6.4.1 Introduction
Testing is required mainly for two reasons:
1. To determine parameters of predictive models and design methods.
2. To verify or augment the predictive calculation. This typically related to cases where structural glass design cannot be solely based on predictive modelling. The difficulties with modelling arise mainly in the following areas:
a) Glass is extremely sensitive to stress concentrations. Numerical models, how- ever, often cannot provide reliable information on stress fields and particularly stress concentrations. This lack of confidence in the numerical models often
arises when there is limited information about the materials being modelled (e. g. liners, gaskets, bushings etc.) and/ or when the assembly process may cause stress raising imperfections (e. g. misalignment, large tolerances etc.). b) Despite recent advances in the field[229], the post-breakage structural capacity
often cannot be reliably predicted by predictive modelling.
c) There is not much experience and quantitative information available concern- ing the surface damage caused by various hazard scenarios.
d) The response of structural elements or entire sub-structures to impact loads is difficult to model.
e) Building owners, insurers and authorities generally have little confidence in glass structures and often ask for full scale tests.
In particular the following issues should be considered:
u It is very important that design and interpretation of tests are based on a thorough
understanding of the material behaviour. The fact that results from tests at ambient conditions represent a combination of both surface condition and time-dependent crack growth is particularly crucial. It is unfortunate that much project-specific testing is performed without taking time-dependent effects properly into account.
u If testing at ambient conditions is unavoidable, subcritical crack growth during the
tests must be modelled. While this can efficiently be done using the model from Section 3.3, dependence of the crack velocity parameters on the environmental conditions and the stress rate still diminishes the accuracy and reliability of the results.
u The problems related to subcritical crack growth in laboratory tests can be addressed
by the near-inert testing procedure summarized in Section 6.4.2. By preventing sub-critical crack growth during tests, it allows substantial improvement in the accuracy and safety of test results.
u Tests on as-received specimens or on specimens with artificially induced homoge-
neous surface damage are unsuitable for assessing the structural performance of glass elements in surface damage hazard scenarios. Such elements should be tested with realistic design flaws. This issue is discussed in Section 6.4.3.
6.4.2 Determination of surface condition parameters
Introduction
Reliable surface condition parameters (θ0, m0) form the basis of random surface flaw population-based modelling and must be derived from glass strength data. The testing procedures used today to obtain glass strength data were explained in Section 3.5.2. While European and North American design methods are based on fundamentally different testing procedures (see also Table 4.8), all current design methods use strength data obtained at ambient conditions, i.e. in normal, humidity containing air. This strength data depends on a specimen’s surface condition and on the subcritical growth of the surface flaws during the tests. It was shown in Section 3.2.3 that the relationship between stress intensity and crack velocity varies widely and depends strongly on the environmental conditions, on the residual stress in the glass and on the stress rate at
which a specimen is loaded. This prevents accurate estimation of the growth of surface flaws during experiments. Inaccurate estimation, however, can result in unsafe design parameters. Glass surface condition data should, therefore, be obtained from laboratory
testing in near-inert conditions.
Creating near-inert conditions in laboratory tests
Considering the chemical background of stress corrosion (see Section 3.2), inert conditions can be achieved in various ways:
1. Testing in a vacuum or in a completely dry environment. 2. Testing in a normal environment with a hermetic coating. 3. Testing in a normal environment at very rapid stress rates.
4. Testing at a sufficiently low temperature, at which the kinetics of environmentally induced reactions are arrested.
Not all possibilities outlined above are equally suitable for structural applications. Options 1. and 4. are difficult and expensive, especially for full-scale testing on large specimens. Options 2. and 3., in contrast, are comparatively simple and inexpensive provided that the conditions do not need to be fulfilled perfectly. Haldimann[187] showed that near-inert conditions in laboratory tests can be achieved by combining a near-hermetic surface coating and a relatively rapid stress rate. The latter reduces the effect of the former’s imperfection and vice versa. The proposed testing procedure is as follows:
1. Drying. The specimens are dried in an oven at 100± 5◦C for 48± 6 hours. The humidity in the oven is maintained below 5% RH by a high performance molecular sieve desiccant.
2. Hermetic coating. To achieve a hermetic coating, a silicone grease is applied to the tension face of the specimens. This grease is highly hydrophobic, impermeable and its viscosity is high enough to ensure that the coating remains intact during handling and testing.
3. Adoption. Specimens are kept at ambient conditions for 2 hours to allow them to adopt ambient temperature.
4. Destructive testing. The specimens are loaded to failure using a high stress rate (about 20 MPa/s is recommended).
If some subcritical crack growth occurs during near-inert tests, the results are conservative. This is a major advantage over ambient testing, in which overestimation of the crack growth during the tests leads to too optimistic surface condition parameters and therefore to unsafe design.
Interpretation of experimental inert strength data
The experimentally determined failure stresses at inert conditions represent the material’s inert strength. The surface condition parameters can be obtained from such data as follows:
u For tests with simple stress fields, such as coaxial double ring tests or four point
bending tests, simple analytical equations can be used (see[187, Section 5.3.3]). Fitting of the Weibull distribution to test results can be done by simple parameter estimation or maximum likelihood fitting.
u For tests with complex stress fields, such as tests on large rectangular glass plates, the
general lifetime prediction model described in Section 3.3 should be used. Failure loads and stresses are influenced by the non-linear load/stress relationship and the location-dependent stress history caused by the complex stress field. Experimental data, which at best provides information about the stress history at a few discrete points on the surface, will therefore generally not follow a Weibull distribution, such that a distribution-independent fitting method such as least-squares fitting or maximum likelihood fitting should be used to determine surface condition parameters.
An example of how this can be done as well as the required algorithms and their implementation in computer software are provided in[187].
Using strength data from ambient tests
If no inert strength data is available, the derivation of surface condition parameters from ambient strength data may be useful. Equations required for this purpose are given in[187]. They are derived from Equation (3.42) by narrowing its range of validity to constant stress or constant and moderate stress rate, uniform stress fields, a constant principal stress ratio and constant crack velocity parameters.
6.4.3 Obtaining strength data for design flaws
Current design methods rely on strength data obtained from specimens with as-received surfaces, specimens with artificially induced homogenous surface damage or weathered specimens. Specimens with such surface conditions are useful when adopting a random surface flaw population-based approach (see Section 6.2 and 6.3). Such strength data is, however, unsuited for design flaw-based design (see the same two sections). To obtain strength data for this approach, i. e. to quantify the damage caused by a surface damage hazard scenario (design flaw) and to assess the structural performance of glass elements that contain such damage, tests need to be performed on specimens with deep close-to-reality flaws. Such flaws have to meet two conflicting requirements:
1. They should be as similar as possible to the surface damage that structural glass elements are likely to undergo in in-service conditions. This includes accidental damage (e. g. due to handling, cleaning, impact of vehicles, tools falling down or impact of heavy wind-borne debris) as well as malicious damage (vandalism). 2. They should be as reproducible as possible.
In order to achieve an optimal compromise between these requirements, Haldimann [187] suggests to use a specially developed surface scratching device. This device may be used to induce long surface cracks on the glass surface by applying a constant force to a 0.33 carat dressing diamond. This scratching tip was found to be well suited because it does not show much wear and because its relatively large opening angle causes some widening of the scratch, which is an effect that is likely to happen with objects commonly used by vandals (e. g. diamond rings). A steel plunger holds the diamond tip. A casing guide ensures that the plunger is positioned exactly perpendicular to the glass plate. Ball bearings are used to minimize the sliding friction between the plunger and the guide. The plunger can be loaded with steel blocks of known weight and creates a constant contact
pressure between the scratching tip and the specimen. In dry diamond on glass scratching, the regularity of the surface flaws is problematic. An evaporating glass cutting oil makes depth and geometry of the flaws more uniform and allows higher loads to be applied.
The following should be considered with regard to deep surface flaw testing (for more details, see[187, 188]):
u The scatter of the strength of deep surface flaws is extremely high.
u The locally fractured glass zone around a surface scratch is significantly deeper than
the open, visible depth. The effective nominal flaw depth that governs strength is, therefore, significantly deeper than the optically measured flaw depth. This phenomenon is less pronounced in heat treated glass where the residual compressive stresses hinder fracture of the glass beyond the zone that is in direct contact with the scratching tip. Therefore, the strength reduction caused by a given surface damaging influence is much less severe in heat treated glass than it is in annealed glass.
u When testing specimens with deep surface flaws, the time to failure is so short that a
high stress rate is sufficient to ensure near-inert conditions (see Figure 6.2). Strength measurements obtained this way, i. e. without drying and hermetic surface coating, can be interpreted as inert strength data without being excessively conservative. This makes such laboratory testing simple and inexpensive, even in the case of large structural elements.3
u The key factor for meaningful results is a close match between the design flaw and
potential in-service damage.
50 100 150 200 250 3000 20 40 60 80 100 50 100 150 200 250 300 0 20 40 60 80 100 F ai lu re s tre ss , σf (M Pa )
Initial crack depth, ai (µm)
v0 = 0.01 mm/s Stress rate: inert 200 MPa/s 20 MPa/s 2.0 MPa/s 0.2 MPa/s Fa ilu re s tre ss , σf (M Pa )
Initial crack depth, ai (µm)
(Y = 1.12, KIc = 0.75 MPa m0.5, n = 16)
v0 = 6 mm/s
Figure 6.2: Failure stress of surface flaws in constant stress rate tests. In common laboratory conditions (right graph), the strength of deep surface flaws measured at ambient conditions and
with a stress rate of 20 MPa/s or above is virtually identical to the inert strength. This was confirmed by experiments in[187].