
speaking, we get a ranking of three different numbers Φ < DΓ < DΓmax .
If the material has a high level of trapping by defects, then the calculated per-
meability can be an order of magnitude higher than the real permeability. The
176 Yury V. Zaika and Ekaterina K. Kostikova
Let us write the solution of the linear boundary value problem using the source
function (Green’s function):
Z tZ `
ĉ(t, x) = G1 (x, ξ, t − τ )f (τ ) dξ dτ,
0 0
∞ n n2 π 2 o
2X nπx nπξ
G1 (x, ξ, t) ≡ exp − 2 t sin sin .
` ` ` `
n=1
4g t n n2 π 2
Z X0 o X0 X
ĉx|0 = − q̇(τ ) exp (τ − t) dτ , ≡ .
` 0 `2 n=1,3,5...
4g X0 t n n2 π 2 Z t
Z o
cx(t, 0) = − q̇(τ ) exp − 2 D(s) ds dτ .
` 0 ` τ
integration by parts lest a divergent series arises. We are concerned with the
interval [t1 , t2 ], which corresponds to the TDS peak. Measurement for t ≈ 0, t∗
yields little information. There is a voluminous body of literature on Riccati-
type equations (including matrix equations for the optimal control theory).
In the meantime, this a priori restriction (arising from the assumption that
stationary “balls” are ordered geometrically on a plane) is highly questionable.
For a dynamics model, it appears that the concentration threshold qmax may be
higher, if it is specified what meaning is attached to the term “bubbling” surface
layer (at a quite high temperature). The real q̄ value is somewhat conventional,
since it is strongly influenced by the experiment pre-treatment (building of vac-
uum before the start of heating). However, most of the hydrogen is in the bulk,
the diffusion equation (and the diffusion process itself) has a smoothing effect,
and we are interested in evident thermal desorption peaks, since the initial and
final time intervals of the experiment offer little information. For initial calcula-
tions it is therefore sufficient to correctly estimate the magnitude of the “effec-
tive” q̄ concentration. High precision requirements are non-critical here (neither
can they be realized in practice).
Let us remark that only one of the possible methods of reaching initial uni-
form saturation was discussed here. It is possible to use alternative methods.
Total hydrogen content in the sample is determined after the experiment (using
the registered desorption flux dynamic). The initial (uniform) concentration and
the corresponding q̄ value are determined in the model (using S, ` and current
g estimate). After all, data on dissolution (c̄ = c̄(p̄,T̄ )) can be used because the
material is multi-method tested.
takes finite values for s > 0. The series is rapidly converging for large s.
Formally, substituting s = 0 (the integration variable τ reaches the upper limit t)
we obtain a divergent series. This can be “fixed” by term-by-term integration.
The order of terms in the series is O(n−2 ) (n = 2i − 1, i ≥ 1), wherefore
convergence is slow.
Let us formulate the problem. The equation has to be approximated by a
low order ODE system to enable application of standard software packages (for
180 Yury V. Zaika and Ekaterina K. Kostikova
example, the authors use Scilab freeware). Let us run some transformations
using the theory of Jacobi theta functions to explicitly extract the integrable
singularity. We consider
X∞
θ3 (t, x) = 1 + 2 exp{−n2 π 2 t} cos(2nπx), t > 0.
n=1
The series on the left is rapidly converging for large t. But the series on the right
is rapidly converging for small t. Let us run auxiliary transformations:
∞ n n2 o ∞
1 X X X0
√ exp − = 1+2 exp{−k2 π 2 4t} + 2
πt −∞ t
k=1
∞ n n2 o
X0 1 X
=2 +√ exp − .
4πt −∞ 4t
Proceeding from here, the last series is subtracted from the first series and the
result is doubled. The following expression is obtained for the values s > 0
q ≡ − exp{−1/(4s)} :
∞
X0 1−Q X 2
exp −n2 π 2 s = √ , qn .
Θ(s) ≡ 4 Q(s) ≡ −2
πs n=1
where the fraction (weak singularity under the integral) rapidly decreases from
infinity (τ = t) to near zero (τ = t − 1). For t > 1, the lower limit of
the integral can be replaced with t − 1. Thus, the neglectable background
can be easily
R t2 identified in the original physical time using the relation 1 =
0 0 2
t2 − t1 = t1 D(T (τ ))dτ /` . The compact functional differential (v̇ ≡ dv/dt)
TDS Equation (15) with initial data v(0) = 1 replaces the original nonlinear
boundary value problem (from section 3.1.) with dynamic boundary conditions
in the sense that formally only the dynamics of the surface concentration (des-
orption) is required for TDS spectrum construction.
Q(s)
Q(s)
s s
1-Q( )
d
0
N
n2
Q(s)= 2 q , q= exp{ 1/(4s)},
n=1
N > 10
s ~ 0.334 s
√
Figure 2. Functions Q(s) and Q(s)/ πs.
The greatest contribution to the integral is made by the value v̇(τ ), τ ≈ t due
to non-limited (but integrable) singularity. Thus, the quadratic approximation
is used due to function v̇ concavity. Let us consider the current segment of time
t ∈ [kh, (k + 1)h], k ≥ 0. The conditions
Z t
v̇(τ )|kh = v̇(kh), v̇(τ ) dτ = v(t) − v(kh)
kh
As the result we obtain an ODE system of third order instead of the functional
differential Equation (15):
Q(t − kh) √
κ̂ 2
v̇(t) = −b̂(t)v 2 (t) + √ + t − kh v̇(kh)
π 15 4
− 4κ̂ w1 (t) + w2 (t) , v(0) = 1, (16)
b̃
b̂ ≡ Q(t−kh) √
,
1+ √κπ 32
15 − 4 t − kh
κ
κ̂ ≡ Q(t−kh) √
.
1+ √κπ 32
15 − 4 t − kh
Ellipsis stands for the right-hand side of the first line of the first equation
(−b̂ . . . v̇(kh)). The series for Q(s) converges very rapidly. It is sufficient to
use 5–8 terms for software implementation.
The sought function v(t) = q(t)/q̄ on the current segment of dimensionless
time (in the system (16) t = t0 , t0 ∈ [kh, (k + 1)h], k ≥ 0) is computed by
the Runge-Kutta 4th order method for integration of ODE systems (the authors
used Scilab software). If the TDS spectrum has two peaks, then w3,4 should be
additionally used to improve the modeling accuracy. Returning to the physical
time, we get the model spectrum J(T ) = b(T )q 2 (t) (t0 ↔ t ↔ T (t)).
Qualitatively, thermal desorption spectra of metal structural materials have
a typical form. Fig. 3 illustrates numerically simulated TDS spectra for the
above-listed hydrogen permeability parameter values of the structural materials
at different heating rates. Even when defects (traps) are not taken into account,
the suggested model can yield different types of two-peak graphs (where the
low temperature peak is more pronounced or where the peaks are comparable).
For reference, the change of so-called transport parameter W = `bvol c̄/D =
`bc̄/(Dg 2) is shown in the top part of the graphs. This parameter is crucial
for the study of membrane hydrogen permeability [5]. We supposed that SLR
corresponds to W < 10−2 and DLR corresponds to W > 104 . For beryllium
and steel, the low temperature TDS peak takes place in the range where diffusion
and surface processes play commensurable roles. The high temperature TDS
peak occurs in DLR. For nickel, surface processes and diffusion have similar
effect throughout the experimental temperature range, and only a peak-like step
appears in the TDS spectrum at a low temperature.
Let us focus on TDS spectra for tungsten because these spectra are qualita-
tively different from the previous ones. The aforementioned parameter values
“for tungsten” are, of course, formal. Besides micro impurities, the parameters
depend on how the sample surface was treated. This is especially true because
in practice a plate is thin, the volume is small and the effect of surface processes
is more vivid. This is one of the reasons for such a high variation of the quanti-
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 185
1 2
TBe TBe TNi
T T
J(T) .10
-12
J(T) .10
-13
8.
6
9.
6
W W
DLR Tungsten 12Cr18Ni10Ti DLR
. .
T = 7, 5, 2.5 T = 2, 1, 0.5
Figure 3. TDS-spectra.
• The low order ODE system of type (16) is numerically integrated step
by step in dimensionless time (on an h length interval). Other ODE ap-
proximations were presented in [24]. This procedure is standard for every
modern mathematical software.
• The graph J(t) = b(T (t))q 2 (t) or spectrum J(T ) is plotted in physical
time using the function v(t0 ) = q(t0 )/q̄.
Beryllium Tungsten
_ _
/ /
T x/ T x/
Nickel 12Cr18Ni10Ti
_ _
/ /
T x/ T x/
(q/q) 2
(q/q) 2 JDiff = Tungsten
Beryllium
b/b0
b/b0 12Cr18Ni10Ti
Tungsten
2
TW1 TSt1 TSt2 TW2 1
TBe TBe TW2
T T
TW1
JDiff . 10 / c
4
(q/q)2 Beryllium
JDiff =
Nickel
12Cr18Ni10Ti
2
(q/q)
Nickel
b/b0
1 b/b0 2
TBe TNi TBe TSt1 TNi TSt2
T T
relates with diffusion in the bulk. The problem of estimating the desorption
coefficient (bvol = b/g 2) for the situation where accumulation on the surface
can be neglected (q̇ ≈ 0) was presented in [22].
Since the function D(t) ≡ D(T (t)) is known, it is reasonable to move
directly to dimensionless time t0 , which is oriented at the characteristic time of
the diffusion transfer `2 /D. The old notation t is retained not to complicate the
formulas. Let us present the nonlinear term in the system (16) in the form
`
b̂(t)v 2 (t) = α(κ) · · J(t), J(t) ≡ b(t)q 2 (t).
D(t)c̄
are usually noisy, but the function J(t) is comprised in the right-hand side of
the ODE, and is smoothed by integration.
Note that the right-hand sides of the equations now depend only on one es-
timated dissolution parameter κ = g`. The identification algorithm can thus
be decomposed. By setting the value of g we can eventually calculate the de-
pendence q(t) = q(t; g) in the original physical time t. Be reminded now that
Eb 103
J = bq 2 ⇒ b = Jq −2 ⇒ ln b0 − · = ln ψ, (17)
R · 103 T
where ψ(t) ≡ J/q 2 , T = T (t). Since heating is monotonic t ↔ T (t), we can
introduce the coordinates X ≡ 103 /T , Y ≡ ln ψ. The parametric curve X(t),
Y (t) is obtained on the plane (X, Y ). Judging by the ratio (17), this curve has
to be a straight line with a negative slope. Hence the criterion for the choice of
the “correct” value for g: the curve X(t), Y (t) = ln{J(t)/q 2 (t; g)} has to be
a straight line segment on the plane (X, Y ). This sub-problem is scalar, where
only g varies. Formally, we prolong the straight line segment until it crosses the
coordinate axes. The crossing with the Y axis (X = 0) yields the ln b0 value.
The crossing with the abscissa x̄ = 103 × R ln b0 /Eb determines the Eb value.
The algorithm is laborious due to iterative use of the procedure of numeri-
cal solving of the initial problem for the ODE system. It requires some effort
and familiarity with mathematical packages, but it is much easier and quicker to
use the standard built-in operation than to perform iterative solution of the orig-
inal nonlinear boundary value problem with three varying parameters (which
requires special software).
The presented linearization method is illustrated in Fig. 6. Model spectra
(for g = const) were plotted to test the algorithm. A system of five equations
(for the variables v(t), w1−4 (t)) was numerically solved to obtain more accurate
estimates for two-peaks spectra. In addition, a series of experiments was per-
formed for spectra with a random error not higher than 20% (we used standard
Scilab uniform random number generator). The identification algorithm based
on ODE system integration demonstrated the noise resistance of experimental
data treatment. The initial value of g yields the mass balance
Z t∗
−2
c̄` + 2q̄ = c̄[` + 2g ] = 2 J(τ ) dτ.
0
The experiments have demonstrated that the accuracy of such estimation de-
190 Yury V. Zaika and Ekaterina K. Kostikova
creases for g 1. Curves based on the “correct” numerical κ value (the re-
sultant curve is close to the straight line segment) and curves for 20% deviation
from κ are shown. The percentage error of the identification algorithm testing
is less than several percent. This error appears due to the numerical error of di-
rect and inverse problem solving. There is high sensitivity (appearing as vertical
“beak” singularity) to situations where the “true” κ value is exceeded. Math-
ematical singularity appears because, formally, the function v(t) changes sign,
taking also negative values. We did not take efforts to avoid this “non-physical”
transition because it is a vivid sign of a wrong κ value. The same numerical
experiments were done for noisy spectra. Activation energies are determined
more accurately because energy parameters more actively influence desorption
during heating. From the mathematical point of view, there is no need to strive
for visual agreement of simulated and experimental curves (because the experi-
mental error is tens of percents). ODE approximation is quite adequate.
103 103
x = 103R y Eb-1 Tungsten T y = ln b0 1.2 Beryllium T
y = lnb0 1.2 noisy data
0.8
0.8
103Ry ~
x= - 2.837
Eb
ln{J(T(t)) q-2(T(t),g)} ln{J(T(t)) q-2(T(t),g)}
103 103
12Cr18Ni10Ti T Nickel T
0.8 0.8
3
10 Ry ~
103Ry ~ x= - 6.122
x= - 2.128 Eb
Eb
ln{J(T(t)) q-2(T(t),g)} ln{J(T(t)) q-2(T(t),g)}
Conclusion
The problem of identification of the spectra of hydrogen thermal desorption
from structural materials is analyzed. This problem is of high relevance for the
nuclear power industry. Qualitatively, the identification consists in revealing the
causes of desorption peaks. It is usually assumed that TDS peaks appear due
to hydrogen release from traps with different binding energies. Here, it was
demonstrated using an elementary diffusion model (for a homogeneous mate-
rial) that if surface processes are taken into account, two-peak spectra can be
obtained even for very thin experimental samples. The tendency to resort to
the “theory of different traps” as the only explanation is understandable, but the
volume of our samples was near zero for the trapping capacity to manifest itself.
Another part of the problem is parametric identification. There are many
hydrogen permeability models, including various boundary conditions, where-
fore diffusion coefficient estimates vary even in orders of magnitude. A TDS
experiment has natural limitations in information capacity. For example, the
results have to be correctly (in a unique way) extrapolated for a higher material
thickness, considering that experimental samples are usually very thin because
the measurement time is limited. In this paper, the nonlinear boundary value
problem (standard diffusion equation with dynamical boundary conditions) is
reduced to the functional differential equation for the surface concentration,
because nothing but desorption dynamics is required to plot a TDS spectrum.
An effective algorithm oriented to the use of mathematical packages (includ-
ing freeware) is proposed. The main final output of the paper is a geometrically
transparent method for solving the inverse problem of surface parameters identi-
fication where desorption and diffusion in the bulk are dynamically interrelated.
Acknowledgments
This work was supported by the Russian Foundation for Basic Research
[Grant 15-01-00744].
References
[1] Hydrogen in metals; Alefeld, G.; Völkl, J.; Eds; Springer-Verlag: Berlin,
192 Yury V. Zaika and Ekaterina K. Kostikova
1978.
[3] The hydrogen economy: opportunities and challenges; Ball, M.; Wi-
etschel, M.; Eds; Cambridge University Press: New York, 2009.
[4] Handbook of hydrogen storage: new materials for future energy storage;
Hirscher, M.; Ed.; Wiley-VCH: Weinheim, 2010.
[7] Varin, R. A.; Czujko, T.; Wronski, Z. S. Nanomaterials for solid state
hydrogen storage; Springer: New York, 2009.
[8] Lototskyy, M. V.; Yartys, V. A.; Pollet, B. G.; Bowman Jr, R. C. Int. J.
Hydrog. Energy. 2014, 39, 5818–51.
[10] McRae, G. A.; Coleman, C. E.; Leitch, B. W. J. Nucl. Mater. 2010, 396,
130–3.
[11] Mieza, J. I.; Vigna, G. L.; Domizzi, G. J. Nucl. Mater. 2011, 411, 150–9.
[12] Yukawa, H.; Zhang, G. X.; Watanabe, N.; Morinaga, M.; Nambu, T.; Mat-
sumoto, Y. J. Alloy Compd. 2009, 476, 102–6.
[14] Zaika, Yu. V.; Rodchenkova, N. I. Appl. Math. Model. 2009, 33, 3776–91.
[15] Evard, E. A.; Gabis, I. E.; Yartys, V. A. Int. J. Hydrog. Energy. 2010, 35,
9060–9.
[16] Schmid, K.; Rieger, V.; Manhard, A. J. Nucl. Mater. 2012, 426, 247–53.
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 193
[17] Legrand, E.; Oudriss, A.; Savall, C.; Bouhattate, J.; Feaugas, X. Int. J.
Hydrog. Energy. 2015, 40, 2871–81.
[20] Zaika, Yu. V.; Bormatova, E. P. Int. J. Hydrog. Energy. 2011, 36, 1295–
305.
[21] Rodchenkova, N. I.; Zaika, Yu. V. Int. J. Hydrog. Energy. 2011, 36, 1239–
47.
[22] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2014, 39, 15819–
26.
[23] Zaika, Yu. V.; Kostikova, E. K. Adv. Mater. Sci. Appl. 2014, 3, 120–9.
[24] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2017, 42, 405–15.
[26] Zaika, Yu. V. Comp. Maths Math. Phys. 1996, 36, 1731–41.
[32] Oya, Y.; Inagaki, Y.; Suzuki, S.; et al. J. Nucl. Mater. 2009, 390–391,
622–5.
In: Computer Simulations ISBN: 978-1-53613-095-9
Editors: Michael D. Pfeffer et al. c 2018 Nova Science Publishers, Inc.
Chapter 7
Abstract
High-purity hydrogen is required for clean energy and a variety of
chemical technology processes. The various alloys potentially well-suited
for use in gas-separation plants for the membrane technology of high-
purity hydrogen production were investigated by measuring specific hy-
drogen permeability. For structural materials one had to estimate the
parameters of diffusion and sorption to numerically model the different
scenarios and experimental conditions of the material usage (including
extreme ones), and identify the limiting factors. The proposed mathemat-
ical model matches the experimental method of hydrogen permeability
and takes into account only the main limiting factors for the applied mem-
brane filtering problem and the informative capabilities of the experiment
modifications.
∗
Corresponding Author Email: zaika@krc.karelia.ru.
196 Yury V. Zaika and Natalia I. Rodchenkova
1. Introduction
Studies on the interaction of hydrogen isotopes with structural materials are
mainly necessitated by problems in the energy industry, metal protection from
hydrogen corrosion and the design of chemical reactors [1–10]. Different alloys,
which may be well-suited for use in gas-separation plants for the membrane
technology of high-purity hydrogen production, were investigated by measur-
ing specific hydrogen permeability. For structural materials one had to estimate
the parameters of diffusion and sorption to numerically model the different sce-
narios and experimental conditions of the material usage (including extreme
ones), and identify the limiting factors. Some particular problems of the hy-
drogen materials science related to the topic of this study were presented and
investigated in [11–15]. Experiments show that the limiting factors are diffu-
sion processes as well as physical and chemical phenomena at the surface [1,2].
Fast Hydrogen Permeability of Structural Materials 197
2. Aggregation of Experiments
Experimental practices employ various modifications of the penetration method
and thermal desorption spectrometry (TDS). The results of measurements de-
pend both on the unit design features and the procedure of preparing samples
for hydrogen permeability testing. A successive use of various methods often
causes impurities to appear on the sample surface, which significantly affects
the reproducibility of the results. These data are the input for the inverse prob-
lems of parametric identification, which are sensitive to the level of error. It is
therefore advisable to aggregate experiments to improve the accuracy and in-
formative value of the measurements. We suggest the following set-up of the
‘cascade’ experiment.
A membrane heated to a fixed temperature served as the partition in the
vacuum chamber. Degassing was performed in advance. A sufficiently high
pressure of hydrogen gas was built up in steps at the inlet side. The penetrat-
ing flux was determined by mass-spectrometry in the vacuum maintained at the
outlet side. This is a classical penetration method. Its advantage is a reliable
determination of the diffusion coefficient by the Daines - Berrer method (based
on so-called lag time). It allows distinguishing between the bulk and the surface
processes in the model, keeping in mind that surface parameters are significantly
more difficult to estimate. When the steady state level of the penetrating flux is
registered, we increase the inlet pressure and wait until a new steady state value
is established. Using (at least) three pressure jumps at the inlet side we record
the steady state flux values at the outlet side, thus determining ‘the degree of rec-
tilinearity’ of the isotherm. Then the pumping for vacuum building is stopped
and the experiment proceeds as the ‘communicating vessels’ method (this stage
is described in detail below). When pressure values become nearly equal (the
sample is almost uniformly saturated with hydrogen) it is possible to turn off
the heating, create the vacuum at both sides of the membrane and begin slowly
reheating the sample (TDS-experiment). In addition, there is no depressuriza-
tion of the diffusion cell and the sample surface remains uncontaminated with
additional impurities. We will clarify the details of the first two stages as we
describe the method of solving the inverse problem of parametric identification.
Fast Hydrogen Permeability of Structural Materials 199
∂c ∂ 2c
= D(T ) 2 − f (T , z, c), (1)
∂t ∂x
∂z h z(t, x) i
= f ≡ a(T ) 1 − c − aout(T )z, (2)
∂t zmax
area S is about 1 cm2 , the inflow pressure p0 (0) is within several hundreds Torr.
The pressure range [pmin , pmax ] is narrow, so we restrict ourselves to the relation
zmax = σc̄, σ 6 0.1. This restriction will not lead to a failure of the Sieverts’
√ √
law c̄ + zmax ∝ ps while c̄ + zmax ≈ c̄ = Γ ps is within the experimental
accuracy range.
It now remains to find the magnitudes of Qin , Qout. Within the time of hy-
drogen transfer through the membrane the gas is in the thermodynamical quasi-
equilibrium with the surface, wherefore we use the formula N = pV /(kT).
Here, N is the number of gas particles occupying the volume V at the temper-
ature T and the pressure p (in the SI system [p] = Pa, [V ] = m3 , [k] = J/K).
Taking into account the relationships Torr = 133.322 Pa and Pa = J/m3 (for-
mally), we get the following expressions for the corresponding pressures and
volumes in the dynamical boundary conditions (3), (4): Q = 2N = αpV /T ,
α ≈ 1.931 × 1019. Here, p, V ,T are the numerical values of the variables in the
selected units (Torr, cm3 , K).
Within the experimental unit the membrane is situated in the tube (which is
heated to a predetermined temperature) between the inlet and the outlet cham-
bers. The tube diameter is large enough to consider the equality of pressures as
the criterion of thermodynamic quasi equilibrium between the gas in the tube
and in the chambers. The membrane temperature should be taken for the for-
mula for the kinetic constant µ(T ). The gas inside the volumes Vin,out (whose
massive walls are at room temperature) may get heated up. During the prelim-
inary experiment it is recommended to fill the chambers with a practically im-
permeable metal membrane between them with ambient gas. Then heat the tube
and record the pressure rises. Within the framework of the ideal gas approxima-
tion (equation of state) this procedure enables estimation of the increments of
the gas temperature inside the chambers. The corresponding gas temperatures
are the ones to be used in the formula given for Q (and the subsequent ones, only
excluding the value µ). The need of such a refinement arises from the character-
istics of this particular experimental unit. Such an adjustment of the values of
T should not cause difficulties in further calculations. Besides, this procedure
has relatively little effect on the final calculation of the model pressures taking
into account the measurement errors and relatively large volumes V .
202 Yury V. Zaika and Natalia I. Rodchenkova
Let us formulate the problem of modelling the concentrations c0,` pusing the
pressures p0,` without the quasi-equilibrium simplification c(t) = Γ p(t).
The quasi-stationary state is achieved within a time t0 , which is short com-
pared to the total experiment time (the traps become saturated and ∂x c =
−[c0 (t) − c` (t)]/`). So, the original model (1)–(4) can be simplified (taking
into account the formula Q = αpV /T ):
−1
ṗ0,` (t) = ∓β0,` c0 (t) − c` (t) , β0,` ≡ SD αVin,out` T, (5)
2sµ p0,` (t)−bc20,`(t) = ±D`−1 c0 (t)−c` (t) , t ≥ t0 > 0. (6)
Therefore, we shall aim to derive differential equations for X0,` , since informa-
tion about the dynamics of the boundary concentrations c0,` is important.
Differentiate the equations (6) with respect to time and substitute the pres-
sure derivatives from (5). For the variables X0,` we get the ordinary differential
equations (ODE)
−1
X` − Vin Vout µST
Ẋ0 (t) = −2sM0 X0 − X` · , M0 ≡ , (8)
X0 X` − 1 αVin
X0 − Vin−1 Vout µST
Ẋ` (t) = 2sM` X0 − X` · , M` ≡ , t ≥ t0 . (9)
X0 X` − 1 αVout
Let us formulate step-by-step the algorithm of modelling the pressures
p0,` (t) (t ≥ t0 ) for the current values of D, b, s (the authors used the Scilab
freeware). We target at the ‘normal’ experimental conditions [16–20], includ-
ing the values of p, T , `, V , S.
1. We fix t = t0 : omit fast transient processes (the duration of the transient
processes is about tens of seconds on the hours-long experimental time scale).
For the variable X ≡ X` we choose the root of the biquadratic polynomial
[X 2 − a` (t0 )]2 − 4[2X + a0 (t0 )]. √
From physical considerations
p it follows that
c` (t0 ) > c̄` (t0 ) and thus X > 1 + 1 + a` = 1 + 2`Γ p` (t0 ) bD −1 .
2. The system of equations a0 +2X` = X02, a` +2X0 = X`2 (t = t0 ) yields
the missing value of X0 (t0 ). Formally, one equation is enough, but we take into
account averaging procedures including determination of the values of p0,` (t0 ).
3. We numerically integrate the ODE system (8), (9) with the obtained initial
data. The change of variables in (7) defines the concentrations c0,`(t), which are
used to calculate the model pressures p0,` (t) (t ≥ t0 ) from the equations (6).
Computational experiments show that the model curves p0,`(t) (t ≥ t0 )
almost coincide with those generated by the originally proposed model, i. e. the
nonlinear distributed initial boundary value problem.
Observe the difference from the quasi-equilibrium model (the Richard-
son approximation), where the only approximation parameter is the complex
Φ = DΓ. All the variable parameters of the original model that influence the
permeability, namely D, b, s, are important when running the above algorithm.
Thus, the fast hydrogen permeability model does not lose in informativeness
concerning transfer parameters.
204 Yury V. Zaika and Natalia I. Rodchenkova
Membranes are usually very thin and have relatively high hydrogen permeabil-
ity coefficients. So, for the concentration gradient one can take cx = [cout −
cin ]/`. Then, according to the Fick’s law, the formula J = −D[cout − cin ]/`
is ‘exact’. Here, J(t) is the density of the penetrating flux of monatomic hy-
drogen. Since near-to-surface concentrations cannot be measured directly, it
is necessary to use the quasi-equilibrium approximation, substituting the equi-
√
librium concentration c̄ = Γ p for c in accordance with the Sieverts’ law. By
virtue of the inequalities c̄in > cin , c̄out < cout such a substitution overstates the
second factor in the expression J = D[cin −cout ]/` since c̄in −c̄out > cin −cout .
To maintain the equality the value of D should be formally understated.
Thus, if we fit the experimental data {p(t), J(t)} to find the value of Φ
(permeability) in accordance with the formula (10), then Φ < DΓ. Taking
the numbers Φ and Γ from the formula Φ = DΓ (which is usually used), we
get the lower estimate of the diffusion coefficient D. Thereafter, the stationary
(quasi-stationary) permeability stage mainly involves dissolved atomic diffusive
hydrogen. The ‘saturation-degassing’ experiment yields the value of the total
concentration ĉ > c̄ and an overstated (for the permeability problem) value of
√
the solubility coefficient Γmax : ĉ = Γmax p.
So, the value of Φ can be estimated by fitting using the formula (10). This
information is of practical value as a factor for converting pressures into flux.
If we take D from one experiment (published) and Γ from another source
(Γmax ), then, strictly speaking, we get a hierarchy of three different numbers:
Φ < DΓ < DΓmax . For a material with a high level of capture by traps the cal-
culated permeability can be overstated by an order of magnitude compared with
the actual one. The permeability Φ (as a parameter in (10)) has an S-shaped (Ar-
rhenius) form of the saturation curve with respect to the magnitude of pressure.
Only at relatively high pressures (when the boundary concentrations approach
the Sieverts’ ones) we have Φ ≈ DΓ.
Fast Hydrogen Permeability of Structural Materials 205
The series on the left is rapidly converging for large t. But the series on the right
is rapidly converging for small t > 0. After some auxiliary transformations for
0 < t0 1 we get
2 X n −m2 o
f (t0 ) = √ exp 0
n ∈ N , f ≈ 1, t0 > 1. (13)
πt0 m=2n−1 4t
These representations enable a correct calculation of the flux J1 (t), t > 0. For
the considered applied problem it is sufficient to use partial sums consisting of
5–6 terms. Stage I ends with c(t∗ , x) = c̃1 (` − x)`−1 .
Stage II (t∗ → t0 = 0 is the t time zero):
ct = Dcxx, c(t, 0) = c̃2 , c(t, `) = 0, c(0, x) = c̃1 (` − x)`−1 .
The penetrating flux is
h ∞
X n n2 π 2 Dt oi
J2 (t) = D`−1 c̃1 + (c̃2 − c̃1 ) 1 + 2 (−1)n exp − .
`2
n=1
−1
Stage II ends with c(t∗ , x) = c̃2 (` − x)` .
Stage III. The formulas are similar to the cyclic interchange c̃1 → c̃2 , c̃2 → c̃3 .
The result of connecting the stages into a single curve of the penetrating flux
(conventionally, t = (t1 , t∗1 + t2 , t∗2 + t3 ), J = (J1 , J2 , J3 )) is shown in Fig. 1.
Fast Hydrogen Permeability of Structural Materials 207
7
J3
6
J2 p3=70 Torr
5
J, 1017 cm-2 s-1
p2=50 Torr
J1
4 1
0.8
3 p1=30 Torr
0.6
f(t')
2 0.4
0.2
1
0
0 0.2 0.4 0.6 0.8 1
t'
0
0 100 200 300 400
t, s
(the curves visually coincide). We wait until the sample is uniformly saturated
(pressure values become nearly equal) before moving to the TDS experiment.
In this article we confine ourselves to aggregating only two types of experiments
and referencing [26–28] (TDS models for a cylindrical sample and for a powder
are presented in [14, 29, 30]).
The intersection of the asymptote with the t axis gives so-called lag time τ0 =
`2 /(6D). Analytically,
Z t∗
τ0 ≈ t∗ − J1 (τ )J¯1−1 dτ, J1 ≈ J¯1 , t ≥ t∗ .
0
Note that a relative magnitude which does not require absolute values of the pen-
etrating flux (J¯1 = sup J1 (t)) is under the integral sign. In addition, the value
τ0 does not depend on c̃1 . It is usually assumed that the locally equilibrium con-
√
centration c̄1 = Γ p̄1 is quicklyp established at the inlet, so one can additionally
estimate the solubility Γ = 2sµ/b and the permeability Φ = DΓ using the
value J¯1 = Dc̄1 /`. This assumption is not used in this article. We assume
that according to the experimental conditions the stationary inlet concentration
c0 (t) ≈ c̃1 < c̄1 (t > ε, ε 1) is considered to be quickly established given
that c` (t) ≈ 0. The value of c̃1 as such is yet to be clarified. Thus, only the
estimate of the diffusion coefficient D is considered reliable at this stage.
With a new zero-time reference, integrating the expression J2 (t) we get
Z t
[J2 (τ ) − J¯1 ] dτ ≈ [J¯2 − J¯1 ] · [t − `2 (6D)−1 ] t∗ → t0 = 0 ,
0
where J¯i = Dc̃i /`, t ≥ t∗ ≡ t∗2 . Formally, we obtain the same expressions for
the lag time and the estimate of D if we change both the zero time and the flux
Fast Hydrogen Permeability of Structural Materials 209
Using the straight-line segment of the isotherm we find DΓ/` (the slope of
the straight line) and knowing the estimate of D we determine the initial ap-
proximation of the solubility coefficient Γ. From the intersection of the straight
line with the ordinate
p axis we find b (formally setting p̄ = 0). Knowing the
values of Γ = 2sµ/b and b compute s and Φ = DΓ. A graphic illustration
(using a minimal required set of J¯1,2,3 values) is presented in Fig. 2. The initial
data are b = 5.84×10−24 , s = 1.22×10−4 , Γ = 1.98×1020 , Φ = 3.98×1015 .
8 D = 2 10-5
7 b = 5.84 10-24
s = 1.22 10-4
6
1020
5 1015
0.1
4
3 0
2
-D2
J=
1 -0.1 22b
-0.1 0 0.1
0
0 1 2 3 4 5 6 8 10
pin , Torr
¯
Figure 2. Extrapolation of the isotherm of the steady state flux J.
70
60
D = 2 10-5
50
b = 5.7 10-24
p0, p, Torr
40
s = 1.2 10-4
= 2 1020
30 = 4 1015
20
10
t0=2 min
0
0 1 2 3 4 5 6
t, 103 s
Conclusion
It is advisable to aggregate the penetration experiments (with and without vac-
uum pumping) to make the measurements substantially more informative for
further estimation of the hydrogen permeability parameters and to improve the
accuracy of the measurements. This paper suggests a cascade experiment tech-
nique and the corresponding software.
The identification algorithm uses only integral operators thus ensuring the
noise resistance of experimental data treatment. The penetration method with
vacuum pumping is characterized by a significant measurement error, and data
on the penetrating flux are required (and this, in turn, requires a more accurate
determination of the vacuum system characteristics). The model of the dis-
solved hydrogen concentration jump at the inlet side is not very precise either.
We are brought to a conclusion that the ‘communicating vessels’ stage, where
molecular hydrogen pressures are measured over a long time, is characterized
by a much higher accuracy of measurements.
The first stage of the aggregated experiment is perceived as preliminary es-
timation of the diffusion D, desorption b and absorption s coefficients. It is
essential that the solution of the inverse problem of parametric identification
is unique, since the results obtained for thin laboratory membranes are extrap-
olated (recalculated) to the dimensions of real-life structures. The results are
‘fine-tuned’ by means of local variation of the preliminary values of D, b, s in
the ODE model (8), (9).
Acknowledgment
This work was supported by the Russian Foundation for Basic Research
[Grant 15-01-00744].
Appendix
We define the function f (t0 ) as follows:
1. t0 ∈ [0, 10−4] ⇒ f (t0 ) ≡ 0; 2. t0 ∈ [10−4 , 10−1 ] ⇒ f (t0 ) : (13);
3. t0 ∈ [10−1 , 1] ⇒ f (t0 ) : (11); 4. t0 ≥ 1 ⇒ f (t0 ) = 1.
The graph for f (t0 ) has an S-shaped (‘Arrhenius-like’) saturation curve form
(see a small inset in Fig. 1).
212 Yury V. Zaika and Natalia I. Rodchenkova
P P
Finally, we denote 0 ≡ 1,3,5,... and present auxiliary transformations:
∞
X X0
f (t0 ) = 1 + 2 (−1)n exp {−n2 π 2 t0 } = 1 − 2 exp {−n2 π 2 t0 }
n=1
∞ ∞ n −n2 o
X
2 2 0 (12) X0 1 X
+2 exp {−4n π t } = −2 ···+ √ exp
2 πt0 −∞ 4t0
n=1
| {z }
even terms
∞ n −n2 oi
X0 1 h X X0
= −2 . . .+ √ 1+ 2 exp = −2 ...
2 πt0 4t0
n=1
n −n2 o ∞ n n2 o i
1 h X0 X
+ √ 2 exp + 1 + 2 exp − 0
2 πt0 4t0 n=1
t
| {z }
even terms
∞
(12)1 X0 n n2 o 1 h√ 0 n X oi
= √ exp − 0 + √ πt 1 + 2 exp{−n2 π 2 t0 }
πt0 4t 2 πt0 n=1
X0 1 X0 n n2 o 1
−2 exp {−n2 π 2 t0 } = √ exp − 0 + f (t0 )
πt0 4t 2
2 X0 n n2 o
⇒ f (t0 ) = √ exp − 0 .
πt0 4t
References
[1] Interactions of hydrogen with metals; Zakharov, A. P.; Ed.; Nauka:
Moscow, 1987 [in Russian].
[2] Hydrogen in metals; Alefeld, G.; Völkl, J.; Eds; Springer-Verlag: Berlin,
1978.
[5] Cherdantsev, Yu. P.; Chernov, I. P.; Tyurin, Yu. I. Methods of studying
metal-hydrogen systems; TPU: Tomsk, 2008 [in Russian].
[7] Handbook of hydrogen storage: new materials for future energy storage;
Hirscher, M.; Ed.; Wiley-VCH: Weinheim, 2010.
[8] Lototskyy, M. V.; Yartys, V. A.; Pollet, B. G.; Bowman Jr, R. C. Int. J.
Hydrog. Energy. 2014, 39, 5818–51.
[9] The hydrogen economy: opportunities and challenges; Ball, M.; Wi-
etschel, M.; Eds; Cambridge University Press: New York, 2009.
[10] Varin, R. A.; Czujko, T.; Wronski, Z. S. Nanomaterials for solid state
hydrogen storage; Springer: New York, 2009.
[11] Evard, E. A.; Gabis, I. E.; Yartys, V. A. Int. J. Hydrog. Energy. 2010, 35,
9060–9.
[13] Zaika, Yu. V.; Bormatova, E. P. Int. J. Hydrog. Energy. 2011, 36, 1295–
305.
[14] Zaika, Yu. V.; Rodchenkova, N. I. Appl. Math. Model. 2009, 33, 3776–91.
[17] Kojakhmetov, S.; Sidorov, N.; Piven, V.; Sipatov, I.; Gabis, I.; Arinov, B.
J. Alloy Compd. 2015, 645, S36–40.
[18] Song, G.; Dolan, M. D.; Kellam, M. E.; Liang, D.; Zambelli, S. J. Alloy
Compd. 2015, 509, 9322–28.
[19] Terrani, K. A.; Balooch, M.; Wongsawaeng, D.; Jaiyen, S.; Olander, D. R.
J. Nucl. Mater. 2010, 397, 61–8.
[20] Zhang, Y.; Maeda, R.; Komaki, M.; Nishimura, C. J. Membr. Sci. 2006,
269, 60–5.
214 Yury V. Zaika and Natalia I. Rodchenkova
[21] Alimov, V. N.; Busnyuk, V. N.; Notkin, M. E.; Livshits, A. I. Tech. Phys.
Lett. 2014, 40, 228–30.
[22] Zaika, Yu. V.; Rodchenkova, N. I.; Sidorov, N. I. Comput. Res. Model.
2016, 8, 121–35 [in Russian].
[23] Voyt, A.; Sidorov, N.; Sipatov, I.; Dobrotvorskii, M.; Piven, V.; Gabis, I.
Int. J. Hydrog. Energy 2017, 42, 3058–63.
[26] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2017, 42, 405–15.
[27] Zaika, Yu. V.; Kostikova, E. K. Adv. Mater. Sci. Appl. 2014, 3, 120–9.
[28] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2014, 39, 15819–
26.
[29] Zaika, Yu. V.; Rodchenkova, N. I. Math. Model Comput. Simul. 2008, 19,
67–79 [in Russian].
[30] Rodchenkova, N. I.; Zaika, Yu. V. Int. J. Hydrog. Energy. 2011, 36, 1239–
47.
INDEX
# B
computer, vii, viii, ix, xii, 1, 6, 8, 54, 57, 63, distribution, viii, x, xi, 5, 36, 39, 41, 42, 53,
103, 108, 109, 118, 124, 131, 146, 168, 54, 114, 116, 130, 131, 132, 141, 150,
196 152, 156, 157, 159, 160, 162, 170, 202
computer simulation, vii, viii, ix, xii, 1, 4,
63, 103, 108, 120, 196
E
computer simulations, ix, 63, 103
configuration, 38, 44, 110, 111, 116, 153,
ecosystem, 75, 76, 77
155, 156
electric field, 111, 112, 113, 114, 115, 116,
constituent materials, 57
117, 118
construction, x, 2, 4, 5, 23, 64, 69, 149, 165,
electrical properties, 138
181
electricity, x, 149, 151
contact pressure, 5, 39, 40, 41, 42, 52, 53,
energy, 13, 15, 19, 28, 29, 30, 38, 50, 150,
54
152, 160, 161, 168, 170, 190, 192, 196,
cornering stiffness, 43, 48
200, 213
energy density, 13, 15, 19, 30
D energy parameters, 190
environmental conditions, 152, 156, 164
defects, 169, 170, 174, 175, 184 equilibrium, 30, 69, 71, 73, 74, 79, 80, 83,
deformation, 13, 14, 15, 23, 42, 43, 44 84, 86, 94, 95, 171, 172, 173, 175, 185,
derivatives, 173, 200, 203 200, 201, 202, 203, 204, 205, 207, 208,
desorption, viii, xi, xii, 167, 168, 169, 171, 209
172, 173, 176, 178, 179, 181, 184, 185, equipment, 109, 123, 173, 205
190, 191, 196, 198, 200, 211 experimental condition, xii, 181, 195, 196,
deviation, 24, 25, 89, 90, 190 200, 203, 205, 208
differential equations, 64, 203
differential games, vii, ix, 63, 64, 65, 66, 67,
F
69, 103, 104
diffraction, vii, ix, 107, 108, 109, 111, 114,
filaments, 130
119, 125, 126, 132, 133, 148
finite element, vii, viii, 1, 2, 4, 5, 8, 9, 12,
diffraction grating(s), vii, ix, 107, 108, 109,
13, 31, 32, 33, 34, 35, 38, 39, 43, 44, 51,
119, 126
54, 57
diffusion, xi, xii, 132, 168, 169, 170, 175,
finite element method, vii, viii, 1, 2, 4, 5, 51
176, 179, 181, 184, 185, 191, 195, 196,
force, ix, 2, 6, 23, 24, 42, 45, 47, 48, 49, 50,
197, 198, 199, 204, 205, 206, 207, 208,
52, 56, 57, 66
211
formula, x, 25, 84, 109, 129, 174, 175, 200,
diffusion process, 168, 169, 179
201, 202, 204
diffusion time, 181
Fresnel cosine and sine integrals, 111, 113,
discretization, 55, 57, 106
114, 116, 117, 118
displacement, x, 11, 12, 24, 31, 38, 108,
friction, 3, 6, 38, 39, 41, 151
109, 116
fuel consumption, 3, 6, 46, 50
Index 217
G L
163, 167, 168, 169, 171, 173, 175, 177, technology(ies), xii, 5, 173, 195, 196
179, 181, 183, 185, 187, 189, 191, 193, temperature, xi, 7, 14, 16, 39, 150, 151, 152,
196, 202, 207 153, 155, 156, 157, 158, 161, 163, 164,
slip angle, 46, 47, 48 165, 169, 170, 171, 172, 173, 178, 179,
smoothing, 25, 179 184, 185, 198, 199, 200, 201, 205, 207
software, vii, viii, xii, 1, 4, 9, 11, 23, 34, 64, tension, 7, 16, 19, 20, 21, 22
109, 114, 121, 124, 141, 179, 184, 186, test data, 5, 17, 19, 20, 23, 26, 80
189, 196, 197, 211 testing, 20, 21, 22, 23, 24, 190, 198
solar chimney power plant(s) (SCPP), viii, texture, 39, 131, 133, 136
x, 149, 150, 152, 153, 154, 156, 159, time increment, 54, 57
163, 164, 165 tires, v, ix, 1, 2, 4, 5, 8, 9, 13, 14, 29, 30, 36,
solar radiation, xi, 150, 156, 158, 161, 163 38, 43, 44, 46, 49, 52, 54, 57, 58, 60
solid state, 192, 213 transformations, 176, 180, 206, 212
solubility, 175, 204, 208, 209 tread, ix, 2, 9, 10, 25, 33, 34, 35, 38, 39, 41,
solution, vii, ix, x, 9, 38, 54, 63, 64, 65, 68, 42, 51, 52
69, 72, 79, 80, 89, 93, 103, 130, 132, treatment, ix, 63, 179, 189, 197, 211
142, 177, 179, 200, 211 turbulent flow, 150
sorption, xii, 171, 172, 178, 195, 196 twist, viii, x, 11, 129, 130, 131, 132, 141,
sorption process, 171 142, 143, 145, 146, 147, 148
steady state rolling, ix, 2, 7, 8, 32, 44, 46, twist angle, viii, x, 129, 130, 131, 142, 143,
47, 50, 51, 55, 57 145, 146, 147, 148
steel, viii, 1, 2, 7, 11, 30, 34, 181, 184, 185 twist level, 130
strain energy density, 13, 15, 19, 30
stress, 4, 6, 18, 19, 20, 23, 25, 26, 27, 28,
U
30, 36
structure, 35, 57, 69, 73, 75, 78, 93, 109,
uniaxial tension, 19
130, 136, 146, 147
uniform, 34, 55, 169, 179, 185, 189
successive approximations, 64, 104
universal test machine, 24
sustainable development, 67, 77, 105
unsteady state, 150
symmetry, 13, 32, 33, 35, 119, 142, 172,
173
V
T vacuum, xi, 167, 173, 176, 197, 198, 199,
205, 209, 211
Talbot carpet, vii, ix, 107, 108, 109, 110,
variables, vii, ix, 38, 39, 63, 65, 73, 76, 78,
120, 121, 122, 123, 124
82, 89, 92, 96, 103, 112, 118, 178, 183,
Talbot distance, ix, 107, 108, 119, 122
189, 201, 202, 203
Talbot effect, vii, ix, 107, 108, 109, 110,
vector, 46, 47, 68, 134
119, 120, 124, 127, 128
velocity, xi, 6, 44, 45, 46, 47, 50, 51, 55, 57,
Talbot, Henry, vii, ix, 107, 108
150, 151, 152, 153, 156, 157, 158, 163
techniques, viii, 1, 5, 9, 30, 57, 66, 127, 169
vessels, xii, 196, 197, 198, 199, 207, 211
220 Index
viscoelastic properties, 6
Y
viscosity, 150, 162, 163
volumetric changes, 14
yarn, v, viii, x, 129, 130, 131, 132, 133, 134,
135, 136, 138, 141, 142, 143, 144, 145,
W 146, 147, 148