Вы находитесь на странице: 1из 20

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/319114519

Analytical modeling to predict thermal shock failure and maximum


temperature gradients of a glass panel

Article · August 2017


DOI: 10.1016/j.matdes.2017.08.021

CITATIONS READS

3 89

1 author:

Paolo Foraboschi
Università Iuav di Venezia
61 PUBLICATIONS   952 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

New methods to increase bond strenth of FRP reinforcement. View project

All content following this page was uploaded by Paolo Foraboschi on 05 September 2017.

The user has requested enhancement of the downloaded file.


Materials and Design 134 (2017) 301–319

Contents lists available at ScienceDirect

Materials and Design

journal homepage: www.elsevier.com/locate/matdes

Analytical modeling to predict thermal shock failure and maximum


temperature gradients of a glass panel
Paolo Foraboschi
Università IUAV di Venezia, Dipartimento di Architettura Costruzione Conservazione, Convento delle Terese, Dorsoduro 2206, 30123 Venice, Italy

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Thermal stress analytical prediction


• Closed-form mathematical formulas for
the design of structural glass
• Upper bound of thermal stresses
• Changes that rectify errors and inaccu-
racies of code prescriptions and recom-
mendations
• Tool to justify the use of annealed and
heat-strengthened glass

a r t i c l e i n f o a b s t r a c t

Article history: The research has tackled thermal cracking distress in glass from a design perspective. The paper addresses a novel
Received 22 May 2017 modeling framework, and provides closed-form solutions to predict thermally-induced stresses and the temper-
Received in revised form 18 July 2017 ature profiles that cause glass to crack. The analytical formulation applies to monolithic glass and to plies of
Accepted 7 August 2017
laminated glass or insulating glass.
Available online 12 August 2017
Results and discussion prove that the ratios between maximum thermal stress and temperature gradient have an
Keywords:
upper bound, which is given, show that empirical approaches are largely inaccurate, and establish evidence that
Differential movements some code stipulations are wrong and some recommendations misleading.
Glass rupture © 2017 Elsevier Ltd. All rights reserved.
Temperature effects
Thermal loadings
Thermal stress

1. Introduction dead loads (self-weight loading), but also environmental loads and live
loads. When a building is challenged by natural hazards such as wind-
Up until two decades or so ago, glass elements in building envelope storms or snowstorms, glass elements in building envelope system
systems were commonly classified as ‘nonstructural’ or ‘architectural’ (windows, overhead glazing, glass façade panels, glazed doors, spandrel
components. Such a classification implied that they were designed, panels, skylights) are at the very frontline of that building's structural
assessed, and built as non-load-bearing, which was, and is, certainly defenses. Moreover, many glass elements (both external and internal
not the case in reality. Those elements must in fact resist not only the components) have to bear considerable live loads.
For that reason, those elements should be tackled in the same way as
structural building components. Classifying glass elements as ‘nonstruc-
E-mail address: paofor@iuav.it. tural’ left them in a gap of design attention between the architect and

http://dx.doi.org/10.1016/j.matdes.2017.08.021
0264-1275/© 2017 Elsevier Ltd. All rights reserved.
302 P. Foraboschi / Materials and Design 134 (2017) 301–319

the structural engineer. As a result, there have been many initiation and fast propagation up to failure, due to rapid changes in
significant examples of poor performance of architectural glass in the temperature).
past [1–4]. A temperature change will always produce either an expansion or a
Those examples prompted a novel design philosophy in order to contraction of a material. If the material has high heat conduction (such
close that gap and to address the need for mitigating the risk of glass el- as metals) the change in size is rather even. If this is accompanied by
ements [5–8], which resulted in the development of new standards and high tenacity (such as metals) then thermal cracks and thermal shock
code requirements for architectural glass, in many countries. The pur- are not likely results [26–28]. If the material has poor heat conduction
pose of those systems of principles, rules and recommendations was together with poor tenacity (such as ceramics, rocks and glass
to provide the building envelope designer with comprehensive resource [29–32]) it heats up or cools down in an uneven manner, which causes
documents, specifications, design guides, and standard practice exam- it to expand or contract differently and, in turn, causes high stress-to-
ples that will enable her/him to design architectural glass elements to strength ratios, so that thermal cracks and thermal shock are possible
resist the naturally occurring phenomena of windstorms, snowstorms events (Section 2).
and earthquakes as well as live loads in accordance with the latest Thermal stresses in glass due to uneven heating or cooling can
model building code provisions and industry standards in the world. be high enough to cause cracks and failure. Prediction of stresses intro-
Tackling glass elements in the same way as structural building com- duced in glass by temperature gradients is therefore a matter of funda-
ponents entails assigning an appropriate class of consequences to every mental interest to the glass designers and manufacturers.
glass element, based on the consequences of failure for occupants and A designer can derive the relationships between thermal stresses
pedestrians (loss of human life and wounded), on economic and social and restrains from basic mechanics, and can generally complete the re-
consequences (including public perception of failure, which in glass el- quired project work with basic and simple models. The relationships be-
ements is important), on potential environmental damage, on expense tween thermal stresses and temperature gradients in unrestrained glass
and procedures necessary to reduce the risk of failure, and on the seem to be far more elusive, even in the common situations. Some codes
possible causes and modes of attaining each limit state (not only and standards about structural glass state that the relationships be-
serviceability and ultimate limit states, but also collapse limit state, tween thermal stresses and temperature distribution are too complex
including post-breakage behavior). to be modeled analytically, even considering the plane stress state,
Nowadays, glass elements are eventually designed by licensed de- and that thermal stresses can only be obtained by means of specific
sign professionals, considering the foregoing possible consequences of and complex numerical modeling.
a structural failure. For instance, in Italy glass elements are categorized The research presented herein has tackled that issue and has devel-
in the consequence classes that the European platform prescribes for oped a general analytical modeling framework to predict thermal
the structures (Eurocodes [9]), which consider not only the role as- stresses induced by non-uniform temperatures in unrestrained glass
sumed by the element in the structural system, but also the possible (in an isostatically restrained glass panel).
outcomes of failure of that element in terms of risk to life, limb (injury
to people) or property (economic losses). For each consequence class, 2. Glass and thermally-induced stress
the acceptable risk is translated into an acceptable level of the failure
probability under the loads that are prescribed for structural systems. This section states exactly the reason for performing the study and
It follows that, in Italy, glass elements must be designed and built in makes the study's statement of purpose deeper.
the same way as structural members [10–12]. Furthermore, glass ele- Toughness of glass is 0.75 MPa·m1/2, which is extremely low. Glass
ments must be assessed using the semi-probabilistic method used for has therefore little ability to resist the propagation of flaws (the energy
building structures [13,14], as well as criteria in agreement with the that is necessary to supply for initial crack growth is very small).
Eurocode ‘Basis of structural design’ (known as Eurocode 0 [9]), where- Tensile constitutive law of glass is linear elastic up to rupture (utterly
as a deterministic method, a procedure specifically developed for struc- elastic behavior). The elastic modulus of glass is 63,000–77,000 N/mm2,
tural glass or an empirical methodology would be inadequate [15,16]. which is relatively high. Glass has therefore little ability to withstand
The way of dealing with architectural glass is shared by the whole high tensile strains and prevent catastrophic failure [33,34].
Europe, mutatis mutandis. As a matter of fact, the European Committee The combination of toughness and tensile constitutive law entails
for Standardization (CEN) is developing a Eurocode on glass. American that glass has little ability to accommodate motion-induced strains.
codes have allowed for architectural glass for many years now. The For that reason designers create isostatic restraints, which are obtained
most current building code enforced in the vast majority of jurisdictions with mechanical or chemical fixing that simply support each glass
in the United States is the International Building Code (IBC), which con- panel. When the glass panel is in a frame, designers provide an adequate
tains specific requirements for architectural glass since the first edition clearance gap around the edges of the glass panel which would
(1997). accommodate the anticipated relative displacements of the building
At the most basic level, the glass subjected only to the loads directly [12,35,36].
applied can be called ‘structural glass’ to differentiate it from the glass Linear temperature expansion coefficient of soda-lime glass is 9.0
that also bears loads not directly applied, which is referred to as ‘glass μm/(m·K), which is rather high. This means that glass subjected to uni-
structures’, but however structural glass and glass structures are now form temperature changes exhibits rather large strains. All the more
designed, analyzed, assessed, and built using a structural engineering reason why a glass panel must be kinematically and statically determi-
approach almost everywhere in the world. nate, and the clearance gap wide: the extremely low toughness, the
The above-depicted scenario shows that glass design is particularly utterly elastic constitutive law with relatively high elasticity modulus
demanding, because it entails addressing both general structural issues and the rather large linear temperature expansion coefficient entail
and specific issues directly related to the nature of the material [17–21]. that glass has little ability to accommodate the thermal movements.
One of the major specific issues is raised by temperature effects [22–25]. The restraints have thus to neutralize the uniform thermal strains.
The main mechanical effects of temperature are thermal stresses Thermal conductivity of glass is 0.9–1.0 W/(m·K), which is very low.
(which are stresses induced either by uniform and non-uniform Glass has therefore little ability to conduct heat away. The very low ther-
temperature changes in hyperstatically restrained bodies or by non- mal conductivity implies that glass plies can be subjected to very high
uniform temperature changes in unrestrained — namely, isostatically temperature gradients, which cause every portion of glass to expand
restrained — bodies), thermal cracks (which are defined as cracking in or contract differently than each other.
a component caused by thermal stresses, with or without mechanical An expansion or contraction of a portion of a solid body with
stresses), and thermal shock (which is defined as thermal crack respect to the rest of the body induces stresses, since that portion is
P. Foraboschi / Materials and Design 134 (2017) 301–319 303

kinematically and statically indeterminate (each portion is 3. Analytical modeling


hyperstatically restrained by the rest of the body). The combination of
toughness, tensile constitutive law, linear temperature expansion coef- As seen in Sections 1 and 2, thermal stress is a stress that is induced
ficient, and thermal conductivity entails that glass has little ability when the temperature of a body or a portion of a body changes in some
to accommodate the stresses induced by non-uniform temperatures. manner, yet the body or the portion of the body is restrained from the
Any type of glass, not only typical solar control glazing, incurs sub- free movement of the case of thermal strain. The body or the portion
stantial thermally-induced stresses on a daily basis. Unfortunately stan- of the body subjected to a temperature rise (drop) tends to expand
dards, codes, and guidelines deal with thermal stresses only marginally (contract), but the restrained boundaries prevent the body or the por-
and do not provide engineers and architects responsible for building tion of the body from expanding (contracting) with a free thermal
envelope performance design with any reliable and accurate method strain.
to predict thermally-induced stresses and maximum temperature gra- It has been noted that thermal stress is associated with thermal load-
dients tolerated by a glass panel. ing in contrast to a mechanical stress which is associated with mechan-
The document that most deepen the thermal stress analysis is a ical loading. Indeed, there is no such thing as thermal stress, but only a
French code, whose importance is however limited by the fact that it thermally-induced stress. Nevertheless, this paper often uses the term
is not written in English (the available English translation is badly ‘thermal stress’ with the understanding that it means thermally-
written) and was not developed by the university but by the industry. induced stress.
Anyway, thermal stress design provisions of that code consist only of All of the above considerations clearly indicate that thermal stress can
overly-conservative empirical formulas, as proven by this research, so be induced not only by uniform or non-uniform temperature changes in a
it does not solve the problem. restrained body, but also by temperature gradients in an unrestrained
Scientific literature is largely deficient about this topic too (impor- body. Research activity was directed at the latter case, whereas the former
tant contributions can be found in [22–25,36]) and expertise of glass case does not belong to good construction practice, as glass panels are
manufacturers in the field of thermal shock has always remained a typically restrained in isostatic condition, so that each glass panel is
trade secret. kinematically and statically determinate, as previously mentioned.
Commercial structural software packages can take temperature gra- The formulation developed herein predicts the thermal stresses,
dients as input, but the typical inputs are loads (or displacements). So, which must be superimposed upon the mechanical stresses obtained
many software packages that implement the finite element method by the relevant model (whereas, in the environmental load combina-
for stress predictions cannot easily and reliably yield the thermally- tions, the latter are often negligible in comparison with the former).
induced stress (little possibility of verifying the output). As a result, The temperature behavior depends on the profile and pattern of the
criteria used to assess acceptable performance against temperature gra- thermal gradient. The temperature distributions that induce strain pro-
dients by professional practitioners often consist in predicting the distri- files either uniform or with a first-order variation do not induce thermal
butions of temperatures that can occur under some design conditions stresses as long as the element is isostatically restrained; thus, modeling
with refined finite element models (often 3D), using software packages developed herein only accommodates higher-order thermal gradients.
devoted to heat transfer [37], and in verifying that the maximum Modeling allows for ring-pattern gradients — namely, temperatures
gradients of temperature lie within acceptable limits. This means that that vary along a radial direction from the origin (pole) of the reference
standard practice for assessment of thermal shock failure ignores the system (Subsection 3.2) — and stripe-pattern gradients — namely, tem-
comparison between the maximum thermally-induced stress and peratures that vary along an abscissa orthogonal to the stripes
glass tensile strength, but employs some limits of the temperature gra- (Subsection 3.3) — which cover almost all the cases. The two models
dients established without any actual scientific support. Although tem- share the mechanical assumptions (set out in Subsection 3.1).
perature distributions throughout the glass element are predicted The thermal stress depends on the elasticity modulus of the
accurately, that remains an empirical approach. glass, which is denoted by E, and by the linear temperature
Future trend in thermal performance requirements and code provi- expansion coefficient of glass, which is denoted by α. The values used
sions for architectural glass in Europe is to assume that the maximum herein (and in practice) for those parameters are E = 70,000 N/mm2
thermally-induced stress is proportional to the maximum temperature and α = 9.0 × 10−6 K−1. The difference between maximum and mini-
difference within the glass panel. However, the coefficient of propor- mum temperatures in the glass panel is denoted by ΔT.
tionality (sometimes called ‘shadow coefficient’) only takes into ac- Strains are taken to be positive when they are dilatation. Stresses are
count the rebate inertia (including the presence of structural sealant) inserted in the formulas with their real direction, which can be deduced
and whether the glazing is submitted or not to a cast shadow (the beforehand. Accordingly, the inequalities contain the absolute value of
presence of solar screens, loggias, vertical or horizontal pivot casements, the stresses.
top boards, reveals, or any mask). As a result, those documents will
not furnish any enhanced tool, but only rough empirical laws to prevent 3.1. Modeling assumptions
thermal shock. Once again, an empirical approach.
This means that there is the need for a model that predicts the stress- The assumptions aim at providing a realistic description of the me-
es induced by non-uniform temperatures, determines the maximum chanical behavior and not a simplified description.
temperature gradients withstood by a glass panel without cracking,
and identifies the most severe thermal loadings. The issues raised 3.1.1. Reference structure
must be addressed by establishing the analytical relationships between The reference structure is a glass panel externally restrained against
temperatures and strains, which yield closed-form solutions expressed rigid motions, but not externally restrained against uniform expansions
by normalized coordinates and dimensionless quantities, so as to obtain or contractions. Thus, the boundary conditions create isostatic re-
not only specific predictions, but also general results and a better under- straints, so that if the glass expands or contracts with a uniform strain
standing of the mechanical behavior. While, conversely, finite element or a constant angle it will not be stressed, while if it is subjected to a
models only yield specific solutions; moreover, if a software package thermal gradient (except for the above-mentioned first-order variation)
is used, the designer has little control over the process. it will be stressed.
This paper presents a modeling framework conceived to tackle ther- The presentation addresses the square glass panel but the formula-
mally-induced stresses. The presentation includes the description of the tions can be directly extended to the rectangular and circular glass
mathematical developments and the application of the formulation to panels (see the closing sentence of Subsections 3.1 and 3.2), as done
obtain information about temperature effects. in Section 6. The side of the square panel is denoted by L.
304 P. Foraboschi / Materials and Design 134 (2017) 301–319

The interlayer can increase considerably the bending stiffness of a As for the number of terms of the summation, N = 2 has been prov-
laminated glass panel, while it increases only marginally the in-plane en to be adequate for the vast majority of the cases (Section 4). A tem-
stiffness [38]. Not even the combination of large differences in in- perature distribution with N = 2 is described by four parameters, i.e.
plane strains between the connected layers and an ionoplastic interlay- T1max, λ1, T2max, and λ2. If those four parameters cannot describe accu-
er can induce considerable in-plane stresses (whereas those differences rately T(x), N shall be increased. In either case, Timax and λi are obtained
are usually small). from the data; i.e., with either N = 2 or N N 2 the function (1) is known.
For that reason, the in-plane thermal stresses of a layer that com- The parameter λi describes the peak of the function; the greater λi
poses a laminated glass depend only on the temperature of that individ- the sharper the peak, and vice versa.
ual layer. Accordingly, these models can be used for any glass system, i.e. The maximum temperature in the glass panel, which is denoted by
a sheet of a laminated glass panel, a pane that composes an insulating Tmax and is reached at x = 0, is equal to
glass system or a monolithic glass.
X
N

3.1.2. Static phenomenon T max ¼ T max


i ð2Þ
i¼1
Thermal stresses occur when one part of the glass panel is expanding
or contracting faster than another part, due to differences in tempera-
ture between the parts. Those differences in expansion (or contraction) Eq. (2) shows that the parameter Timax gives the maximum temper-
rates create non-uniform normal and shear strains, which induce ther- ature in the glass panel.
mal stresses. Hence, thermal stresses derive from different rates of ex- The uniform temperature profile Tu that composes the function (1)
pansion/contraction of the parts. After some time, the glass panel will and that does not induce thermal stresses is:
reach an equilibrium temperature and the thermal stresses eventually
vanish. X
N
Tu ¼ T max
i  e−λi ð3Þ
Since the cause is a transient thermal condition, thermal stress is a i¼1
dynamic function, connected to the rate of the differential movements.
Then again, the duration of such transient thermal stresses is long in The constant function (3) represents the temperature of the glass
comparison to the natural frequencies of the system, seeing that the panel before being subjected to the temperature gradient.
thermal conductivity of glass is very low. Moreover, the mass involved The temperature gradient Tg that composes the function (1) and that
is small. From this it ensues that the influence of the rate and mass on induces thermal stress independently of the restraints is:
the thermal stress is marginal.
Modeling assumes therefore that the phenomenon is static; i.e., the N 
X 
 e−λi x −T max  e−λi
2
inertia of the phenomenon is neglected. T g ðxÞ ¼ T max
i i ð4Þ
i¼1

3.1.3. Thermal loading


A thermal load is defined by the relationship between the tempera- The variable function (4) represents the thermal loading. According-
ture and the position, while the time is neglected (Subsection 3.1.2). ly, thermal loading and thermal gradient (or temperature gradient)
Every temperature distribution can be decomposed into a uniform are expressions that have the same meaning in all senses herein; i.e.,
temperature profile and a non-uniform temperature profile (tempera- Eqs. (1) and (4) are equivalent. That is, a thermal loading consists indif-
ture gradient). Thermal stress is induced by the latter, while the former ferently of either (3) plus (4) or only (4). Likewise, ring-pattern, radial,
does not induce any stress in the reference structure, since the restraints and axi-symmetric have the same meaning, as well as stripe-pattern
can accommodate any uniform strain or first-order strain variation of and axial-symmetric.
the glass panel. Ultimately, the glass panel exhibits a profile of temperature that
The presentation addresses the thermal loading composed of a tem- ranges from the maximum value Tmax, which may occur at any point of
perature gradient that adds to the uniform temperature of the glass the square (it defines the origin of the reference system), to the minimum
panel. The temperature gradient that subtracts from the uniform tem- value, which occurs at one point of one edge (it defines x = 1) or at some
perature of the glass panel can be obtained with easy (a specific case points of one or more edges (depending on the temperature distribution).
is shown in Subsection 4.1). The maximum temperature is given by Eq. (2), the minimum by Eq. (3),
The small thickness of a glass ply and the boundary restraints render and the gradient by Eq. (4). Thermal stress is induced by Eq. (4), while Eq.
the out-of-plane thermal gradients uninfluential, although the low ther- (3) does not induce any stress. Nor any other uniform temperature profile
mal conductivity of glass. Consequently, temperatures are assumed to added or subtracted to Eq. (1) induces stresses.
be constant across the thickness of the glass panel [23].
The ring-pattern is axi-symmetric and the stripe-pattern is axial- 3.1.4. Linearity between the thermal loading and stress
symmetric. Both of them depend, therefore, on one coordinate only, The constitutive law of glass is linear-elastic up to cracking. The dis-
which lies in the plane of the glass panel. It follows that the temperature placements of structural glass under thermal loadings are marginal.
profiles of the two patterns can be described by the same mathematical Thus, modeling assumes that the behavior is linear up to failure
function: (physical and geometric linearity).
Linearity assumption means that superposition of solutions may be
used. The thermal stress produced at a given point by several thermal
X
N
loadings acting simultaneously on the glass panel can be found by
 e−λi x
2
T ðxÞ ¼ T max
i ð1Þ
i¼1
superimposing thermal stresses at the same point produced by thermal
loadings acting individually.
Linearity assumption does not mean that the relationship between
where T denotes the temperature at the coordinate x and is expressed in ΔT and stress is linear, because the thermal loading is defined by the
degrees Kelvin [K]; x denotes either the normalized radial coordinate of temperature distribution and not by ΔT.
a polar coordinate system or the normalized abscissa of a Cartesian co- Linearity enables the superposition of the thermal stress state
ordinate system. Normalization implies that x ranges between −1 and 1 and the mechanical stress state, unless the mechanical behavior is geo-
(or between 0 and 1). The parameters Timax, N, and λi are calibrated to metrically non-linear [33,39]. The total stress state of the glass panel can
describe the profile of the distribution, before using the model. thus be obtained by two individual structural analyses, whereas the
P. Foraboschi / Materials and Design 134 (2017) 301–319 305

mechanical stress state is often insignificant when environmental Let us consider a temperature uniformly distributed over a circle of
actions are the leading variable loadings. glass inside the square glass panel, with center placed at the center of
the glass panel. That circular uniform temperature distribution adds or
3.1.5. Material strength subtracts to the uniform temperature of the glass panel, which is the
Failure occurs when tensile stresses reach glass tensile strength, temperature of the rest of the square.
whereas compressive stresses cannot cause failure because glass The difference between the temperature of the circle and the tem-
compressive strength is very high (not even shear stresses, besides the perature of the remaining part of the panel gives rise to a discontinuity
tensile stresses that they generate). (i.e., there is a temperature jump between the glass circle and the rest of
Glass tensile strength depends on the type of glass, sensitivity of the the glass panel). For the sake of clarity that jump is here considered to
glass to thermal shock, duration of load, and can depend on glass edge be finite, while in the relevant model will be infinitesimal.
finishing treatment. Glass strength independently of edge working is The circle subjected to the increase (or decrease) in temperature ex-
herein denoted by fg. pands (or contracts) with respect to the portion comprised between the
The four main types of glass used in buildings around the world are circle and the square boundaries. In turn, the portion of glass comprised
annealed, heat-strengthened, tempered, and chemically-strengthened between that circle and the circumference tangent to the four sides (i.e.,
glass. The difference between those glass types consists in whether inscribed in the square) encircles the glass comprised in the enclosed
and how glass is precompressed [13,18]. Heat-strengthened glass and circle (excuse the pun). Concisely, the circle is encircled by a ring tan-
tempered glass are also referred to as ‘thermally toughened’ (the latter gent to the sides of the panel (i.e., by the annulus inscribed in the
as ‘fully tempered’). square).
The strength of a glass panel along the edges is weaker than away The portions of glass comprised between the four sides of the
from the edges. Edge strength is herein denoted by fge. panel and the circumference tangent to the sides play a marginal
The ratio between fge and fg only depends on how the glass edges are role in the resisting system, because those portions do not form a closed
finished. There are a number of standard options for the worked shape section, so that those portions have no ability to encircle the other
of glass edges — namely, cut (clean; sharp-edged), arrised, swiped, portions.
ground (flat; smooth ground; ground miter), ogee, beveled, round, The maximum tensile stress occurs along the circumference tangent
seamed, and polished edges [13]. to the four sides (boundary of the circle inscribed in the square), as
Thermal shock sensitivity depends on the nature of glass (monolith- proven in the relevant section. The minimum strength along that cir-
ic glass of thickness lower than 12 mm, between 15 and 19 mm or cumference is at the points where it is tangent to the four sides, since
25 mm; symmetric laminate or non-symmetric laminate; wired glass; the strength of those points is fge while the strength of the rest of the cir-
patterned glass) and on its transformation (sawed glass; as cut or cumference is fg N fge. Thermal cracking is therefore dictated by the tan-
arrised; smooth ground or polished). Those factors imply different sen- gent points.
sitivity of glass to thermal shocks, which must be taken into account Let us consider two rectangles that complement each other to pro-
when calculating fg and fge. Thermal shock sensitivity can be measured vide the whole square glass panel, and a temperature uniformly distrib-
with standard test methods. uted over one of the two rectangles. That rectangular uniform
Precompression of tempered, chemically-strengthened, and heat- temperature distribution adds or subtracts to the uniform temperature
strengthened glass at the edges is lower than away from the edges, of the glass panel, which is the temperature of the other rectangle.
and the difference depends on how precompression is induced. Espe- The difference between the temperature of two rectangles gives rise
cially for heat-strengthened glass, but sometimes also for chemically- to a discontinuity (i.e., there is a temperature jump between the two
strengthened glass, the real value of fge can be considerably lower than rectangles). For the sake of clarity that jump is here considered to be fi-
the nominal value, so that fge can be substantially lower than fg. nite, while in the relevant model will be infinitesimal.
To be more specific, in heat-strengthened glass, the cooling rate The rectangle subjected to the increase (or decrease) in temperature
is lower than in tempered glass (lower cooling quench), which expands (or contracts) with respect to the rectangle by which it is
entails less control of the precompression at the edges. In chemically- complemented. In turn, the other rectangle restrains the rectangle
strengthened glass the provision of the state of compression in that has changed the temperature. Concisely, the two rectangles that
the surface strongly depends on the actual depth of penetration complement each other to provide the whole square interact along
relative to the thickness of the lite, the error-function profile of the the common boundary.
ion distribution, and stress relaxation. Thermal stress analysis has thus The strain profile induced by that interaction in each cross-section of
to gain adequate insight into fge guaranteed by each lite of heat-strength- the two rectangles (i.e., in the transverse sections orthogonal to the com-
ened and chemically-strengthened glass, especially when ΔT is high. mon side) is assumed to be a straight line. The sum of the straight-line
The duration of load reduces the fraction of glass strength that is not profiles of stresses induced by all the pairs of rectangles associated with
created by precompression. That phenomenon must be taken into all the temperature jumps within the glass panel (if the thermal gradient
account when calculating fg and fge: the fraction of the instantaneous is continuous, the pairs are infinite) produces a highly curved strain pro-
tensile strength not created by precompression must be multiplied by file over a cross-section that cuts the entire glass panel. Then again, a lin-
the strength modification factor Kmod. ear strain profile would not induce any stress, so thermal stresses need a
The maximum value of Kmod for thermal loadings ranges from 0.5 to strain profile on the entire cross-section that is not a straight-line. This as-
0.6. For annealed glass, where no fraction of strength derives from sumption is supported by experimental evidence [33,39].
precompression, the actual strength against thermal loadings is up to
50% the instantaneous strength. For precompressed glass the reduction 3.2. Model for the ring-pattern thermal gradient
is lower, but not negligible.
Thermal stresses from heating and cooling conditions can create This model predicts the maximum tensile stress induced by an
significant tensile stresses at the glass edges, which can surpass fge. axi-symmetric variable temperature profile; i.e., a temperature that
changes from one radial location to another with respect to a pole
3.1.6. Elastic internal restraints of the plane thermal strain (Fig. 1).
The temperature profiles described by Eq. (1) are constant through- The formulation presented herein yields the thermal stress at the
out the depth of the glass panel. Accordingly, the thermal strains are edges. Nonetheless, the model can be easily extended to predict the
plane. It follows that the thermal stresses produced by Eq. (1) consist thermal stress at any point of the glass panel induced by a radial thermal
in a plane stress state. gradient (extended model of Subsection 4.1).
306 P. Foraboschi / Materials and Design 134 (2017) 301–319

According to Subsection 3.1.3, N = 2 is sufficiently representative


(i.e., two terms). With N = 2 Eq. (2) becomes

T max ¼ T max
1 þ T max
2 ð8Þ

ρ and Eq. (6) becomes:

T e ¼ T max
1  e−λ1 þ T max
2  e−λ2 ð9Þ
θ
Eqs. (8) and (9) are the maximum and minimum temperatures, re-
spectively, for the two-term temperature function.
The easiest way to get the parameters of Eq. (5) with N = 2 is to de-
scribe the temperature at the edges with the first term (i = 1) and the
peak with the second term (i = 2) or vice versa. Accordingly, λ1 shall
be chosen relatively low, so that it represents the uniform temperature
of Eq. (6), while λ2 shall be chosen much higher than λ1, so that it rep-
resents the gradient of Eq. (7). In so doing, T1max and T2max turn out to be:
8
> Te
< T max
1 ¼ −λ −Φ
e 1 ð10Þ
Fig. 1. Glass panel subjected to a radial temperature distribution: square ply of side L. Polar >
: T max ¼ T max − T e
coordinate system: radial coordinate R, normalized radial coordinate ρ with 0 ≤ ρ ≤ 1 2
e−λ1
(where ρ = 2 · R/L, with 0 ≤ R ≤ L/2), and angular coordinate θ.

Eqs. (9) and (10) give T1max and T2max as long as λ1 and λ2 have been
defined and Φ is tuned. The calibration factor Φ, which is a temperature,
3.2.1. Radial temperature distribution has to compensate for the edge temperature T2max ⋅ e−λ2; i.e., Φ has to
For the sake of clarity the presentation addresses the temperature neutralize the temperature at the edge produced by the 2nd term. The
distributions whose maximum is at the center of the square, which is use of Φ does not increase the number of parameters that define the
the most demanding position of the peak. Yet the model can be used function (5), which remains four, but helps define those four parame-
for temperature distributions with maximum in any point of the glass ters, in order to adequately represent both the shape of the peak and
panel. Moreover, the model can be rearranged to predict the thermal the extreme values of the function.
stress when the temperature decreases (Subsection 4.1).
Eq. (1) takes the form:

X
N X
N
 e−λi ρ
2
T ðρÞ ¼ T i ðρÞ ¼ T max
i ð5Þ
i¼1 i¼1

in which ρ is the normalized radial coordinate of a polar coordinate


system with origin at the center of the glass panel (Fig. 1). More specif-
ically, ρ = 2 · R/L, where R is the radial coordinate (ρ = 0 identifies the ρ
center and ρ = 1 the middle point of each edge).
The absolute temperature T(ρ) cannot be negative by definition.
That condition could also be satisfied if some terms of the summation
are negative. However, the formulation requires that Timax N 0, otherwise
some mathematical operators would not be defined. No term of Eq. (5) ρ
can thus be negative, which represents a preliminary condition to verify.
The maximum temperature in the glass panel Tmax (which occurs at
ρ = 0) is given by Eq. (2). The minimum temperature in the glass panel,
which occurs at the edges (i.e., at ρ = 1) and is denoted by Te, is:

X
N X
N
Te ¼ T ei ¼ T max
i  e−λi ð6Þ Fig. 2. Panel subjected to an infinitesimal radial thermal loading: reference structure. The
i¼1 i¼1 temperature of the dark circle is infinitesimally greater than the temperature of the bright
ring; i.e. the temperature of the former and the latter differ from one another by dT. The
temperature of the ring, in turn, is equal to the temperature of the rest of the glass panel
which coincides with the uniform temperature component of T, (the latter plays no role). The diagram shows the resisting system for that infinitesimal
described by Eq. (3). temperature jump — namely, the dark circle is encircled by the bright ring. The resisting
The difference between the maximum temperature in the glass system is composed of the circle and the ring, while the four portions between the
panel Tmax and the minimum temperature Te, which has been denoted external circumference of the ring (outer boundary of the ring: solid line) and the sides
of the square (boundaries of the panel: pointed lines) do not belong to the resisting
by ΔT and has been described by Eq. (4), can be rewritten in a different
system; i.e., the four white portions do not bear any stress. The resisting system for a
form: radial temperature distribution (ring-pattern profile) is the composition of infinite
encircling actions equal to that in the diagram. The resisting system derives so from the
superposition of infinite circles and rings, along with infinite temperature jumps dT. The
inner boundary of each ring is the circumference of the complementary circle and varies
X
N   from R = 0 to R = L/2. The outer boundary of each ring is the circumference tangent to
ΔT ≡ T max −T e ¼ T max
i  1−e−λi ð7Þ the square in four points Q. Cracking is expected to initiate at one of the four tangent
i¼1 points Q.
P. Foraboschi / Materials and Design 134 (2017) 301–319 307

3.2.2. Thermal stress induced in the glass panel Considered that the hydrostatic stress state of the circle is a uniform
The glass panel is envisioned as the composition of infinite circles bi-axial compression, that the radial stress σr in the encircling ring is a
(Fig. 2). The generic circle with radius R = ρ · L/2 is subjected to the compressive stress, and that the circumferential stress σθ in the
temperature T(ρ) while the remaining part of the panel to the temper- encircling ring is a tensile stress, cracking (failure) is dictated by σθ in
ature T(ρ + dρ). Thus, the temperature in the circle R is greater than in the encircling ring, since σθ is the only stress that can crack the glass.
the rest of the glass panel, and the difference is infinitesimal, because of The radial stress σr(ρ) transferred across the circumference of radius
continuity of temperatures; i.e., T(ρ) − T(ρ + dρ) = dT. ρ is the amount necessary to compress the circle and to expand the sur-
Due to dT, the circle with radius R tends to expand differently than rounding ring, in order to have deformation compatibility between the
the rest of the glass panel. Compatibility requires that those infinitesi- circle and the surrounding ring under the thermal load. Accordingly,
mal differential in-plane displacements are zero. It ensues that compat- σr(ρ) derives from the following compatibility equation, where ν is
ibility engenders the thermal stress in order to nullify the differential the Poisson's ratio:
displacements due to the difference in temperature. That is, without
thermal stress, the circumference of radius R = ρ · L/2 if is attached to L dσ r ðρÞ L ν L
α2πρ  dT−  2  π  ρ  þ  dσ r ðρÞ  2  π  ρ  ð11Þ
the circle undergoes in-plane displacements different than if is attached 2 E 2 E 2
to the surrounding ring. dσ θ ðρÞ L ν L
¼  2  π  ρ  þ  dσ r ðρÞ  2  π  ρ 
According to the resisting system of Subsection 3.1.6 (Fig. 2), the in- E 2 E 2
finitesimal thermal strains induced by the infinitesimal difference in
temperature are elastically restrained by the ring that surrounds the cir- The left-hand side of Eq. (11) is the circumferential strain of the cir-
cle, which exerts an encircling action. cumference of radius R = ρ · L/2 attached to the circle, while the right-
The thermal stresses in the glass panel exhibit the same axial sym- hand side of Eq. (11) is the circumferential strain of that circumference
metry of the thermal gradient. It ensues that the thermal stresses consist attached to the surrounding ring.
of radial and circumferential (hoop) stresses, which are denoted by σr As expected, dσr(ρ) depends on dT, E and α, while it does not depend
and σθ, respectively (Fig. 3). Owing to axi-symmetry those stresses on the length of the circumference, since 2 · π · ρ · L/2 can be eliminat-
can be identified by ρ only, while the angular coordinate θ takes no role. ed. Moreover, the Poisson's effects counterbalance each other.
The encircling action exerts uniform radial stress at the boundaries Thermal stress and temperature are state functions. The thermally-
of the circle and of the surrounding ring. Consequently, the circle and induced stress is thus a process independent of the path taken; i.e., the
the ring transmit radial stress σr(ρ) from each other across the circum- differentials of thermal stress and temperature are path-independent.
ference of radius R = ρ · L/2 (Fig. 3). For that reason the infinitesimal thermal stress is an exact differential
The radial thermal stress σr(ρ) induces a uniform stress state in the in the domain of temperatures. So, Eq. (11) takes on the form:
circle of normalized radius ρ. The circle is therefore subjected to a
two-dimensional hydrostatic stress state; i.e., for ρ′ ≤ ρ: σr(ρ′) = σr(ρ) dσ r ðρÞ dσ θ ðρÞ
þ ¼Eα ð12Þ
and σθ(ρ′) = σr(ρ′) = σr(ρ). The encircling action induces thus only dT dT
normal stresses in the circle, but no shear stresses. Seeing that the ther-
The differential equation expressed by Eq. (12) involves two un-
mal strain causes the circle to expand, the hydrostatic stress state σr and
knowns, i.e. the radial stress σr and the circumferential stress σθ. The lat-
σθ consists of compressive stresses (Fig. 3).
ter is connected to the former by the following relationship, which can
The radial thermal stress σr(ρ) induces non-uniform normal stresses
be obtained from the Airy function [40]:
in the ring that surrounds the circle of normalized radius ρ. The
encircling action induces both radial stresses σr and circumferential
1 þ ρ2
stresses σθ in the surrounding ring, which decrease with ρ; i.e., for ρ″ σ θ ðρÞ ¼ σ r ðρÞ  ð13Þ
1−ρ2
N ρ′ ≥ ρ: 0 b σr(ρ″) b σr(ρ′) and 0 b σθ(ρ″) b σθ(ρ′). Seeing that the ther-
mal strain causes the circle to expand, σr in the ring is a compressive
Eq. (13) shows that, with reference to the modulus: σθ(ρ) N σr(ρ),
stress and σθ a tensile stress (Fig. 3). This means that thermal stresses
apart from ρ = 0 where σθ = σr (note that σθ is a tensile stress and σr
in the ring are composed not only of normal stresses but also of shear
is a compressive stress).
stresses.
According to Eq. (13), for ρ that approaches unity σθ approaches in-
finity, because ρ → 1 entails that the thickness of the surrounding ring is
infinitesimal, and a profile of finite σθ on an infinitesimal thickness can-
not balance any finite radial pressure σr. If the radius R approached in-
finity, ρ would approach zero, and the circumferential stress would
σθ approach the modulus of the radial stress, yet remaining greater.
Replacing Eq. (13) in Eq. (12):
σ σ  
dσ r ðρÞ 1 þ ρ2
 1þ ¼Eα ð14Þ
σθ dT 1−ρ2

which leads to:


 
dσ r ðρÞ 1−ρ2
¼ Eα ð15Þ
dT 2
Fig. 3. Stresses of the resisting system shown in Fig. 2: plane stress state. On the left:
stresses in the ring (bright annulus). The diagram shows the uniform radial stress σr
acting on the inner boundary of the ring and the circumferential (hoop) stress σθ acting
on an infinitesimal element at the inner boundary of the ring (dark square). On the right 3.2.3. Encircling tensile stress
(dark circle): stresses in the circle. The diagram shows the uniform radial stress σr The encircling behavior that restraints the expansion of the circle by
acting on the boundary of the circle and the stresses acting on an infinitesimal element means of σr(ρ) acting on the circumference of that circle implies radial
in the circle. The plane stress state of the ring is composed of tensile and compressive
stresses. The plane stress state of the circle is composed of two-dimensional hydrostatic
stress equal in magnitude and opposite in direction acting on the inner
compressive stresses. The figure shows that the tensile stresses in the glass panel are boundary of the ring, which induces circumferential tensile stress σθ in
produced by the circumferential stresses σθ, which hence dictate the thermal cracking. that ring. It can be proven that: σθ(ρ)/σθ(ρ = 1) = (1 + ρ2)/2 · ρ2 [40].
308 P. Foraboschi / Materials and Design 134 (2017) 301–319

For ρ b 1, thus, σθ(ρ)/σθ(ρ = 1) N 1, which confirms that the σθ profile, According to Eq. (19), the derivative of σe is the sum of N terms σie,
induced by σr at ρ, decreases from ρ to the edge (ρ = 1). As anticipated each one induced by the term Ti(ρ) of the N terms that compose the
in Subsection 3.2.2, hence, σθ(ρ) N σθ(ρ = 1). temperature distribution.
However, the edges have the minimum strength, i.e. fge. Moreover, Eq. (18) proves that σe is the antiderivatives (primitive function) of
σθ at the edge derives from the contributions of the circles in the ρ2(T) · E · α, as well as Eq. (19) proves that σei is the antiderivative of
range 0–1, while σθ at ρ derives only from the contributions of the cir- ρ2(Ti) · E · α. The antiderivative can be obtained only once the right-
cles in the range 0–ρ. As a result, the maximum ratio between σθ in- hand side of Eq. (19) is expressed as a function of Ti. Then, the
duced by a temperature distribution and the strength occurs at the function (5) has to be inverted, in order to obtain ρ as a function of Ti.
edge, where the former is the maximum and the latter is the minimum. Inverting the generic term Ti(ρ) that composes the function (5) gives:
Ultimately, the σθ that dictates failure is the one at the edge (Fig. 4).  
The model provides σθ(ρ = 1), which is denoted by σe, whereas it can ln T max
i − ln ðT i Þ
ρ2 ¼ ð20aÞ
be used to predict any σθ(ρ). λi
The amount of circumferential stress at the edge, i.e. σθ(ρ = 1), in-
where T takes on values of:
duced by σr at ρ is equal to [40]:
T ei ≤T ≤T max ð20bÞ
2  ρ2 i
σ θ ðρ ¼ 1Þ ≡ σ e ¼ σ r ðρÞ  ð16Þ
1−ρ2
Plugging Eq. (20a) into Eq. (19):
Recalling that the thermally-induced stresses σr and σθ are exact dif-  
dσ ei ln T max
i − ln ðT i Þ
ferential on the T domain: ¼ Eα ð21Þ
dT i λi
dσ e dσ θ ðρ ¼ 1Þ 2  ρ2 dσ r ðρÞ
¼ ¼  ð17Þ 3.2.5. Maximum thermal stress at the edges
dT dT 1−ρ2 dT
The generation of the thermal stress is modeled using the superpo-
sition property (Subsection 3.1.4). The stress at the edge results from
Plugging Eq. (15) into Eq. (17), the circumferential stress at the edge
the summation of the N stresses induced by the N terms that T has
caused by the movement of the circle of radius ρ with respect to the sur-
been decomposed into, and each of those N stresses, in turn, results
rounding ring turns out to be (Fig. 4):
from the definite integral of the infinitesimal stresses induced by the
dσ e infinite dTi that Ti can be split into. Hence, the calculation of σe entails
¼ ρ2  E  α ð18Þ determining σie.
dT
Seeing that dTi and dσie are exact differentials, the maximum
thermal stress induced at the edges by the i-th term can be obtained
3.2.4. Breaking down and inversion of the T(ρ) relationship by integrating all the contributions of dTi from Tie to Timax:
The temperature T is the summation of N terms Ti. This implies that
TZmax TZmax  
σe as well as its derivative must be decomposed into the contributions i i
ln T max − ln ðT i Þ
i
induced by each of the N components of Eq. (5). The i-th component σ ei ¼ dσ ei ¼  E  α  dT i ð22Þ
λi
of σe is hereafter denoted by σie: T ie T ie

dσ ei Hence, σie is the difference in the evaluation of the antiderivative at


¼ ρ2 ðT i Þ  E  α ð19Þ
dT i the endpoints of the temperature interval.
The linearity of the integral operator allows Eq. (22) to be split into
two integrals. The antiderivative of the first member of the right-hand
side of Eq. (22) integrated from Tie to Tmax
i is:
     
σ σ
TZmax
i
ln T max ln T max ln T max
i
 E  α  dT i ¼ i
 E  α  T max
i − i
 E  α  T ei ð23Þ
λi λi λi
T ei  
ln T max  
σ σ ¼
λi
i
 E  α  T max
i  1−e−λi

The antiderivative of the second member of the right-hand side of


Eq. (22) integrated from Tie to Timax is:
ρ TZmax
i
ln ðT i Þ Eα T max
 E  α  dT i ¼  jT i  ln ðT i Þ−T i jT ie ð24Þ
λi λi i

T ei
E  α  max    
¼  T i  ln T max
i −T max
i −T ei  ln T ei þ T ei
λi
σ σ
The right-hand side of Eq. (23) minus the right-hand side of Eq. (24)
σ σ leads to the final result:

TZmax
i
Eα   max    
σ ei ¼ dσ ei ¼  ½ ln T max
i  T i  1−e−λi −T max
i  ln T max
i
Fig. 4. Circumferential (hoop) stress at the edges of the glass panel: tensile stress σe which λi
may cause thermal crack. The diagram shows the greatest circumference (the boundary of T ei
   
the bright circle), which is tangent to the sides of the glass panel at the four points Q þ T ei  ln T ei þ T max
i 1−e−λi 
(Fig. 2), the four infinitesimal elements at those points (dark squares), and the
circumferential stress σe that acts on those elements. ð25Þ
P. Foraboschi / Materials and Design 134 (2017) 301–319 309

The components of Eq. (25) are the two constants E and α which de-
fine the material properties, the 2 · N parameters Timax and λi which de-
fine the thermal loading, and the N edge temperatures Tie each of which
is given by Eq. (6), i.e. Tie = Timax ⋅ e−λi. All the components of Eq. (25) are
thus known, so that it yields directly the stress at the edge σie induced by
the i-th term of the function (5).
The stress at the edge σe induced by the temperature distribution
described by the function (5) is (Fig. 4):

X
N
σe ¼ σ ei ð26Þ
i¼1

in which σei is provided by Eq. (25).


The stress σe is the maximum tensile stress induced by the temper-
ature distribution represented by Eq. (5). Since σe occurs where the
strength is minimum, Eq. (26) solves the problem. Fig. 5. Glass panel subjected to an infinitesimal stripe-pattern thermal loading. Cartesian
coordinate system: normalized abscissa x, with 0 ≤ x ≤ 1 (where x = X/L, with 0 ≤ X ≤ L).
Eq. (26) can be used directly to predict σe induced by a given radial
The ordinate is never used. The transverse side at x is the common side of two
distribution of temperature, in order to compare that result with fge, rectangles that the square is split into. Accordingly, x defines two rectangles that
which is the allowable stress (and includes the safety factor). complement each other to provide the whole square glass panel. The temperature of the
Eq. (26) can also be used indirectly to predict the maximum gradi- dark rectangle differs from that of the white rectangle by an infinitesimal value dT.
ents of temperature tolerated by a glass panel having allowable stress Compatibility equations are written for the common side of the two rectangles. The
model focuses on the behavior along the axis of symmetry (i.e., along each midspan
fge. To that end, Eq. (25) has to be inverted. cross-section of the two complementary rectangles).
In the case of a rectangular or a circular glass panel, Eq. (25) can be
applied to the square glass panel with sides of length equal to twice
the minimum distance between the point with the highest temperature the infinitesimal differential displacements due to dT in the infinite pairs
and the edges. of rectangles.
The latter rectangle (the one of the pair with the lowest tempera-
3.3. Model for the stripe-pattern thermal gradient ture) exerts infinitesimal in-plane shear stresses dτ along the common
side (Fig. 6), which elastically restrain the infinitesimal thermal strains
This model predicts the maximum stress induced at the edges of the induced by dT in the former rectangle (Subsection 3.1.6). Those shear
glass panel by an axial-symmetric variable temperature profile with stresses change the direction at the midpoint of the common side
axis of symmetry parallel to two sides of the square, herein called ‘lon- shared by the two rectangles.
gitudinal sides’, while the two other sides are called ‘transverse sides’. The in-plain shear stresses along the common side induce infinites-
The results presented in Subsection 4.2 demonstrate that the imal normal stresses in the glass panel. Those stresses vary along both x
greatest tensile stresses are induced by temperature distributions with and the direction orthogonal to x. For any x the normal stresses reach
maximum on one transverse side and minimum on the facing trans- the maximum on the symmetry axis of the glass panel shown in Fig. 5
verse side. The presentation is therefore developed for that temperature
distribution. Nevertheless, the model can be used for any axial-symmet-
ric temperature distribution (extended model of Subsection 4.2).

3.3.1. Axial-symmetric temperature distribution


The function (1) can directly describe the temperatures without
changing the mathematical form, as long as x is the normalized abscissa
of a reference system with origin at the edge with the highest tempera-
ture (Fig. 5) and 0 ≤ x ≤ 1 (x = 1 identifies the edge with the lowest
temperature). If the highest temperature occurs inside the glass panel,
−1 b x b 1.

3.3.2. Strains induced by compatibility shear stresses


The glass panel is envisioned as the composition of infinite pairs of
τ
rectangles. The two rectangles of a pair complement each other to pro-
vide the whole square glass panel (Subsection 3.1.6). A pair is identified τ
by the abscissa of the common side shared by the two rectangles
(Fig. 5). The rectangle that spans from 0 to x is subjected to the temper-
ature T(x), while its complement to the temperature T(x + dx). Thus,
the temperature in the former is greater than that in the latter, and
the difference is infinitesimal, because of temperature continuity; i.e.,
T(x) − T(x + dx) = dT.
Due to dT, the former rectangle tends to expand with respect to the
latter rectangle, so that the common side at x if is attached to the former Fig. 6. The abscissa x defines two glass rectangles that complement each other to provide
undergoes in-plane displacements different than if is attached to the lat- the whole square panel of side L. The diagram shows the resisting system for the
infinitesimal thermal loading of Fig. 5. The dark rectangle is subjected to an infinitesimal
ter (Fig. 5). Compatibility requires that those differential in-plane dis- thermal expansion (dilatation) with respect to the white rectangle, so the latter
placements are zero, which entails generating thermal stresses. From restrains the movement of the former. The restraint consists of infinitesimal in-plane
this, it springs that compatibility engenders the thermal stress to nullify shear stresses dτ exchanged through the common side, i.e. transmitted at x.
310 P. Foraboschi / Materials and Design 134 (2017) 301–319

(while they are zero at the longitudinal edges). The model focuses on implies that dε′ = dε″. Setting Eqs. (27a) and (27b) equal to one
the normal strains and stresses on the cross-section cut along the sym- another:
metry axis (at midspan) of each rectangle on that account (Fig. 7).
The resultant force of the infinitesimal shear stresses from the longi- E  α  L  dT dF dF
− ¼ ð28Þ
tudinal side to the midpoint of a common side shared by the two rectan- 4 x ð1−xÞ
gles of a pair is denoted by dF (Fig. 7); obviously dF(x). Note that dF(x) is
the maximum value of the in-plane resultant force produced by the The process is independent of the path taken; i.e., the differentials
shear stresses dτ on a common side (the value of the in-plane resultant are path-independent. Rearranging:
force at the longitudinal side is zero).
According to the assumptions of Subsection 3.1.6, dF induces a dF E  α  L
¼  x  ð1−xÞ ð29Þ
straight strain profile on the cross-section of each rectangle of the pair. dT 4
The strain in the midspan cross-section of the rectangle with higher
temperature is equal to the thermal strain plus the strain induced by the
infinitesimal force dF. At the midpoint of the common side that sum 3.3.3. Maximum thermal stress in the glass panel
gives rise to dε′ (Fig. 7): The analysis of the stresses induced by dF shows that the maximum
tensile stress occurs at x = 0. That stress, which is denoted by σe (as for
the radial gradient), is equal to (Fig. 7):
4  dF
dε0 ¼ α  dT− ð27aÞ
ExL 2  dF
dσ e ¼ ð30Þ
xL

The strain in the midspan cross-section of the other rectangle of the


On the contrary the opposite edge of the glass panel (x = 1) is
pair is induced only by the infinitesimal force dF. At the midpoint of the
subjected to compression stress.
common side that force gives rise to the strain dε″ (Fig. 7):
Concisely, dT induces dF, which, in turn, induces a stress state whose
maximum is σe given by Eq. (30), which occurs at the edge with the
4  dF highest temperature.
dε″ ¼ ð27bÞ The side of the glass panel with the highest temperature simulta-
E  ð1−xÞ  L
neously reaches the maximum stress and has the minimum strength.
The maximum temperature gradient tolerated by a glass panel and
The compatibility condition requires that the strains along the com- the shock failure are therefore dictated by the stress at the edge with
mon side shared by the two rectangles of the pair are equal, which the highest temperature. The model provides σe, whereas it can be
used to predict any stress in the glass panel.
Replacing Eq. (30) into Eq. (29):

dσ e Eα Eα
¼− xþ ð31Þ
dT 2 2

3.3.4. Breaking down and inversion of the T(x) relationship


The temperature T is the summation of N terms Ti. This implies that
σe as well as its derivative must be decomposed into the contributions
induced by each of the N components of Eq. (1). As for the radial pattern,
the i-th component of σe is hereafter denoted by σei . Eq. (31) takes on
the form:

dσ ei Eα Eα
¼−  xðT i Þ þ ð32Þ
dT i 2 2

According to Eq. (32), the derivative of σe is the sum of N terms σie,


each one induced by the term Ti(x) of the N terms that compose the
temperature distribution.
Eq. (31) proves that σe is the antiderivatives (primitive function) of
the right-hand side, as well as Eq. (32) proves that each term σei is the
antiderivative of the right-hand side. The antiderivative can be obtained
only once the right-hand side of Eq. (32) is expressed as a function of Ti.
Fig. 7. Total infinitesimal strains (thermal and mechanical strains), and infinitesimal Then, the function (1) has to be inverted, in order to obtain x as a func-
internal forces and couples along the midspan of the resisting system shown in Fig. 6. tion of Ti. Inverting the generic term Ti(x) that composes the function
The two complementary rectangles are cut by the axis of symmetry shown in Fig. 5, so (1) gives:
as to show the behavior of the midspan cross-sections. The diagram shows the strain
profile on each transverse section, induced by the infinitesimal force dF, which is the
  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
resultant force of the infinitesimal shear stresses dτ shown in Fig. 6 at the midpoint of
2 ln T max
i − ln ðT i Þ 2 ln T max i − ln ðT i Þ
the common side. The diagram also shows the thermal strain profile and the total strain x ¼ →x ¼ ð33aÞ
λi λi
profile. The strain profiles indicate the midspan strains ε′ and ε″ on the common side at
x of the pair of rectangles, and the maximum dilatation in the glass panel, denoted by εe,
which is reached at the side with the highest temperature and gives rise to σe (all of where T takes on values of:
them are infinitesimal). The facing side of the square panel is subjected to compressive
stress. The diagram also shows the two infinitesimal forces dF, together with the
internal forces and moments acting on the two midspan cross-sections. T ei ≤T ≤T max
i ð33bÞ
P. Foraboschi / Materials and Design 134 (2017) 301–319 311

3.3.5. Maximum thermal stress at the edge the most severe ring-pattern and stripe-pattern thermal gradients
Plugging Eq. (33a) into Eq. (32) the derivative of the stress at the (axi-symmetric and axial-symmetric thermal loadings, respectively).
edge σe is explicitly expressed as the sum of N terms σie, each one in- Insight into the most demanding thermal loadings gave evidence
duced by the term Ti(x) of the N terms that compose the temperature that slightly different values of ΔT, or even the same ΔT, may entail sig-
distribution: nificantly different temperature profiles, which, in their turn, entail sub-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi stantially different values of the greatest stresses.
 
dσ ei Eα 2 ln T max − ln ðT i Þ E  α Thus, the difference between the maximum and minimum temper-
i
¼−  þ ð34Þ ature ΔT does not define a ring-pattern or a stripe-pattern thermal load-
dT i 2 λi 2
ing; to wit: ΔT cannot be used standalone, because the stresses also
depend on the temperature distribution. The most demanding thermal
The generation of the thermal stress is modeled using the superpo-
loadings obtained with the extended models consisted hence of func-
sition property (Subsection 3.1.4) in the same way as for the radial tem-
tions expressed in the form Eq. (1) or Eq. (5), or else Eq. (4), and not
perature (Subsection 3.2.5). By the same token, the contribution to the
of ΔT values.
maximum thermal stress σe due to the i-th term of Eq. (1) — namely,
That result suggested seeking a simpler way to describe a thermal
the stress σie — is obtained by integrating all the contributions of dTi
loading, in order to make the models easier to use. The results from
from Tie to Timax (Fig. 7):
the extended models revealed that the maximum stresses induced by
8 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi9 a given ΔT are basically governed by the shape of the peak, while the
TZmax
i TZmax
i
<   =
Eα 2 ln T max i − ln ð T i Þ shape of a temperature distribution away from the peak influences mar-
σ ei ¼ dσ ei ¼  1−  dT ð35Þ
2 : λi ; ginally the zones of the stress profiles that reach the greatest values.
T ei T ei That postulation indeed fits the ring-pattern thermal loadings better
than the stripe-pattern thermal loadings, but is sufficiently accurate
The components of Eq. (35) are the two constants E and α which de- for the latter too.
fine the material properties, the 2 · N parameters Timax and λi which de- The results from the extended models also revealed that N = 2 is the
fine the thermal loading, and the N edge temperatures Tie each of which minimum number of terms of Eqs. (1) and (5) that ensures adequate ac-
is given by Eq. (6), i.e. Tie = Timax ⋅ e−λi, and occurs at the edge facing the curacy; i.e., N N 2 improves accuracy only marginally.
edge that reaches the maximum temperature Timax. All the components The two outcomes make it possible to identify the profiles of the
of Eq. (35) are thus known, so that it yields directly the stress at the edge ring-pattern and stripe-pattern thermal loadings using only the shape
σie induced by the i-th term of the function (5). of the peak of each pattern, which gives the opportunity to reduce the
The maximum thermal tensile stress σe induced by the N terms σie is thermal loadings to a relatively limited number of peak shapes, and to
obtained by plugging the N values obtained from Eq. (35) into the define a thermal loading using only two terms, which simplifies the
summation (26). calculations.
The integral of Eq. (35) is the sum of the infinite contributions en-
gendered by the infinite pairs of rectangles, in which one tends to ex-
pand and the complement acts as an elastic restraint. It follows that 4.1. Thermal stresses induced by ring-pattern thermal gradients
the easiest way to solve the definite integral (35) is to split the panel
into a finite number of pairs of rectangles and calculate Δσe produced The extended model was used to calculate the stress profiles in-
by each pair of rectangles. The sum of the Δ σe terms due to all the duced by many axi-symmetric temperature distributions with a suffi-
pairs is σe. ciently complete set of peak shapes. The analysis showed that the
Eq. (35) holds true for the rectangular glass panel as well. In the case most demanding radial thermal loadings can be reduced to four func-
of a circular glass panel, Eq. (35) can be applied to the square with side tions (Fig. 8), all of them with the highest temperature at the center of
equal to two-thirds of the diameter.

4. Application of the mathematical models to the glass panel

This section presents the results obtained from the two models and
the following section the discussion of these results.
The models were applied to the glass panel subjected to the most se-
vere thermal gradients. The objective of that analysis was threefold —
namely, 1 - to evaluate the maximum tensile stress that can be induced
by given values of ΔT, 2 - to determine if, and under which conditions, a
linear relationship can be established between the maximum tensile
stress and ΔT; 3 - to ascertain whether a thermal loading alone can
cause the glass panel to crack only at the edges or also away from the
edges.
That analysis implied performing two preliminary tasks in succes-
sion. First, the mathematical formulations of Subsections 3.2 and 3.3
were extended to enable predictions of the stresses throughout the
whole glass panel (extended models), while Eqs. (25) and (35) only pre-
ρ
dict the stresses at the edges. The extended models were obtained sim-
Fig. 8. Temperature T, in °C, as a function of the normalized coordinate ρ (i.e., ρ = 0
ply by modifying the limits of integration of Eqs. (22) and (35). They are indicates the center of the glass panel, while ρ = ± 1 indicates the edges). Most
not presented herein since the results (shown in the following) have demanding axi-symmetric variable temperature profiles: four radial thermal gradients
proven they are of limited use, so that professional practitioners shall described by the function (5), with N = 2. The parameters that define each two-term
use the extended models only occasionally. function are shown in Table 1, which also shows the maximum thermal stress at the
edges σe (last column). Each temperature profile ranges from 60 °C at the center to 20
Then, the extended models were applied to determine the tempera- °C at the edges (ΔT = 40 °C). Symbols in the graph: F = flat peak function; R = round
ture profiles that induce the greatest tensile stresses (most demanding hill peak function; S = sharp peak function; P = pointed peak function. The function F
thermal loadings), in order to provide the thermal stress analysis with is the most severe.
312 P. Foraboschi / Materials and Design 134 (2017) 301–319

the glass panel: flat peak (F), round hill peak (R), sharp peak (S) and In order to define an operative framework adapted to the sphere of
pointed peak (P). design, the model was used to determine when the linear assumption
The functions F, R, S and P cover all the radial variations of tempera- guarantees sufficient accuracy. The answer derived directly from the
tures that induce the greatest stresses either at the edges or inside the paramount role performed by the peak shape of the temperature distri-
glass panel, for any significant value of ΔT. bution in the stresses at the edges.
Ultimately, those four functions, which basically differ from each The formula σe2 = σe1 · ΔT2/ΔT1 (proportional relationship) accu-
other in the sharpness of the peak, represent the most demanding radial rately predicts the stress at the edges σe2 induced by ΔT2 if the peak of
thermal gradients that a glass panel can be subjected to. the radial thermal gradient whose difference between maximum and
The comparison between the thermal stress profiles obtained from minimum temperature is ΔT2 has a shape that is similar to the shape
the extended model for those four functions and the glass strength of the peak of the radial thermal gradient whose difference between
throughout the panel has proven that cracking is dictated by the stress- maximum and minimum temperature is ΔT1, which induces the edge
es and strength at the edges, while stresses and strength away from the stress σe1. The degree of accuracy that is achieved using that formula
edges perform no role in glass rupture even under the most severe ther- for the functions F is shown in Table 2 (Fig. 9).
mal loadings. That result prompted a systematic use of the model devoted to
It follows that cracking can be dictated by the stresses and strength obtaining σe induced by the peaks that the radial temperature distribu-
away from the edges only if the thermal stresses act combined with sub- tions can exhibit independently of ΔT. In so doing, the results consisted
stantial mechanical stresses. Then again, each load combination related of σe/ΔT ratios associated with a specific peak shape. The model was
to environmental loads has usually no live load as part of the loading hence used to conduct a systematic application to figure out σe induced
(i.e., the load factor on the live loads that is typically specified for ther- by axi-symmetric temperature distributions, which extended far be-
mal stress analysis is zero or, however, low [41]). yond the limits covered by the functions F, R, S, P and included a wide
According to that result, stresses and strength away from the edges, range of peak shapes that a glass panel can be subjected to, due to differ-
as well as the extended model, were ignored in all the subsequent anal- ent ways of sun irradiations and cast shadows (Fig. 10).
yses of this Subsection, which only employed the formulas of Subsection That application was intended to make σe available to all for the pos-
3.2, as previously anticipated. sible practical cases (Table 3; curves F, R, S, P of Fig. 8, and a, b, c, d, e, f, g,
Mechanical stresses at the edges can be significant only if they are in- h of Fig. 10).
duced by a mechanical loading that both acts in the plane of the panel Once the given temperature distribution is associated to the function
and has a considerable load factor, while lateral mechanical loadings of Fig. 8 or Fig. 10 with the most similar peak shape, the actual value of
usually induce no more than moderate edge stresses regardless the σe can be obtained by multiplying σe found in the relevant table by the
load factor. ratio between the actual ΔT and ΔT = 40 °C (i.e., by using the above lin-
Ultimately, combined thermal and mechanical stresses are impor- ear equation). The use of Tables 1 and 3 together with Figs. 8 and 10
tant only if the latter are mainly induced by in-plane dead load, which shall allow σe to be obtained for almost any radial gradient.
is not the case in most of the applications of structural glass. As shown by the relevant figures, the peak of those profiles occurs at
The maximum stresses at the edges predicted by Eqs. (25) and (26) the center. However, the results can also be applied to profiles with peak
for those four thermal profiles with ΔT = 40 °C (Fig. 8) are shown in away from the center, as long as ρ = 1 is taken as the minimum distance
Table 1, which also furnishes the parameters that describe each function between the peak and the edges.
Ti(ρ), so that anyone can repeat the calculations and get the same out- The most important case not directly described by Eqs. (25) and (26)
comes. Table 1 shows that the greatest stress at the edges is induced is the glass panel with a small circular shadow. When a panel is subject-
by the flat-peak function; i.e., the maximum tensile stress of the four ed to solar irradiance the temperature increases in the whole glass ex-
maxima is that of F, and is equal to σe = 12.8 N/mm2. Function F is cept for the circular portion in shadow. The temperature of the glass
hence the most demanding radial thermal gradient. that has been subjected to irradiance and the temperature of the circle
The model was then applied to calculate the maximum stress at the in shadow can be considered as uniform, so that there is a temperature
edges σe induced by functions T(ρ) with different ΔT. That analysis jump between the circle and the rest of the panel. The measure of that
aimed at determining the degree of non-linearity reached by the rela- temperature discontinuity is denoted by ΔT.
tionship between σe and ΔT (it is to note that the Subsection 3.1.4 ap- The entire glass panel expands except for the circle, so that the latter
plies only to predetermined temperature profiles). As was expected elastically restrains the former, which engenders tensile stresses. If the
from what referred to at the beginning of this Subsection, the analysis circular shadow is tangent, intersects or is close to one edge, it causes
results have shown that the σe–ΔT relationship regularly exhibits significant tensile stresses just where glass tensile strength has the low-
large deviations from linearity due to the strong dependence of the tem- est value.
perature profile on ΔT. That case is described by the same mathematical formulation of
Subsection 3.2.2 as long as the radius of the circular shadow is assumed
to be so small that it can be neglected (ρ approaches zero). The circle in
shadow is therefore subjected to constant tensile stresses σθ and σr (hy-
drostatic bi-dimensional tensile stress state):
Table 1
Maximum thermal stress at the edges, σe, induced by each of the four radial thermal gradi- Eα
σθ ¼ σr ¼  ΔT ð36Þ
ents shown in the graph of Fig. 8, each one of them ranges from 20 °C at the edges to 60 °C at 2
the center of the glass panel (ΔT = 40 °C). 1st column: symbol that identifies each type of
function (F = flat peak; R = round hill peak; S = sharp peak; P = pointed peak). 2nd–5th col-
umns: coefficients that define the two-term function (5) for each axi-symmetric variable
temperature profile (N = 2). 6th column: factor Φ of Eq. (10). Last column: σe. Eq. (36) provides the maximum tensile stress in the glass panel, be-
cause the glass outside the circle is subjected to compressive circumfer-
Tmax
1 λ1 Tmax
2 λ2 Φ σe
ential stresses and radial tensile stress that decrease moving from the
K / K / K N/mm2 circle. Part of the tensile stress state described by Eq. (36) acts on the
F 332.5159 0.1322 0.6342 0.915 2.0000 12.8 portion of glass shared between the circle and the edge. If σθ of Eq.
R 288.5800 0.0302 44.5700 1.220 13.5000 10.6 (36) reaches fge the glass panel cracks.
S 293.2086 0.0002 39.9414 20.5758 0.0 1.24 The result of Eq. (36) does not depend on the radius of the shadow,
P 299.0700 0.0200 34.0780 15.5000 0.0 3.21
as long as it is small. Eq. (36) proves that the tensile stress per difference
P. Foraboschi / Materials and Design 134 (2017) 301–319 313

Table 2
Maximum thermal stress at the edges, σe, induced by each of the three radial thermal gra-
dients shown in the graph of Fig. 9, which range from 10 °C at the edges to 30, 50, 70 °C at
the center of the glass panel (ΔT = 20, 40, 60 °C, respectively). 1st column: ΔT. 2nd–5th col-
umns: coefficients that define the two-term function (5) for each axi-symmetric variable
temperature profile (N = 2). 6th column: factor Φ of Eq. (10). Last column: σe.

Tmax
1 λ1 Tmax
2 λ2 Φ σe

K / K / K N/mm2

ΔT = 20° 297.0595 0.0550 6.0905 1.1820 2.1 6.0


ΔT = 40° 321.1048 0.1320 2.0452 0.9150 2.0 12.6
ΔT = 60° 301.9048 0.1320 41.2452 0.8150 21.2 17.8

of temperature between the portion in shadow and the portion subject-


ed to solar irradiance is constant:

70; 000  9  10−6 N ρ


σθ ¼ σr ¼ ¼ 0:315 ð37Þ
2  ΔT mm2  ° C
Fig. 10. Temperature T, in °C, as a function of the normalized coordinate ρ. Each axi-
symmetric variable temperature profile ranges from 20 °C to 60 °C: eight radial thermal
At the most basic level, each degree of temperature discontinuity be- gradients, identified by the labels a, b, c, d, e, f, g, h. Every curve is described by the
tween the part of the circular portion of glass covered by the shadow function (5), with N = 2. The parameters that define each two-term function are shown
and the rest of the glass panel induce 0.315 N/mm2 in the circular in Table 3, which also shows the maximum thermal stress at the edges σe (last column).
glass portion.
= σe1· ΔT2/ΔT1 can be used for the stripe-pattern temperature distri-
4.2. Thermal stresses induced by stripe-pattern thermal gradients butions too. That is, if the peaks of the stripe-pattern thermal gradients
whose difference between maximum and minimum temperature are
As is the case with axi-symmetric thermal loadings, the results from ΔT1 and ΔT2 have similar shapes, σe2 induced by ΔT2 can be obtained
the extended model obtained for the most demanding axial-symmetric from σe1 induced by ΔT1 using the above proportional relationship.
thermal loadings have proven that thermal cracks are dictated by the The accuracy that can be achieved is lower than for the ring-pattern gra-
stresses and strength at the edges, while the stresses and strength dients being equal the degree of similarity between the two peaks, but
away from the edges take no role in glass rupture, unless the thermal acceptable.
stresses act in combination with substantial mechanical stresses in- The model of Subsection 3.3 was then used to define analytically the
duced by dead load (which is typically not the case in structural glass). stripe-pattern gradient that induces the maximum tensile stress at one
Thus, stresses and strength away from the edges were ignored, and edge, for a given ΔT; i.e., to identify the most severe axial-symmetric
the subsequent analyses of this Subsection only employed the formulas thermal loading T(x) for a predetermined difference between maximum
of Subsection 3.3, as previously anticipated. and minimum temperature ΔT.
Likewise, mechanical stresses at the edge resulted to be negligible in A thermal loading is defined by the position of the maximum, which
the vast majority of the cases, for the same reason as the radial is hereinafter identified by the coordinate xm, and by the parameters
gradients. T1max, λ1, T2max, and λ2 (since N = 2).
The σe–ΔT relationship was also found to exhibit large deviations The values of xm, T1max, λ1, and λ2 can be obtained solving the
from linearity. Considering that the reason for non-linearity is the following system of four equations:
same as for radial gradients, the selfsame proportional relationship σe2
8  
>
> ∂σ e xm ; T max max
1 ; λ1 ; T 2 ; λ2
>
> ¼0
>
> ∂xm
>
>
>
>  
>
> ∂σ e xm ; T max max
>
> 1 ; λ1 ; T 2 ; λ2
>
< max ¼0
∂T 1
ð38Þ
> ∂σ x ; T max ; λ ; T max ; λ 
>
>
> e m 1 1 2 2
>
> ¼0
>
> ∂λ1
>
>  
>
>
>
> ∂σ e xm ; T max max
1 ; λ1 ; T 2 ; λ2
>
: ¼0
∂λ2

The value of T2max can be obtained from Eq. (4) by replacing Tg = ΔT.
The solution of the system (38) for ΔT = 40 °C is:

xm ¼ 0; T max
1 ¼ 293:15 K; λ1 ¼ 0:0; T max
2 ¼ 40:00 K; λ2 ¼ 1096:0 ð39Þ

ρ
The function described by those parameters (Eq. (39)) is shown in
Fig. 9. Temperature T, in °C, as a function of the normalized coordinate ρ. Three axi- Fig. 11. Inserting that function into the model, the Eqs. (35) and (26)
symmetric variable temperature profiles: radial thermal gradients described by the give σe = 12.17 N/mm2. This value is the maximum stress that a stripe-
function (5), with N = 2. The parameters that define each two-term function are shown pattern temperature profile with ΔT = 40 °C can induce in a glass panel.
in Table 2, which also shows the maximum thermal stress at the edges σe (last column). The solution xm = 0 was found to be independent of ΔT. Ergo, the
Every profile in this graph has the shape of the flat peak curves (Fig. 8; symbol F). Each
profile starts from 10 °C at one edge and reaches its own peak at the center of the glass
profiles of the most severe stripe-pattern thermal loadings have the
panel, which is 30, 50, 70 °C. Namely, each curve has its own ΔT; i.e., 20, 40, 60 °C, maximum at one edge and the minimum at the facing edge (i.e., the
respectively. most demanding axial-symmetric temperature distributions for a
314 P. Foraboschi / Materials and Design 134 (2017) 301–319

Table 3 thermal cracking and thermal shock, depending on ΔT and fge, as well
Maximum thermal stress at the edges, σe, induced by the eight radial thermal gradients as on the actual temperature distribution for a given ΔT.
shown in the graph of Fig. 10, each one of them ranges from 20 °C at the edges to 60 °C
at the center of the glass panel (ΔT = 40 °C). 1st column: label of each curve (Fig. 10).
That result highlights the importance of thermal stress assessment
2nd–5th columns: coefficients that define the two-term function (5) for each axi-symmet- for annealed glass both with and without mechanical stresses.
ric variable temperature profile (N = 2). 6th column: factor Φ of Eq. (10). Last column: σe. Considering the product of the upper bound multiplied by the possi-
ble ΔT, tempered and chemically-strengthened glass subjected to ther-
Tmax
1 λ1 Tmax
2 λ2 Φ σe
mal loadings alone do not suffer from thermal cracking and thermal
K / K / K N/mm2
shock as long as the percent deviation of the actual edge strength
a 34.6194 0.0050 298.5306 0.1450 260.0 12.4 from the nominal edge strength is moderate (i.e., provided that the de-
b 285.5711 0.0250 47.5789 1.1820 15.0 10.6 crease in precompression at the edges is not irregular). For tempered
c 287.6793 0.0001 45.4707 2.1450 5.5 8.4
d 292.1259 0.0006 41.0242 3.5550 1.2 6.4
glass, the requirement that fge is not excessively lower than fg is typically
e 294.0661 0.0038 39.0839 6.1520 0.2 4.3 met, while fge of chemically-strengthened glass is sometimes far less
f 319.1574 0.0850 13.9926 41.1500 0.0 8.3 than expected (Subsection 3.1.5).
g 299.1318 0.0202 34.0182 100.0000 0.0 2.1 Ultimately, tempered and chemically-strengthened glass suffer from
h 295.2092 0.0070 37.9408 450.0000 0 0.7
thermal cracking and thermal shock only if they are simultaneously
subjected to substantial thermal and mechanical loadings, which is a
load combination that needs in-plane dead load, while otherwise it is
given ΔT imply that the temperature at one transverse side minus the not typical.
temperature at the other transverse side is equal to ΔT). Considering the product of the upper bound multiplied by the possi-
As done for the ring-pattern temperature distributions, the model ble ΔT, heat-strengthened glass subjected to thermal loadings alone
was used to conduct a systematic application in order to figure out σe in- does not suffer from thermal cracking or thermal shock as long as the
duced by the possible peak shapes of the stripe-pattern temperature ratio between the edge strength and the strength away from the
distributions independently of ΔT. Each result consisted thus of a σe/ edges is not considerably low. Unfortunately, the type and quality of
ΔT ratio associated with a specific peak shape. That application was edge working, and the actual permanent precompression that can be in-
intended to make σe available to all for the possible practical cases duced at the edges weigh on that ratio much more than for tempered
(Table 4), from the most demanding (Fig. 11) to the less demanding glass. Likewise, the percent deviation of the actual edge strength from
(Fig. 12). the nominal edge strength weighs much more.
Once the given temperature distribution is associated to the function As a result, fge of heat-strengthened glass could turn out to be lower
of Fig. 11 or Fig. 12 with the most similar peak shape, the actual value of than thermally-induced stresses. Because of that, heat-strengthened
σe can be obtained by multiplying σe found in Table 4 by the ratio be- glass needs information about the actual strength of the edges, especial-
tween the actual ΔT and ΔT = 40 °C (to wit, by using the above linear ly if ΔT is high or the mechanical stresses at the edges are substantial.
equation). The use of Table 4 together with Figs. 11 and 12 shall allow In brief, annealed glass always requires an assessment: it must be
σe to be obtained for almost any stripe-pattern thermal gradient. checked for thermally-induced stress at design loadings, to guarantee
that the maximum thermal stress plus the mechanical stress acting si-
multaneously if existing is lower than fge. Tempered glass and chemical-
5. Discussions ly-strengthened glass typically require an assessment of thermal stress
only when the thermal loading acts in combination with mechanical
The maximum tensile stress at the edges σe induced by a thermal loadings that induce substantial mechanical stresses. Heat-strength-
loading strongly depends not only on whether the temperature distri- ened glass lies somewhere in between; it may require an assessment
bution whose total gradient is ΔT is a ring-pattern or a stripe-pattern even if it is subjected to thermal loading alone.
gradient, but also on the profile of the temperature distribution. That
dependence can be expressed using the shape of the peak in lieu of
the entire function, but cannot ever be ignored, because a constant of
proportionality between σe and ΔT can only lead to largely inaccurate
predictions of σe.
It follows that all empirical formulas, which can be nothing but
linear, are radically inaccurate. Moreover, all code formulas, which are
empirical and must err nothing but on the side of caution, must be dras-
tically over-conservative.
When accurate predictions have to be obtained, thermal stresses
must be derived from analytical modeling (Subsections 3.2 and 3.3).
When the peak shape of the thermal loading is included among the
graphs of Figs. 8, 10–12, σe can be obtained by using Tables 1, 3, 4.
The σe/ΔT ratios lie between an upper bound and a lower bound. The
upper bound is 0.32 N/mm2·°C−1, to wit: no temperature distribution
can induce an edge stress greater than 0.32 N/mm2 per degree of tem-
perature difference between two points of the glass panel.
The lower bound depends on whether the temperature distribution
is a radial or a stripe-pattern thermal loading (Subsections 5.1 and 5.2).
Fig. 11. Temperature T, in °C, as a function of the normalized coordinate x (i.e., x = 0 and x =
In both the cases, however, the range is too wide to ignore the actual 1 indicate the transverse facing edges of the glass panel). Most demanding stripe-pattern
value of σe and replace it with the upper bound. temperature profile with ΔT = 40°. The graph shows the axial-symmetric variable
Considering the product of the upper bound of the σe/ΔT ratios mul- temperature profile that induces the greatest possible value of the tensile stress when T =
tiplied by the possible difference between the highest and lowest tem- 60 °C at the edge that reaches σe (upper edge of Figs. 5–7; i.e., x = 0) and T = 20 °C at the
facing edge (lower edge; i.e., x = 1), which is subjected to compressive stress. The thermal
peratures within a glass panel ΔT, the maximum thermal stress at the gradient is described by the function (1), with N = 2. The parameters that define the two-
edge σe can be up to ten times the allowable edge stress fge of annealed term function are shown in the text, i.e. Eq. (39); the maximum thermal stress at the edge
glass. Stresses can thus lie within or beyond the threshold dictating σe is shown in the test too.
P. Foraboschi / Materials and Design 134 (2017) 301–319 315

Table 4 5.1. Ring-pattern thermal gradients


Maximum thermal stress at the edge, σe, induced by the eight stripe-pattern thermal gra-
dients shown in the graph of Fig. 12, each one of them ranges from 60 °C at one edge (the
edge that reaches σe) to 20 °C at the facing edge of the glass panel (the edge that is subject-
The maximum tensile stresses at the edges are induced by the distri-
ed to compressive stress). Namely, ΔT = 40 °C for every axial-symmetric variable temper- butions with the highest temperature at the center of the glass panel.
ature distribution of this table. 1st column: label of each curve (Fig. 12). 2nd–5th columns: Nevertheless, substantial tensile stresses at the edges are also induced
coefficients that define the two-term function (1) for each axial-symmetric variable tem- by a small circular shadow nearby an edge (cast shadow).
perature profile (N = 2). 6th column: factor Φ of Eq. (10). Last column: σe.
The flatter the peak of the temperature distribution the greater σe,
Tmax
1 λ1 Tmax
2 λ2 Φ σe and vice versa.
K / K / K N/mm2 The lower bound is close to zero. The sharper the peak the closer σe
to the lower bound independently of ΔT. It is possible to say that σe
a 293.4433 0.0010 39.7067 1106.0000 0 12.1
b 298.4148 0.0178 34.7352 386.0000 0 11.0
ranges from zero to ΔT × 0.32 N/mm2.
c 304.3518 0.0375 28.7982 186.0000 0 9.7 It is to remark that codes and recommendations do not deal
d 322.5262 0.0955 10.6238 123.0000 0 6.3 adequately with radial thermal gradients and circular shadows.
e 309.8798 0.5550 23.2707 223.0000 0 8.7
f 319.3170 0.0855 13.8330 423.0000 0 7.0
g 328.2201 0.1130 4.9299 43.0000 0 5.2
5.2. Stripe-pattern thermal gradients
h 332.1825 0.1250 0.9675 53.0000 0 4.5
The maximum tensile stresses at the edges are induced by the distri-
butions with the highest temperature at one edge of the glass panel and
the lowest temperature at the facing edge. The maximum tensile stress
occurs at the edge with the highest temperature.
Heat-strengthened glass is very commonly used worldwide for the In the vast majority of the cases, the sharper the peak of the temper-
plies of laminated glass as it gives the best tradeoff for glass strength ature distribution the greater σe, and vice versa. The greatest tensile
and a favorable break pattern. Yet the upper bound of thermal stresses stresses at the edges are hence induced by the temperature distribu-
raises the interesting dilemma where a weaker heat-strengthened ply tions with a sharp peak at one edge. The stripe-pattern thermal gradi-
is more likely to break than tempered glass, but, when all the glass ents imply thus a thermal behavior opposite to the ring-pattern
plies in a laminated glass unit are broken, the larger particle sizes of a thermal gradients.
heat-strengthened ply will offer better post-breakage load-carrying ca- It follows that a rectangular shadow with one side that coincides
pacity than tempered glass. with a side of the glass panel and the facing side of the rectangle inside
The solution to that dilemma must consider all facets of the structur- the glass and quite close to the facing side of the square panel is the
al behavior of architectural laminated glass [12]. On one hand, breakage most dangerous (i.e., a large shadow that leaves a rather narrow
of tempered glass is a safe failure mode, because, if tempered glass unshadowed stripe of glass).
breaks, it will shatter into a mode of ‘dice’ about the size of cubes of It is to remark that some codes state that cracking occurs at the edge
the glass thickness, while breakage of heat-strengthened glass is a dan- with the lowest temperature, and that the sharper the peak the lower σe
gerous failure mode, because if heat-strengthened glass breaks, it will (and vice versa). The truth is exactly the opposite, while those state-
fracture into large, sharp pieces that can become glass projectiles. On ments are wrong.
the other hand, the post-breakage load-carrying capacity of tempered Some codes also state that a narrow rectangular shadow with one
glass is marginal, because it offers negligible resistance to falling out as side that coincides with the side of the panel is one of the most severe
a complete ‘wet blanket’, while the post-breakage load-carrying capac- thermal loading. Again, the truth is exactly the opposite: the most se-
ity of heat-strengthened glass is significant, because when heat- vere rectangular shadow is the large one, while the narrow one induces
strengthened glass breaks, it typically has a few fully developed cracks compressive stress at the edge in the shade and marginal tensile stress
and most, if not all, of the pieces have one or more edges connected to at the facing edge of the glass panel. The narrow shadow could fracture
the frame.
After weighing up advantages and disadvantages, the author
suggests employing heat-strengthened glass for one of the layers of
laminated glass systems, although heat-strengthened glass does not
cost less than tempered glass and that layer may entail an extra lamina-
tion. The layers whose broken glass shards can fall off and drop to the
ground must be made of tempered glass, and the layers subjected to im-
pacts must be made of tempered glass too. But one of the other layers of
laminated glass systems should be made of heat-strengthened glass,
which implies that, if necessary, glass design should add a specific
inner layer made of heat-strengthened glass to the laminated glass unit.
That layer can be made of annealed glass instead of heat-strength-
ened glass, because, when broken, annealed glass will have a break pat-
tern of relatively large pieces, similar to heat-strengthened glass.
Designers often entertain the thought of using annealed glass in lieu
of heat-strengthened glass (sometimes in lieu of tempered glass too), as
it gives the best tradeoff for a favorable structural behavior and low cost.
In fact, it consists in the standard float glass product that has been slow-
ly cooled after forming in the molten tin float bath, without any
precompression process. So, it is relatively cheap. Fig. 12. Temperature T, in °C, as a function of the normalized coordinate x. The graph
As a result of this research, annealed glass can be used in lieu of shows eight stripe-pattern thermal gradients with ΔT = 40 °C, identified by the labels a,
heath-strengthened glass only if design and assessment are based on re- b, c, d, e, f, g, h. Every axial-symmetric variable temperature profile ranges from 60 °C at
one edge (where the tensile stress reaches the maximum) to 20 °C at the facing edge
liable predictions of the thermal stresses, while the design thermal load- (subjected to compression). Each curve is described by the function (1), with N = 2.
ings prescribed by the general structural codes together with finite The parameters that define each two-term function are shown in Table 4, which also
element analyses pose risk to designers. shows the maximum thermal stress at the edges σe (last column).
316 P. Foraboschi / Materials and Design 134 (2017) 301–319

the glass panel only if the facing side of the shadow were so close to the across the facing side cannot be substantial, although the low thermal
edge that the glass strength is equal to fge throughout the entire area conductivity of glass.
covered by the rectangular shadow (i.e., the width of the rectangular The lower bound is approximately ΔT × 0.10 N/mm2 (a little bit
shadow is less than five times the thickness of the ply) and the jump lower). The flatter the peak the closer σe to the lower bound indepen-
of temperature across the facing side were substantial (i.e., N 15 °C). dently of ΔT. Hence, σe ranges from ΔT × 0.10 N/mm2 to ΔT ×
But if the width of the shadow is so small the jump of temperature 0.32 N/mm2.

6. Practical application

The formulation was applied to the glass panels of an Italian swimming-pool center whose envelope was a glazing system composed of large glass
panels (insulating double glazing), supported by a frame system composed of multi-chamber hollow aluminum profiles (a thermally insulated self-
supporting aluminum façade system).
Shortly after construction twenty-three glass panels of the building envelope cracked. At the suggestion of the designers, manufacturer, builder,
and an Italian glass research center, they were replaced with identical glass panels. But they cracked once again in a relatively short time. Yet again
they were replaced with identical glass panels. As should have been expected they cracked once more, after few months (Fig. 13).
This time, the cracked panels were not replaced. The decision taken by the owners of the building was to entrust the author with the task of iden-
tifying the cause of the systematic failure of the glass panels, performing a structural assessment to determine if unsafe structural conditions were
present in the whole glazing system, and redesigning the elements that had cracked as well as the elements that would have turned out to be inad-
equate. This section presents the first of those activities — namely, the identification of the root cause of the systematic cracking — while the other
activities will be presented in a future paper devoted to engineering failure analysis.
Nine of those twenty-three glass panels had square shape with sides of length 2.12 m, and the other fourteen had rectangular shape with dimen-
sions 2.12 m and 3.37 m, or 1.32 m and 2.86 m.
Each glass panel of the building envelope system consisted in an insulating glass unit, inserted into the aluminum frame with adequate clearances
between the vertical or horizontal glass edges and the glazing frame (large gaps between the glass edges and the aluminum framing).
Every insulating glass unit consisted in insulating laminated glass, composed of a monolithic glass ply on the exterior side, hermetically sealed
space in the middle, and a two-layer laminated glass on the interior side. The external lite was a 8 mm tempered glass. The space in the middle
was a 13.2 mm gap filled by argon. Each lite of the two-layer internal laminated glass was a 4 mm annealed glass, while the interlayer was a
0.76 mm polyvinylbutyral sheet.
All the lites were finished with polished edges. Both the lites of the laminated system were made of low-emissivity glass (low-E glass).
Cracks had happened in the annealed laminated glass on the interior of the building, while the tempered monolithic glass on the exterior did not
crack. In all the twenty-three panels both the plies that composed the laminated glass cracked. Crack initiation and propagation in the two plies of the
laminated glass were identical to one another. Cracking had initiated at the middle point of the bottom edge and had propagated into several
branches toward the other edges of the glass panel (a branch passed through the center). The glass plies had broken off into large, sharp shards,
which the interlayer held in place (Fig. 14).
The author established immediately the cause of cracking, without the need to use the models of this paper. Cracking and glass breakage had de-
rived from excessive thermal stresses at the bottom edge (Figs. 13, 14). The models were used to perform a quantitative analysis and assessment ac-
cording to the consequence classes, in order to ascertain whether the glass panels had an acceptable level of safety against thermal failure or the risk
of collapse was excessive.
The digital image correlation technique identified no major flaws in the glass plies [42]. The tensile strength of the annealed glass at the edges,
obtained from the material tests performed on the pieces of broken glass, resulted to be fge = 7.6 N/mm2.
That value of fge allowed for glass type (annealed), edge working (polished), duration of thermal loading (1 h and a half), and sensitivity to ther-
mal shocks (the standard tests showed that the glass had no particular sensitivity to thermal cracking), of the glass that the cracked plies were made
of [23]. It was the average value of the results from the tests and did not include any partial safety factor. Thus, 7.6 N/mm2 was the expected glass
strength at the edges (most probable value), not the design glass strength.
The structural analysis was first directed at determining the thermal loadings experienced by the glazing system. The glass panels that had
cracked were all at the inferior level of the building envelope. As such, the lower side of those glass panels was supported by aluminum beams at
the inferior level of the façade system, which were in contact with the reinforced concrete structures of the foundation. The building foundation
prevented those aluminum beams from changing fast the temperature, and the latter, in turn, prevented the bottom edge of those glass panels
from changing fast the temperature.
The high thermal inertia of the bottom edge of the glass panels that had cracked with the addition of plies made of low-E glass implied that solar
irradiance caused the temperature to increase in a highly uneven way throughout those glass panels. The glass panels that had cracked had thus been
exposed to rather high thermal gradients [22].
The thermal loading that the author choose as being representative for the highest temperatures really experienced by the glass panels that
had cracked was a radial temperature distribution (ring-pattern gradient) with the profile of the curve R (Fig. 8), defined by the following parameters
(N = 2): T1max = 279.8382 °C; λ1 = 0.0302; T2max = 73.3118 °C; λ2 = 1.2200; Φ = 22.3 °C.
Plugging T1max = 279.8382 °C and λ1 = 0.0302 as well as T2max = 73.3118 °C and λ2 = 1.2200 into Eq. (6) gives:

T 1e ¼ 279:8382  e−0:0302 ¼ 271:5134 °C; T 2e ¼ 73:3118  e−1:2200 ¼ 21:6439 °C ð40Þ

Plugging those values, i.e. T1max = 279.8382 °C, T2max = 73.3118 °C, T1e = 271.5134 °C and T2e = 21.6439 °C, into Eq. (7) gives:

ΔT ¼ 279:8382 þ 73:3118–271:5134–21:6439 ¼ 59:993 °C ð41Þ


P. Foraboschi / Materials and Design 134 (2017) 301–319 317

Fig. 13. Swimming-pool center located in Italy. Photos of 2 of the 23 façade glass panels that cracked systematically. Each glass panel was an insulating laminated glass unit: the exterior
component was a fully tempered monolithic glass; the interior component was a two-layer annealed laminated glass.

Hence, the difference between the maximum temperature (at the center of the glass panel) and the minimum temperature (at the edges) from
Eq. (41) turned out to be ΔT = 60 °C. The glass panels that had cracked, in all likelihood, underwent such thermal gradient, although rather high,
whereas the other glass panels experienced lower thermal gradients.
Plugging T1max = 279.8382 °C, λ1 = 0.0302 and T1e = 271.5134 °C into Eq. (25) yields σ1e:

70000  9:0−6     
σ e1 ¼  ln ð279:8382Þ  279:8382  1−e−0:0302 −279:8382  ln ð279:8382Þ þ 271:5134  ln ð271:5134Þ þ 279:8382  1−e−0:0302
0:0302
N
σ e1 ¼ 20:8609  ð46:9035−1576:6674 þ 1521:5642 þ 8:3248Þ ¼ 2:609
mm2
ð42Þ

Plugging T2max = 73.3118 °C, λ2 = 1.2200 and T2e = 21.6439 °C into Eq. (25) yields σ2e:

70000  9:0−6     
σ e2 ¼  ln ð73:3118Þ  73:3118  1−e−1:2200 −73:3118  ln ð73:3118Þ þ 21:6439  ln ð21:6439Þ þ 73:3118  1−e−1:2200
1:2200 ð43Þ
N
σ e2 ¼ 0:5164  ð221:8996−314:8539 þ 66:5489 þ 51:6680Þ ¼ 13:045
mm2

Plugging σ1e and σ2e into Eq. (26) yields the stress at the edge σe induced by the thermal loading:

N
σ e ¼ σ e1 þ σ e2 ¼ 2:609 þ 13:045 ¼ 15:655 ð44Þ
mm2

The edge stress σe obtained from Eq. (44) occurred at the midpoint of the bottom edge of all the panels that had cracked; i.e., the thermal loading
induced 15.7 N/mm2 in all the 23 square and rectangular glass panels.
As 15.7 N/mm2 N 7.6 N/mm2, the glass could not tolerate the thermal loading chosen as being representative for the highest temperatures really
experienced by the cracked panels. As the capacity-to-demand ratio of every glass panel resulted to be 7.6 / 15.7 = 0.48, which was not only inad-
equate but also low, the model results demonstrated that thermal cracking was an unavoidable event for the insulating glass unit that had been de-
signed and built at the inferior level of the building envelope.
The designers of the glazing system had performed a finite element analysis under ΔT = 40 °C. The maximum stress at the edges from their finite
element model was 4.1 N/mm2. As the designers estimated a glass tensile strength at the edge of 7.9 N/mm2, the glass panels satisfied their thermal
stress assessment procedure. Note that their procedure would have also been satisfied by ΔT = 60 °C.
The maximum edge stress σe in those glass panels calculated by utilizing the draft of the document ‘Glass in building – Thermal stress calculation
method’, which will become a European Standard, is given by the formula σe = kt · E · ΔT · α, in which kt is the shadow coefficient. For those glass
panels that document prescribes kt = 1.1. Thus, σe = 1.1 · 70,000 · 60 · 9 · 10−6 = 41.6 N/mm2.
As 41.6 / 15.7 = 2.65, that document leads to highly over-conservative predictions. Furthermore, by utilizing that document, the entire glazing
system would turn out to be unsafe. On the contrary, the author did not replace the glass panels that had not cracked, since the models demonstrated
that they guaranteed adequate performance against design thermal loadings.

7. Conclusions help engineers, architects and manufacturers choose the most suitable
glass systems and glass types, as well as gain better understanding
This paper has developed increased insights into real-world glass of their designs and components without the need to over-rely on
problems through novel mathematical modeling, new applications, laboratory mock-up test results, which are difficult to do for thermal
and a combination of these. This final section reviews the most signifi- loadings.
cant implications of the information presented in the body of the article. The analytical form of the two models, together with the normaliza-
The easy-to-use closed-form models presented in this paper have tion of the abscissa, has led to general results about thermal stresses.
proven that the common belief (shared by codes) according to which The relationship between the maximum stress at the edges (or in
thermal stresses cannot be tackled analytically was unfounded. The the panel) and the difference in temperature within the glass panel
key point is to represent the thermal behavior with an adequate refer- (between center and edges or between two facing edges) has been
ence structure, governed by realistic basic assumptions. The two models proven to depend heavily on the type of distribution (ring-pattern or
presented in this paper shall allow the glass designers to predict accu- stripe-pattern) and on the shape of the temperature profile (sharp or
rately the thermally-induced stresses. Moreover, those models shall flat peaks).
318 P. Foraboschi / Materials and Design 134 (2017) 301–319

Fig. 14. Cracks always occurred in the internal laminated glass (made of annealed glass), while the external monolithic glass (made of tempered glass) never cracked. Both the plies of the
laminated glass cracked. Crack initiation and propagation in the two layers that composed the laminated glass unit were the same.

That mechanical dependence entails cascade effects. The relationship The upper bound, combined with the temperature excursions expe-
between the maximum stress at the edges (or in the panel) and the dif- rienced by glass panels, entails that a thermal gradient induces thermal
ference in temperature within the glass panel (temperature gradient) is stresses that may lie far beyond the edge strength of annealed glass. In
strongly non-linear. The operative assumption of linearity between stress some cases a thermal gradient acting alone can cause annealed glass
and temperature gradient is largely inaccurate. An empirical approach is to crack under the typical temperature excursions experienced by
inadequate to thermal stress analysis, since the empirical coefficients ig- glass panels, while in other cases design thermal gradients can be safely
nore distribution and profile of temperatures. The combination of an em- tolerated and thermal shock cannot occur.
pirical coefficient and linearity leads to extremely rough formulas. The upper bound reveals a wide gap between safety conditions rec-
In the allowable stress approach of codes, standards and recommen- ognized by codes and real safety, regarding thermal stresses. Annealed
dations, the maximum thermally-induced stress is equal to the difference glass often satisfies the latter but only seldom the former; the same
in temperature multiplied by the elasticity modulus, by the linear tem- more than sometimes occurs to heat-strengthened glass. In the end,
perature expansion coefficient, and by an empirical coefficient that de- the above-mentioned cascade effects due to the influence of tempera-
pends on the incident solar flow and shadowing conditions of the glass ture distribution strongly reverberate in annealed glass and in part in
panel (shadow coefficient). That is, code formulas are empirical, and ac- heat-strengthened glass. Disregarding non-linearity implies leaving lit-
curacy depends on a coefficient whose possible values have been defined tle room to the use of annealed glass and limiting the use of heat-
with no representative tests. Then again, the values of that empirical strengthened glass, which admit no more than little approximation in
coefficient cannot be refined since they disregard the thermal stress anal- predictions.
ysis. Those formulas have therefore to include large safety factors to allow That problem is particularly acute with today's thermally efficient
for uncertainties about the coefficient of proportionality and to compen- glasses using spectrally selective compositions and coatings to control
sate for what is disregarded. As a result, the formulas provided by codes, unwanted solar heat gain, and with the addition of low emissivity coat-
standards and recommendations about structural glass to calculate ings to further improve thermal performance. In those glass types, ther-
thermal stress are overly-conservative. This paper has found an mal stresses due to sun exposition are high and empirical formulas
empirical-to-analytical edge stresses ratio equal to 2.7. almost always predict a maximum value that can only be tolerated by
Besides being greatly inaccurate, although on the side of caution, that tempered glass (or by chemically strengthened glass).
type of formulas is incapable of capturing the critical aspects of the phe- As a result, engineers and architects responsible for building enve-
nomenon and taking into account the different conditions. At the most lope performance design have little latitude in deciding the type of
basic level, those formulas are virtually the same as ignoring thermal glass in compliance with codes, since recommended provisions for the
stresses and referring to allowable thermal gradients, which is exactly structural use of glass often lead to glazing systems entirely made of
what some practitioners do using a best practice approach. tempered glass to prevent breakage under thermal loadings. Unfortu-
The results from the models have confirmed that thermal shock is nately, the post-breakage behavior of that configuration is poor.
typically dictated by the thermal stresses at the edges, while the me- The models presented in this paper have proven that annealed
chanical stresses at the edges are often negligible when thermal loading glass and even more so heat-strengthened glass can be used in many
is the leading variable action. Moreover, thermal stresses away from the applications, including energy efficient glass, as long as thermally-
edges may play a role in failure only if combined with substantial me- induced stresses are accurately predicted and assessed. This paper
chanical stresses, which is not typically the case. aims at providing designers with a new tool which shall replace the
It has been found that the maximum thermal stress induced by radi- overly-conservative prescriptions and inaccurate formulas of codes.
al or stripe-pattern thermal gradients (axi-symmetric or axial-symmet- The paper has shown not only that code predictions about thermal
ric variable temperature profiles) is always lower than 0.32 N/mm2 per stresses are inadequate, but also that some information included
degree of difference between maximum and minimum temperatures into codes and recommendations is wrong, especially regarding the
within the glass panel. shadows, and that some prescriptions imply incorrect design decisions.
The upper bound, combined with the temperature excursions As a final remark, finite element modeling has not been considered
experienced by glass panels, entails that a thermal gradient acting herein since it would have brought forth no advance in knowledge to
alone (i.e., without substantial mechanical loads, which are actually thermal stress analysis. To replace a system of differential equations
negligible in the vast majority of the load combinations that include en- with a system of algebraic equations only allows specific results to be
vironmental loads) cannot cause tempered or chemically-strengthened obtained, with almost no control over the numerical outcomes. Nor
glass to crack. it allows the equations that govern the mechanical behavior to be ob-
The upper bound, combined with high temperature excursions, en- tained, whereas analytical modeling has yielded the general solution
tails that a thermal gradient can cause heat-strengthened glass to that describes the mechanical behavior and can be directly applied to
crack (i.e., also without mechanical loads). any case.
P. Foraboschi / Materials and Design 134 (2017) 301–319 319

References [23] M.B. Kadri, S. Nisar, S. Zia Khan, W.A. Khan, Comparison of ANN and finite element
model for the prediction of thermal stresses in diode laser cutting of float glass,
[1] D. Callewaert, J. Belis, M. Vandebroek, R. Van Impe, Spontaneous failure of a passable Optik – Int. J. Light Electron Optics 126 (19) (2015) 1959–1964 October.
laminated glass floor element, Eng. Fail. Anal. 18 (7) (2011) 1889–1899 October. [24] C. Louter, J. Belis, F. Veer, J.-P. Lebet, Durability of SG-laminated reinforced glass
[2] M.S. Shetty, L.R. Dharani, J. Wei, D.S. Stutts, Failure probability of laminated architec- beams: effects of temperature, thermal cycling, humidity and load-duration, Constr.
tural glazing due to combined loading of wind and debris impact, Eng. Fail. Anal. 36 Build. Mater. 27 (1) (2012) 280–292 February.
(January) (2014) 226–242. [25] Y. Wang, Q. Wang, G. Shao, H. Chen, J. Sun, L. He, K.M. Liew, Experimental study on
[3] E. Speranzini, S. Agnetti, M. Corradi, Experimental analysis of adhesion phenomena critical breaking stress of float glass under elevated temperature, Mater. Des. 60
in fibre-reinforced glass structures, Compos. Part B 101 (2016) 155–166. (August) (2014) 41–49.
[4] Q.D. To, Q.-C. He, M. Cossavella, K. Morcant, A. Panait, J. Yvonnet, Failure analysis of [26] A. Benallal, D. Bigoni, Effects of temperature and thermo-mechanical couplings on
tempered glass structures with pin-loaded joints, Eng. Fail. Anal. 14 (5) (2007) material instabilities and strain localization of inelastic materials, J. Mech. Phys.
841–850 July. Solids 52 (3) (2004) 725–753 March.
[5] E. Axinte, Glasses as engineering materials: a review, Mater. Des. 32 (4) (2011) [27] A.M. Korsunsky, T. Sui, E. Salvati, E.P. George, M. Sebastiani, Experimental and
1717–1732 April. modelling characterisation of residual stresses in cylindrical samples of rapidly
[6] L. Biolzi, S. Cattaneo, G. Rosati, Progressive damage and fracture of laminated glass cooled bulk metallic glass, Mater. Des. 104 (August) (2016) 235–241.
beams, Constr. Build. Mater. 24 (4) (2010) 577–584 April. [28] T. Wejrzanowski, M. Grybczuk, M. Chmielewski, K. Pietrzak, K.J. Kurzydlowski, A.
[7] W. Gao, J. Xiang, S. Chen, S. Yin, M. Zang, X. Zheng, Intrinsic cohesive modeling of Strojny-Nedza, Thermal conductivity of metal-graphene composites, Mater. Des.
impact fracture behavior of laminated glass, Mater. Des. 127 (August) (2017) 99 (2016) 163–173.
321–335. [29] M. Aiello, F. Focacci, A. Nanni, Effects of thermal loads on concrete cover of fiber-re-
[8] W. Smetana, R. Reicher, Preventing failure of soda lime cover glasses by design op- inforced polymer reinforced elements: theoretical and experimental analysis, ACI
timization, Eng. Fail. Anal. 7 (2) (2000) 87–99 April. Mater. J. 98 (4) (2001) 332–339 July.
[9] European Committee for Standardization – CEN, Eurocode: basis of structural de- [30] Z.Q. Chen, L. Huang, F. Wang, P. Huang, T.J. Lu, K.W. Xu, Suppression of annealing-in-
sign, European Standard EN 1990, 2002 (Brussels, Belgium). duced embrittlement in bulk metallic glass by surface crystalline coating, Mater.
[10] D. Baraldi, A. Cecchi, P. Foraboschi, Broken tempered laminated glass: non-linear Des. 109 (November) (2016) 179–185.
discrete element modeling, Compos. Struct. 140 (April) (2016) 278–295. [31] K. Hazirbaba, Y. Zhang, J.L. Hulsey, Evaluation of temperature and freeze–thaw ef-
[11] L. Biolzi, E. Cagnacci, M. Orlando, L. Piscitelli, G. Rosati, Long term response of glass- fects on excess pore pressure generation of fine-grained soils, Soil Dyn. Earthq.
PVB double-lap joints, Compos. Part B 63 (July) (2014) 41–49. Eng. 31 (3) (2011) 372–384 March.
[12] P. Foraboschi, Optimal design of glass plates loaded transversally, Mater. Des. 62 [32] R. Livaoglu, The numerical and empirical evaluation of chimneys considering soil
(October) (2014) 443–458. structure interaction and high-temperature effects, Soil Dyn. Earthq. Eng. 66 (No-
[13] M. Badalassi, L. Biolzi, G. Royer-Carfagni, W. Salvatore, Safety factors for the structur- vember) (2014) 178–190.
al design of glass, Constr. Build. Mater. 55 (March) (2014) 114–127. [33] P. Foraboschi, Experimental characterization of non-linear behavior of monolithic
[14] L. Biolzi, M. Orlando, L.R. Piscitelli, P. Spinelli, Static and dynamic response of pro- glass, Int. J. Non-Linear Mech. 67 (December) (2014) 352–370.
gressively damaged ionoplast laminated glass beams, Compos. Struct. 157 (Decem- [34] L. Galuppi, S. Massimiani, G. Royer-Carfagni, Buckling phenomena in double curved
ber) (2016) 337–347. cold-bent glass, Int. J. Non-Linear Mech. 64 (September) (2014) 70–84.
[15] L. Biolzi, S. Casolo, V. Diana, C.A. Sanjust, Estimating laminated glass beam strength [35] C.C. Lawrence, G.J. Lake, A.G. Thomas, The deformation and fracture of balloons, Int.
via stochastic rigid body-spring model, Compos. Struct. 172 (July) (2017) 61–72. J. Non-Linear Mech. 68 (January) (2015) 59–65.
[16] E. Speranzini, S. Agnetti, Flexural performance of hybrid beams made of glass and [36] K. Martens, R. Caspeele, J. Belis, Performance of statically indeterminate reinforced
pultruded GFRP, Constr. Build. Mater. 94 (2015) 249–261. glass beams – experimental comparison with determinate systems and effect of a
[17] K.L. Edwards, E. Axinte, L.L. Tabacaru, A critical study of the emergence of glass and discontinuous glass section, Constr. Build. Mater. 146 (August) (2017) 251–259.
glassy metals as “green” materials, Mater. Des. 50 (September) (2013) 713–723. [37] L. Qin, J. Shen, G. Yang, Q. Li, Z. Shang, A design of non-uniform thickness mould for
[18] L. Galuppi, G. Royer-Carfagn, A homogenized analysis à la Hashin for cracked lami- controlling temperature gradient and S/L interface shape in directionally solidified
nates under equibiaxial stress. Applications to laminated glass, Compos. Part B 111 superalloy blade, Mater. Des. 116 (February) (2017) 565–576.
(February) (2017) 332–347. [38] P. Foraboschi, Analytical model for laminated-glass plate, Compos. Part B 43 (5)
[19] H.D. Hidallana-Gamage, D.P. Thambiratnam, N.J. Perera, Numerical modelling and (2012) 2094–2106 July.
analysis of the blast performance of laminated glass panels and the influence of ma- [39] P. Foraboschi, Laminated glass columns, Struct. Eng. 87 (18) (2009) 20–26
terial parameters, Eng. Fail. Anal. 45 (October) (2014) 65–84. September.
[20] P. Rama Subba Reddy, T. Sreekantha Reddy, V. Madhu, A.K. Gogia, K. Venkateswara [40] G.B. Airy. On the strains in the interior of beams. Proc. R. Soc. Lond.; 12, 1862, pp.
Rao, Behavior of E-glass composite laminates under ballistic impact, Mater. Des. 84 304–306.
(November) (2015) 79–86. [41] P. Foraboschi, Versatility of steel in correcting construction deficiencies and in seis-
[21] J. Xu, Y. Sun, B. Liu, M. Zhu, X. Yao, Y. Yan, Y. Li, X. Chen, Experimental and macro- mic retrofitting of RC buildings, J. Build. Eng. 8 (December) (2016) 107–122.
scopic investigation of dynamic crack patterns in PVB laminated glass sheets subject [42] E. Speranzini, S. Agnetti, The technique of digital image correlation (DIC) to identify
to light-weight impact, Eng. Fail. Anal. 18 (6) (2011) 1605–1612 September. defects in glass structures, Struct. Control. Health Monit. 21 (6) (2014) 2015–2029.
[22] H. Chen, Q. Wang, Y. Wang, H. Zhao, J. Sun, L. He, Experimental and numerical study
of window glass breakage with varying shaded widths under thermal loading, Fire.
Technol 53 (1) (2017) 43–64 January.

View publication stats

Вам также может понравиться