1 s2.0 S0045794922001158 Main
1 s2.0 S0045794922001158 Main
1 s2.0 S0045794922001158 Main
a r t i c l e i n f o a b s t r a c t
Article history: The paper presents a novel framework, for the optimization of seismic retrofitting design of existing rein-
Received 10 February 2022 forced concrete (RC) frame structures. The framework is oriented to the to minimization of retrofitting-
Accepted 18 June 2022 related costs, simultaneously controlling the associated expected annual loss (EAL). The proposed proce-
Available online 25 July 2022
dure makes use of the capabilities offered by artificial intelligence (AI) techniques, adopting a genetic
algorithm (GA) based optimization routine, handling constrains with a non-penalty approach through
Keywords: the definition of innovative parent and survival selection operators. The framework implements multiple
Seismic retrofitting
retrofitting techniques optimization for the same structure, so that both serviceability and ultimate limit
Structural optimization
Genetic algorithm
states are simultaneously controlled. In the paper, the optimization procedure is applied to a case study
Expected annual loss structure, considering carbon fiber-reinforced polymers (CFRP) wrapping of columns and steel braces
FRP bracing as potential retrofitting interventions. For both, the framework provides the optimal position
Steel bracing (topological optimization) and design (sizing optimization). Results show that retrofitting costs and
EAL are effectively controlled by the proposed GA-based optimization approach.
Ó 2022 Elsevier Ltd. All rights reserved.
https://doi.org/10.1016/j.compstruc.2022.106855
0045-7949/Ó 2022 Elsevier Ltd. All rights reserved.
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
have addressed the issue of the optimization of fiber-reinforced paradigm, the assessment of the expected annual loss is evaluated
polymers (FRP) jackets [12–14] or dissipative bracings for RC frame by considering the reference design limit states and their probabil-
structures [15], fluid viscous dampers [16], or both [17]. More ity of occurrence during the nominal service life. Within the pro-
recent studies focused the topic of the optimization of seismic ret- posed framework, the optimal retrofitting configuration is
rofitting costs Among them, Papavasieliou et al. [18] proposed a determined in terms of reinforcement amount (sizing optimiza-
genetic algorithm (GA)-based optimization framework for encased tion) and position (topological optimization), simultaneously
steel–concrete composite columns through three different retrofit- assessing and controlling the resulting EAL through the definition
ting techniques: steel bracing, concrete jacketing, and steel jacket- of proper constraints. The GA-based optimization procedure has
ing, Falcone et al. [19] implemented a framework for the been specifically redefined to face these specific issues by introduc-
optimization of the costs for FRP jacketing and steel bracings of ing a hybrid design vector, made of Boolean and natural discrete
existing reinforced concrete (RC) frame structures through a variables. Moreover, a non-penalty approach was implemented
genetic algorithm (GA). A similar approach was pursued by Di Tra- to effectively handle the constraints with by two novel survival
pani et al. [20] who proposed an innovative framework based on a selection operators. Since EAL assessment involves different limit
genetic algorithm aimed at minimizing steel jacketing seismic ret- states fulfilment, the proposed framework can take also take into
rofitting costs for ductility deficient RC structures. This last frame- account multiple retrofitting interventions to provide additional
work was further enhanced in Di Trapani et al. [21] to tackle the strength, stiffness or ductility when needed. The framework is
optimization of steel jacketing for both ductility-critical and tested with the case study of a non-seismically conforming RC
shear-critical frame structures. Finally, Minafò and Camarda [22] frame building where two different retrofitting techniques are
provided a GA-based optimization procedure, for the minimization simultaneously considered, namely FRP wrapping of columns (to
of costs for the implementation of buckling-restrained braces on increase ductility and shear resistance) and steel bracings (to
reinforced concrete frames. enhance global lateral stiffness). The optimization process is car-
These very recent research activities focused on defining new ried out by connecting the GA optimization routine, which has
efficient algorithms for optimizing retrofitting interventions costs been implemented in MatLabÒ with a 3D fiber-section model anal-
under the constrain of checking structural safety as a feasibility ysed through the OpenSees software platform [26]. The structural
condition. However, a comprehensive retrofitting design, should performance of each tentative retrofitting configuration is evalu-
also address the expected annual loss (EAL) [23,24], as also ated starting from the results of static pushover analyses in the
requested by some recent technical guidelines (e.g. the Italian framework of the N2 method [27]. Results of the application of
guidelines for the classification of seismic risk of buildings, D.M. the proposed framework show that seismic retrofitting costs and
28/02/2017,n. 58 [25]). The expected annual loss has proved being resulting expected annual loss can be effectively controlled thanks
valid parameter for assessing seismic structural performance dur- to the aid of artificial intelligence.
ing the service life [23], estimating the mean annual expected eco-
nomic losses caused by seismic events that could occur during the
2. Optimization framework
reference service life of a structure. In this context, retrofitting
interventions are designed to maintain under a certain limit repair
2.1. Working principle
costs over the service life, and therefore, the achievement of every
limit state inducing an economical loss is checked.
The optimization framework herein proposed is based on a
Conceiving a retrofitting design framework gathering both costs
genetic algorithm (GA) optimization routine developed in
and EAL optimization is not simple, as in general, minimizing costs
MatLabÒ. The optimization algorithm relates a structural model
under the constrain of achieving ultimate limit state safety does
implemented in the OpenSees software platform with the GA opti-
not ensure a feasible performance in terms of expected annual loss
mization process. Genetic algorithms are a class of metaheuristic
(Fig. 1). However, controlling EAL in retrofitting design results
algorithms inspired by the evolution theory. This optimization pro-
quite important to maintain the annual rate of exceedance of ser-
cedure analyzes the search space by evaluating the objective func-
vice and ultimate limit states under a design limit.
tion by points, and proceeds in the search for the minima based on
Based on these considerations, this paper presents a new GA-
the mixing of the set of design parameters (called genome) that
based framework aimed at optimizing retrofitting interventions
have had the best results in each iteration. This provides generat-
on RC frame structures to minimize retrofitting costs and simulta-
ing populations of tentative solutions (individuals) that, in the cur-
neously controlling service-life costs by checking EAL performance.
rent case, represent different retrofitting arrangements. Each
In agreement with the performance-based earthquake engineering
individual handled by the algorithm is characterized by a design
2
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
vector gathering all the design variables to be optimized. Design 2.2. Evaluation of the expected annual loss
parameters can define the position of a reinforcement intervention
(i.e. in which structural element the reinforcement is applied) or its The expected annual loss represents the annual loss of eco-
amount (i.e. how much to reinforce a structural element). The opti- nomic value of a structure in its reference service-life, taking into
mization involves the definition of a proper objective function that account the associated seismic risk, which essentially depends on
estimates the intervention costs of each tentative solution. The fea- the site hazard and the structural vulnerability. Economic losses
sibility of each solution is assessed by performing a pushover anal- are intended as the average annual cost needed to repair the dam-
ysis of the structure and evaluating the performance in terms of ages and cover losses induced by seismic events. The determina-
safety indexes for the different limit states (1E,LS). EAL is simultane- tion of the EAL requires therefore the assessment of the
ously evaluated as a function of the obtained seismic safety performance of the structure with respect to different limit states
indexes and compared to the reference one for a code-compliant (LS), associated with the respective return periods TR,LS and mean
structure (EALcc). In this way, only the retrofitting cost is mini- annual rate of exceedance, expressed as the inverse of the return
mized during the iterations, while EAL is indirectly included in periods (kLS = 1/ TR,LS). The achievement of a limit state is associated
the optimization as a constraint. The outcome of the optimization with a specific repair cost, which can be computed using different
will provide the minimum cost retrofitting solution among those approaches.
feasible both in terms of safety and EAL. A flowchart of the pro- A simplified method to compute EAL has been proposed by
posed framework is depicted in Fig. 2. As it can be observed, the Cosenza et al. [24]. According to this approach, EAL curve connects
optimization algorithm is the core of the framework, while engi- the annual rate of exceedance for each limit state with the respec-
neering decision phases are up on the designer. The ‘‘Initial engi- tive repairing cost with straight segments. The major advantage is
neering decision” allows defining a restricted design space (e.g. that repairing costs associated with each limit state are defined as
limiting the optimization to a portion of building instead of the fixed percentages of the costs for the complete reconstruction of
whole or to a restricted number of variables. This can dramatically the structure.
reduce the computational cost of the optimization. Finally, the According to [24], eight limit states are considered (Fig. 3). Ulti-
optimal solution found by the algorithm should not be intended mate limit states are the reconstruction limit state (RLS), the col-
as an absolute one, but the intent of the framework is providing lapse limit state (CLS) and the life safety limit state (LSLS), that
the designer with a reduced number of cost and EAL effective ret- are associated with a repair cost (%RC) equal to 100%, 80% and
rofitting configurations to support the ‘‘final engineering decision”. 50% respectively. Service limit states are the damage limitation
3
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
kOLS ¼ 1:67 kDLLS
ð4Þ
kCLS ¼ 0:49 kLSLS
Therefore, EAL can be simply assessed once kDLLS and kLSLS are
obtained from structural analysis. These two parameters, accord-
ing to Eqs. (2–3), are direct functions of the safety indexes evalu-
ated for LSLS and DLLS (e.g. by performing a pushover analysis).
In this framework, EAL compliance is directly related to the out-
comes of seismic structural performance with respect to the refer-
ence limit state demands. Fig. 4 summarizes the relationship
between safety assessment in the acceleration-displacement (AD)
format and EAL according to the above described simplified
approach for EAL assessment. In the context of the proposed opti-
mization algorithm, examples represented in Fig. 4, can be inter-
preted as possible outcomes from potential tentative retrofitting
solutions iteratively analysed within the optimization process.
The objective function and the assessment process will be
described in detail in the following sections.
Fig. 3. Reference EAL curve for a code-compliant building.
limit state (DLLS) and the operational limit state (OLS) with a %RC 2.3. Encoding of the design vector
of 15% and 7% respectively. The initial damage limit state (IDLS) is
characterized by a repair cost equal to zero and a mean annual fre- EAL code-compliance requires different limit state verification,
quency of exceedance that is conventionally assumed hence retrofitting design should simultaneously consider more
askIDLS ¼ 10%. Table 1 summarizes the reference limit states than one retrofitting system (e.g. to provide additional stiffness,
together with the respective return periods (TR,LS) and mean annual additional ductility or additional strength). To this goal the pro-
frequencies (kLS) for a structure having a nominal life (VN) of posed framework considers both the potential application of CFRP
50 years. (or simply called FRP) wrapping of RC columns (providing addi-
EAL is evaluated as the area under the curve connecting kLS - % tional confinement and shear resistance and concentric steel brac-
RCLS points for each limit state (Fig. 3), so that it can be calculated ing (providing additional stiffness). The decision variables that
as: encode the position and sizing of both retrofits are collected into
the so-called design vector. The choice of the parameters to opti-
X
5
EAL ¼ 0:5 kLSði1Þ kLSðiÞ %RC LSðiÞ %RC LSði1Þ þ kCLS %RC RLS mize and those to fix as a problem data is made in the first decision
1¼2 phase (Fig. 2). With reference to the considered retrofitting devices
ð1Þ the most meaningful design variables have been individuated as
the number of braced frame fields (nbr), the cross-section area of
For a code-compliant building, namely a building having a braces (Abr), the CFRP strips spacing (sFRP), the number of FRP layers
capacity exactly coincident with the demand for each limit state (nFRP), and the position of the columns retrofitted by the CFRP
(TrC,LS = TrD,Ls), Eq. (1) results in EAL = EALcc = 1.13 %RC. (Fig. 5). All the decision variables are gathered in the design vector
The evaluation of the EAL for a generic structure provides the b so defined:
determination of the capacity return period for each limit state
T
(TrC,LS) based on the determination of the safety index (fE;LS ), as b ¼ ð nbr Abr sFRP nFRP pÞ ð5Þ
follows:
in which the term p is an array of binaries representing the
g
T rC;LS ¼ T rD;LS fE;LS ð2Þ position of the FRP retrofitted columns defined as:
T
where the parameter g can be fixed, as g = 1/0.41 [24]. The cor- p ¼ ... ... cij ... ð6Þ
responding mean annual frequencies of exceedance are then
obtained as: where the generic element cij, is a Boolean variable assuming
the value 1 if the column is retrofitted and 0 if not. The subscript
kLS ¼ 1=TrC;LS ð3Þ i indicates the position of a column in plan and j the storey. The
According to [24], the annual rates of exceedance for the oper- variable nFRP, is a natural value indicating the number of overlap-
ational and collapse limit states can be obtained as a function of ping layers of FRP on each column belonging to the interval:
those evaluated for DL and LS limit state, so that: nFRP 2 ½1; nFRP;max ð7Þ
where nFRP;max is the maximum allowed number of FRP layers.
Table 1 The variable sFRP is a natural discrete values representing the FRP
Mean annual frequency of exceedance (k) and repair costs (%RC) associated with each
LS for a code-compliant building.
strip spacing (intended as the interaxis distance) and belonging
to the interval:
Limit state %RC TrD,LS kLS = 1 / TrC,LS EALcc
[%] [years] sFRP 2 ½bFRP ; sFRP;max ð8Þ
[%]
where sFRP;max is the maximum allowed spacing between two
RLS 100 1 0.00 1.13%
CLS 80 975 0.10 strips, and bFRP is the width of the CFRP fabric. If sFRP = bFRP the
LSLS 50 475 0.21 wrapping is continuous along the column.
DLLS 15 50 2.00 Each assignment of the design variables represents a candidate
OLS 7 30 3.33
retrofitting solution. Candidate solutions are generated and han-
IDLS 0 10 10.0
dled by the GA as described in detail in the following sections. In
order to reduce the research space dimensions and so the required
4
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Fig. 4. Relationships between capacity demand assessment for LSLS and DLLS and resulting EAL curve: a) DLLS not satisfied; b) LSLS not satisfied; c) DLLS and LSLS not
satisfied; d) DLLS and LSLS satisfied.
Fig. 5. Representation of the design variables: (a) Generic FRP and bracing retrofitting configuration; (b) Typical arrangement of bracings; (c) Typical arrangement of CFRP.
computational effort to perform the optimization, the following where C br is the cost related to the bracing arrangement and
simplifying hypotheses are introduced: CFRP is the one for the realization of the FRP wrapping of columns.
Both the terms consider material and manpower costs and the nec-
i) FRP fabrics have fixed width (bFRP) and thickness (tFRP). essary works for the demolition and restoration of adjoining plas-
ii) The number of FRP layers and the spacing between strips is ter and infills. The first term Cbr can be evaluated as:
constant for all the retrofitted columns.
iii) Bracing cross-section dimensions are the same for every
frame. X
nbr
C br ¼ W br;i cbr þ nbr cbr;m þ nbr cshear ð10Þ
iv) Mechanical properties of the materials used for both the ret-
i¼1
rofitting systems are constant.
where cbr is the manpower and material cost per unit weight
Simplifying hypotheses allow reducing computational time but (estimated in cbr = 6 €/kg), cbr,m is the fixed cost related to the
can be removed if needed. At the same time, more simplifying demolition and restoring works to include bracings within an infill
hypotheses can be added when dealing with large problems. (2000 € per braced frame field) and W br;i is the weight of the brac-
ings in the i-th frame field evaluated as:
2.4. Definition of the objective function
2
The objective function evaluates the retrofitting costs of each /br
W br ¼ 2 Lbr p cs ð11Þ
tentative solution as a function of the design vector. Therefore, in 2
is general form, the functional to minimize is:
in which Lbr is the length of one steel brace and cs is the unit
F ðbÞ ¼ C br þ C FRP ð9Þ weight of steel (78:5 kN=m3 ). The term cshear is the fixed cost asso-
ciated with the shear reinforcement of the ends of the columns
5
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
adjacent to the bracing systems to support the additional shear 3.2. Parent selection operator
demand. As regards CFRP reinforcement, the term C FRP is computed
as: Parent selection is the subroutine entrusted with selecting par-
ent individuals to be forwarded to the genetic operators (mutation
X
nc
and crossover) for the generation of a new population. Parent indi-
C FRP ¼ AFRP;i cFRP þ nc cFRP;m ð12Þ viduals are individuated with a tournament selection, directly
i¼1
comparing the individuals performance. Within the individuals of
where nc is the number of retrofitted columns, cFRP is the unit the initial populations, k individuals are randomly selected (where
cost for the arrangement of the CFRP wrapping (estimated in k is called tournament size), among them the best individual is
cFRP = 300 €/m2), cFRP,m is the cost per column for the demolition chosen as the parent for the ensuing mating process (Fig. 7). To
and reconstruction of adjacent masonries and plasters (cFRP, make the algorithm effective in selecting performing parent indi-
m = 1000 € per column) and AFRP,i is the area of the FRP fabric used viduals, the parent selection operator has been implemented to
to retrofit the generic i-th column, evaluated as: compare both individual fitness Eq. (9) and constraint violations
(Eq. (14)). The comparison is carried out firstly by assessing the
lc;i bf number of constraints violated (which according to Eq. (14) can
AFRP;i ¼ þ 1 nFRP 2 ðbi þ hi Þ ð4 pÞ r 2c bf ð13Þ
sFRP be 0, 1 or 2) and then with respect to the fitness value. In this
way, individuals with the lowest number of violated constraints
where lc;i is the length of the i-th column bi and hi are the geo-
are preferred. Then, for the individuals having the same number
metric dimensions of each column cross-section, and rc is the
of constraints violations the individual with the best fitness is cho-
rounding radius of columns edges. It is noteworthy observing that
sen (Fig. 7). Tournament size is the parameter that rules the selec-
by fixing some parameters on the characteristics of the retrofitting
tive pressure provided by parent selection. An increase in size leads
systems (such as the width of the FRP and the unitary costs) the
to a more thorough selection of best individuals, a reduction of k
objective function strictly depends on the parameters gathered in
allows a greater probability of been selected even for individuals
the design vector. The optimization problem is then formalized as:
with worse performances.
8
> min FðbÞ
>
> 3.3. Crossover and mutation
< s:t:
ð14Þ
> k
> DLLS 6 kccDLLS Crossover and mutation operators are employed to generate
>
:
kLSLS 6 kccLSLS child individuals with the aim to improve the chromosomes of
the genomes of each current generation. Crossover operator works
meaning that cost minimization is constrained by limiting the by mixing chromosomes of two selected parent individuals to gen-
annual rated of exceedance of each limit state (kDLLS, kLSLS) to that erate a new offspring. Mutation operator is instead used to intro-
of the code compliant building (kccDLLS, kccLSLS), that implies: duce a random variation in parent genome.
For the crossover subroutine, a new specific procedure has been
EAL 6 EALcc ð15Þ
defined to correctly handle heterogeneous genomes such as those
The feasibility (or not) of each tentative solution is performed of the proposed framework, which is composed of natural numbers
by a non-penalty approach, making use of a specifically defined and Boolean variables. The crossover of natural decision variables
survival selection operator, as described in the following sections. is carried out by randomly selecting a value among the chromo-
This approach is more affective in this case, as it allows to handle somes of the parents. In particular, the proposed crossover func-
multiple constraints without performing a specific calibration of tion, which can be called random intermediate value crossover, is
the penalty for each of them. implemented to smoothly mix the parent’s chromosomes. In this
way, each new chromosome is obtained as a random value
between those belonging to parents including extreme values
3. Optimization algorithm subroutines (Fig. 8). The choice of implementing this special crossover function
is justified by the fact that, for the current problem, the employ-
3.1. Algorithm working principle ment of traditional crossovers, such as the widely used one-
point, multi-point, or uniform crossover functions, can be disrup-
The search for the optimal solution follows a GA workflow. tive, creating child individuals that have characteristics too differ-
Specific modifications are introduced within standard genetic ent from the parents.
operators in order to improve their effectiveness on the optimiza- As regards Boolean variables, a single point crossover is
tion process. A scheme of the GA framework working principle is employed (Fig. 8). The operating principle provides the random
illustrated in Fig. 6. A population of random individuals encoded selecting of a position along the string called crossover point. The
by the design vectors is initially generated. By analysing each of child individual is constructed by picking, from the beginning of
the population’s tentative solutions, the fitness value and the num- the chromosome to the crossover point the genome of the first par-
ber of violated constraints are estimated. Then, the best individuals ent, while the rest is copied from the second one. In this way, each
are selected by the parent selection operator to generate a new parent passes down a sequence of genetic information (a substring
population. Two different genetic operators are employed to of binaries) to the offspring ensuring a adequate mixing of the
improve the genes of the individuals, the crossover, which mixes genomes.
the genomes of the best individuals, and the mutation that pre- As regards mutation, this is applied by selecting one (or more)
vents from stuck on local optima by introducing random slight random positions along the design vector and changing the value
changes in the design vectors. Lastly, the best individuals of the of that decision variable. For discrete variables a new random value
new generation, together with those of the initial population are among the possible ones is drawn. For Boolean variables included
selected through the survival selection operator. The framework in the p sub-vector the mutation of a gene is simply a switch from
proceeds until one of the stopping criteria (maximum number of 0 to 1 or vice-versa. It should be finally observed that crossover and
generations or stall) is achieved. The genetic operators’ subroutines mutation are not used sequentially (as it is usually done) but are
are described in detail in the following sections. alternatively used to generate new individuals (Fig. 6).
6
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Mutation
Crossover
# violated
Genome Picking k Selection from the
Fitness constrains
individuals at tournament the
random best individual
6 2
4 0 4 0
3 1 3 1
4 0
2 2
5 0 5 0
9 1
3 0
2 2
3.4. Survival selection operator The operating principle provides a double sorting process, first
ordering individuals based on the number of violated constraints,
Survival selection operator is used to allow the best individuals and then, among the individuals with the same number of con-
to spread their genome in subsequent generations. The initial pop- straint violations sorting is based on the fitness value. In this
ulation is merged with child individuals, and both undergo the sur- way, the individuals are sorted in ascending order starting from
vival selection process. In this way individuals of the initial individuals with the minimum (or null) number of violated con-
population having good fitness are forwarded to the next genera- straints and the minimum fitness to individuals with maximum
tion, together with most performing children. numbers of constraints violation and a high fitness value. At the
A new survival selection function called sorting and truncation end of the process, only the best individuals are taken for the next
selection is implemented to compare individual both from the fit- generation by eliminating the genomes at the end of the sorted list.
ness (Eq. (9)) and constraints (Eq. (14)) point of view. Considering A scheme of the proposed survival selection operator is illustrated
Eq. (14) violated constrains can be 0, 1 or 2. in Fig. 9
2 10 4 300 1 0 1 0 0 1 0 0 1 0
Parent 1
2 10 4
300
3 20 3
200 4 40 3 300 1 0 1 0 1 1 0 1 0 1
4 30 2
100 Child
5 40 1
5 40 1 100 0 0 0 1 1 1 0 1 0 1
Parent 2
7
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Population dimension
4 0 1 0 1 0
1 0 6 0 6 0
2 1 1 1 1 1
4 1 2 1 2 1
1 1 4 1 4 1
4 2 2 2 2 2
2 2 4 2 4 2
This proposed sorting and truncation selection function has solution. Any FE software accomplishing non-linear static analyses
proved to be a valid operator to manage constrained optimization can be connected to the optimization framework. For the current
problems without involving penalty approaches. In fact, penalty application, the OpenSees software platform [26] has been used.
approach requires a burdensome calibration procedure of the The frame elements (beams and columns) are modelled using
weight of the penalties to be assigned to the boundaries. On the distributed plasticity force-based elements with five Gauss-
other hand, the proposed approach could result less performing Lobatto integration points (Fig. 11). Steel braces and infill equiva-
but certainly more robust, as it allows a proper implementation lent struts are modelled with fiber-section trusses. For concrete
of the optimization without expensive calibrations and specific elements the Concrete02 uniaxial material model is used. The latter
GA expertise from users. is calibrated according to confinement action of stirrups and FRP
wrapping (for reinforced elements) as shown in the following sec-
3.5. Heuristic repair tion. Steel rebars and braces are modelled using the Steel02 Giuffrè-
Menegotto-Pinto model (elastoplastic with linear strain harden-
Heuristic repair of the design vector is used to force some vari- ing). Infills are modelled as equivalent struts according to the
ables to assume a specific value if some conditions occur. In the model by Di Trapani et al. 2018 [28] also using Concrete02 material.
current application this was necessary to provide shear reinforce- Finally, in order to simulate crushing of the cross-section fibers,
ment of side columns of braced spans. To introduce CFRP reinforce- Concrete02 and Steel02 materials are combined with MinMax mate-
ment into the design vector (if not already present in the rial, which removes the contribution of a fiber when a specified
retrofitting configuration) the heuristic repair routine controls strain threshold is achieved (Fig. 11).
the position of the bracings and adjusts vector p so that columns
adjacent to the braces are reinforced with the same spacing and 4.2. Modelling reinforced concrete elements with and without CFRP
number of layers that are encoded into the design vector. reinforcement
A sample of the working principle of the heuristic repairs sub-
routine is illustrated in Fig. 10. The confined concrete model adopted for RC elements with and
without retrofitting is the standard confined parabola rectangle
4. Reference structural model constitutive model (Fig. 11). According to the CNR-DT200 [29]
and Eurocode 8 [30], this model can be adapted for concrete ele-
4.1. General assumptions ments confined by only stirrups or stirrups and FRP wrapping.
For the sake of brevity, the formulation of the model is here
The proposed framework is interfaced with a FE software to briefly recalled only for the RC elements wrapped by FRP to high-
perform structural analysis and assessment of each tentative light the dependence with the parameters included in the design
1 40 1 200 0 1 1 0 0 1 0 0 0 1 40 1 200 0 1 1 1 0 1 1 1 0
Original design vector Modified design vector
8
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Fig. 11. Definition of the fiber-section elements in OpenSees, concrete elements with and without FRP wrapping, masonry infills, and steel bracings.
vector, namely nFRP and sFRP. In detail, the confined peak stress (fcc) where b and h are the cross-section dimensions (Fig. 12a) and
is obtained as: tFRP is the thickness of FRP wrapping that can be evaluated as:
f l;eff tFRP ¼ nFRP tFRP;1 ð21Þ
f cc ¼ f c 1 þ 2:6 ð16Þ
fc
where nFRP is the number of FRP layers, also encoded in the
where fc is the peak stress of the concrete confined by stirrups design vector (Eq. (5)), while tFRP,1 is the thickness of the fabric.
and fl,eff is the effective lateral confinement pressure that can be In Eq. (17) the coefficient keff represents the confining efficiency
evaluated as: exerted by the FRP system that can be calculated as the product of
f l;eff ¼ keff f l ð17Þ three coefficients:
where fl is the confinement pressure exerted by the FRP that can keff ¼ kv ka kh ð22Þ
be calculated as:
1
fl ¼ q Ef efd;red ð18Þ
2 f
in which Ef is the elastic modulus of the FRP fabric (along the
fiber direction), efd;red is the reduced FRP peak strain that, in case
of ductility evaluations, can be obtained as:
efk
efd;red ¼ ga 6 0:6 efk ð19Þ
cf
where efk is the design rupture strain of FRP reinforcement, ga is
the environmental reduction factor (set as 0.90), and cf is a partial
safety factor of the FRP (set as 1.10). In Eq. (18), qf is the geometric
reinforcement percentage that, in the case of a rectangular cross-
section is:
2 t FRP ðb þ hÞ bFRP
qf ¼ ð20Þ
b h pFRP
Fig. 12. Reference configuration and geometric parameters of a column retrofitted
by FRP wrapping: a) in plan; b) along the height.
9
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
where kv and kh are the vertical and horizontal confining effec- response in compression for a reference column cross-section are
tiveness coefficients and ka is an effectiveness is related to the fab- reported in Fig. 13 by varying the design variables nFRP and sFRP.
ric tilt. According to Eurocode 8 [30], for a rectangular cross-
section, these coefficients can be evaluated as. 5. Structural analysis and results post-processing
p0f 2
kv ¼ 1 2minfb;hg ; ka ¼ 1
; 5.1. Feasibility assessment by pushover analysis
2
1þðtan af Þ ð23Þ
kh ¼ 1 ðb2rc Þ3bh
2
þðh2rc Þ 2
Safety indexes (1E,LS) are used to estimate EAL through the eval-
uation of the mean annual rates of exceedance (kLS) (Eqs. (2–3)).
In previous equations af is the tilt angle with respect to the lon- Nonlinear static analysis (pushover) combined with N2 method
gitudinal direction of the RC element, rc is the radius of the column [27] is used to assess safety indexes for DL and LS limit states,
edge rounding and s0FRP (Fig. 12b) is the distance between FRP strips which are computed by assessing the following capacity / demand
that can be easily calculated as: ratios:
8 dDL;c
s0FRP ¼ sFRP bFRP ð24Þ < fE;DLLS ¼ S
de;DLLS ðT Þ
ð26Þ
As it can be observed, the design variable sFRP appears in Eq. :f ¼
PGAc;LSLS
E;LSLS PGAd;LSLS
(24). Finally, the confined peak strain (ecc )ultimate strain (eccu )
are evaluated as: where T* is the period of the bilinear equivalent single degree of
freedom system, Sde;DLLS ðT Þ is the displacement demand from the
2
elastic DLLS spectrum and dDL;c is the top displacement associated
ecc ¼ ec0 f cc
fc
qffiffiffiffiffiffi ð25Þ with the achievement of the damage limitation condition. The lat-
eccu ¼ 0:0035 þ 0:015 f l;eff
fc
ter is conventionally achieved when the stress on the most com-
pressed equivalent stuts reaches 50% of the maximum resistance
where ec0 is the peak strain of the concrete confined by stirrups. of the strut (fmd0) (Fig. 14a). In (Eq. (26)), PGAd;LSLS is the peak
The effect of FRP retrofitting is introduced into reinforced con- ground acceleration demand associated with the reference seismic
crete elements by simply modifying the constitutive model of con- hazard (LS limit state), while PGAc;LSLS is the peak ground accelera-
crete fibers. For the sake of simplicity it is assumed that the tion associated with the earthquake inducing life safety limit state
confining effect exerted by the FRP wrapping is extended to the (shear collapse or maximum base shear reduction larger than 15%)
entire cross-section fibers. Samples of the resulting stress–strain (Fig. 14b).
Fig. 13. Samples of stress–strain response in compression of concrete with and without FRP retrofitting: a) by varying the number of FRP layersðnFRP Þ; b) by varying the
spacing of stripsðsFRP Þ.
Fig. 14. Limit state conditions: a) Damage limitation (DLLS); b) Life safety (LSLS).
10
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Once 1E,LS are obtained, mean annual rates of exceedance (kDLLS 1 bFRP
V Rf ¼ 0:9 d f fed 2 tf ðcot hÞ ð31Þ
and kLSLS) are evaluated by Eqs. (2–3). After EAL is computed by Eq. cRd sFRP
(1). Constraint violations are finally assessed by Eq. (14).
where cRd = 1.2 is the partial safety factor, d is the are effective
width of the concrete cross section, h is the inclination of the com-
5.2. Shear verification of RC elements with and without CFRP
pressed concrete struts, ffed is the effective design strength depend-
reinforcement
ing on the debonding strength of the FRP.
Fig. 15. Geometrical dimensions of the reference structural model: (a) 3D frame view; (b) Plan view and RC element cross-section dimensions.
11
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
masonry infills are summarized in Table 3. The structure has dou- fyb = 275 MPa, elastic modulus Esb = 210 GPa and strain hardening
ble symmetry in-plan, and is regular in elevation. Vertical loads are ratio g = 0.01.
modelled as point loads applied to the top node of each column as In order to limit the design space dimension, and so the compu-
a function of the respective tributary areas in-plan. Rigid dia- tational burden, the following restrictive hypotheses are also
phragm behaviour is imposed at every floor. made:
Table 2
Geometrical dimensions and reinforcement details of RC elements.
Table 3
Geometric and mechanical modelling parameters for RC elements and infill equivalent struts.
12
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Fig. 16. As-built configuration preliminary analysis: (a) SR capacity / demand assessment; (b) SR EAL curve; (c) SC capacity / demand assessment; (d) SC EAL curve.
Table 4
Results of the assessment of the as-built structures.
6.4. Optimization results and discussion The convergence histories of the optimization analyses for SR
and SC structural configurations are illustrated in Fig. 18. It can
The proposed optimization framework has been tested with the be observed that a proper definition of the genetic operators
two case-study structural configurations above described. The allowed a gradual transition between the exploration and exploita-
effectiveness of the algorithm in providing suitable retrofitting tion phases. The convergence trends are quite fast and regular,
design solutions in terms of minimum cost with a code- without significant stalls in local minima due to potential prema-
conforming EAL has been assessed. The analyses have been carried ture convergences. As regards SR configuration, the solution with
out using an initial population (P) of 100 tentative solutions. The minimum cost at the first generation was 58.532 €. The optimal
generation of the initial population was performed as suggested one was 28.686 € (-51%) after 20 generations. For SC configuration,
by Di Trapani et al. [21] for GA retrofitting optimization of shear- as it was expected, the optimization required larger computational
deficient structures. In detail, a subspace (ppr) constituted by 50% effort (25 generations), however the initial minimum-cost solution
the population is reserved to individuals with a high probability (155.121 €) was reduced by 60% to 62.620 €, denoting a better per-
(Pr = 90%) that each column of the design vector is retrofitted. formance of algorithm despite the major number of generations
The algorithm proceeds by generating 100 new children every gen- required. In Fig. 18, it also noteworthy observing the trend of the
eration through the previously described routine, involving parent tentative solutions for each generation, which are clearly ordered
selection, crossover and mutation. A tournament size k = 3 is used as a function of their fitness. This is due to the application of the
to for the parent selection operator. Stopping criteria have been set proposed survival selection operator, putting in order the tentative
to at a maximum of 25 generations (Gmax) and a stagnation of 5 solutions by considering their fitness and the number of violated
generations (Smax), representing the maximum number of genera- constrains.
tions in which the algorithm does not improve the solution. GA In Fig. 19, the outcomes of the optimization are analysed in
parameters set-up is summarized in Table 5. terms of EAL. For both SR and SC configurations it can be observed
13
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Table 5
GA analysis parameters set up for the case studies.
Dimension of the Population Number of Tournament Mutation Max Max Dimension of the Prob. of retrofitting an element in
design vector size offspring size ratio generations stall ppr space the ppr space
dim(b) P O k Pm Gmax Smax ppr Pr
44 100 100 3 2% 25 5 50 90%
Fig. 19. Feasible individuals EAL values trend: a) shear-resistant configuration; b) shear-critical configuration.
14
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Fig. 20. Optimal solutions retrofitting configurations (deformed shape): (a) shear-resistant case; (b) shear-critical case.
Table 6
Case studies optimization analysis results.
Structural nbr (#) Abr nc sFRP nFRP fE;DL fE;LS EAL Fitness
configuration (cm2) (#) (mm) (#) (-) (-) (%RC) (€)
obtained EAL values were similar for the SR (EAL = 1.016%) and the The optimal solution found to the SC configuration (Fig. 20b), is
SC (EAL = 1.006%) configurations. characterized by FRP wrapping of 8 columns in the external trans-
The finally obtained optimal solutions are illustrated in Fig. 20, verse frames of the first two floors. In this retrofitting configura-
where the structures are depicted in their deformed shape under tion, FRPs are implemented with two layers (nFRP = 2) and a strip
the lateral loads applied along Z direction. Both SR and SC optimal spacing sFRP of 200 mm. It is noteworthy observing that this opti-
solutions are characterized by steel bracing of the external frames mal solution is consistent with the expected structural deficiencies,
for the first two stories (nbr = 2) with a cross-section area of the since retrofitted columns are those adjoining masonry infills. This
braces (Abr) of 2.83 cm2. This result is the same for both the config- means that the solution found by the algorithm has effectively pin-
urations, because this solution is the minimum cost-one allowing pointed the need for an increase in increase shear capacity of these
damage limitation limit state compliance. For the SR configuration columns to face the additional shear demand due to the frame-
it is found that bracing is the only retrofitting intervention needed infill interaction. Parameters obtained at the end of the optimiza-
(FRP reinforcement of columns shown in Fig. 20a is still associated tion are reported in Table 6. A representation of the convergence
with bracings). This result is consistent with the fact that SR struc- history of the retrofitting design parameters for both the structures
ture already satisfied LS limit state in the as-built configuration. is also depicted in Fig. 21. To provide a concise representation,
Fig. 21. Convergence history of the design vector parameters: a) SR configuration; b) SC configuration.
15
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
Fig. 22. Performance of the optimal solution for the SR configuration: (a) Capacity / demand assessment in AD format; (b) EAL curve.
Fig. 23. Performance of the optimal solution for the SC configuration: (a) Capacity / demand assessment in AD format; (b) EAL curve.
design vector parameters are here summarized by the total volume wrapping of the columns (used to the increase shear capacity),
of steel bracing (Vbr), the ratio between the number of FRP layers which influence the overall performance providing additional
and their spacing (nFRP/sFRP), the number of retrofitted columns strength and ductility.
(nc). It can be observed that parameters related to bracings and EAL curves associated with the optimal solutions are finally
FRP arrangement converged within the first 10 generations, while depicted in Figs. 22b and 23b and compared with those of the
the topological optimization of the retrofitted columns required respective as-built configurations. As it can be observed code-
more than double the generations. conforming EAL was obtained for both the cases. It should be also
The capacity curves of the two structural configurations are noted that the final EAL of both the structures (1.016% for the SR
depicted in Fig. 22a and 23a in the AD space. The latter show both structure and 1.006 for the SC structure) was quite close to the
an increase in lateral strength, stiffness and ductility. For what con- maximum allowed one (1.13%).
cerns DL limit state, since the bracing system was the same for SR
and SC configurations, identical safety indexes (close to unit) were 7. Conclusions
found (1E,DLLS = 1.024). On the other hand, LS limit state safety
indexes were significantly higher than the unit. In detail they were The paper presented a new GA-based optimization framework
1E,LSLS = 2.3 for the SR structure and 1E,LSLS = 2.7 for the SC structure. aimed at minimizing seismic retrofitting costs of RC frame struc-
The significant increase of LS safety is a secondary effect of the tures and simultaneously controlling service-life costs. The pro-
bracings (used to reduce lateral deformability) and of the FRP posed algorithm implements cost optimization of CFRP wrapping
16
F. Di Trapani, Antonio Pio Sberna and Giuseppe Carlo Marano Computers and Structures 271 (2022) 106855
of columns (used to improve ductility and shear resistance) and [5] Lagaros ND, Papadrakakis M, Kokossalakis G. Structural optimization using
evolutionary algorithms. Comput Struct 2002;80:571–89.
steel bracing (used to enhance lateral stiffness) providing the posi-
[6] Govindaraj V, Ramasamy JV. Optimum detailed design of reinforced concrete
tion and the amount of reinforcement as output (topological and frames using genetic algorithms. Comput Struct 2005;84:34–48.
sizing optimization). The effectiveness of the proposed optimiza- [7] Govindaraj V, Ramasamy JV. Optimum detailed design of reinforced concrete
tion framework has been tested with a non-conforming RC frame continuous beams using Genetic Algorithms. Engineering Optimization 2007,
39(4): 471–494.
structure with two different configurations (shear-resistant and [8] Babaei M, Mollayi M. Multi-objective optimization of reinforced concrete
shear-critical) subject to pushover analysis. Based on the obtained frames using NSGA-II algorithm. Engineering structures and technologies
results, the following concluding remarks can be stated: 2016;8(4):157–64.
[9] Mitropoulou CC, Lagaros ND, Papadrakakis M. Life-cycle cost assessment of
optimally designed reinforced concrete buildings under seismic actions. Reliab
Seismic retrofitting costs and expected annual loss, can be effec- Eng Syst Saf 2011;96:1311–31.
tively controlled during the design process by means of the [10] Papavasileiou GS, Charmpis DC. Seismic design optimization of multi–storey
steel–concrete composite buildings. Comp Struct 2016;170:49–61.
capabilities offered by the proposed artificial intelligence- [11] Pham TD, Hong WK. Genetic algorithm using probabilistic-based natural
based framework. selections and dynamic mutation ranges in optimizing precast beams. Comput
The use of a non-penalty approach is relevant to handle multi- Struct 2022;258:106681.
[12] Chisari C, Bedon C. Multi-Objective Optimization of FRP Jackets for Improving
ple constraints by means of the specifically defined parent the Seismic Response of Reinforced Concrete Frames. Am J Eng Appl Sci 2016;9
selection and survival selection operators. (3):669–79.
The algorithm has shown to be sufficiently robust and flexible [13] Seo H, Kim J, Kwon M. Optimal seismic retrofitted RC column distribution for
an existing school building. Eng Struct 2018;168:399–404.
in finding optimal retrofitting solutions even in the case of
[14] Mahdavi G, Nasrollahzadeh K, Hariri-Ardebili MA. Optimal FRP Jacket
structures with significantly different needs (e.g. stiffness lacks, Placement in RC Frame Structures Towards a Resilient Seismic Design.
shear deficiencies, ductility deficiencies). Sustainability 2019;11:6985.
The major flexibility of the algorithm is mostly due the fact that [15] Braga F, Gigliotti R, Laguardia R. Intervention cost optimization of bracing
systems with multiperformance criteria. Eng Struct 2019;182:185–97.
it implements multiple limit state verifications as well as mul- [16] Pollini N, Lavan O, Amir O. Minimum-cost optimization of nonlinear fluid
tiple retrofitting techniques. viscous dampers and their supporting members for seismic retrofitting. Earthq
Even if the algorithm is not aware of the geometrical and Eng Struct D 2017;46:1941–61.
[17] Lavan O, Dargush GF. Multi-objective evolutionary seismic design with passive
mechanical features of the structure, except for the values that energy dissipation systems. J Earthq Eng 2009;13(6):758–90.
design vector components can assume, the obtained optimal [18] Papavasileiou GS, Charmpis DC, Lagaros ND. Optimized seismic retrofit of
solutions are consistent with the engineering expectations in steel-concrete composite buildings. Eng Struct 2020;213:110573.
[19] Falcone R, Carrabs F, Cerulli R, Lima C, Martinelli E. Seismic retrofitting of
terms of position of the reinforced elements. existing rc buildings: a rational selection procedure based on genetic
Cost minimization associated with EAL control combined multi- algorithms. Structures 2019;22:310–26.
ple reinforcement techniques to match each limit state. This has [20] Di Trapani F, Malavisi M, Marano GC, Sberna AP. Greco, Optimal seismic
retrofitting of reinforced concrete buildings by steel-jacketing using a genetic
shown a significant increase of safety indexes for some limit algorithm-based framework. Eng Struct 2020;219:110864.
states because of the beneficial interaction of all the imple- [21] Di Trapani F, Sberna AP, Marano GC. A new genetic algorithm-based
mented retrofitting systems. framework for optimized design of steel-jacketing retrofitting in shear-
critical and ductility-critical RC frame structures. Eng Struct
The presented framework has been specialized for RC structures
2021;243:112684.
with some specific reinforcement techniques. However, the [22] Minafò G, Camata G. An open-source GA framework for optimizing the
overall conceiving has a general validity and can be further seismic upgrading design of RC frames through BRBs. Eng Struct
developed for different structural typologies and the respective 2022;251:113508.
[23] Calvi GM. Choices and criteria for seismic strengthening. J Earthq Eng 2013;17
retrofitting techniques. (6):769–802.
AI-guided design can be a quite useful tool to move from a trial- [24] Cosenza E, Del Vecchio C, Di Ludovico M, Dolce M, Moroni C, Prota A, et al. The
and-error design approach to an engineered one, providing Italian guidelines for seismic risk classification of constructions: technical
principles and validation. B Earthq Eng 2018;16:5905–35.
solutions which are suitable in terms of costs and performance. [25] Ministerial Decree 28/02/2017 n. 58, Guidelines for the classification of the
However, the outcomes of a so-defined approach should be seismic risk of buildings as well as the procedures for the certification, by
interpreted as an aid to the designer, who has to maintain the qualified professionals, of the effectiveness of the interventions, 2017. In
Italian.
control on final engineering decision. [26] McKenna F, Fenves GL, Scott MH. Open system for earthquake engineering
simulation. University of California Berkley, 2000.
[27] Fajfar P. A nonlinear analysis method for performance-based seismic design.
Earthq Spectra 2000;16(3):573–92.
[28] Di Trapani F, Bertagnoli G, Ferrotto MF, Gino D. Empirical equations for the
References direct defnition of stressstrain laws for fber-section based macro-modeling of
inflled frames. J Eng Mech 2018;144(11):04018101.
[1] De Martino G, Di Ludovico M, Prota A, Moroni C, Manfredi G, Dolce M. [29] Italian National Research Council. CNR-DT 200; Instructions for design,
Estimation of repair costs for RC and masonry residential buildings based on execution, and control of strengthening interventions through fiber-
damage data collected by post-earthquake visual inspection. Bull Earth Eng reinforced composites. 2013.
2017;15(4):1681–706. [30] European Committee for Standardization. Eurocode 8. Design of structures for
[2] Del Vecchio C, Di Ludovico M, Pampanin S, Prota A. Repair costs of existing RC earthquake resistance - Part 3: Assessment and retrofitting of buildings, 2005.
buildings damaged by the L’Aquila earthquake and comparison with FEMA P- [31] Di Trapani F, Malavisi M. Seismic fragility assessment of infilled frames subject
58 predictions. Earthq Spectra 2018;34(1):237–63. to mainshock/aftershock sequences using a double incremental dynamic
[3] Quaranta G, Lacarbonara W, Masri SF. A review on computational intelligence analysis approach. B Earthq Eng 2019;17:211–35.
for identification of nonlinear dynamical systems. Nonlinear Dyn [32] Biskinis DE, Roupakias GK, Fardis MN. Degradation of shear strength of
2020;99:1709–61. reinforced concrete members with inelastic cyclic displacements. ACI Struct J
[4] Falcone R, Lima C, Martinelli E. Soft computing techniques in structural and 2004;101(6):773–83.
earthquake engineering: A literature review. Eng Struct 2020;207:110269.
17