Beruflich Dokumente
Kultur Dokumente
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/316643901
CITATIONS READS
0 11
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Arun Menon on 03 May 2017.
Keywords: Dynamic filtering effect of URM buildings, seismic input for out-of-plane walls, inelastic response
of primary resisting structure
ABSTRACT:
Response of URM walls to out-of-plane seismic excitation is a complex, yet usually, inadequately addressed
theme in seismic analysis. Modern seismic design codes for new masonry buildings provide dimensioning
and detailing rules that exclude out-of-plane failure even under severe seismic load. In contrast, out-of-plane
collapse of peripheral walls is a recurring mechanism in existing URM buildings. A first step towards under-
standing the phenomenon must entail estimation of the seismic demand on walls considering the dynamic fil-
tering effect of the building. Ground excitation that a URM building is exposed to is modified by the response
of in-plane walls and that of floor diaphragms before reaching the out-of-plane wall. The response accelera-
tion of the floor diaphragm is the seismic input on the out-of-plane wall which, largely comprises of energy at
the fundamental period of vibration of the building. Within linear response, resonance occurs when the out-
of-plane wall period coincides with the building’s fundamental period. Non-linear response of the building
can alter the response of a secondary system either by significantly reducing or substantially amplifying the
response compared to linear response. Interactions between the response of the primary resisting structure and
secondary elements (out-of-plane walls, non-structural elements) can be fully studied only through inelastic
dynamic analysis.
This paper deals with the development of a semi-analytical formulation to predict the seismic load to the out-
of-plane wall through parametric inelastic time-history analyses performed to different levels of displacement
ductility demand. SDOF oscillators have been employed in the parametric study implying that the global
structural response is mainly controlled by first mode characteristics. The study differs from previous ones in
that it explicitly considers the inelastic response of the primary resisting structure. Based on statistical exami-
nation of time-history analyses outputs, a formulation has been proposed for the estimation of seismic load (in
terms of spectral acceleration) on an out-of-plane wall or a non-structural element.
IN
2
2.00
IN
3 20% Damping 3
1.50 IN
4
2 70% Damping IN
1.00 5
1 0.50
0
0 0 20 40 60 80 100 120 140
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 T/T
1
Period ratio, T/T1
Figure 5. Transfer functions from time-history analyses (THA)
Figure 4. Amplification of ground motion for steady-state har-
using the Takeda hysteresis model for SDOF oscillator M1 for
monic excitation.
all levels of ductility demand.
Morphologically, the transfer function resembles
6
the amplification of response from steady-state elas- EL
tic behaviour where the amplification depends on the 5 IN
1
Transfer Functions
Transfer Functions
5 IN2 IN2
4
IN3 IN3
4
IN4 3 IN4
3
IN5 IN5
2
2
1 1
0 0
0 5 10 15 20 25 30 35 0 2 4 6 8 10
T/T1 T/T1
Figure 7. Transfer functions – Takeda – M5 - all μΔ. Figure 12. Transfer functions – Takeda – M12 - all μΔ.
6 6
EL EL
5 IN 5 IN
1 1
Transfer Functions
Transfer Functions
IN IN
2 2
4 4
IN IN
3 3
3 IN 3 IN
4 4
IN IN
2 5 2 5
1 1
0 0
0 5 10 15 20 0 2 4 6 8
T/T T/T
1 1
Figure 8. Transfer functions – Takeda – M7 - all μΔ. Figure 13. Transfer functions – Takeda – M13 - all μΔ.
6 6
EL EL
5 IN1 5 IN
1
Transfer Functions
Transfer Functions
IN2 IN
2
4 4
IN3 IN
3
3 IN4 3 IN
4
IN5 IN
2 2 5
1 1
0 0
0 2 4 6 8 10 12 14 0 1 2 3 4 5 6 7
T/T1 T/T
1
Figure 9. Transfer functions – Takeda – M8 - all μΔ. Figure 14. Transfer functions – Takeda – M14 - all μΔ.
6 6
EL EL
5 IN 5 IN
1 1
Transfer Functions
Transfer Functions
IN IN
2 2
4 4
IN IN
3 3
3 IN 3 IN
4 4
IN IN
2 5 2 5
1 1
0 0
0 2 4 6 8 10 12 14 0 1 2 3 4 5 6 7
T/T T/T
1 1
Figure 10. Transfer functions – Takeda – M9 - all μΔ. Figure 15. Transfer functions – Takeda – M15 - all μΔ.
6 6
EL EL
5 IN 5 IN
1 1
Transfer Functions
Transfer Functions
IN IN
2 2
4 4
IN IN
3 3
3 IN 3 IN
4 4
IN IN
2 5 2 5
1 1
0 0
0 2 4 6 8 10 0 1 2 3 4 5 6
T/T T/T1
1
Figure 11. Transfer functions – Takeda – M11 - all μΔ. Figure 16. Transfer functions – Takeda – M17 - all μΔ.
4.2 Interpretation of Numerical Results tion of stiffness and period with increasing ductility
demand.
In the preceding plots in Figures 5-16 (from time-
Regression of data from non-linear time history
history analyses using Takeda hysteresis), transfer
analyses (shown in Figure 18) establishes the fol-
functions for various levels of ductility are illus-
lowing relationship between displacement ductility
trated based on increasing oscillator period (T1).
and the ratio Teff/T1 (effective post-yield period to
The transfer functions for very rigid primary
the initial elastic period).
structures are close to the input ground motion’s ac-
Teff
celeration response spectrum (Figures 5 and 6). As = k ∗ μΔ , where k = −0.192 ln(μ Δ ) + 1.00 Eq. 1
the structure becomes flexible (increasing values of T1
T1), effect of the input ground motion reduces and The relationship is bound by the condition of
effect of interaction of the dynamic response of the elastic response, i.e. when the displacement ductility
secondary element with that of the primary structure is equal to 1.0,
becomes evident (as seen in Figures 7-16). Teff
With increasing levels of ductility demand, the = [− 0.192 ln(1.0 ) + 1.00]∗ 1.0 = 1.0 Eq. 2
T1
appearance of a second peak in the transfer function
plot becomes evident for semi-rigid and flexible 4
Acceleration [m/s2]
primary structures. The presence of a second peak in
2
Acceleration [m/s2]
2
-4
characteristics of the natural ground motion records, 50 100 150 200 250
1.50
4.3 Identification and Definition of an Effective
1.40
Inelastic period
1.30
Subsequent to the identification of formation of the 1.20
second peak in the transfer functions, it was essential 1.10
to investigate the reason behind its formation. The 1.00
post-yield inelastic “effective” period of the SDOF 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
oscillator is identified by signal processing of the re- μΔ
sponse acceleration time-histories using Spectro- Figure 18. Scatter of the ratio Teff/T1 with respect to the dis-
grams (time-frequency analysis) and Short-Time- placement ductility demand and the regression curve repre-
Fourier-Transform (STFT) In STFT, the time- sented by “k”.
frequency resolution is fixed over the entire time-
frequency space by preselecting a window length.
Therefore, a resolution in seismic data analysis be- 4.4 Comparison of TREMURI and Takeda
comes dependent on the user specified window In Figure 19, the transfer functions estimated from
length (Sinha et al., 2006). the time-history analysis using TREMURI with the
The time-frequency plot aids in identifying ine- typical masonry shear-sliding hysteresis is compared
lastic structural cycles rich in relevant frequency with those using the modified Takeda hysteresis in
content as shown in Figure 17. STFT performed Ruaumoko. For different levels of displacement duc-
over these inelastic cycles helps in identifying an ef- tility demand, the basic form of the transfer function
fective post-yield period in the selected inelastic is apparently similar irrespective of the two different
phase. The procedure is repeated for all the accelera- hysteresis rules. The appearance of the second peak
tion response histories in order to identify the varia- is evident in the TREMURI analyses as well.
6.00
EL
ciated with relative motion of the structure to the
5.00 M13 - TREMURI EL_1 ground and the viscous damping energy in the struc-
IN_1 ture) in the analysis using the modified Takeda rule
Transfer Functions
4.00 IN_2
IN_3
are much higher than those using the EPP rule. This
3.00 IN_4 is attributable to the degree of stiffness degradation
IN_5 offered by the Takeda hysteresis and the lack of the
2.00
same in the EPP rule. This explains the lower ampli-
1.00 fications in the transfer function using the modified
Takeda rule.
0.00
EPP_A.E.
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 250
EPP_D.E.-K.E.
T/T1 TAK_A.E.
6.00 200 TAK_D.E.-K.E.
EL
5.00 M13 - TAKEDA IN_1
150
Energy
Transfer Functions
IN_2
4.00
IN_3
100
IN_4
3.00
IN_5
50
2.00
1.00 0
0 5 10 15 20 25 30 35 40
0.00
Time (seconds)
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00
T/T1
Figure 21. Comparison of energies using the modified Takeda
Figure 19. Comparison of transfer functions from time-history and the Elastic-Perfectly Plastic hysteresis (AE: applied en-
analyses using TREMURI and Takeda hysteresis models. ergy, DE: damping energy, KE: kinetic energy).
different initial periods (T1 = 0.54 s). The seismic sup3or Average
Transfer Function
4.00
energy plots of time-history analysis using the two NATURAL RECORDS - M10
different hysteretic rules are compared in Figure 21. 3.00
7.00 2.00
EPP 1.00
6.00
TAKEDA
Transfer Function
0.00
5.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00
T/T1
4.00
3.00 6.00
sim1 sim2 sim3
sim4 sim5 sim6
2.00 5.00 sim7 sim8 sim9
sim10 Average
Transfer functions
0.00 3.00
rule show slightly higher amplification compared to Figure 22. Comparison of transfer functions using natural and
that of the modified Takeda rule. As seen from the synthetic records.
energy plots, the strain energies (applied energy to
the structure minus the kinetic energy of mass asso-
The thin grey lines correspond to the individual are defined using a harmonic response-type formula
time-history analyses whereas the thick line is the based on the initial elastic period of the primary
average transfer function for the set. The scatter in structure T1 and its effective post-yield period of vi-
the time-history analyses is appreciably more with bration Teff, respectively.
the natural records. The forms of the transfer func- The relationship between the initial elastic period
tions are comparable and the appearance of the sec- of vibration of the structure T1 and an effective post-
ond peak can be identified in the transfer functions yield period Teff has been investigated and the semi-
using artificial records as well. The amplification in empirical formula (Eq. 1) dependent on the dis-
the transfer function plot is over a wider period placement ductility demand is used in the analytical
range for the synthetic records and this is attribut- formulation.
able to the typically unrealistic energy content in The two dynamic interaction components are
synthetic records. combined using the proportionality coefficients DT1
and DTeff. The proportionality coefficients DT1 and
DTeff are clearly dependent on the displacement duc-
5 FORMULATION TO PREDICT SEISMIC tility demand on the primary structure in the formu-
INPUT lation. The “input ground motion component” in the
formulation is combined with the “dynamic interac-
Based on inferences from the parametric inelastic tion component” through a proportionality coeffi-
seismic analyses, the seismic input for a secondary cient C. Coefficients DT1, DTeff, coefficients of the
element is predicted using a semi-analytical formu- harmonic function (α, a and β) and Teff are the duc-
lation that takes into consideration the displacement tility-dependent terms in the formulation.
ductility demand (which can be directly related to The formulation captures the physical interpreta-
the structural behaviour factor) in terms of spectral tion of the filtering effect of the URM building
acceleration. rather elegantly. For secondary element attached to a
rigid primary resisting structure, the contribution to
⎧⎪ ⎛T ⎞ ⎪⎫ the seismic input to the secondary element is princi-
S ae ,o (T ) = {S ae ,i (T ) ∗ C}+ ⎨S a ,i (T1 )* TFhar ⎜⎜ ⎟⎟ * DT1 ⎬
⎪⎩ ⎝ T1 ⎠ ⎪⎭ pally from the “input ground motion component”,
Eq. 3 the “dynamic response component” diminishes and
⎧⎪ ⎛ T ⎞ ⎫⎪
( )
+ ⎨S a ,i Teff * TFhar ⎜ ⎟
⎜ Teff ⎟ Teff ⎬⎪
*D
the response spectrum resembles that of the input
ground motion. On the other hand, as the primary re-
⎪⎩ ⎝ ⎠ ⎭
sisting structure becomes more flexible, the contri-
⎛T ⎞ 1 bution from the “dynamic interaction component”
TFhar ⎜⎜ ⎟⎟ = Eq. 4
⎝ T1 ⎠ ⎛ ⎛ T ⎞ α ,a ⎞
2
2 starts increasing with a simultaneous reduction from
⎜1 − ⎜ ⎟ ⎟ + ⎛⎜ βζ T ⎞⎟ the “input ground motion component”.
⎜ ⎜T ⎟ ⎟ ⎜ T ⎟
⎝ ⎝ 1⎠ ⎠ ⎝ 1⎠ A residual contribution from the input ground
⎛ T ⎞ motion component has been consistently observed
1
TFhar ⎜ ⎟= Eq. 5 for all the models (different T1) and for all levels of
⎜ Teff ⎟ 2
⎝ ⎠ ⎛ ⎛ ⎞
α ,a ⎞ ⎛ ⎞
2 displacement ductility demand. This contribution
⎜ ⎜ T ⎟ ⎟ ⎜ T ⎟
⎜1 − ⎜ T ⎟ + ⎜ βζ T stabilises to approximately 10% beyond T1 of 0.25
⎜ ⎝ eff ⎟ ⎟ ⎟
⎝ ⎠ ⎠ ⎝ eff ⎠ seconds (as represented by Equation 7).
For purely elastic response of the primary resist-
⎧⎪ S (T ) ⎫⎪ ⎧⎪ ⎛T ⎞ ⎫⎪
TF = ⎨ ae ,i * C ⎬ + ⎨TFhar ⎜⎜ ⎟⎟ * DT1 ⎬ ing structure, the “dynamic interaction component”
⎪⎩ S a ,i (T1 ) ⎪⎭ ⎪⎩ ⎝ T1 ⎠ ⎪⎭ is constituted only by the “elastic dynamic response
Eq. 6
+⎨
( )
⎧⎪ S a ,i Teff ⎛ T
* TFhar ⎜
⎞
⎟ * DT
⎫⎪ component” and the “inelastic dynamic response
⎬ component” is equal to zero. As the inelastic re-
⎪⎩ S a ,i (T1 ) ⎜ Teff ⎟ ⎪⎭
eff
⎝ ⎠ sponse increases, the contribution from the “inelastic
C = x + (1 − x )* e − AT1 x = 0.10 Eq. 7 dynamic response component” increases with a si-
multaneous decrease in the “elastic dynamic re-
Teff = k μ Δ T1 Eq. 8
sponse component”.
Equation 3 gives the spectral acceleration ex-
The formulation (Equations 3-8) is made up of pected on the secondary element. The dynamic in-
two components that are superimposed and charac- teraction components are calculated using Equations
terise the contributions from the input ground mo- 4 and 5. The proportionality coefficient C is defined
tion and the dynamic interaction between the pri- using Equation 7 and the effective post-yield period
mary resisting structure (idealised by the in-plane Teff is defined using Equation 8. The “true” transfer
shear response) and secondary element. The “dy- function is evaluated in analytical terms, as per the
namic interaction component” is in turn separated original definition (in § 4.1), in Equation 6.
into the “elastic dynamic response component” and
the “inelastic dynamic response component” which
6 COMPARISON OF NUMERICAL AND SEMI- 1.40
ANALYTICAL MODEL TRUE Sa,o (T)
1.20
CALC Sa,o (T)
In Figures 22-26, the predictions of the spectral ac- 1.00 M13
celeration expected on the secondary element using
Sa,o (g)
0.80
the semi-analytical formulation is compared to the
numerical analysis for various ductility classes. The 0.60
1.60
0.60
TRUE Sa,o (T)
1.40
CALC Sa,o (T)
0.40
1.20 M16
0.20
1.00
Sa,o (g)
0.00 0.80
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00
T (s) 0.60
0.40
Figure 22. Comparison of predicted spectral accelerations to
those from the numerical analysis for SDOF oscillator M1 and 0.20
ductility class I (μΔ = 1.0-1.5).
0.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00
0.70
TRUE Sa,o (T)
T (s)
0.60 CALC Sa,o (T)
Figure 26. Oscillator M16 and ductility class IV (μΔ = 2.5-3.0)
0.50 M3
0.30
prediction is well below 10% for the indicators at T
= 0 seconds, T = T1 seconds, T = 2 seconds. The er-
0.20
ror is the prediction at T = 4 seconds is apparently
0.10 much higher, but the values of the spectral accelera-
0.00
tion themselves are quite small in this range.
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 Apart from being mainly ductility demand de-
T (s) pendent, the proportionality coefficients DT1 and
DTeff are also slightly sensitive to the variation of the
Figure 23. Oscillator M3 and ductility class II (μΔ = 1.5-2.0) initial period of vibration of the building. Neverthe-
less, despite neglecting the variation of these con-
1.40 stants with respect to the initial period, the errors in
TRUE Sa,o (T) the prediction at wall periods at T = 0 seconds, T =
1.20 CALC Sa,o (T) T1 seconds, T = 2 seconds are still well under 15%.
1.00 M6 This formulation seems to produce satisfactory re-
0.80 sults up to a wall period of 2 seconds.
Sa,o (g)
0.60
Table 2: Error in the prediction by the semi-analytical formula.
0.40
T1 ERRORS (%)
0.20 μΔ MODEL
(sec) T=0 T = T1 T = 2.0 T = 4.0
0.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00
I M1 0.038 5.1 -4.8 -0.2 -0.2
T (s)
II M3 0.082 12.4 1.8 6.3 31.2
III M6 0.225 -2.2 0.9 2.3 75.3
Figure 24. Oscillator M6 and ductility class III (μΔ = 2.0-2.5) V M13 0.654 3.6 1.0 2.4 79.0
IV M16 0.854 4.2 8.4 -0.7 63.1
7 CONCLUDING REMARKS CEN–EN–1998-1, 2005. Eurocode-8 Design of Structures for
Earthquake Resistance, Part 1: General rules, seismic ac-
Seismic input to a secondary element (out-of-plane tion and rules for buildings.
Doherty, K., Griffith, M.C., Lam, N., Wilson, J., 2002. Dis-
wall, non-structural component, etc.) is defined by a placement based seismic analysis of out-of-plane bending
semi-analytical formulation that explicitly considers of unreinforced masonry walls. Earthquake Engineering
the inelastic response of the primary system (in and Structural Dynamics, Vol. 31, pp. 833-850.
terms of displacement ductility demand that can Emori, K., Schnobrich, W.C., 1978. Analysis of Reinforced
eventually be related to the “structural behaviour Concrete Frame-Wall Structures for Strong Motion Earth-
quakes, Civil Engineering Studies, Structural Research Se-
factor”) in terms of the spectral acceleration. ries No. 434, University of Illinois, Urbana, Illinois.
The proposed formulation superimposes an “input Fajfar, P., 1999. Capacity Spectrum Method Based on Inelastic
ground motion component” and a “dynamic interac- Demand Spectra, Earthquake Engineering and Structural
tion component” between the primary structure (ide- Dynamics, 28, pp. 979-993.
alised by shear response) and the secondary element. FEMA 356, 2000. Prestandard and Commentary for the Seis-
The latter is further separated into the “elastic dy- mic Rehabilitation of Buildings, Federal Emergency Man-
agement Agency, Washington D C.
namic response component” and the “inelastic re- Galasco, A., Lagomarsino, S., Penna, A., Resemini, S., 2004.
sponse component”, related respectively to the initial Non-linear Seismic Analysis of Masonry Structures, Pro-
period T1 and a post-yield effective period Teff, and ceedings of the 13th World Conference on Earthquake Engi-
combined using proportionality coefficients depend- neering, Vancouver, B.C., CANADA.
ent on the displacement ductility demand. The rela- Gambarotta L., Lagomarsino, S., 1996. On Dynamic Response
of Masonry Panels, Proceedings of the National Confer-
tionship between T1 and Teff is represented using a ence “La meccanica delle murature tra teoria e progetto”,
semi-empirical formula dependent on the μΔ. The er- Messina, (ed.) L. Gambarotta (in Italian).
rors in the prediction of the spectral acceleration at Griffith, M.C., Magenes, G., Melis, G., Picchi, L., 2003.
wall periods T = 0, T = T1 and T = 2 seconds, is well Evaluation of Out-of-Plane Stability of Unreinforced Ma-
within 15% compared to the numerical results. sonry Walls Subjected to Seismic Excitation, Journal of
The formulation can be considered valid up to a Earthquake Engineering, Vol. 7, Spl. issue 1, pp. 141-169.
Lagomarsino, S., Penna, A., Galasco, A., 2006. TREMURI
secondary element period of T = 2 seconds beyond 1.0.6: Seismic Analysis Program for 3D Masonry Build-
which an error correction term might be required. ings, University of Genoa and EUCENTRE.
The current parametric study and the formulation Magenes, G., 2006. Masonry building design in seismic areas:
have been based on time-history analyses using ac- Recent experiences and prospects from a European stand-
celeration records compatible with an elastic re- point, Keynote lecture 9, Proceedings of the 1st European
Conference on Earthquake Engineering and Seismology,
sponse spectrum for stiff soil. There is the need to Geneva, Switzerland.
investigate the effect of soft soil on the dynamic fil- Mondal, G., Jain, S.K., 2005. Design of Non-Structural Ele-
tering effects of URM buildings. ments for Buildings: A Review of Codal Provisions, The
The current investigation is based on evaluation Indian Concrete Journal, Vol.79, No.8, pp. 22-28.
of the acceleration spectra. The study has to be ex- NZS 4203, 1992. New Zealand Standard – Code of Practice for
tended to evaluate transfer functions in terms of the General Structural Design and Design Loading of Build-
ings, Vol. 1 Code of Practice.
displacement spectra and validate the semi- NZS 4203, 1992. New Zealand Standard – Code of Practice for
analytical formulation for the same. General Structural Design and Design Loading of Build-
The parametric study developed on SDOF oscil- ings, Vol. 2 Commentary.
lators, implicitly considers a first mode shape in the Otani, S., 1974. SAKE A Computer Program for Inelastic Re-
inelastic response for the URM structure. This is a sponse of RC Frames to Earthquakes, Report UILU-Eng-
74-2029, Civil Engineering Studies, University of Illinois at
rather acceptable assumption for 3-4 storied existing Urbana-Champaign.
URM buildings, common in Italy. Nevertheless, in Pampanin, S., Christopoulos, C., Priestley, M.J.N., 2002. Re-
taller and more flexible structures, effects of higher sidual Deformations in the Performance-Based Seismic As-
modes in the response are not insignificant, hence, sessment of Frame Structures, ROSE Research Report
variation and distribution of accelerations demands 2002/02, IUSS Press, Pavia, ITALY.
on the secondary elements over the height of a struc- Priestley, M.J.N., 1985. Seismic Behaviour of Unreinforced
Masonry Walls, Bulletin of the New Zealand National Soci-
ture have to be investigated and accounted for. In a ety for Earthquake Engineering, Vol. 18, No. 2, pp. 191-
following stage, the parametric study will be ex- 205.
tended to multi-degree-of-freedom systems. Ray Chaudhuri, S., Hutchinson, T.C., 2004. Distribution of
URM buildings with rigid diaphragms and shear- Horizontal Peak Floor Acceleration for Estimating Non-
dominated behaviour were assumed in the current Structural Element Vulnerability, Proceedings of the 13th
World Conference on Earthquake Engineering, Vancouver,
study. Influence of flexible floor diaphragms on the B.C., Canada.
dynamic filtering effect needs to be investigated. Satish Sinha, Partha S. Routh, Phil D. Anno, John P. Castagna,
2006. Spectral decomposition of seismic data with continu-
REFERENCES ous-wavelet transform, Geophysics, Vol. 70, pp 19-25.
Carr, A.J., 2004. RUAUMOKO Computer Program Library, Tomaževiĉ, M., 1999. Earthquake-resistant design of masonry
Department of Civil Engineering, University of Canterbury, buildings, Series on innovation in structures and construc-
New Zealand. tion, Vol. 1, Imperial College Press, London.