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

net/publication/325226428

Exploring the application domain of adaptive structures

Article  in  Engineering Structures · July 2018

DOI: 10.1016/j.engstruct.2018.03.057

CITATION READS

1 178

3 authors:

Gennaro Senatore Philippe Duffour

École Polytechnique Fédérale de Lausanne University College London
16 PUBLICATIONS   61 CITATIONS    49 PUBLICATIONS   389 CITATIONS   

SEE PROFILE SEE PROFILE

Pete Winslow
Expedition Engineering
17 PUBLICATIONS   78 CITATIONS   

SEE PROFILE

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

EngD Thesis: Decision-based optimisation framework of building structures View project

Design and control of adaptive structures: an energy-based approach beyond lightweight design View project

All content following this page was uploaded by Gennaro Senatore on 04 June 2018.

The user has requested enhancement of the downloaded file.

 Engineering Structures 167 (2018) 608–628

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Exploring the application domain of adaptive structures T

a,⁎ b c
Gennaro Senatore , Philippe Duffour , Pete Winslow
a
Swiss Federal Institute of Technology (EPFL), School of Architecture, Civil and Environmental Engineering (ENAC), Applied Computing and Mechanics Laboratory
(IMAC), Station 18, CH-1015 Lausanne, Switzerland
b
University College London, Gower Street, WC1E 6BT London, United Kingdom
c
Expedition Engineering, 4 Maguire St, SE1 2NQ London, United Kingdom

A R T I C LE I N FO A B S T R A C T

Keywords: Using a previously developed design methodology it was shown that optimal material distribution in combi-
Adaptive structures nation with strategic integration of the actuation system lead to significant whole-life energy savings when the
Active structural control design is governed by rare but strong loading events. The whole-life energy of the structure is made of an
Shape control embodied part in the material and an operational part for structural adaptation. Instead of using more material
Load-path optimisation
to cope with the effect of loads, the actuation system redirects the internal load-path to homogenise the stresses
Whole-life energy minimisation
and change the shape of the structure to keep deflections within limits.
Structural optimisation
This paper presents a systematic exploration of the domain in which adaptive two-dimensional pin-jointed
structures are beneficial in terms of whole-life energy and monetary costs savings. Two case studies are con-
sidered: a vertical cantilever truss representative of a multi-storey building supported by an exoskeleton
structure and a simply supported truss beam which is part of a roof system. This exploration takes five directions
studying the influence of: (1) the structural topology (2) the characteristics of the load probability distribution
(3) the ratio of live load over dead load (4) the aspect ratio of the structure (e.g. height-to-depth) (5) the material
energy intensity factor. Results from the main five strands are combined with those from the monetary cost
analysis to identify an optimal region where adaptive structures are most effective in terms of both energy and
monetary savings. It was found that the optimal region is broadly that of stiffness-governed structures. For the
cantilever case, the optimal region covers most of the application domain and it is not very sensitive to either
live-to-dead-load or height-to-depth ratios thus showing a wide range of applicability, including ordinary
loading scenarios and relatively deep structures.

1. Introduction integration of actuators has been shown to be an effective way to

suppress vibrations in high stiffness/weight ratio truss structures [9].
Adaptive structures are defined here as structures capable of Actuation has been used to modify the membrane stress state in
counteracting actively the effect of external loads via controlled shape shell structures which are usually designed via shape optimisation
changes and redirection of the internal load path. These structures are methods achieving ideal geometry under permanent load. To deal with
integrated with sensors (e.g. strain, vision), control intelligence and rarely occurring loading conditions different to the permanent load,
actuators. additional material is distributed locally which is, therefore, only uti-
In civil engineering, active control has focussed mostly on the lised during peak demands. In addition, in the event of cuttings or re-
control of vibrations for building or bridges to improve safety and sidual stress formed after formworks removal [10], the load carrying
serviceability during exceptionally high loads (i.e. strong winds, capacity is reduced significantly. In the event of such disturbances,
earthquakes) [1,2]. Active brace systems have been tested using hy- actuation in the form of induced strain distributions or induced support
draulic actuators fitted as cross-bracing elements of the structure, displacements (actively controlled bearings) has been used to homo-
controlling directly its response using actively controlled forces [3–5]. genise the stress field and in so doing minimising the maximum stress
Cable stayed bridges have been controlled using the stay cables as ac- governing the design [11,12].
tive tendons to reduce displacements [6,7]. Active cable-tendons have Active structural control has also been used in applications for shape
also been used to change the amount of pre-stress in reinforced concrete control. Some all-weather stadia use deployable systems [13] for ex-
beams and in steel trusses to limit displacements under loading [8]. The pandable/retractable roofs e.g. the Singapore National Stadium [14].

⁎
Corresponding author.
E-mail address: gennaro.senatore@epfl.ch (G. Senatore).

https://doi.org/10.1016/j.engstruct.2018.03.057
Received 28 February 2017; Received in revised form 12 March 2018; Accepted 19 March 2018
0141-0296/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Tensegrity structures consist of a set of compression and tension distribution. This is done by varying the Material Utilisation Factor
members whose stability relies on self-stress [15–17]. Tensegrity (MUT) which can be thought of as a scaling factor on the cross-sections.
structures have been used for deployable systems in aerospace appli- Varying the MUT changes the design from a least-weight structure with
cations [18] and they have been investigated for force/displacement small embodied but large operational energy, to a stiffer structure with
control [19–21] and frequency tuning [22] in civil engineering. Com- large embodied and smaller operational energy. This is shown dia-
pliant structures can be thought of as structures which act like mono- grammatically in Fig. 2 which describes the notional variation of the
lithic mechanisms. Compliance can be discrete or continuous. In the total energy with the MUT.
former, motion is allowed using flexural hinges (i.e. hinges that allows The inner optimisation itself consists of two main steps. The first
motion by bending) [23] while in the latter it is achieved through the step finds the optimum load path and corresponding material dis-
flexibility of the constituent elements of the structure [24]. Active tribution ignoring geometric compatibility and serviceability limit
compliant structures have been used for the deployment of antenna states thus yielding a design that represents a lower bound in terms of
reflectors [25], for the control of aircraft wings to improve on man- material mass. The optimisation is subject to equilibrium and ultimate
oeuvrability [26] as well as for the control of direct daylight in build- limit state constraints including member buckling. The members of the
ings [27]. structure are sized so that they have the capacity to meet the worst
Adaptive structures have a good potential for mitigating strong expected ‘demand’ from all load cases for strength only. Under external
hazard events and control of displacements and vibrations in deflection- loads however, the compatible forces are in general different from the
sensitive structures [28]. Because of uncertainties regarding the long- optimal forces and the resulting displacements might be beyond servi-
term reliability of sensor and actuator technologies combined with long ceability limits. For this reason, the second step finds the optimal ac-
service lives of buildings and long return periods of loads, the recent tuator layout to manipulate the internal forces by changing the shape of
trend has been to develop active structural control to help satisfying the structure. The actuators are devices which can either reduce or
serviceability requirements rather than contributing to strength and increase their length and are integrated in the structure by replacing
safety improvement [29]. part of their elements. Via controlled actuator length changes, geo-
The potential of using adaptation to save material has been in- metric compatibility is satisfied and at the same time deflections are
vestigated by some [30–33] but whether the energy saved by using less controlled. For indeterminate structures, it is possible to control both
material makes up the energy consumed through control and actuation the internal load-path and shape. Instead, if the structure is determi-
is a question that has so far received little attention. A novel design nate, the active system can only control the shape because there is no
methodology for adaptive structures was presented in Senatore et al. self-stress state.
[34]. This method is based on improving structural performance Once the actuator layout is known, a control strategy is determined.
through the reduction in the energy embodied in the material at the If a change in the loads causes a state of stress that violates a service-
cost of a small increase in operational energy necessary for structural ability limits state (SLS), the load path is redirected and displacements
adaptation and sensing. In [35] it was shown that adaptive structures are controlled by the active system. In case of a power outage or ac-
designed with this method can achieve up to 70% energy savings when tuation system failure and concurrent occurrence of a strong event, the
compared to identical passive structures designed using state art opti- structure might not be serviceable but load carrying capacity is not
misation methods. The examples studied so far range from planar portal compromised (i.e. fail-safe). In other words, the structure is designed
frames and catenary arch bridges to spatial configurations of complex not to collapse under the worst load case even without the contribution
layout including doubly curved grid-shells and exoskeleton structures. of the active system.
A large scale prototype designed using this methodology was success- The structure is designed to take permanent loads as well as ran-
fully tested validating key assumptions and numerical predictions [36]. domly fluctuating live loads. The methodology is based on the prob-
These conclusions are based on a set of assumptions including the ability of occurrence of the live loads. In a real design situation, this
structural layout, the live load probability distribution and the material probability should be based on empirical data or commonly used sta-
energy intensity. The purpose of this paper is to investigate how energy tistical models for the load considered. For illustrative purposes an
and monetary costs vary are as these inputs are changed via a para- example of one such probability distribution function is shown in
metric study. Fig. 3(a) and (b). The method identifies the load activation threshold
(the dashed line in Fig. 3a) above which actuation is needed for com-
2. Background: adaptive structures design methodology pensation of internal forces and displacements. Fig. 3(b) plots the hours
of occurrence for each level of the load obtained by discretising the
In conventional design situations, members are capacity designed probability density distribution scaled by the total number of hours of
and the highest demand is dictated by a worst load case. However, service. The introduction of the load activation threshold shows how
generally building structures experience loading significantly lower passive and active design can be combined to reach a higher level of
than the design load, meaning that they are effectively overdesigned for efficiency.
most of their working life.
If the structure relies on an active system for deflection control, its 3. Parametric exploration
stiffness can be distributed strategically such that the passive-active
configuration achieves higher efficiency in terms of whole-life energy. 3.1. Scope
The whole-life energy (also referred as total energy) is here understood
as the sum of the embodied energy in the material and the operational The parametric study carried out in this paper has five main ob-
energy used by the active control system. Senatore et al. [34] proposed jectives:
a new design method whereby the active system is only used when
necessary to ensure that the whole-life energy of the structure is kept to 1. Compare statically indeterminate against determinate structures to
a minimum. The method is briefly summarised here, the reader is re- appreciate the influence of the load-path redirection on operational
ferred to [37] for a detailed presentation. The method has so far been energy consumption;
implemented for reticular structures and this paper only deals with such 2. Appreciate the sensitivity of the energy savings to features of the
structures. The process comprises two nested optimisation stages as probability of occurrence of external loads;
shown by the flowchart in Fig. 1. 3. Appreciate the sensitivity of the energy savings to the live load to
The outer optimisation stage identifies a structure with minimal dead load (L/D) ratio;
overall energy (embodied + operational) for a given load probability 4. Study the importance of the slenderness of the structure by varying

609
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Inputs

Layout, material Design load & Control nodes &

& element type probability distribution serviceability limits

Whole-life energy optimisation

Load path & material Actuator layout

optimization optimization
0% < MUT(i) < 100%

Operational energy Load path redirection Load activation

computation & shape control threshold detection

Minimum whole-life energy design

Optimal material Optimal actuator

Optimal load-paths
distribution layout

Embodied energy Operational energy

MUT = Material Utilisation
optimisation computation
Fig. 1. Design methodology flowchart. @IOP Publishing. Reproduced with permission. All rights reserved [36].

In addition, a monetary cost analysis is carried out to investigate

whether the energy savings translate into monetary cost savings. The
results from the main strands will be combined with those from the
monetary cost analysis to identify an optimal region where adaptive
structures are most effective in terms of both energy and monetary cost.

3.2. General assumptions

3.2.1. Actuator type, dynamics and control system energy consumption

It is assumed that linear actuators are used with a mechanical effi-
Active design Passive design ciency of 80%. For civil engineering structures, the force magnitudes to
be controlled will likely orient the choice to hydraulic actuators. The
Fig. 2. Embodied, operational and whole-life energy as a function of the ma-
terial utilisation factor (MUT). @IOP Publishing. Reproduced with permission. mechanical efficiency of hydraulic actuators is in a range of 90–98%
All rights reserved [36]. [38]. For this reason, the assumption on the mechanical efficiency of
the actuators is conservatives.
The frequency response of the actuators is set to the 1st natural
main geometrical features such as height-to-depth (H/D) or span-to-
frequency of the structure as this is likely to be the frequency that will
depth (S/D) ratio;
dominate the response of most lightly damped structures excited by
5. Appreciate the importance of the embodied energy compared to the
wind, earthquakes or pedestrians. This assumption is conservative be-
operational energy by varying the material energy intensity (MEI)
cause it implies that even for quasi-static or low frequency response, the
factor.
actuators will always work at the 1st natural frequency of the structure.

(a) (b)

Fig. 3. (a) Live load cumulative distribution; (b) live load hours. @IOP Publishing. Reproduced with permission. All rights reserved [36].

610
G. Senatore et al. Engineering Structures 167 (2018) 608–628

I assumed that non-active means are used to control vibrations (e.g. an optimisation routine described in Senatore [39] that outperforms the
tuned mass dampers) if required. This is the case when vibrations are Modified Fully Utilised Design Method (Patnaik et al. [40]). The energy
caused by loads below the activation threshold (ULS and SLS re- savings are defined as the difference between the embodied energy of
spected). Whilst the active system could be used to compensate this the passive structure and the total energy of the adaptive one divided by
effect, it may come at the expense of a significant additional operational the former.
energy since vibrations can occur very often.
The energy it takes to power the control system (e.g. sensors and 3.3.1. Structural elements and material energy intensity (MJ)
signal processing) is modelled here as a linear function of the number of All structural elements have a cylindrical hollow section. To limit
structural elements and actuators. This assumption is based on em- the optimisation process complexity, the wall thickness is set to 5% of
pirical knowledge gained via experimental testing on a large scale the external diameter. The mass of an actuator is assumed to be a linear
adaptive truss prototype [36]. Note that the control system energy is function of the required force with a constant 0.1 kg/kN (e.g. an ac-
part of the total operational energy but it is not related to the energy tuator with a push/pull load of 1000 tons weighs 1000 kg) [41].
needed for structural adaptation (i.e. for force and shape control). The The energy analysis is carried out using a material energy intensity
former is usually substantially lower than the latter. The control system factor (MEI) to convert the material mass into embodied energy. The
energy requirement becomes important for small structures because in material utilised in the simulations described in this paper is steel in the
this case it is comparable with the embodied energy savings. form of rods obtained from predominantly virgin materials (no recycled
content) whose energy intensity is 35 MJ/kg [42].
3.2.2. Control system integration
It is assumed that the structures are fitted with as many strain 3.3.2. Ultimate and serviceability limit state
sensors as necessary to be able to compute the displacement field with Both passive and adaptive structures are subject to the same load
sufficient accuracy. factors and ultimate limit state constraints including member buckling.
The actuators are thought of as integrated into the structure by The limits on deflection used in the examples described in this paper are
replacing part of its elements. As shown in Senatore et al. [37], the those commonly used for the design of civil engineering structures. To
actuators positions can be determined selecting those elements whose make a fair comparison between adaptive and passive structures, the
length changes contribute most efficiently to correct the internal forces displacements are assumed to be caused by the live load only. The
and displacements. This analysis requires as input the selection of a passive structure is thought of as perfectly pre-cambered under the
certain number of degrees of freedom to be controlled. The choice of permanent load. Hence the material distribution is driven to compen-
the controlled degrees of freedom is up to the designer and is usually sate for deflections caused by the live load only. For the adaptive
dictated by serviceability. The minimum number of actuators n ACTs to structure, instead, the actuators reduce completely the displacements
control exactly the desired displacements is equal to the number of caused by the permanent load. In this way, both the passive and the
controlled degrees of freedom nCDOFs plus the degree of indeterminacy adaptive structures will already be stressed when the live load is ap-
of the structure (n ACTs = nCDOFs + r ). This is the minimum number of plied.
actuators needed to turn the structure into a controlled mechanism. If
fewer actuators are utilised, displacements can be controlled approxi- 3.4. Case studies
mately.
Throughout this parametric exploration, the structural configura-
3.2.3. Statistical modelling of the load tions under consideration will be a vertical cantilever and a horizontal
The structures are designed to take a generic permanent load (e.g. simply supported truss.
self-weight + cladding) as well as a randomly fluctuating one such as For the cantilever case, each truss can be thought of as the exos-
wind, earthquakes, crowd loading or moving loads such as trains. For keleton of a multi-storey building shown in Fig. 4(a). Several heights
simplicity, these loads are all considered as live loads. The statistics of (H) and height-to-depth (H/D) ratios will be tested. The floor perimeter
this live load are modelled using a log-normal probability distribution of the building is kept square. For simplicity, due to symmetry the
(Fig. 3) because this distribution is closely related to the normal dis- design of the structure is reduced to that of one planar truss. Fig. 4(b)
tribution, hence it is general only taking positive real values and thus it shows the statically indeterminate sample with a H/D ratio of 10.
provides the desired bias toward the lower values of the random vari- Fig. 4(c) shows the corresponding statically determinate topology.
able. The horizontal displacements of all the unconstrained nodes are set
The live load magnitudes used in this paper are commensurate with as controlled degrees of freedom (CDOFs) which are indicated by circles
the loads used by practicing design engineers. The design load (ex- in Fig. 4(b) and (c). The total building drift is limited to height/500
cluding safety factor), thereafter called the characteristic load is nor- [43].
mally set to the 95th percentile of the load probability distribution. There are three load cases, L1 is self-weight + dead load; L2 and L3
However, other characteristic loads (e.g. 99th or 85th percentile) will represent wind loads (live loads) acting in opposite directions. Their
be considered in Section 5. The probability distribution only describes intensity varies parabolically with height. The dead load is distributed
the occurrence of the live load. For simplicity, the mean of the under- on the floor area ADead = D ·D and it is applied every 4 m along the
lying normal distribution is set to zero. Once the mean and the char- height of the structure. The wind loads are applied on the façade area
acteristic load are set, the standard deviation can be determined. For all ALive = H ·D on both sides using a pressure coefficient of 0.5 for the
case studies discussed in this paper the structure service life is set to downwind side [44]. Table 1 gives the three load combinations con-
50 years. sidered for this case study.
For the simply supported case, each truss is part of roof supporting
3.3. Comparison adaptive vs passive system shown in Fig. 5(a). Several spans (S) and span-to-depth (S/D)
ratios will be tested. Fig. 5(b) shows a statically indeterminate sample
The metrics considered for the adaptive vs passive comparison are with an (S/D) of 20 and (c) show the corresponding statically de-
mass and total energy savings. The passive structure is designed using terminate topology. The vertical displacements of all the nodes except

611
G. Senatore et al. Engineering Structures 167 (2018) 608–628

L1 L1
L2 L3 L2 L3

D D
(a) (b) D
(c) D
Fig. 4. (a) Exoskeleton structure; (b) indeterminate, (c) determinate; controlled nodes indicated by circles.

Table 1 between 100 km/h and 200 km/h). The hurricane intensity refers to the
Cantilever|load combination cases. Saffir-Simpson scale whose highest category is 5. This way, the live load
Case Load factor Permanent load Load factor Live load
relates to the dead load representing realistic loading scenarios.
In general, the geometry of the truss bay is not an important para-
LC1 1.35 L1 = dead load + self 1.5 n/a meter as long as extreme aspect ratios are avoided. For this reason, the
LC2 1.35 L1 = dead load + self 1.5 L2 truss bay aspect ratio is set as a linear function of the height or the span
LC3 1.35 L1 = dead load + self 1.5 L3
with a proportional constant of 1/10 (H/10 in Fig. 4 and S/10 in Fig. 5).
This avoids very long truss elements which would be impractical for
the supports of the top chord are set as controlled degrees of freedom transportation and conversely too short elements which would make
which are indicated by circles in Fig. 5. Deflections are limited to span/ fabrication extremely difficult. Note that because member instability is
360. part of the optimization constraints, all element capacities satisfy ulti-
There are two load cases, L1 is self-weight + dead load; L2 is an mate limit states.
upward load which is representative of the suction caused by a wind
type load. Each bay of the truss supports a subsidiary area A = S·S Op 4. Energy savings vs topology
out of plane as shown in Fig. 5(a) where S Op is the length of the out of
plane span. Table 2 gives the two load combinations aconsidered for In this section a statically determinate truss is compared to a stati-
this case study. cally indeterminate one for both cantilever and simply supported case.
For both structural configurations, it is assumed that lateral stability The aim is to assess the effect of the structural topology and the load
in the other direction is provided by some other means. In both cases, path redirection on material distribution and the implications on op-
the dead load is set to 3 kN/m2 and the live load maximum intensity is erational energy consumption.
set proportional to the dead load via the live-to-dead-load (L/D) ratio.
The intensity of the live load is determined via the L/D ratio which will 4.1. Cantilever case|determinate vs indeterminate
be varied in a range from 0.1 to 2. This gives a 0.3 kN/m2 to 6.0 kN/m2
live load which can be thought of as resulting from wind pressure The structure considered in this study is a 400 m tall cantilever truss
caused by a category 1 to 3 hurricane respectively (wind velocity whose H/D ratio is set to 10. Both the height and H/D ratio are mid

612
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a)
L2
L1

(b)

L2
L1

(c)
Fig. 5. (a) Roof structure; static (b) indeterminate, (c) determinate−controlled nodes indicated by circles.

(indeterminacy is 10) structures respectively (Section 3.2.2). The dead

Table 2 load is set to 3 kN/m2 each truss supporting 40 m of out of plane span
Simply supported|load combination cases. resulting in a UDL of 120 kN/m. The live load cases described in table 1
Case Load factor Permanent load Load factor Live load and illustrated in Fig. 4 are applied. The L/D ratio is set to 1.
Fig. 6 shows the live load cumulative distributions functions (CDFs)
LC1 1.35 L1 = dead load + self 1.5 n/a for the two cases LC2 and LC3 plotted on the same chart (the loads have
LC2 0.9 L1 = dead load + self 1.5 L2
opposite directions but identical probability distributions). The vertical
dashed line represents the activation threshold for the indeterminate
case which is identical for both L2 and L3 because of symmetry. Si-
milarly, the dash-dotted line represents the activation threshold for the
determinate case which is slightly lower compared to that of the in-
determinate case. Consequently, the total actuation time required to
compensate for displacements is 3.6 and 4 years for the indeterminate
and determinate case respectively.
Fig. 7(a) shows the embodied, operational and total energy as
functions of the MUT, the dashed lines representing the indeterminate
case. For the determinate case, the optimal adaptive configuration is
obtained for an MUT of 68% whilst the passive structure corresponds to
an MUT of 31%. This means that the adaptive and passive structures are
designed so that the maximum stress under the worst load combination
is 68% and 31% of the yield stress respectively. Similar considerations
apply for the indeterminate case. In terms of total energy savings, the
Fig. 6. Live load cumulative distribution function (CDF) and load activation indeterminate and determinate adaptive structures achieve 33% and
threshold.
36% respectively compared to the passive structures as shown by the
bar chart in Fig. 7(b); mass savings are 55% and 58% respectively.
value in the range that will be investigated in Section 6. There is a total Fig. 8 compares the optimised passive structure (a) with the adap-
of 20 controlled degrees of freedom indicated by circles in Fig. 4. The tive structure (b). The actuators are represented by bigger diameter
minimum number of actuators to control exactly the controlled nodes is cylinders replacing the central part of the elements they are fitted onto.
20 and 30 for the statically determinate and indeterminate Fig. 8(c) shows both the controlled shape and the deformed shape

613
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
100% 68% 61% 35%
MUT
Fig. 7. Cantilever case; (a) Total energy vs material utilisation factor (MUT) for indeterminate (dashed) and determinate (solid) case; (b) passive vs adaptive total energy.

under LC2. Without active displacement compensation (i.e. shape well as deformed and controlled shape (c) and optimal forces (d) for the
change), the tip deflection is 1969 mm which is beyond serviceability determinate case. Note that for statically determinate structures op-
limit (height/500 = 800 mm). Fig. 8(d) shows the optimal load path timal and compatible forces are identical hence no load-path redirec-
and the force redirection (e), difference between optimal and compa- tion occurs.
tible load paths for LC2. Referring to Fig. 8(e), it can be appreciated the Comparing Fig. 8 with Fig. 9, it can be observed that material dis-
internal forces are redirected alternating tension and compression in tribution as well as controlled and non-controlled displacement fields
the bracers every other bay. are similar between the indeterminate and determinate trusses. There
Fig. 9 shows the comparison between passive (a) and adaptive (b) as are more elements (50) in the indeterminate truss with respect to the

800 mm

L2

- 355 N/mm2 355 N/mm2

(a) (b) (c) (d) (e)
Fig. 8. Cantilever truss statically indeterminate (a) passive and (b) adaptive solution; (c) controlled and non-controlled shape under LC2, mag. ×20; (d) optimal load-
path and (e) force redirection under LC2. Scale 1:4000.

614
G. Senatore et al. Engineering Structures 167 (2018) 608–628

800 mm

L2

-355 N/mm2 355 N/mm2

(a) (b) (c) (d)
Fig. 9. Cantilever truss statically determinate (a) passive and (b) adaptive solution; (c) controlled and non-controlled shape under LC2, mag. ×20; (d) optimal load-
path under LC2. Scale 1:4000.

determinate one (40) resulting into a slightly higher embodied energy Similar to the cantilever case, the indeterminate and determinate
for the former because the cross-section area lower bound is set iden- adaptive simply supported trusses have similar activation thresholds.
tical for both cases. The load path redirection occurring in the in- The total time during which actuation is required to compensate for
determinate structure has little effect on the operational energy. displacements is 2.5 and 2.1 years for the indeterminate and determi-
nate structure respectively. Fig. 10(a) shows the embodied, operational
4.2. Simply supported case|determinate vs indeterminate and total energy functions of the MUT. Total energy savings are 25%
and 30% for the indeterminate and determinate case respectively
The structure considered in this study is a 100 m span simply sup- compared to their corresponding passive structures as shown by the bar
ported truss with a S/D ratio set to 20. Both the span and S/D ratio are chart in Fig. 10(b); mass savings are 45% and 42% respectively.
mid value of the range that will be investigated in Section 6. There is a Figs. 11 and 12 compare the optimised passive structures (a) with
total of 9 controlled degrees of freedom (CDOFs) indicated by circles in the adaptive structures (b) showing the optimal load-paths in (c).
Fig. 5. The minimum number of actuators to control exactly the desired Fig. 11(d) and (e) show the stress redirection and non-controlled stress
displacements is 9 and 17 for the determinate and indeterminate (in- flow respectively. Internal forces are redirected by adding compressive
determinacy of 8) case respectively (Section 3.2.2). The dead load is set forces in the bracers located at mid span (Fig. 11d) which would be in
to 3 kN/m2 each truss supporting a 10 m out of plane span (S Op ) re- tension without active control due to the upward external load
sulting in a UDL of 30 kN/m. The live load cases described in table 2 (Fig. 11e). In addition, tensile forces are added to all vertical elements
and illustrated in Fig. 5 are applied, the L/D ratio is set to 1.5 (live load between top and bottom chords. Without active displacement com-
max intensity of 4.5 kN/m2). pensation (i.e. shape change), the mid-span deflections are 492 mm and

615
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
100% 60% 55% 30%
MUT
Fig. 10. Simply supported case; (a) Total energy vs material utilisation factor (MUT) for indeterminate (dashed) and determinate (solid) case; (b) energy comparison.

(a)

(b)

L2

(c)

(d)

(e)

-355 N/mm2 355 N/mm2

Fig. 11. Simply supported truss beam statically indeterminate (a) passive and (b) adaptive solution; (c) optimal (controlled) load-path, (d) force redirection under
LC2 and (e) non-controlled forces. Scale 1:800.

(a)

(b)

L2

(c)

-355 N/mm2 355 N/mm2

Fig. 12. Simply supported truss beam statically determinate (a) passive and (b) adaptive solution; (c) optimal load-path under LC2. Scale 1:800.

528 mm for the determinate and indeterminate case respectively which 4.3. Conclusions topology study
is beyond serviceability limit (span/360 = 277 mm).
As for the cantilever case, the indeterminate and determinate Following from the results shown in this section, it can be said that
adaptive simply supported trusses have a similar behaviour both in the energy performances of a planar truss will be similar whether it is
terms of displacement compensation and energy savings. However, in statically determinate or indeterminate. Due to the similarity of the
this case the force redirection has a substantial effect on the operational energy savings between the two cases, the remainder of this parametric
energy which is almost double that of the determinate case. exploration will be carried out on the statically determinate topology.
The simplicity of the statically determinate structure will help assess

616
G. Senatore et al. Engineering Structures 167 (2018) 608–628

relationships between parameters of interest (Section 3.1) avoiding Fig. 14(a) shows the operational, embodied and total energy as
potential numerical issues that can arise with more complex layouts. functions of the MUT for the cantilever case. The curves correspond to
the live load probability distributions with a characteristic load set to
5. Energy savings vs characteristic loads the 85th (dashed), 95th (solid) and 99th (dash-dot) percentile. As ex-
pected, if higher levels of the live load are more likely to occur (e.g.
The characteristic load considered so far has been set to the 95th characteristic load set to the 85th percentile), the optimal configuration
percentile of the load probability distribution (a log-normal function) is obtaeind for a lower MUT because it is more effective to rely on
for each load case (Section 3.2.3). In this section, three load probability passive resistance (material mass) to minimise the whole-life energy.
distributions with identical mean but characteristic loads set to the The opposite happens when the characteristic load is set to 99th per-
85th, 95th and 99th percentile are considered. Setting the characteristic centile. Fig. 14(b) shows the bar chart of the total energy for the passive
load to the 85th percentile means that high levels of the live load are structure and the three adaptive solutions obtained for each char-
more likely to occur compared to a characteristic load set to the 95th or acteristic load. There is a 15% loss in energy savings when moving from
99th percentile. The structures considered in this study are the same the 99th to the 95th percentile but only 5% loss when moving from the
statically determinate cantilevered and simply supported trusses con- 95th to the 85th percentile. Similar results are obtained for the simply
sidered in the previous section. The L/D ratio is kept at 1 for the can- supported case as shown by the energy curves as functions of the MUT
tilever and 1.5 for the simply supported truss. in Fig. 15(a) and by the bar chart comparing passive and adaptive total
Fig. 13 shows the live load cumulative distribution functions and energy in Fig. 15(b).
the activation threshold (indicated by dashed lines) for the cantilever Note that for both cases (cantilever and simply supported), if the
(a) and simply supported case (b). As expected, it is found that the characteristic loads are set lower than the 85th percentile, there is no
activation threshold is pushed towards higher values the higher the further significant loss in the energy savings. This is because although
probability of occurrence (CL = 85%) of the live load to reduce the higher levels of the load are more likely to occur, the load distribution
operational energy. covers a smaller part of the service life and thus the operational energy

(a) (b)
Fig. 13. Live load cumulative distribution functions (CDF) and activation thresholds. (a) cantilever, (b) simply supported.

(a) 94% 69% 63% 35% (b)

MUT
Fig. 14. Cantilever case; (a) total energy vs material utilisation factor (MUT) dashed CL = 85%, solid CL = 95%, dash-dot CL = 99% (CL = characteristic load); (b)
comparison total energy.

(a) 100% 73% 60%56% 20%

(b)
MUT
Fig. 15. Simply supported case; (a) total energy vs material utilisation factor (MUT) dashed CL = 85%; solid CL = 95%; dash-dot CL = 99% (CL = characteristic
load); (b) comparison total energy.

617
G. Senatore et al. Engineering Structures 167 (2018) 608–628

consumption does not increase. Therefore, the energy savings sensi- case. For both cases, the L/D ratio will be varied in a 0.1 to 2 range to
tivity to the probability of occurrence of the live load is low. Given a benchmark performances from low level loads to strong hazard events.
certain load probability distribution, the characteristics loads do not For the cantilever case, the height will be varied from 50 m to 800 m
represent a critical consideration for adaptive designs. and the H/D ratio will be varied in a 2–20 range going from deep to
very slender high-rise structures. The span of the simply supported truss
6. Energy savings vs live-to-dead-load and geometry will be varied from 20 m to 200 m and the S/D ratio will be varied in a
4–40 range going from deep to shallow roof truss systems.
6.1. Settings
6.2. Cantilever case | live-to-dead-load (L/D) ratio vs height
In this section the performances of the same statically determinate
structural configurations described in 3.4 are studied in relation to the An optimum configuration minimising the total energy is obtained
live-to-dead-load (L/D) ratio as well as the height-to-depth (H/D) for for each height sample in the range 50–800 m with a step of 50 m and
the cantilever or the span-to-depth (S/D) ratio for the simply supported the L/D ratio in the range 0.1–2 with a step of 0.1. The H/D ratio is kept

800 m
600 m
400 m
200 m
100 m
50 m

Fig. 16. Cantilever trusses 50 m to 800 m height, live-to-dead-load ratio L/D = 1. Scale 1:3000.

618
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Fig. 17. (a) Embodied, Operational and Total energy vs live-to-dead-load (L/D) ratio and material utilisation factor (MUT); (b) Total vs L/D ratio and MUT. Height = 400 m.

constant at 10 because it is mid value in the range (2–20) that will be

analysed in Section 6.4. The H/D ratio applies to both the in-plane and
out-of-plane depths of the structure (refer to Fig. 4). To illustrate some
of the configurations discussed in this section, Fig. 16 shows the opti-
mised adaptive solutions drawn to scale at six different heights for a L/
D ratio of 1.
Fig. 17(a) shows a surface plot of the whole-life energy as a function
of the MUT and the L/D ratio for the 400 m height sample. This plot is a
three-dimensional extension of the two-dimensional embodied-opera-
tional energy vs MUT graph shown previously. For clarity, the total
energy is shown again in Fig. 17(b) as a contour map where the thick
curve is the locus of the optimal MUT (minimum energy) for each L/D
ratio. The MUT decreases as the L/D ratio increases because deflections
must be limited using an optimal combination of structural mass and Fig. 19. Live load cumulative distribution function (CDF) and activation
active control. The contribution of the structural mass to limit deflec- thresholds at live-to-dead-load ratio L/D = 1.
tions is efficient at high L/D ratios and therefore the minimum energy
design is obtained for lower levels of the MUT as the L/D ratio in- saving curves are very similar and therefore are not shown here. Both
creases. energy and mass savings increase substantially with the height up to
Fig. 18 (a) and (b) shows a surface plot and a contour map of the 45% and 65% respectively. Because all energy saving curves have a
total energy as a function of the MUT and the height for a L/D ratio of 1. maximum, it means that as the live load increases it is effective to use
The MUT increases substantially as the height increases going from a actuation for deflection control until the operational energy becomes
minimum of 20% for the 50 m sample up to 100% for the 800 m sample. dominant over the embodied energy after which passive stiffness (i.e.
Referring to Fig. 18(a), it can be seen that for tall structures, as the MUT structural mass) must be used to lower the whole-life energy.
decreases, the embodied energy increases at much faster rate than the The energy savings reach maxima indicated by dots in Fig. 20(a) for
decrease in operational energy hence shifting the optimum towards a higher L/D ratio as the height increases. This means that low height
higher MUTs. Conversely, for low height structures it is more effective structures perform better than taller ones at a low L/D ratio thus
to use structural mass to limit deflections rather than operational en- matching real-world scenarios in which the probability of occurrence of
ergy. a high intensity live load (i.e. wind) is higher for tall buildings.
Therefore, to minimise the whole-life energy, the load activation The total time of actuation as a function of the L/D ratio has a very
threshold must be pushed to high values for low height structures as similar behaviour to that of the energy savings. The actuation time
indicated in Fig. 19 which shows the live load CDF and the activation maxima also increase as the height increases going from approximately
thresholds represented as dashed-dot lines. In other words, the taller 1 year for the 50 m height sample at a L/D ratio of 0.1 to 6 years for the
the structure the better utilised is the material. 800 m height sample at a L/D ratio of 2. Fig. 20(b) shows the maximum
From the observation that operational energy becomes dominant actuator length change as a function of the L/D ratio for each height
over embodied energy at high L/D ratios and material is better utilised sample. Although the operational energy is the highest at large spans
in tall structures, it follows the energy saving behaviour as a function of and high L/D ratios, the actuator length changes remain small (less than
the L/D ratio shown in Fig. 20(a) for each height sample. The mass

Fig. 18. (a) Embodied, Operational and Total energy vs Height and material utilisation factor (MUT); (b) Total energy vs Height and MUT. Height = 400 m.
619
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
Fig. 20. (a) Energy savings and (b) actuator max length change vs live-to-dead-load ratio (L/D).

400 mm). Minimum actuator control effort is obtained as a result of the function of the MUT and the span at a L/D ratio of 0.5 and 1.5 re-
optimal actuator placement method formulated in Senatore et al. [37]. spectively. In the first case (Fig. 22a), the MUT remains fixed at 100%
(indicated by a dash-dot line). However, when the L/D ratio is larger
6.3. Simply supported case | live-to-dead-load (L/D) ratio vs span than 1.0 (Fig. 22b) the MUT increases as the span increases. To explain
this behaviour, it is useful to refer to Figs. 23 and 24 which show the
An optimum configuration minimising the total energy is obtained controlled shape (a), non-controlled (or deformed) shape (b) and load-
for each span and L/D ratio. The S/D ratio is kept constant at 20 be- path (c) for a 100-m span simply supported truss under LC2 at L/D
cause it is mid-value in the range that will be analysed in Section 6.5. ratios of 0.5 and 1.5 respectively. As explained in Section 3.3, the
To illustrate some of the configurations discussed in this section, Fig. 21 adaptive structure takes the live load already stressed due to the dis-
shows the optimised adaptive solutions drawn to scale at six different placement compensation under permanent load. In this case the
spans for a L/D ratio of 1. structure is kept perfectly flat under the permanent load. The diagonal
The whole-life energy as a function of the MUT and L/D ratio is members hosting the actuators are in compression having made posi-
similar to the cantilever case (see Fig. 17a). Given a certain span, the tive length changes (expansion) to compensate for the downward per-
contribution of the structural mass to limit deflections is efficient at manent load. When the live load is lower than the dead load little
high L/D ratios because the operational energy becomes dominant over displacement compensation is needed (Fig. 23b and c). The actuators
the embodied energy and therefore the minimum energy design is ob- are required to make a length reduction under compressive forces
tained for lower levels of the MUT as the L/D ratio increases. (negative work) and thus very little or null operational energy is
Fig. 22(a) and (b) show the contour map of the total energy as a needed. For this reason, the MUT can stay fixed at 100% because in this

20 m

35 m

50 m

100 m

150 m

200 m
Fig. 21. Simply supported trusses 20 m to 200 m span, live-to-dead-load ratio L/D = 1. Scale 1:1500.

Fig. 22. Total energy vs material utilisation factor (MUT) and Span, live-to-dead-load ratios (a) L/D = 0.5 (b) L/D = 1.5.

620
G. Senatore et al. Engineering Structures 167 (2018) 608–628

L2 1.5 kN/m2

L1 3.0 kN/m2

(a)

(b)

(c)

-355 N/mm2 355 N/mm2

Fig. 23. 100-m span simply supported truss (a) controlled shape, (b) non-controlled shape and (c) optimal load-path under LC2; live-to-dead-load ratio L/D = 0.5.
Mag. ×10, scale 1:800.

L2 4.5 kN/m2

L1 3.0 kN/m2

(a)

(b)

(c)

-355 N/mm2 355 N/mm2

Fig. 24. 100-m span simply supported truss (a) controlled shape, (b) non-controlled shape and (c) optimal load-path under LC2; live-to-dead-load ratio L/D = 1.5.
Mag. ×10, scale 1:800.

case it is more efficient to use little operational energy than to add The change in behaviour that occurs when the L/D ratio becomes
structural mass regardless of the span of the structure. When the live larger than unity can also be seen by analysing the activation threshold.
load is larger than the dead load, displacement compensation becomes Fig. 25(a) and (b) show the activation thresholds (dashed-dot lines) on
substantial and the stresses in the diagonal members hosting the ac- the plot of the live load cumulative distribution functions for a L/D
tuators reverse from compression to tension (Fig. 24b and c). Since the ratio of 0.5 and 1.5 respectively. In the first case the activation
actuators are required to reduce their length whilst being in tension thresholds are similar regardless of the span because the MUT remains
(positive work), the operational energy increases substantially. For this fixed at 100%. By contrast, when the L/D ratio is 1.5, as expected the
reason, for L/D ratios larger than 1, the MUT increases as the span activation threshold is higher for small span structures because it takes
increases going from a minimum of 20% to 100% for large spans higher loads to reach the deflection limits.
(> 150 m) as shown in Fig. 22(b). Fig. 26(a) plots the energy savings as a function of the L/D ratio for

(a) (b)

Fig. 25. Activation thresholds at (a) live-to-dead-load ratio L/D = 0.5 and (b) live-to-dead-load ratio L/D = 1.5.

621
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
Fig. 26. (a) Energy savings and (b) actuator max length change vs live-to-dead-load ratio (L/D).

each span. As observed for the cantilever case, the energy savings in- explains why for the cantilever case the operational energy increases at
crease for all spans as the L/D ratio increases. The savings reach a lower rate as the L/D ratio increases and thus the energy savings
maxima (indicated by dots) in Fig. 26(a) at higher L/D ratios as the decrease more gradually with respect to the simply supported case.
span increases. After reaching maxima, the energy savings decrease As for the cantilever case, the total time of actuation as a function of
more rapidly than what was observed for the cantilever case. This is the L/D ratio has very similar behaviour to that of the energy savings.
caused by different boundary conditions and the way the live load re- The actuation time maxima increase as the span increases going from
lates to the dead load for the two cases. As explained in Section 3.4, the approximately 3 years for the 20 m span sample at a L/D ratio of 1 to
L/D ratio is the ratio between the live load and dead load without 4.4 years for the 200 m height sample at a L/D ratio of 1.5. Fig. 26(b)
considering their application area. For the simply supported case shows that the actuator length changes remain small (less than
(which is thought of as a roof support system), both live load and dead 250 mm) even for long spans and high L/D ratios.
load application areas are identical. For the cantilever case instead
(which is thought of as a multi-storey building) the dead load appli-
6.4. Cantilever case|height-to-depth ratio vs height
cation area is the sum of all floor areas and the live load one is the
façade area. For this reason, as the height of the cantilever structure
This section presents a study concerning the influence of the height-
increases, the dead load becomes dominant over the live load. This
to-depth (H/D) ratio (i.e. slenderness) on the energy savings for the
means that for the same L/D ratio, the live load resultant is compara-
cantilever case. When studying the influence of the L/D ratio on the
tively higher with respect to the dead load resultant at low heights. This
energy savings, the H/D ratio was kept constant at 10. In this study the

400 m

160 m 80 m 40 m 26 m 20

H/D=2.5 H/D=5 H/D=10 H/D=15 H/D=20

Fig. 27. 400-m height cantilever trusses, L/D = 1, height-to-depth ratio H/D from 2.5 to 20. Scale 1:4000.

622
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Fig. 28. (a) Total energy vs Height-to-depth (H/D) ratio and material utilisation factor (MUT), height = 400 m; (b) Total energy vs Height and MUT, H/D = 10.

(a) (b)

Fig. 29. Live load cumulative distribution and activation thresholds (a) height = 50 m, (b) height = 400 m.

(a) (b)
Fig. 30. (a) Energy savings and (b) actuator max length change vs height-to-depth ratio (H/D).

H/D ratio will be varied between 5 and 20 going from deep to very towards higher MUTs. In other words, for tall structures it is effective to
slender structures. The L/D ratio will be kept at 1 because it is mid- control deflection using active control resulting in a very efficient
value of the range that has been explored in Section 6.2. An optimum material utilisation.
configuration that minimises the total energy is obtained for each H/D Fig. 29(a) and (b) show the plot of the live load CDF and the acti-
ratio and height. Fig. 27 shows the optimised adaptive solutions drawn vation thresholds represented by dash-dot lines for the 50 m and 800 m
to scale for a height of 400 m. height samples respectively. In both cases, the load activation threshold
Fig. 28(a) shows a contour map of the total energy as a function of decreases as the structure becomes more slender. As the H/D ratio in-
the MUT and H/D ratio. The thick curve is the locus of optimal MUTs. creases, deflection limits are reached for lower levels of the live load.
Increasing the H/D ratio favours adaptive structures because the ac- The load activation threshold varies substantially for the 800 m height
tuators can control deflections no matter how slender the structure is. sample going from almost the design load value for deep structures (H/
However, at high H/D ratios the operational energy is high because D = 2.5 and H/D = 5) to a very low value for more slender structures.
deflection limits are reached for lower levels of the live load. For this This is because for tall and deep structures, the dead load resultant is
reason, the contribution of the structural mass to limit deflections is much larger than the live load resultant and thus little displacement
effective at high H/D ratios and therefore the minimum energy design is compensation is needed. When the H/D ratio increases, the live load
obtained for lower levels of the MUT. resultant increases substantially up to 40% the value of the dead load
Fig. 28(b) shows a contour map of the total energy as a function of resultant and therefore the load activation threshold also decreases
the MUT and the height for a H/D ratio of 10. The MUT increases substantially.
substantially as the height increases going from a minimum of 20% for Fig. 30(a) shows the plot of the energy savings as a function of the
low heights (50 m) to 100% for tall structures (500–800 m). This is H/D ratio for each height sample. As expected the energy savings in-
because for tall structures the operational energy decreases at a slower crease as the H/D ratio increases for all height samples. The large dif-
rate than the increase in embodied energy hence shifting the optimum ference in the load activation threshold between deep and slender tall

623
G. Senatore et al. Engineering Structures 167 (2018) 608–628

structures (Fig. 29b) explains the drastic increase in the energy savings the minimum energy design is obtained for a MUT of 100% at all S/D
for the height samples above 400 m. There are no substantial savings ratios and spans. This is because the L/D ratio is kept at 1 and the live
for deep and low height structures (S/D of 2.5 and H = 50). However, load is identical but opposite in direction to the dead load. For this
for H/D ratios larger than 5, the savings become substantial at all reason, little displacement compensation is needed and thus the ma-
heights reaching 70% for the 800 m height sample. terial can be fully utilised without causing any substantial increase in
The actuation time is substantial at all heights from a H/D ratio operational energy to reduce deflection actively.
larger than 5 reaching almost a quarter of the service life (50 years) for Fig. 33 shows the plot of the live load CDF and the load activation
tall structures. For deep (H/D = 5) and low height structures no ac- thresholds represented by dash-dot lines for (a) the 100-m span sample
tuation is needed. Fig. 30(b) shows the plot of the maximum actuator at all S/D ratios and (b) for an S/D ratio of 40 at all spans. The load
length changes as a function of the H/D ratio for each height sample. activation threshold decreases as the S/D ratio increases because the
The maximum length change does not exceed 700 mm even for the structure becomes more slender and therefore it takes lower levels of
800 m span sample. the live load to reach deflection limits. However, the activation
threshold increases with the span because the minimum energy design
6.5. Simply supported case | span-to-depth ratio vs span is obtained for an MUT of 100% regardless of the span. This is because
as the span increases the cross section area lower bound must be in-
This section presents a study concerning the influence of the span- creased to obtain feasible solutions resulting in a higher value of the
to-depth (S/D) ratio (i.e. slenderness) on the energy savings for the load activation threshold.
simply supported truss case. The S/D ratio will be varied from 5 to 40 Fig. 34(a) shows the plot of the energy savings as a function of the
going from deep to very slender structures. The L/D ratio is set to 1 S/D ratio for each span sample. Although the energy savings are neg-
because it is mid value of the range that was analysed in Section 6.3. To ligible for deep structures (S/D of 5), they increase rapidly above 20%
illustrate some of the configurations discussed in this section Fig. 31 for all spans as the S/D ratio increases to 10. The savings continue to
shows the 100-m span adaptive solutions for five S/D ratios. increase steadily reaching 50% for an S/D ratio of 20 and up to 70% for
Fig. 32 shows the contour map of the total energy as: (a) a function an S/D ratio of 40. In this case, the span has little influence on the
of the MUT and S/D ratio for the 100-m span sample; (b) a function of energy savings because for L/D ratios smaller or equal to unity, the
the MUT and span for a S/D ratio of 20. Both contour maps show that minimum energy design is obtained for an MUT of 100% regardless the

S/D = 5

20 m
S/D = 10

10 m
S/D = 20
5m

S/D = 30
3.3 m

S/D = 40
2m

100 m
Fig. 31. 100-m span simply supported trusses, L/D = 1, span-to-depth ratio S/D from 5 to 40. Scale 1:800.

Fig. 32. Total vs Span-to-depth ratio and material utilisation factor (MUT), span = 100 m; (b) Total vs Span and MUT, span-to-depth ratio S/D = 10.

624
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
Fig. 33. Live load cumulative distribution function CDF and activation thresholds, (a) span = 100 m and (b) span-to-depth ratio S/D = 40.

(a) (b)

Fig. 34. (a) Energy savings and (b) operational energy vs span-to-depth ratio (S/D).

S/D ratio and the span. assumed to be identical because both structures require fabrication
The actuation time is substantial at all spans from a S/D ratio larger of an equal number of joints and elements. Because the passive
than 10 reaching a fifth of the service life (50 years) for slender struc- structure is generally substantially heavier than the adaptive one, it
tures. For deep structures (S/D = 5) no actuation is needed. The max- is likely that fabrication costs will be higher for the former due to
imum actuator length change does not exceed 200 mm even for the logistics.
200 m span at a S/D ratio of 40 as shown in Fig. 34(b). • Control system maintenance costs include inspections as well as
replacements. It is assumed that an inspection is scheduled once per
7. Monetary cost analysis year taking one engineer a full-day to examine 5 actuators and 10
sensors. The cost of one inspection is set to £500 per day. In addi-
7.1. Introduction tion, it is assumed that up to 20% of the actuators and sensors will
have to be replaced once during service life and thus an additional
A monetary cost analysis is carried out for the same set of structures cost for replacement is accounted for.
considered in Section 6. The aim is to appreciate how the adaptive
solution compares to the passive one in monetary cost terms. Input 7.2. Monetary cost analysis
parameters are identical to the previous study on energy savings.
Monetary costs are evaluated in relation to the live-to-dead-load (L/D) Fig. 35(a) shows the plot of the monetary cost difference (adaptive
ratio, the height-to-depth H/D or span-to-depth (S/D) ratio as well as minus passive) as a function of the height and the L/D ratio for the
the height or span for the cantilever and simply supported case re- cantilever case. For L/D ratios larger than 0.5, the adaptive solution
spectively. To run the monetary cost analysis some assumptions have competes and eventually becomes less expensive as the height of the
been made: structure and the L/D ratio increases. For tall structures, the large dif-
ference in mass between the passive and the adaptive solution results in
• Actuation is hydraulic. The cost of a hydraulic actuator is assumed a monetary gain that outweighs the extra expenditure for sensors, ac-
to be linearly proportional to the required force [45] at a rate of tuators and control system. The adaptive structure becomes less ex-
0.97 £/kN (e.g. an actuator with a push/pull force of 1000 ton costs pensive the higher the L/D ratio because it is increasingly more difficult
£ 9700). The cost of the hydraulic system (e.g. pumps, loading for the passive structure to meet deflection limits by adding more ma-
manifold assembly) and driver electronics is estimated at £5000/ terial.
actuator [45]. This monetary cost trend is also caused by the variation of the ac-
• Element stresses are monitored using strain gauges. The average tuator and sensor density that is the number of actuators and sensors
cost per strain gauge sensor (full-bridge type) is set to £500/unit per cubic meter of structure (the volume occupied by the material used
including lead wires and signal amplification [46]. by the structure). Fig. 35(b) shows the plot of the actuator density as a
• The cost for data acquisition (i.e. monitoring) and processing is set function of the height and the L/D ratio. The actuator density decreases
to £500 per channel – one channel per strain sensor, two channels rapidly as the height and the L/D ratio increases because the structural
(position feedback and power output) per actuator. topology remains fixed while the mass increases substantially. The
• The cost of construction material (in this case steel) is set to £3000/ sensor density as a function of the height and L/D ratio has a similar
tonnage [47]. This figure mostly depends on the building type behaviour. Therefore, the control system cost share becomes less im-
usually varying between £1000 and £5000 per tonnage depending portant as the height and the L/D ratio increases. The same applies
on specifications (e.g. office vs landmark building). when varying the height-to-depth (H/D) ratio, the higher the H/D ratio
• The cost of fabrication for the passive and the adaptive structure is the more the adaptive structure becomes competitive in terms of

625
G. Senatore et al. Engineering Structures 167 (2018) 608–628

(a) (b)
Fig. 35. Monetary cost (adaptive-passive); (a) cantilever; (b) simply supported.

(a) (b)

Fig. 36. Cost (£) of energy savings vs cost (£) of photovoltaics (PV); (a) cantilever; (b) simply supported.

monetary costs. Above an H/D ratio of 5, the adaptive cantilever truss is cost, the results shown in Section 6 and 7 are combined. Energy and
less expensive than the passive one at all heights. Similar results are monetary cost savings in percentage terms are averaged thus forming a
obtained for the simply supported case. For L/D ratios larger than 1, unique value.
span-to-depth (S/D) ratios larger than 20 and spans longer than 100 m, Regarding the cantilever case, Fig. 37(a) shows the contour map of
the adaptive solution is less expensive than the passive one. the combined energy plus monetary cost savings as a function of height
If adaptive structures are intended as energy saving devices, the and L/D ratio for a H/D ratio of 10. The zero level set is indicated by a
monetary cost of saving energy using structural adaptation can be thick curve. The optimal region covers the majority of the domain ex-
compared to the cost of producing energy using other technologies. For cluding L/D ratios lower than 0.2 and heights lower than 150 m.
instance, the cost per kWh of energy saved using actuation can be Fig. 37(b) shows the combined energy plus monetary cost savings as a
benchmarked against the cost of producing energy using photovoltaic function of span and H/D ratio for a L/D ratio of 1. In this case the
panels (PV). In fact, the adaptive minus passive monetary cost differ- optimal region includes deep structures (H/D = 2) for heights above
ence (Fig. 35a) divided by the energy difference between the two so- 200 m.
lutions represents the cost of saving energy using structural adaptation. Regarding the simply supported case, Fig. 38 shows a contour map
Fig. 36(a) and (b) shows the plot of the cost of saving energy for the of the combined savings as a function of: (a) the span and L/D ratio for
cantilever case as a function of the height at different L/D and H/D a S/D ratio of 20 and (b) the span and S/D ratio for a L/D ratio of 1. The
ratios respectively. The cost of saving energy using PVs (the dashed optimal region in this case occupies the upper right hand corner. This is
line) is taken from the National Renewable Energy Laboratory (NREL) because the energy savings reach maxima for L/D ratios in a range from
using a performance calculator for grid-connected PV systems [48]. 0.5 and 1.5 (Fig. 26a) and the adaptive solution is generally more ex-
Setting the location to London gives an average cost of 0.09 £/KWh. pensive than the passive one except for spans above 100 m and L/D
The cost of saving energy via actuation is higher than that of producing ratios larger than 0.7 and S/D ratios larger than 20.
energy via PVs for structures lower than 50 m regardless the L/D ratio. Although the optimal regions shown in Figs. 37 and 38 are different,
The two systems compare above a height of 50 m and L/D ratios above similar conclusions can be drawn for the cantilever and a simply sup-
0.5. Adaptive structures are more efficient than PVs for heights above ported case study: it is most beneficial to use structural adaptation in
200 m regardless the L/D ratio. The same applies when varying the H/D energetic and monetary terms for tall or long span structures as well as
ratio. For deep and short structures, the cost of saving energy via ac- for slender structures. This is broadly the type of structures that are
tuation is higher than that of producing energy using PVs. The two stiffness governed.
systems compare even for deep structures as the height increases. For The simulations described in previous sections were carried out
heights above 200 m, adaptive structures are more efficient than PVs using a material energy intensity factor (MEI) of 36.5 Ml/kg which is
regardless the H/D ratio. Similar results are obtained for the simply that for steel obtained from predominantly virgin materials (i.e. no
supported case. The thresholds beyond which the cost of saving energy recycled contents) [42]. The same set of structures has been analysed
via structural adaptation is lower than that of producing energy using using an MEI of 15 MJ/kg which corresponds to secondary steel ob-
PVs are 100 m span, L/D ratio of 0.5 and S/D ratio of 20. tained from predominantly recycled contents. For both cantilever and
simply supported case, although savings are reduced, the optimal re-
8. Application domain optimal region gions preserve almost identical boundaries thus showing little sensi-
tivity to the MEI.
To determine the boundaries of the region where adaptive struc- Table 3 gives the degree of sensitivity of the energy and monetary
tures outperform passive structures in terms of energy and monetary cost savings in relation to the input parameters. The degree of

626
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Fig. 37. Optimal region cantilever truss (a) height-to-depth ratio H/D = 10, (b) live-to-dead-load ratio L/D = 1. MEI = 36.5 MJ/kg; thick curves indicate zero level
sets.

Fig. 38. Optimal region simply supported truss (a) span-to-depth ratio S/D = 20; (b) live-to-dead-load ratio L/D = 1. MEI = 36.5 MJ/kg, thick curves indicate zero
level sets.

9. Conclusions
Table 3
Energy and monetary cost savings sensitivity. From the results of the parametric study presented in this paper it
Case Live-to- Span-to- Characteristic load MEI
can be concluded that:
dead-load depth

Cantilever
• Given a planar truss topology, the degree of static indeterminacy has
little influence on the energy savings (Section 4);
Height < 150 m
150 m < Height < 800 m
+++
+
+++
+
+
+
+
+ • Given a certain load probability distribution, the characteristic loads
do not represent a critical consideration for adaptive designs
Simply supported
(Section 5);
•
Span < 100 m +++ +++ + +
100 m < Span < 200 m ++ ++ + + The energy savings as a function of the L/D ratio have a maximum.
This is because for very strong live loads (high L/D ratios), the op-
Sensitivity: + little ++ moderately +++ large. erational energy increases substantially resulting in lower energy
savings. This is an important result because it shows that adaptive
sensitivity is a measure of how sensitive (i.e. small changes in the inputs structures outperform passive ones not only for strong loading sce-
generate big changes in the output) the performance metric is with narios but also for ordinary cases (Section 6);
respect to the input parameters.
The optimal region for the cantilever case is not very sensitive to
• Adaptive structures are more expensive than passive ones in terms
of monetary costs for low heights or small spans as well as for deep
either L/D or H/D ratios which is an important result because it shows structures. However, as either the height or the span or the slen-
that for this structural configuration there is a wide range of applic- derness increases or when stringent deflection limits are required,
ability including ordinary loading scenarios and deep structures. the adaptive solution becomes increasingly less expensive than the
Generally, it can be said that for high L/D ratios, low height or short passive one (Section 7).
span structures as well as for deep structures (i.e. low H/D or S/D ra-
tios), it is not effective to use active control. When the L/D ratio is high
• Even in those cases when the adaptive design is more expensive, the
extra cost with respect to a passive structure is not wasted but rather
the operational energy becomes dominant resulting in lower energy it is used to reduce the environmental impact of the structure. In this
savings. In addition, for low height or small span structures as well as regard, adaptive structures can be thought of as energy saving de-
for deep structures, the monetary cost of the adaptive solution is gen- vices. A comparison of the cost of saving energy using structural
erally higher than that of the passive one. Conversely for tall or long adaptation and that of producing energy using other technologies
span structures as well as for slender structures (i.e. high H/D or S/D e.g. PV, shows that adaptive structures are efficient energy saving
ratios), it is very effective to control deflections actively. In these cases, devices (Section 7).
the adaptive solution is competitive in monetary terms with the passive
solution and becomes cheaper as the height or span and the slenderness
• The optimal region in which adaptive structures outperform passive
structures in energy and monetary cost terms is broadly the region
increase. of stiffness-governed structures (Section 8).

627
G. Senatore et al. Engineering Structures 167 (2018) 608–628

Adaptive structures also fulfil other functions such as being ex- engineering structure. Comput Struct 2008;86(23–24):2215–23.
tremely slender and being capable of reducing deflections completely [21] Veuve NSS, Smith I. Deployment of a tensegrity footbridge. J Struct Eng
2015;141(11):1–8.
thus meeting strict deflection limits at the expense of a small amount of [22] Santos FRA, Micheletti A. Design and experimental testing of an adaptive shape-
operational energy. This could bring several benefits including (1) morphing tensegrity structure, with frequency self-tuning capabilities, using shape-
buildings can be taller, roofs wider and bridges longer (2) buildings can memory alloys. Smart Mater Struct 2015;24:1–10.
[23] Campanile LF. Initial thoughts on weight penalty effects in shape-adaptable sys-
have increased floor space via reduction of structural cores (3) people tems. J Intell Mater Syst Struct 2005;16:47–56.
comfort and the overall structural integrity can be improved reducing [24] Hasse A, Campanile LF. Design of compliant mechanisms with selective compliance.
deflections in real-time. Smart Mater Struct 2009;18(11):115016.
[25] Jenkins C. Compliant structures in nature and engineering, 1st ed., WIT Press; 2005.
The structures taken into consideration in this article are reticular. [26] Previtali F, Ermanni P. Performance of a non-tapered 3D morphing wing with in-
Future work could look into whether these conclusions hold for frames tegrated compliant ribs. J Smart Mater Struct 2012;21:1–12.
as well as continuous structural types (e.g. slabs and shells). [27] Lienhard J, Schleicher S, Poppinga S, Masselter T, Milwich M, Speck T, et al.
Flectofin: a hingeless flapping mechanism inspired by nature. Bioinspiration
Biomimetics 2011;6:1–7.
Acknowledgements [28] Soong T, Cimellaro G. Future directions in structural control. Struct Control Health
Monit 2009;16:7–16.
The authors gratefully acknowledge the Engineering and Physical [29] Korkmaz S. A review of active structural control: challenges for engineering in-
formatics. Comput Struct 2011;89:2113–32.
Sciences Research Council (EPSRC) who provided core funding for this [30] Teuffel P. Entwerfen adaptiver strukturen Doctoral dissertation Struttgart:
project through UCL Doctoral Training Centre in Urban Sustainability University of Stuttgart - ILEK; 2004.
and Resilience (Grant EP/G037698/1), as well as Expedition [31] Begg D, Liu X. On simultaneous optimization of smart structures - Part II:
Algorithms and examples. Comput Methods Appl Mech Eng 2000;184:25–37.
Engineering the project industrial partner who provided significant [32] Sobek W, Teuffel P. Adaptive Systems in Architecture and Structural Engineering.
additional resources. EPFL Applied Computing and Mechanics In: Proc SPIE 4330, Smart Structures and Materials 2001: Smart Systems for
Laboratory (IMAC) is thankfully acknowledged for their support during Bridges, Structures, and Highways, Newport Beach, CA, United States; 2001.
[33] Utku S. Theory of Adaptive Structures: incorporating intelligence into engineered
the review process of this article. products. Boca Raban, Florida, U.S: CRC Press LLC; 1998.
[34] Senatore G, Duffour P, Hanna S, Labbe F, Winslow P. Adaptive structures for whole
References life energy savings. International Association for Shell and Spatial Structures (IASS),
vol. 52, no. 4 December n. 170; 2011, p. 233–40. URL: < https://www.
researchgate.net/publication/261724310_Adaptive_Structures_for_Whole_Life_
[1] Soong T, Chang J. Active vibration control of large flexible structures. Shock Vib Energy_Savings > .
Bull 1982;52:47–54. [35] Senatore G, Duffour P, Winslow P. Whole-life energy and cost analysis of adaptive
[2] Soong TT. State of the art review: active structural control in civil engineering. Eng structures - case studies. J Struct Eng (ASCE); 2018 [in press]. http://dx.doi.org/10.
Struct 1988;10:74–84. 1061/(ASCE)ST.1943-541X.0002075.
[3] Abdel-Rohman M, Leipholz H. Active control of tall buildings. J Struct Eng [36] Senatore G, Duffour P, Winslow P, Wise C. Shape control and whole-life energy
1983;109(3):628–45. assessment of an “Infinitely Stiff” prototype adaptive structure. Smart Mater Struct
[4] Reinhorn AST, Lin R, Riley M. Active bracing sytem: a full scale implementation of 2018;27(1):015022.
active control. Buffalo: National Center for Earthquake Engineering Research; 1992. [37] Senatore G, Duffour P, Winslow P, Hanna S, Wise C. Designing adaptive structures
[5] Bani-Hani K, Ghaboussi J. Nonlinear structural control using neural networks. J Eng for whole life energy savings. In: Proceedings of the fifth international conference
Mech 1998;124(3):319–27. on structural engineering, mechanics & computation, Cape Town, Taylor & Francis
[6] Rodellar J, Mañosa V, Monroy C. An active tendon control scheme for cable-stayed Group, London; 2013, p. 2105–10. URL: < https://www.researchgate.net/
bridges with model uncertainties and seismic excitation. Struct Control Health publication/261724301_Designing_adaptive_structures_for_whole_life_energy_
Monit 2002;9(1):75–94. savings > .
[7] Xu B, Wu S, Yokoyama K. Neural networks for decentralized control of cable-stayed [38] Huber JE, Fleck NA, Ashby MF. The selection of mechanical actuators based on
bridge. J Bridge Eng (ASCE) 2003;8:229–36. performance indices. Proceed Roy Soc A 1997;453(1965):2185–205.
[8] Schnellenbach MH, Steiner D. Self-tuning closed-loop fuzzy logic control algorithm [39] Senatore G. Adaptive building structures Doctoral Dissertation London: University
for adaptive prestressed structures. Struct Eng Int; 2013. p. 163–72. College London; 2016.
[9] Preumont A, de Marneffe B, Deraemaeker A, Bossensb F. The damping of a truss [40] Patnaik S, Gendy A, Berke S, Hopkins D. Modified Fully Utilized Design (MFUD)
structure with a piezoelectric transducer. Comput Struct 2008;86(3–5):227–39. method for stress and displacement constraints. Int J Numer Meth Eng
[10] Sobek W. Auf pneumatisch gestützten Schalungen hergestellte Betonschalen 1998;41:1171–94.
Doctoral dissertation Stuttgart: University of Stuttgart; 1987. [41] ENERPAC. “E328e Industrial Tools - Europe,” 2016. [Online]. Available: < http://
[11] Weilandt A. Adaptivität bei Flächentragwerken Doctoral dissertation Stuttgart: www.enerpac.com/en-us/downloads > [accessed 12 07 2017].
ILEK, University of Stuttgart; 2007. [42] Hammond G, Jones C. Embodied energy and carbon in construction materials.
[12] Neuhäuser S. Untersuchungen zur Homogenisierung von Spannungsfeldern bei Proceed Inst Civil Eng – Energy 2008;161(2):87–98.
adaptiven Schalentragwerken mittels Auflagerverschiebung Doctoral dissertation [43] Griffis L. Serviceability limit states under wind load. Eng J 1993;30(1):1–16.
Stuttgart: ILEK, University of Stuttgart; 2014. [44] Eurocode 1 Actions on structures, “Eurocode 1: Actions on structures — General
[13] Pellegrino S. Deployable structures. NewYork: SpringerWien; 2001. actions — Part 1–4: Wind”; 1991.
[14] Henry A, Kam C, Smith M, Lewis C, King M, Boulter N, et al. Singapore sports hub: [45] Colin T, Alister C. Interviewees, Quotation HQQ49103. [Interview]. 14 06; 2017.
engineering the national stadium. The Structural Engineer, vol. 94, no. 9; 2016. [46] Vishay. [Online]. Available: < http://www.vishaypg.com/micro-measurements/
[15] Fuller B. Tensile-Integrity Structures. U.S. Patent 3,063,521; 1962. > ; 2015. [Accessed December 2015].
[16] Calladine C. Buckminster Fuller's 'Tensegrity' structures and Clerk Maxwell's rules [47] TATA Steel. Sections Publications; 2012. [Online]. Available: < http://www.
for the construction of stiff frames. Int J Solids Struct 1978;14:161–72. tatasteeleurope.com/en/products_and_services/products/long/sections/sections_
[17] Skelton RE, de Oliveira MC. Tensegrity systems. New York: Springer; 2009. publications/ > [accessed 18 March 2013].
[18] Tibert G. Deployable tensegrity structures for space applications Doctoral dis- [48] N.R.E.L. NREL, A Performance Calculator for Grid-Connected PV Systems, [Online].
sertation Stockholm: Royal Institute of Technology; 2002. Available: < http://rredc.nrel.gov/solar/calculators/PVWATTS/version1/ > ;
[19] Fest E, Shea K, Domer B, Smith F. Adjustable tensegrity structures. J Struct Eng 2012 [accessed 02 April 2012].
2003;129:515–26.
[20] Adam B, Smith IF. Active tensegrity: a control framework for an adaptive civil-

628

View publication stats