Sie sind auf Seite 1von 16

The application of the Crack Size Design method to edge- loaded structural glass members with corner cracks

Mark Porter, The University of Oxford Prof. Guy Houlsby, The University of Oxford

Summary

The use of glass beams as long term structural load-bearing members is relatively new. Recent work by the authors [1] has suggested a new method, "Crack Size Design", which moves away from traditional allowable stress techniques. It was developed for glass with abraded edges, where the cracks are of fairly constant depth across the whole width of the glass plate. In freshly cut glass, however, such cracks are rare. Typically the cracks resulting from wheel cutting are localised at the corner of the cross-section. The current work investigates the behaviour of corner cracks under sustained loading using computer simulations. A model is proposed which is easily incorporated into the existing Crack Size Design method, and thus allows structural design of glass members with corner cracks.

Introduction

The use of annealed glass as structural elements is becoming more common [2]. Applications of glass beams and columns are generating interesting new architectural forms which are gaining increasing use. The structural design of such members differs from standard glass design because the loading is no longer transient and lateral but is sustained and in-plane. Thus the well known problem of "static fatigue" strength degradation becomes acutely important.

In recent work by the authors [1] a new design method for the types of members described above was proposed. It was tailored towards members whose edges had been ground, a common practice that gives a more uniform distribution of cracks along the loaded edge, and therefore more confidence in the failure stress. These cracks extend across the whole width of the glass plate and an assumption of uniform depth was made. The new design

method was based on a fracture mechanical analysis of these edge cracks, and was termed Crack Size Design.

σ σ σ c a t σ σ
σ
σ
σ
c
a
t
σ
σ

σ

Figure 1 Geometry of corner crack

The aim of this paper is to extend the applicability of the Crack Size Design method. This will be done by considering glass edges with corner cracks, as represented in Figure 1. Pilkington conducted a study [3] of the types of cracks produced at the glass edge by wheel cutters at the end of the float glass production line. It was found that corner cracks formed the majority, and in most cases worst, of the cracks. In addition it was shown that the depths of crack were rarely larger than 15% of the glass thickness. In highly polished, square edged glass members it is also highly unlikely that a crack will extend the full way across the edge. In this case the critical crack will also be a corner crack.

Being able to design for corner cracks could reduce the cost of structural glass construction. If a glass member may be designed for a non-abraded, as-cut surface, then the cost of edge processing may be eliminated. Since this is a high expense compared to the material cost, economic savings may result. There is no loss in aesthetic quality from using as-cut glass, as the edge away from the initial scoring marks is very smooth and has a polished appearance. However, corner cracks also occur in polished square-edged members, and so a more accurate design method for these will also be of benefit.

The Crack Size Design method is based on modelling the long term growth of the critical

crack in the glass member over the lifetime of the structure. The long term growth patterns

of corner cracks, and a way in which to design for them, is the subject of this paper. It is

also necessary to determine the size of the initial crack as a starting point for design. Since cracks in glass are too small to be inspected continuously in commercial production, the

only information that is available from testing of the glass is the short-term failure stress. In the initial Crack Size Design method this failure stress could be converted to a failure- initiating crack size by making use of the fact that the edge crack geometry is determined by

a single parameter. Corner cracks, if considered simply as quarter ellipses, have two

dimensions that define their shape, their depth up the plate and their width across the glass

thickness. It is not possible to determine from the single piece of information (the failure stress) what these two dimensions are. A way to overcome this indeterminancy in the design process is sought in this paper.

Review of Crack Size Design method

Crack Size Design is outlined here. In [1] we show that all the structural properties of glass can be explained through fracture mechanics. These properties included the variability in short-term strength and its loss over time. Failure occurs when the stress intensity factor, a

fracture mechanics variable, reaches a critical value, K IC . This critical stress intensity factor

is a constant for glass and provides a fixed basis on which to conduct design, rather than

using an allowable stress which has to vary with time. The loss in strength over time is a result of slow crack growth which occurs when the stress intensity factor is less than the critical value. There is also a threshold stress intensity factor limit below which no slow crack growth occurs.

For a fracture mechanics analysis the crack size is required in conjunction with the applied stress. In structural glass applications where the edge is most highly stressed, it is often abraded or ground to introduced a more uniform distribution of cracks, and therefore a less variable failure stress. An added advantage is that all the cracks are of the same form. In the original paper the Crack Size Design method was developed for cracks of relatively uniform depth across the whole width of the glass edge, resulting from edge abrasion. The only variable associated with the crack is therefore its depth into the glass. The growth of this crack is then modelled over time, allowing an accurate calculation of strength.

Design begins with the derivation of the initial design crack size.

The Weibull short-term

failure stress distribution is used to convert a stress of an acceptable probability of glass survival into a crack of acceptable probability of occurrence. The growth of the crack over the lifetime of the structure is then modelled based on the expected stresses. Failure can then be predicted through combining the crack size and applied stress into the stress intensity factor.

Fracture mechanics of a corner crack and the modelling method

Figure 1 shows a quarter ellipse corner crack under far field tension. This is an idealisation of the cracks that often occur on glass edges. Since structural glass members are deep compared to the size of the crack it is not necessary to include bending effects (other than the far field tension arising from bending stresses).

Newman and Raju [4] performed numerous finite element analyses of the crack geometry of Figure 1. Their results formed the basis for empirically derived equations which are lengthy and are therefore not reproduced here. These results were later supported by mathematically derived solutions to similar problems by Zhao and Sutton [5]. The Newman and Raju results show that the stress intensity factor varies smoothly around the crack perimeter and is dependent on the aspect ratio a/c and the crack to plate width ratio a/t. Figure 2 shows a typical stress intensity factor distribution around the edge of a corner crack.

1.2 1 0.8 0.6 0.4 a/c=0.3 a/t=0.1 0.2 0 K I / aσ√π
1.2
1
0.8
0.6
0.4
a/c=0.3
a/t=0.1
0.2
0
K I / aσ√π

0

10

20

30

40

50

60

Angle φ in degrees

70

80

90

Figure 2 Stress intensity factor variation around crack perimeter

Glass fails when the stress intensity factor reaches the critical stress intensity factor (K IC ).

This failure is sudden and results in very rapid crack growth leading usually to loss of structural integrity. If the stress intensity factor is less than the critical value then the crack grows. The speed of this growth is given by the following equation [6]. There is also a threshold stress intensity factor, K th , below which no crack growth occurs.

v

= v

K

I

IC

0

K

n

(1)

In equation (1) n is the static fatigue constant and typically has a value between 12 and 20 (often 16) depending on environmental conditions. The constants of v 0 , K th and K IC can usually be obtained from the glass manufacturer.

As the stress intensity factor varies around the perimeter of the corner crack, equation (1) implies that the speed of crack growth will also vary around the crack. Hence the dimensions of the crack will change so that the aspect ratio will not remain constant. The analysis pursued in this paper assumes that although the aspect ratio changes, the quarter- elliptical profile of the crack remains. This is supported by various evidence. Ohji et al. [7] experimentally tracked the growth of quarter-elliptical cracks in steel. Crack propagation in steel is described by Paris' Law, which is of the same form as equation (1). The experiments showed that the assumption that an elliptical profile is maintained was accurate. Dai et al. [8] modelled the growth of a series of crack profiles, not just quarter ellipses. The results show that the crack maintains a smooth profile, and more interestingly that even an initially "rough" crack propagates so that the profile becomes smooth. Therefore the assumption of a crack maintaining a smooth quarter-elliptical profile during loading is acceptable.

With an assumed quarter-elliptical crack, analysis is based on the fracture mechanics of Newman and Raju. The growth of the two perpendicular crack dimensions, a and c, are modelled over time, taking account of the variation in aspect ratio. For each time point the Newman and Raju stress intensity factors at the two ends of the crack are used to determine the respective growth speeds, using equation (1), and hence the new crack dimension and aspect ratio. This continues until the stress intensity factor at either end of the quarter-ellipse reaches the critical value when sudden failure is assumed to occur.

a/c

Crack growth behaviour

Figure 3 shows the degradation in strength with duration of loading due to crack growth, often referred to as "static fatigue". The case analysed is for an initial value of a/c=0.2. It demonstrates the well known behaviour that the stress that a piece of glass can sustain reduces with the period over which it is being loaded. The plot for the quarter elliptical crack was based on the crack growth algorithm described above. The figure shows that the current model produces results that agree with the empirically based relationship σ n t=const

[9].

1.9 Quarter Ellipse Empirical 1.8 1.7 1.6 1.5 1.4 1.3 -2 -1 0 1 2
1.9
Quarter Ellipse
Empirical
1.8
1.7
1.6
1.5
1.4
1.3
-2
-1
0
1
2
3
4
5
6
7
Log Stress (Log MPa)

1.2

1

0.8

0.6

0.4

0.2

0

Log Time (Log seconds)

Figure 3 “Static fatigue” strength degradation with time Initial values a/c=0.2 a/c=0.3 a/c=0.5 a/c=1.0 0
Figure 3 “Static fatigue” strength degradation with time
Initial values
a/c=0.2
a/c=0.3
a/c=0.5
a/c=1.0
0
0.2
0.4
0.6
0.8
1
1.2

Tim e/T im e to F ailu re

Figure 4 Variation in crack aspect ratio with time

Figure 4 shows a typical set of histories of the variation in crack aspect ratio over the lifetime of the cracks when subjected to constant stress. In this instance it is assumed that the thickness of the glass plate is sufficiently large compared with the crack size so that a quarter plane may be used to model the glass. The effect of thickness is discussed later.

The figure clearly shows that the aspect ratio tends to unity with time, that is the profile approaches a quarter circle. The failure times for each crack represented in Figure 4 were different, with that for a/c=0.2 being very short. That is, for this particular case the stress used in the model for this crack size caused failure after a matter of seconds. It is consistently found that for low a/c the failure times are significantly short that failure intervenes before the aspect ratio of 1 is reached.

0.0008 0.0006 0.0004 0.0002 a c 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 a,c
0.0008
0.0006
0.0004
0.0002
a
c
0
0.0
0.2
0.4
0.6
0.8
1.0
1.2
a,c (m)
K a ,K c (MPa.m 1/2 )

Time/Time to Failure

0.75

0.50

0.25

0.00

Ka Kc 0.0 0.2 0.4 0.6 0.8 1.0 1.2
Ka
Kc
0.0
0.2
0.4
0.6
0.8
1.0
1.2

Time/Time to Failure

Figures 5a and 5b Crack size and stress intensity factor histories

The convergence to a quarter circle profile can be explained through an investigation of the fracture mechanical processes. Let K a and K c denote the stress intensity factors corresponding to the dimensions a and c of a quarter-elliptical crack. Figures 5a and 5b give example histories for the growth of a and c and their respective stress intensity factors. It is found that the smaller dimension has the highest corresponding stress intensity factor. For example, if c is twice as large as a then K a will be larger than K c . The dimension of higher stress intensity factor will grow at a faster rate, as given in equation (1). This equation shows that the speed is a power function of stress intensity factor with the exponent n. Since n is generally of the order of 16, any difference between the stress intensity factors K a and K c will result in a proportionally much higher difference in the crack growth speed. Therefore the crack dimensions tend to grow to a point where the stress intensity factors become equal. For an infinite quarter plane this configuration is a quarter circle. The process of convergence can be seen in Figures 5a and 5b.

aspect ratio, a/cCrack

If the stress is sufficiently high that the initial stress intensity factor at one end of the crack is near K IC , then failure can occur before there has been much opportunity for crack growth. This results in failure at an aspect ratio lower than 1, as seen in the bottom curve in Figure

4.

Effect of finite thickness

The introduction of a finite third boundary to the quarter plane idealisation affects the stress intensity factors around the crack. The effect was included in Newman and Raju’s work. Figure 6 gives a number of crack aspect ratio histories for a crack, but in each case with a different glass plate thickness. The a/t ratio given is the ratio of the initial crack size to thickness. The figure shows that as the plate thickness reduces the crack propagates to a ratio less than 1. It is important to note that for these cases the final aspect ratio is still one that equalises the stress intensity factors at each end of the crack, but due to the new free boundary this point occurs for a different aspect ratio.

1.00 0.75 0.50 0.25 a/t=0.01 a/t=0.1 a/t=0.25 0.00 0.00 0.25 0.50 0.75 1.00
1.00
0.75
0.50
0.25
a/t=0.01
a/t=0.1
a/t=0.25
0.00
0.00
0.25
0.50
0.75
1.00

Time/Time to Failure

Figure 6 Effect of finite thickness on the crack aspect ratio history

Effect of n

It has been discussed earlier that the progression towards an aspect ratio of unity is based

on the different speeds of crack growth at either end of the crack when the stress intensity factors are unequal. Crack growth speed is dependent on n, as given in equation (1). Figure 7 shows the effect that different values of n have on the aspect ratio history of a crack. The values of 12 and 20 are the typical limits encountered for n in glass design. The figure demonstrates that these values of n are all sufficiently large to result in a migration towards a quarter circle profile, only having an effect on the history of the aspect ratio.

1.00 0.75 0.50 n=12 0.25 n=16 n=20 0.00 0.00 0.25 0.50 0.75 1.00 aspect ratio,
1.00
0.75
0.50
n=12
0.25
n=16
n=20
0.00
0.00
0.25
0.50
0.75
1.00
aspect ratio, a/cCrack

Time/Time to failure

Figure 7 Effect of n on the crack aspect ratio history

Proposed design modelling method

It has been discussed earlier that testing is not able to provide sufficient information to determine both dimensions a and c of corner cracks in glass. However, it is now known that regardless of the initial aspect ratio, the crack will propagate so that it approaches a quarter circle.

It is proposed that a design corner crack be assumed which initially has a quarter circular profile, and that it grows maintaining this shape. This assumption is based on the above observation that cracks usually approach this profile. Only one dimension needs to be considered, the radius. The short term strength data can therefore be used to calculate an equivalent crack radius. Design then continues as described in the initial Crack Size Design [1] where the crack dimension being analysed is the radius rather than the through thickness depth. The accuracy of design that can be achieved with this approximation will be addressed.

An earlier section demonstrated that the migration to an aspect ratio of unity is not accurate for short failure times (say up to 5 seconds). However, structural engineering applications are usually concerned with very long service lives, often up to 50 years. For these sort of time scales there is certainly adequate time for the propagation towards a quarter circle to occur, and thus for the proposed modelling method to be applicable and accurate.

Assumption of a quarter-circle, and therefore fixed aspect ratio, uncouples the dependence between the aspect ratio and crack growth rate which resulted from substituting the Newman and Raju functions into equation (1). The stress intensity factor can now be

described in the standard form [1], K

I

=

Y

σ

π
π

a , where Y now becomes 0.722431.

The remainder of this paper is concerned with investigating the accuracy of the proposed model via a series of numerical analyses of glass members with quarter elliptical cracks (of different aspect ratios) and their corresponding design cracks.

Initial conditions

Having established that the new modelling technique will be based on a quarter circle equivalent of the quarter ellipse, it is necessary to determine how to calculate the initial size of this design crack. This will be done by using the information provided by the short term failure stress test.

In standard testing methods glass is loaded with a linearly increasing tensile stress (see for example [3] or [11]). The stressing rate might range from 0.5 to 5 MPa/s. It is well established that slower stressing rates result in lower failure stresses, as there is a longer time for subcritical crack growth to occur. Hence, although the failure stress results are recorded for short time periods they do not represent an instantaneous failure stress.

The equivalent quarter circle crack size used in this work is one that would grow to fail at the same stress after the same duration of loading as the crack which led to failure in the short term test. To determine the initial crack size the stress, time to failure as well as the stressing rate from the test are required. For a quarter circle crack in an infinite quarter space the fracture mechanics equations given earlier can be manipulated to give,

a f

=

where

2

  2 − n n + n  n + 1  2 −
 2 − n
n +
n
n + 1
2 − n
2 − n
Y
σ
π
K
rate
 n + 1
th
n
v
t
+ a
0
f
0
 2(
 
1)
K
Y
σ
π
a
IC
 
rate
0
2
  
1
K
IC
a
= crack size at failure =
f
π
σ
t
  Y
rate
f
σ
= stressing rate for short term test
rate

t

f

=

K th

time to failure of the test specimen

= threshold stress intensity factor below which no slow crack growth occurs

a

0

= initial crack size

(3)

Equation (3) gives the final crack size at failure and may be used to find the equivalent initial size which led to that failure. This initial equivalent crack size is then the one that defines the size of the design crack at the start of the life of the glass member and it is the modelling of the subsequent growth of this crack which is the basis of Crack Size Design.

Effect of thickness

In earlier sections it was shown that, even with finite width, the assumption of progression towards an aspect ratio of 1 was still reasonably accurate. However, the third edge also has an impact on the stress at which failure occurs. Figure 8 shows the stress degradation curve for a series of simulated glass members. Each has the same initial crack size, but the glass plate thickness varies. The figure shows that there is little effect for the depth to thickness ratios investigated, with only the plot for the largest ratio being clearly discernible as separate on the graph with logarithmic axes. There is, however, a general but slight trend of decreasing strength with increasing crack to thickness ratio.

Pilkington's report [3] states that the crack depth to thickness ratios rarely exceed 0.15. For the range of crack aspect ratios considered (the full range allowed by the Newman and Raju equations) the typical maximum decrease in strength from the infinite quarter plane was generally of the order of 10-20%.

1.70 a/t=0.005 1.65 1.60 a/t=0.1 1.55 a/t=0.25 1.50 1.45 1.40 1.35 1.30 1.25 1.20 Log
1.70
a/t=0.005
1.65
1.60
a/t=0.1
1.55
a/t=0.25
1.50
1.45
1.40
1.35
1.30
1.25
1.20
Log Stress (Log MPa)

01234567

Log Time (Log seconds)

Figure 8 Static Fatigue curves for glass specimens with varying thickness

1.60 Stress rate = 0.2MPa/s 1.55 Stress rate = 3MPa/s 1.50 Stress rate = 6MPa/s
1.60
Stress rate = 0.2MPa/s
1.55
Stress rate = 3MPa/s
1.50
Stress rate = 6MPa/s
1.45
Original elliptical crack
1.40
1.35
1.30
1.25
Log Stress (Log MPa)

1234567

Log time (Log seconds)

Figure 9 Static fatigue plots for crack with initial size based on different stressing rates

The question is now whether the proposed modelling technique can account for the loss of strength with thickness. In the simulations the elliptically-cracked, finite thickness plate member was modelled by a circular crack in an infinite quarter space. The only difference between this and earlier models is that the initial crack size is determined from the short

Percentage error

term failure characteristics of the crack in a finite thickness, which are now lower than the original quarter space specimens. Figure 9 shows a typical set of stress degradation curves comparing the behaviour of a series of design cracks, each with initial crack size based on a different stressing rate in the short term test, with the quarter-elliptical, finite thickness original. Even though the crack to thickness ratio of a/t=0.25 is larger than would normally be encountered, the difference is still remarkably small. Extensive simulation revealed that errors of less than 1% (for n=16) were standard for most of the lifetime of the member, although larger errors were found near the subcritical growth threshold limit. Hence the proposed model can be easily applied to finite thickness glass members, and the thickness itself does not need to be accounted for explicitly.

6% a/c=0.2 a/c=0.3 4% a/c=0.5 a/c=0.9 a/c=1.2 2% a/c=1.5 a/c=2.0 0% 0 0.05 0.1 0.15
6%
a/c=0.2
a/c=0.3
4%
a/c=0.5
a/c=0.9
a/c=1.2
2%
a/c=1.5
a/c=2.0
0%
0
0.05
0.1
0.15
0.2
0.25
0.3
a/t
-2%
-4%
-6%

Figure 11 Error between subcritical threshold stress for elliptical and equivalent circular cracks

Effect of n

In a similar way to the variation in thickness, the variation in n results in different failure stresses, even though the cracks still propagate to an aspect ratio of 1. Typical values for n range from 12 to 20. Previous results given in this paper have been for n=16. Figure 10 shows the percentage errors between the times to failure of the design model and original crack for a range of values of n. For a high n there is very little error at all between the elliptical crack and the circular design equivalent since the high n accentuates the

difference in crack speeds for different stress intensity factors. Errors of less than 0.1% are

Reducing n reduces the accuracy. This is demonstrated by the

larger scatter in Figure 10. For the lower limit of 12 for n there is an error of roughly 10% in some cases, which is still acceptable. The lack of smoothness in the curves of Figure 10 is

a result of the numerical modelling.

easily achievable if n is 20.

4% n=12 n=14 n=16 n=18 2% n=20 0% 20 25 30 35 40 45 50
4%
n=12
n=14
n=16
n=18
2%
n=20
0%
20
25
30
35
40
45
50
-2%
-4%
Percentage Error

Stress (MPa)

Figure 10 Percentage error for various values of n

Subcritical threshold

It has been noted that an asymmetrical crack has different stress intensity factors at each

end. Since the design crack is an averageof this crack the stress intensity factor in this circular equivalent will probably be between the values at each end of the original crack. (For large ratios of a/t this may not always be the case). There is an implication here that the modelled crack will start growing at a different stress to its design crack equivalent. It is important to investigate this problem, as it may lead to the designer assuming safety below the threshold stress intensity factor when in fact the crack is growing, and therefore heading to failure.

Figure 11 shows the percentage error between the stress at which the original elliptical crack would start growing compared to that of its circular equivalent. The results are given for a range of aspect and crack size to thickness ratios. Trends between the plots are not

easily made as there is not necessarily a constant, such as a, common between them. However, it can be see that the errors are small, and all less than 10%. It is suggested that in design the threshold stress intensity factor physical constant used for design be scaled down by this margin, in addition to any reductions due to uncertainty in measurement.

Conclusion

This paper has proposed an extension of the Crack Size Design model for application to corner cracks in structural glass members. The design method may be invoked for design of glass elements with as-cutor highly polished square edges, or any other cases where the cracks present are predominantly corner cracks on a square edge.

The extension to Crack Size Design accounts for the two dimensions of a quarter elliptical crack by modelling an equivalent crack which is a quarter circle in profile. This method is suggested because numerical modelling showed this to be the natural profile that corner cracks grow towards.

Further numerical analysis considered the effect of varying glass plate thickness and the subcritical crack growth constant n. It was shown that there was only a small loss in accuracy in the worst cases. A maximum error of approximately 10% is achievable with this method, with the normal errors generally being more of the order of 1% or less. Such accuracy is considered favourable for glass design which traditionally employs large safety factors.

Having shown that elliptical cracks can be modelled well as equivalent circular cracks the design of glass members with corner cracks can proceed as outlined in the initial Crack Size Design paper.

References

[1] Porter, M.I., Houlsby, G.T., Development of Crack Size and limit state design methods for edge-abraded glass members, The Structural Engineer, in press. [2] Stansfield, K., Glass as a structural engineering material, The Structural Engineer, Vol. 77, No. 9, pp. 12-14, 1999. [3] Williams, J.M., McKenzie, H.W., Float glass edge strength overview, Group Research Report GR/97/126, Pilkington, Lathom, UK, 1997. [4] Newman, J.C. Jr, Raju, I.S., Stress-intensity factor equations for cracks in three- dimensional finite bodies subjected to tension and bending loads, NASA Technical Memorandum 85793, NASA, Hampton, Virginia USA, 1984. [5] Zhao, W., Sutton, M.A., Elastic solutions for corner cracks undergoing uniform displacement-controlled loading, International Journal of Fracture, Vol. 70, pp. 335-346,

1995.

[6] Evans, A.G., Slow crack growth in brittle materials under dynamic loading conditions, International Journal of Fracture, Vol. 10, No. 2, pp. 251-259, 1974. [7] Ohji, K., Kubo, S., Tsuji, M., Life Prediction of surface crack propagation in plates with or without residual stresses under fatigue loadings, Mechanical behaviour of materials - VI, Vol. 2, Edited by M. Jono, T. Inoue, Pergamon Press, Oxford, 1992. [8] Dai, D.N., Hills, D.A., Nowell, D., Modelling of growth of three-dimensional cracks by a continuous distribution of dislocation loops, Computational Mechanics, Vol. 19, pp. 538- 544, 1997. [9] Charles R.J., Static fatigue of glass II, Journal of Applied Physics, 29, No.11, 1958, pp.

1554-1560.

[10] Ritter, J.E., Service, T.H., Guillemet, C., Strength and fatigue parameters for soda-lime glass, Glass Technology, Vol. 26, No. 6, 1985.