archives of civil and mechanical engineering 17 (2017) 738–749

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: http://www.elsevier.com/locate/acme

Original Research Article

Heuristics in optimal detailed design of precast road
bridges

Victor Yepes *, José Vicente Martí, Tatiana García-Segura,
Fernando González-Vidosa
Institute of Concrete Science and Technology (ICITECH), Universitat Politècnica de València, 46022 Valencia, Spain

article info abstract

Article history: This paper deals with the cost optimization of road bridges consisting of concrete slabs
Received 4 May 2016 prepared in situ and two precast-prestressed U-shaped beams of self-compacting concrete.
Accepted 20 February 2017 It shows the efficiency of four heuristic algorithms applied to a problem of 59 discrete
Available online variables. The four algorithms are the Descent Local Search (DLS), a threshold accepting
algorithm with mutation operation (TAMO), the Genetic Algorithm (GA), and the Memetic
Keywords: Algorithm (MA). The heuristic optimization algorithms are applied to a bridge with a span
Optimization length of 35 m and a width of 12 m. A performance analysis is run for the different heuristics,
Metaheuristics based on a study of Pareto optimal solutions between execution time and efficiency. The best
Bridges results were obtained with TAMO for a minimum cost of 104 184s. Among the key findings of
Overpasses the study, the practical use of these heuristics in real cases stands out. Furthermore, the
Structural design knowledge gained from the investigation of the algorithms allows a range of values for the
design optimization of such structures and pre-dimensioning of the variables to be recom-
mended.
© 2017 Politechnika Wrocławska. Published by Elsevier Sp. z o.o. All rights reserved.

The basic goal of structural optimization is to find a design
1. Introduction
having lowest cost, and ensuring predicted constraints.
Additionally, in most structural optimization problems, the
Precast-prestressed concrete (PPC), that is, pretensioned main drawback appears to be related to the constructability of
concrete beams with cast-in situ slabs, has been commonly the proposed design such as reinforcement placement, rules of
used by designers when building road bridges [1]. In this good design practice, construction management plan, and so
context, structural engineers have taken advantage of precast on [2]. The decisions taken in such a complex environment
technology by specifying designs that utilize standard beams require the development of new decisional tools and methods
of comparatively short spans, typically ranging from 10 to over that provide more effective and realistic solutions [3,4]. Among
40 m. Moreover, reducing the material weight through the available computational methods that can be used to solve
prestressing is decisive due to transportation and elevation optimization problems, heuristics and metaheuristics are
costs. This is where structural optimization of this type of large approximate methods that are considered as particularly
and repetitive structure becomes especially relevant. useful algorithms in structural engineering [5].

* Corresponding author. Fax: +34 963877569.
E-mail addresses: vyepesp@cst.upv.es (V. Yepes), jvmartia@cst.upv.es (J.V. Martí), tagarse@cam.upv.es (T. García-Segura),
fgonzale@cst.upv.es (F. González-Vidosa).
http://dx.doi.org/10.1016/j.acme.2017.02.006
1644-9665/© 2017 Politechnika Wrocławska. Published by Elsevier Sp. z o.o. All rights reserved.

concrete. 3 shows the the geometric variables.. Harmony Search (HS) [18.12]. i¼1.14]. Variable Neighbor- hood Search (VNS) [17]. xn Þ (1) limited knowledge on the optimization of prestressed concrete gj ðx1 . . and r is the total number of construction for the beams. among others. and steel that minimize the main geometric variables considered in this analysis. [28] focused on the measurements of the units in which the construction of the optimal-cost design of concrete I-girder bridges. evolutionary algorithms [13. The PPC bridge system used as well as the labor. Recent research has been conducted with regard to heuristics and metaheuristics [9]. mi are the feeling.. most Note that x1. and resources necessary studied consists of two PC U-beams with a span length of 35 m for the construction. . García-Segura et al. where pi are the unit prices. such as Descent Local Search (DLS) [10].. However. (4). . Ohkubo et al. Eagle Strategy (ES) [22]. 2 – Cross section of the beam-slab deck. x2. represented by the objective (GSA) [21]. units. [25] have noted. Sirca and Adeli [27] and Ahsan et al.. nevertheless. . . oped specifically for this work. this paper focuses on the cost The cost of the bridge depends on the material volumes optimization of PPC road bridges. and subsequently metaheuristics that are and Dinovitzer [8] identified a great gap between theoretical capable of finding cost-optimized solutions have been devel- work and the practical application of structural optimization. The cost function is obtained by adding and a top slab formed by a 0. Fig. archives of civil and mechanical engineering 17 (2017) 738–749 739 The earliest studies into structural reinforced concrete (RC) cost.19]. In this line of work. 1).. di2 . (1) while satisfying the constraints in Eq. x2 . Martí C¼ pi mi (4) et al. . The top slab allows vehicle traffic constructability. Glowworm Swarm Algorithm optimization of the structure. (3). Beams are made ments (Table 1). the latter also used post-tensioned tendons. Definition of the problem Genetic Algorithms (GAs) [11. xn are the design variables whose approaches for PC bridges found in the literature are not combination is to be optimized. The optimization searches for The analysis includes 59 design variables. Cohn implemented. As discrete values are used to guarantee the of self-compacting concrete. Despite the aforementioned research works. 2. . [29] proposed a hybrid HS for the design of X post-tensioned concrete box-girder pedestrian bridges. Simulated Annealing (SA) [15. Each variable can take on the suitable for application in real-life engineering. algorithm (BB-BC) [23]. (2).r beam road bridges. Hassanain and Loov [24] reviewed xi 2 ðdi1 . this is a combinatorial optimization problem.:xn Þ0 (2) (PC) bridge structures. . x2 .. [26] studied PC box girder bridges and proposed a multi-criteria The objective function considered is the cost function fuzzy optimization of the total construction cost and esthetic defined in Eq. A module evaluating all relevant limit states has been optimization date back to the late 1990s [6. Fig. there is still C ¼ f ðx1 . and has a width of 12 m (Fig. as Hernandez et al. Ant The mathematical problem is comprised of economical Colony Optimization (ACO) [20]. . 2). diqi Þ (3) research on cost optimization of concrete bridge structures. Both used PPC PCC bridge is split.06 m precast RC slab used as the price of each unit multiplied by the respective measure- formwork for the cast-in-place RC slab (Fig. and the Big Bang–Big Crunch function f in Eq.16]. 1 – Longitudinal profile of the bridge.7]. [30] used a hybrid SA to minimize the cost of PC precast U. The Fig. machinery. discrete values listed in Eq. .

where L is the span length.40 Concrete bridge barrier loads 2  5. top flange tablet ns3 = 3 m3 of concrete HA-30 – 69. This large number of minimum of L/18 due to esthetic ground and specific road solutions justifies the application of a heuristic algorithm to transportation considerations.740 archives of civil and mechanical engineering 17 (2017) 738–749 Table 1 – Unit prices of the materials in euros.11 – PC precast bridge width W = 12. Active prestressing steel (Y1860-S7) fpk = 1700 N/mm 2 the thickness of the slab (e4). All of the variables of this durability requirements are those demanded by the concrete analysis are discrete and can adopt a range of values.89 – Reinforcement parameters Passive reinforcing steel (B-500-S) fyk = 500 N/mm2 seven geometric variables comprise the depth of the beam (h1). Fig. reinforcement. inclination of the webs. bottom flange tablet ni3 = 3 m3 of concrete HA-40 – 79.09 m kg reinforcing steel (B-500-S) 2. and exposure parameters. web (t2) and the top flange (t3) of beam. There are 46 Transport distance (one way) Td = 50 km variables that define the standard reinforcement setup in the Active prestressing steel crops 25% beams and slab.00 m m3 of concrete HA-25 – 64.40 kN. the width and thickness of the Beam surface reinforcement Fr = 8 mm flanges (b3 and e3). and the trans. and number of sections with strand sheaths (non-bonded steel) in steel types.40 – Minimum beam slenderness L/18 m3 of concrete HP-45 142.38 – Geometric parameters m of beam mold 75. span length.12 Bottom flange division i4 = 4 m3 of concrete HP-35 122. These include (1) the number of strands in the top flanges (level 4). Table 2 – Input parameters for the analysis. Description Price of Price of Loading parameters beam (s) slab (s) Concrete bridge barrier width 2  0. Regarding Strand sheaths Levels 2 and 3 Vertical slenderness of stirrups 200 (length/diameter) the material strength.00 Spacing between beams Sv = 6. particularly for short Fig. Each adjustment of their design to the existing precast molds. there are two variables that define the Cost parameters concrete grades used for the slab and the girder. dead loads. The slenderness of the beam is limited to a rise to 1. (2) the the width of the deck.03 Inclination.15 – Bearing center to beam face distance 0.47 m m3 of concrete HP-50 152.6  1065 possible solutions. External ambient conditions IIb (EHE) verse reinforcement of the bottom flange (t1). the width of the soffit of the beam Strand diameter Fs = 0. continuously the depth of the beam.50 m Thickness of the wearing surface tws = 0. giving code EHE-08 [31]. number of strands in the bottom flange (levels 1–3). . the optimization algorithm tends to increase time. 3 – Geometric variables of the beam structure.6-inch strands.0 kN/m kg prestressing steel (Y-1860-S7) 3. and (4) third layers of the bottom flange. the top (t4) and Code regulation EHE/IAP-98 bottom (t5) transverse reinforcement of slab. (3) the slenderness of the beam. find cost-effective results within a reasonable computation Otherwise. The prestressing steel. by 0. 3 shows the top (As6) and bottom (As7) Legislative and exposure parameters longitudinal passive reinforcement of slab.600 (b1) and its thickness (e1). is defined by four variables representing: cost.25 – Web inclination 808 m3 of concrete HP-40 133. loading. and the thickness of the webs (e2). which is formed The main parameters are divided into geometry.99 Inclination. The strand is prestressed to 189. The beam parameters are chosen to facilitate the the second.63 1.95 Top flange division s3 = 3 m3 of concrete HA-35 – 74. transport distance.00 m m2 of slab formwork – 30.

and torsion as well as fatigue. Further.0 kN/m2. archives of civil and mechanical engineering 17 (2017) 738–749 741 span bridges. The general exposure class was IIb. The program has been programmed with combination. [30]. . and considered to design the elements and analyze the structural deflection resulting from the dead and traffic loads.5 m analysis of the beam before being connected to the slab. The structural constraints of IAP-98 [32] include the The construction sequences and the long-term interaction verifications of the ultimate and serviceability limit states between the precast beam and the cast-in-place concrete were for bending. In this 4. Fortran 95 and Compaq Visual Fortran Professional 6. This distance between axes). 5 shows a detail of the during the manufacture and placement of the beams. which are connected with three bar program that considers different stress states that occur elements for each of the 21 sections. The program uses two tested both flexural and shear minimum amounts of rein. This grillage model was used to represent to the phase being studied. The second model bridge barrier rails installed along the edge of the deck. decompression does not occur in the mechanical characteristics of the bars in which the the concrete in any fiber in the section. Fig. models for the analysis according to the work by Martí et al. which was checked at each Martí et al. iteration. superimposed with a uniform load of model is composed of 20 beam elements and 21 nodes.6. as well as the minimum geometric requirement. shear. time-dependent deflections were limit. Fig.09 m as well phase. forcement. 5 – Second model detail for deck long-term analysis. The entire bar model with 103 bar elements and 84 nodes. 4 – First model used for the analysis of the beam before being connected to the slab.5 kN/m for concrete calculating the short-term prestress loss. Fig. The dead load is a wearing surface of 0. It is also response of the bridge in each phase. Fig. The program using a 2-D mesh with 20 bar elements and 21 stresses and reactions are obtained from a self-developed sections for each beam. cracking.06 m width were considered for the formwork of long-term prestress loss due to creeping and shrinkage of the top concrete slab. [30] (Table 2). concrete and prestressing steel relaxation by a stiffness matrix according to the Spanish code on structural concrete [31]. Temporary deflections longitudinal stresses due to the distortion of the cross-section were limited to 1/250 of the free span length for the frequent were considered. The details of the parameters can be found in to the working life design. for calculations for the decompression limit state comprise which a linear elastic analysis including gross section verifying that under the combination of actions corresponding properties was used. the elastic shortening of concrete was considered when as a uniformly distributed load of 2  0.0 on a ed to 1/1000 of the free span length for the quasi-permanent personal computer that contains an Intel Core TM2 Quad CPU combination. 4 shows the first model used for a linear elastic The design live load consists of three axis of 200 kN each (1. The durability limit state was checked according Q6600 2.40 GHz processor. Precast calculated stress resultants and reactions taking into account RC slabs of 0.

7 – Cost evolution of the DLS algorithm for different movements and termination criteria. the computing time increases significantly by 72. V9-5000 iterations exhibit a resulting cost of 134 529s and need 790 seconds of computation time. 10. respectively. Fig. [33]. It is known that this heuristic becomes trapped in a local optimum. The movements set out in this study modify 5. 6) begins with obtaining a random solution. Note that the nomenclature Vn-i corre- sponds to the number of variables (n) and the number of consecutive iterations without cost improvement (i). 25. and V12-15000 iterations. The termination criterion is defined as the number of consecutive iterations without cost improve- ment. the working solution replaces the previous one.1. V9 is chosen as the best movement. The best solution (115 459s) was obtained by varying 15% of the variables (V9). Applied heuristic search methods 3. the process is repeated nine times according to the methodology proposed by Payá-Zaforteza et al. This process is repeated until no further improved solution can be found for a given number of iterations. Although further cost reductions of 1. If the new solution has lower cost and is feasible after checking the restrictions. V12-10000 iterations. The Pareto optimal solutions with the best trade-off between cost and time correspond to V9-5000 iterations. A small movement. changes each variable by either increasing or decreasing it by a step or unit.6 and 2. .8% are achieved by the next Pareto solutions (V12-10000 iterations and V12-15000 iterations). or 30% of the variables. 6 – Flowchart of the DLS process. Descent Local Search (DLS) This algorithm (Fig. 20. which modifies a fixed number of randomly chosen variables. 7 shows the average cost values according to the movement and termination criteria. but the main aim of this method is to choose the most efficient moves for incorporation into the threshold acceptance algorithm.742 archives of civil and mechanical engineering 17 (2017) 738–749 3. 15. Fig.2% and more than 200%. Fig. To account for the random component of the method.

This means that the average cost is 3. allows U0 to be fixed in two chains.83%. Fig. 8 – Flowchart of the TAMO process. 9 shows a typical curve The lowest cost. The TA method is also capable of surpassing local optima and gradually converges as the threshold value decrease to zero. and the extreme value estimated using the three-parameter Weibull distribution that fits 400 TAMO results is less than 1. This criterion the mean with respect to the minimum for the costs.2. 8 shows a flowchart of this algorithm. 3. 9 – Typical cost variation and threshold reduction for strategy based on the mutation operator that is used in the GA. Fig. Individuals of a population evolve through genetic crossover and mutation. For the first random solution. a threshold lower than 2% of the The application of the TAMO algorithm requires the initial threshold U0. All of them belong to TAMO. The difference checked between the minimum cost obtained with the nine TAMO runs of the heuristic number T4 (Table 3). The procedure followed for the calibration was the the stop criterion. two chains without improving definition of the threshold coefficient value (aTA) as well as the best solution.2% lower when applying TAMO. After applying a movement. and mutation. and deviation (MD) of temperature equal to 0. Five Pareto solutions are highlighted.85. Fig.3.95 (T4). crossover. The new population is . archives of civil and mechanical engineering 17 (2017) 738–749 743 3. which is 104 184s. according to the methodology proposed by Payá- Zaforteza et al. TAMO. The best movement was V9.95. Genetic Algorithm (GA) This metaheuristic technique was proposed by Holland [37]. As abutments. it is possible to run the algorithm nine times for each span length instead of 400 times in order to achieve a balance between the quality of the results and the amount of computing time required to obtain them. However. which consists in halving or doubling the initial temperature until the accep- tance range is between 20 and 40%. 10 shows the average cost values versus time of the results obtained by applying different parameter combinations for SAMO and TAMO strategies. The initial threshold (U0) is adjusted by Medina's criterion [35]. The GA explores the solution space using a population of solutions and operators such as selection. Table 3 shows the results of a 16 case-study the iteration chain (LI). In addition. [36] to optimize bridge of cost reduction and threshold decrease following TAMO. The average costs of the 16 combinations of SAMO and TAMO are 115 709s and 112 021s respectively. TAMO obtained similar computing times to the SAMO algorithm. mean. TA hybridization was previously applied by Luz et al. The new solution is accepted if the conditions of feasibility and threshold are satisfied. the hybrid Simulated Annealing with Mutation Operation (SAMO) achieved the lowest cost of 108 008s (S12) with a Markov Chain (MC) = 5000 and coefficient reduction (aSA) = 0. inspired by the process of natural evolution. simultaneous compliance with two require- same as that applied in DLS. This value was obtained after running TAMO algorithm 400 times. Fig. Threshold Accepting algorithm with Mutation Operation (TAMO) The Threshold Accepting (TA) algorithm was originally described by Dueck and Scheuer [34]. creating better-adapted individuals. the average cost is also the minimum. the current solution is then checked against structural restrictions. 107 856s. was obtained using LI = 5000 and aTA = 0. For these parameters. while the best average cost was 110 477s (S3) with MC = 2500 and aSA = 0. and second. TAMO combines the TA algorithm with a local search Fig. ments is implemented: first. For the same problem [30]. [33].5% of the cost is taken. a series as well as the minimum. The threshold decreases gradually during the optimization process according to a reduction coefficient (aTA).

0. and 500 individuals therefore.10 T14 0. Results are taken for 25.95 2500 109 312 18 458 107 970 1. The method.67 T15 0. which is obtained as the ratio between the resistance of the structural response and the load effect of actions. worsening their aptitude according to a penalty function (Eq.25. Penalties are implemented for the infeasible solutions.02 selected in an elitist way. Heuristic number aTA LI Average cost (s) Average time (s) Minimum cost (s) Deviation (%) T1 0.99 T11 0. Fig.92 T2 0.90 1000 110 305 3711 107 825 5.85 5000 110 139 12 977 107 894 2.50.89 T16 0.80 1000 114 232 2091 109 139 4. 10 – Average costs of SAMO and TAMO for the 16 different combinations.80 2500 112 214 5074 108 015 3.85 2500 109 552 6413 106 191 3.85 1000 114 370 2661 107 908 5. The crossover operation 150. and ‘‘f’’ is a penalty factor that depends on the feasibility of the solution.95 500 111 073 3667 107 919 2. 100.83 T6 0. 50. 175. . Other penalty functions were tried without improving the convergence to the minimum. individual is proportional to its aptitude.71 T5 0.744 archives of civil and mechanical engineering 17 (2017) 738–749 Table 3 – Results of the TAMO algorithm for 16 parameter combinations. the randomly changes some characteristics of the new solution.80 500 118 589 1074 110 728 7.75 and population sizes of 100. 75. nine executions with crossover values of 0. 11 – Flowchart of the GA process. and 0. For this first analysis.24 T4 0.08 T9 0. The probability of selecting each First. (5)). 125. The penalty factor is smaller than one for an unfeasible solution.17 T12 0. The penalty factor is equal to the minimum limit state factor. C(k) is the unpenalized cost.56 T10 0.05 T3 0. Fig. The mutation operator selected as optional. and 200 generations.85 500 117 418 1314 109 169 7. Fig.80 5000 111 134 9747 107 874 3.99 T7 0. 11 shows a flowchart of this algorithm.17 T8 0.90 2500 109 897 9318 106 569 3.08 T13 0.90 5000 110 452 19 635 108 363 2. Elitism and mutation have been individuals to the new solution. Regarding the population size.95 5000 107 856 36 958 106 042 1. solutions of the current population.90 500 114 587 1931 108 279 5. gives higher probability to the selection of the best have been performed.   1 CpðkÞ ¼ CðkÞ 1:1 þ (5) f Cp(k) is the penalized cost of the unfeasible ‘‘k’’ solution. mutation decides which information is transferred from these two has not been considered. 250. This coefficient reduces the divergence caused by the high sensitivity of the unfeasible PC structures.95 1000 111 200 7463 107 910 3.

8 2 133 992 22 479 126 366 6. solutions are 3.90 75 147 387 16 803 139 702 5.50 75 145 189 16 365 127 325 14.60 150 143 741 18 985 126 450 13.3 4.70 125 146889 19001 136076 7.80 146 160 22 164 137 442 6.60 146 337 21 056 138 290 5. The number of variables subjected to mutation is tracked sequentially.00 142 436 15 498 130 041 9. and 200 generations. 12 – The evolution of average and minimum costs of a GA. the number of individuals increases.30 150 146516 20099 136 076 7.10 138 885 20 879 132 618 4.20 Table 5 – Application of mutation in GA for 500 individuals.00 141 470 16 583 128 945 9. These are 500 individuals.70 200 145 105 22 335 137 207 7. elitism reduces the minimum value.9 4.80 138 743 21 960 132 618 4.0 3. Crossover Generations Without elitism With elitism Average Time (s) Best % Desv. a crossover value of 0. Memetic Algorithm (MA) MA [38] combine population-based global search and local search metaheuristics.00 140 903 17 667 128 618 9.6 – 1 134 846 22 469 130 424 3.40 153 560 13 758 148 212 3.5.60 146825 17734 140179 4. archives of civil and mechanical engineering 17 (2017) 738–749 745 Table 4 – GA results for a population size of 500 individuals.40 175 143 551 19 858 126 450 13.1 4 132 678 22 499 123 078 7.10 155 907 13 745 149 470 4.40 147512 16627 141425 4.4 3 133 091 22 487 125 187 6.50 148 592 15519 145186 2.90 148 711 14 397 143 798 3.30 25 151 442 14 618 133 683 13. The best parameters obtained in the first analysis are selected.4 5 132 707 22 131 125 345 5. Fig.8 4. On some purpose of getting closer to a local optimum.90 146567 18842 139137 5.50 140 890 16 558 134 412 4.60 0.80 200 146 285 22 296 135 709 7. 13 shows a flowchart of this Fig.90 175 146 408 21 197 136 076 7. The best average cost (138 743s) is obtained for 200 generations.20 139 285 19 799 132 618 5.40 50 147 701 15 491 127 325 16.5.50 142 763 15 477 137 458 3.70 100 144 271 17 238 126 613 14. Fig. Mutation Average cost (s) Time (s) Best cost (s) % MD % Improvement 0 138 743 21 964 132 618 4.40 139 830 18 719 133 332 4. cost (s) cost (s) cost (s) cost (s) 0. Compared to quality population initiates the GA procedure.10 125 145 771 19 016 138 275 5. The goal of non-elitism.75 10 156 620 13952 149816 4.80 100 146 284 17 909 139 702 4.50 154134 13763 147789 4. .40 140 084 22 005 128 261 9.60 25 154 493 14 607 144 327 7. Table 4 summarizes the Each individual is improved through a local search with the results for a population size of 500 individuals.90 150 145 492 20 122 138 275 5. Table 5 shows the results for each mutation step.4.30 200 143 314 20 732 126 418 13.25 10 157 175 13 948 147 762 6. 0.5 10 156 951 13 928 149 382 5.70 140 117 17 638 133 377 5.30 0.5 crossover.70 146433 19949 138290 5.40 50 148 806 15 696 139 770 6. The best improvement with regard to the cost value without mutation is mutation = 4.4 average cost improves and the computing time increases as algorithm. Average Time (s) Best % Desv.30 100 147537 17902 141039 4.70 140 354 19 836 128 261 9.40 75 149304 16804 144327 3.50 140 217 20 920 128 261 9. and 200 generations.4 2.1% cheaper. a crossover value of 0. The process starts with population generation.80 140 574 18 752 128 293 9.50 50 150 758 15 705 144 327 4.00 151 048 14 412 147 340 2. 3. The new better- occasions.60 125 144 061 18 112 126 613 13. 12 represents the evolution of the best GA heuristic (mutation = 4) for both average and minimum costs. and elitism.30 25 151 264 14 590 145 577 3. A second analysis is carried out to examine the benefits of mutation.30 146 274 14 414 136 229 7.00 175 145 266 21 228 138 275 5.

the three stop criteria./GA 50 10 129 718 43 639 124 446 4. A behavior similar to that of other metaheuristics can be observed. the achieves the best results is TAMO.0%. a crossover value GA2. However. 25. and mutation = 4. The TAMO technique has tion = 4). and 50 chains.2 2. The heuristic that the cost. compared to GA. respectively. elitism).5. Comparing the results of MA with those of TAMO. Chain Number Average cost (s) Seconds Best cost (s) % MD % Improv. elitism. The parameters increases the cost by 2. Nine executions are run for each of 28.2.746 archives of civil and mechanical engineering 17 (2017) 738–749 Fig.8 . 19. Fig. the cost results decrease much more slowly until they become practically constant. Likewise. are 500 individuals. Then. the values obtained with SAMO been used in this algorithm for the local search strategy. 14 shows the evolution of average and minimum costs as the number of generations increases. Regarding the computing time. 200 generations. 6. where Table 8 summarizes the geometry. and GA1 follow with cost increments of 11.7. [30] are incorporated in the tables. However. improve- ments of 2. GA1 (without global search). respectively. 25. MA.8% (S12). It is worth noting that the computing time beam depth varies between 1. increases by 194–575%. which are then combined to improve Table 7 summarizes the values of the best parameters the quality of the following generations (population-based obtained from the applied heuristics: DLS. respectively.69 and 1. This algorithm use the same operators as GA. 14 – The evolution of average and minimum costs of MA. TAMO.0.1. 4.8% are observed for 10. and increment of the average search: MC equal to 50. and MA.1% higher than those achieved with TAMO.0.4 6. the MA heuristic needed more than seven times the processing time of TAMO. Additionally.7 and 10. and 50 chains. concrete.2 25 124 564 76 934 119 332 4. 13 – Flowchart of the MA process. and reinforce- the average cost refers to the average of the values of the nine ment for the best solutions obtained by each heuristic.0. Fig. Comparison of the results combining both strategies is to acquire good genes for the parents (local search).4% (S3) and 3. and stop criteria of 10. initial temperature (T0) equal to 5% of costs with respect to TAMO are included. and 9.94 m.1 50 119 703 129 441 116 933 2.4 9. where the average values decrease rapidly in the beginning. The following parameters are considered for the local minimum cost. GA3 (with elitism and muta- which are crossover and mutation. The average costs. The executions. the average cost (119 703s) and the minimum cost (116 933s) were 9. SAMO only best heuristic obtained by the GA is chosen. 23. average time. Markov C. GA3. The three local Table 6 – Results of MA. and 33. of 0. DLS. The results are shown in Table 6. GA2 (with elitism).

20 45 30 35 29 0 6 GA3a 123 078 1. Table 9 best results are achieved with minimal prestressing steel in shows the cost differences in the passive reinforcement. This is reflected by high beams and the GA does not affect the geometry of the best solutions.0%).27 1.93 m.7 a GA1 without elitism.0 11.31 45 40 35 26 0 2 SAMOS12 108 008 1.16 0. concrete and active reinforcement.23 1.0 a GA1 without elitism.4 3. The best solution has a the slab. no strands are needed in the DLS and SAMO.145 165.80 0. GA3 with elitism and mutation = 4.149 171. The lower depth values compensated by concrete of greater grade (HP-45). In addition. Table 10 – Best results: concrete in beams and slab. DLS SAMOS3 SAMOS12 TAMO GA1a GA2a GA3a MA Average cost (s) 128 309 110 477 111 903 107 856 143 414 138 863 132 678 119 703 Best cost (s) 115 459 109 004 108 008 106 042 126 418 127 969 123 078 116 933 Average time (s) 8122 18 322 12 986 36 233 20 732 21 960 22 499 129 441 % Average cost/TAMO cost 19.25 0. The larger thicknesses of the slabs obtained strength concrete.33 0.78 0.446 81.22 1. GA2 with elitism. but enough to prevent cracking during the lowest total steel quantities correspond to TAMO.906 78.943 112. This value is close to the limit imposed highest one and the highest concrete grade (HA-40). search heuristics (DLS.575 61.507 94.10 0.9 a GA1 without elitism.882 99.194 Total (m3) 191.93 0.41 0.263 66. TAMO gives a thickness (0.0 33.7 53. Steel DLS SAMOS12 TAMO GA1a GA2a GA3a MA In beams (kg) 6325 5562 6803 8554 8496 7167 6359 In slab (kg) 17 341 19 561 10 860 22 105 22 181 20 015 19 419 Total (kg) 23 666 25 123 17 664 30 659 30 677 27 182 25 778 % Steel/record 34. The the values of the variables are the same. Regarding coincide with the worst solutions.33 0. beams and slab.94 0. However.906 67.92 0.18 0.413 173.33 0. In the strands of the TAMO solution are similar to those of the SAMO beam.2%) and DLS (34. is used from TAMO lead to a greater concrete volume compared with in the beams. higher. GA3 with elitism and mutation = 4.78 0. volume.98 0.78 0.507 103.9 45.9 15.378 In slab (m3) 116.32 40 35 36 34 0 6 MA 116 933 1.7 20.572 % Concrete/record 33.38 0.6 73.30 50 35 32 31 0 2 a GA1 without elitism.133 103. Cost (s) h1 (m) e4 (m) b1 (m) b3 (m) e1 (m) e2 (m) e3 (m) fc. but the minor thickness of the bottom flanges b1 is depths with reduced thickness.4 0. GA3 with elitism and mutation = 4.14 0.27 m. GA2 with elitism.0 42. the thickness e4 acquires values between 0.24 1.19 and beam depth of 1.93 0. Table 8 – Best results: geometry.23 0.beam fc. GA3 with elitism and mutation = 4. as small thicknesses.20 45 30 35 29 0 6 GA2a 127 969 1.22 0.slab p1 (n) p2 (n) p3 (n) p4 (n) (MPa) (MPa) DLS 115 459 1. In an attempt to lighten the structure.16 0. of the flanges (b3) should be short since it does not improve the There is a trend toward economization of the concrete flexure capacity.15 0. GA2 with elitism.1%).8 0.17 0. . This allows the amount of reinforcement to be The beam depth h1.69 0.413 171.1. and number of reduced compared to SAMO (42.26 2. Table 9 – Best results: passive reinforcement. The the top flanges. archives of civil and mechanical engineering 17 (2017) 738–749 747 Table 7 – Summary of the heuristics.886 171.24 1.26 m) that is almost the span length of 1/18. the volume of concrete given by TAMO is the smallest.0 28.17 0.2 0.90 0.92 0.28 45 40 37 20 0 2 GA1a 126 418 1. and SAMO) search for big beam solution.7 23.0 20. SAMO (20. Concrete DLS SAMOS12 TAMO GA1a GA2a GA3a MA 3 In beams (m ) 74.19 1.10 0. which is equivalent to a ratio of depth to 0. TAMO.0 73.23 40 40 36 20 0 2 TAMO 104 184 1.0 2.7 19. GA2 with elitism.38 0. followed by tensioning phase.10 0. providing the best inertia while at the same time It is worth noting that the use of elitism or non-elitism in being as light as possible.00 0. leading to less need for prestressing.99 0.021 143.753 67.1 19. with values between 40 and 50 MPa.206 59.10 0. Table 10 shows the concrete volume in the third layer. bottom flange width b1. The width (1/18).29 0.

GA3. Nigdeli. Turskis. Yee. 202. Balling. Optimal arrangement of  In the attempt to lighten the structure. [12] H. Kesy. González-Vidosa. these two are acceptable for utilization in real cases. Alcalá. Christiansen. Lemonge. Arsenis.7. TAMO.Z. and GA1 increase the hydrodynamic torque converter with application of genetic cost results by 2. Archives of Civil and Mechanical Engineering 11 respectively. the lower [15] V. Moreno-Jiménez. TAMO gives smaller beams and a larger amount of intelligence methods. 617–649. Z. Yepes. Yao.  Comparing MA with GA. 11.  The most efficient move obtained in DLS. García-Segura. Computer- Aided Civil and Infrastructure Engineering 13 (1) (1998) 43–52. J. Zavadskas. Engineering Structures 125 references (2016) 325–336. and SAMO) show a trend toward the maximum cable system of cable-stayed bridges. A genetic algorithm-based multi-objective  The structural results of the three local search heuristics optimization for hybrid fiber reinforced polymeric deck and (DLS. Construction optimization of  SAMOS3. Journal of Structural Engineering 120 (2) (1994) 5000 iterations and a reduction coefficient of 0. reinforcement. Burczyński. González-Vidosa. J. A. Use of optimization in the top flanges. T.J. Dinovitzer. The calibra. A. Cai. Kripka. Yepes. Francisca Molina-Moreno. Journal of Structural Engineering 123 (2) (1997) 193– local search heuristic.M. Multi- 5. [14] P. F. Medeiros. Metaheuristics in combinatorial incorporated. Geem. namely V9 [6] R. G. Turskis. Genetic algorithms in optimal are similar and are smaller than the amount given by SAMO. Blum. [18] S. the solutions are 3. The authors are grateful for the loaded columns. S. Structural Engineering and Mechanics 45 (6) technology. GA2.K. the greater the depth.F. Fedeliński. Martínez-Martín. Beluch. Structural Engineering and Mechanics 54 (6) thorough revision of the manuscript by Phillip Zastrow and (2015) 1097–1109. Optimization Acknowledgments of high-performance concrete structures by variable neighborhood search.  The greater slab thicknesses obtained by TAMO.95. A novel harmony with FEDER funding (research project BIA2014-56574-R) finan- search based optimization of reinforced concrete biaxially cially supported this research. Conclusions criteria analysis of projects' performance in construction. Archives of Civil and Mechanical the top flanges of the beam and minimal prestressing steel Engineering 15 (4) (2015) 1024–1036. Multi- precast-prestressed U-beam road bridges. and 33.D. V. but enough to prevent cracking. was used for the other frames. ACM Computing Surveys 35 (3) (2003) 268–308. A. S. Górski. 23. Vilutiene. DLS. 28. Yepes.8%. detailed design of reinforced concrete members. TAMO.C. Pellicer.8. In addition. 3. A survey of non-gradient to non-elitism. tion recommends a population size of 500 individuals and a [10] A. (4) (2011) 905–920. Two-scale identification of to SAMO. J. F. J. the first raises the computing time Advances in Engineering Software 42 (4) (2011) 151–159. Carbonell. with respect to the average achieved by TAMO. the processing times for Engineering with Computers 13 (4) (1997) 185–196. and MA [2] V. by 194–575% but improves the cost result by 9. Z. Saparauskas. Nutini. Engineering Structures 99 (2015) 311–318. Design of crossover value of 0. Optimization of reinforced concrete (random variation of nine variables). A simple genetic  TAMO was shown to be the most efficient procedure. MA. [20] F. Archives of Civil and higher-strength concretes of between 40 and 50 MPa for the Mechanical Engineering 15 (2) (2015) 525–531. Advances in Engineering Software 59 (2013) 19–28. A. Santos.J. García-Segura. G. Yepes. E. Z. DLS. A.J.W. V. F. [17] C. it is advisable to use reinforcement in composites. T. [7] C.748 archives of civil and mechanical engineering 17 (2017) 738–749 The amounts of reinforcing steel given by TAMO. V.1% cheaper.C. The optimization tends toward short dimensions for structural problems. CO2 emissions. . [11] A. X. (2013) 723–740.0. Kim. and safety.S. Compared [9] W. A cognitive the concrete strength.0. Archives of Civil and Mechanical Engineering 14 (4) (2014) 636–646. Hare. International Journal of Civil The Spanish Ministry of Economy and Competitiveness along Engineering 11 (2) (2013) 90–99. Coello. The best values range between 35 and approach for the multi-objective optimization of RC 40 MPa. R. A.0. Koumousis. The [8] M. E. [19] T.4. Torres-Machí. Multiobjective optimization of post-tensioned concrete box-girder road bridges considering cost. Tesfamariam. Structural and beam depth with minimum thicknesses. The results achieved by SAMO [30] for the same problem are [5] C. Multidisciplinary Optimization 52 (3) (2015) 583–594. are compensated with less passive reinforcement. Kadziela. T. with respect [13] W. elitism improves the cost results. Roli. 19. Application of structural best results for TAMO were achieved with a chain length of optimization.M. SAMOS12. closely algorithm for the design of reinforced concrete beams. and criteria decision-making system for sustainable building MA are used to find cost-optimized solutions automatically. [16] M. S. Social and environmental benefits of precast concrete bridge viaducts. GA.  Regarding GA. assessment/certification.0%. Hospitaler. followed by SAMO.5. composites' material constants by means of computational However.J. PCI Journal 46 (3) (2001) 14–19.K. V. Cohn. Archives of Civil and Mechanical Engineering 14 (1) (2014) 114–121. optimization methods in structural engineering. Four algorithms are compared for the efficient design of [4] M. reinforced concrete road vaults by heuristic optimization. Archives of Civil and Mechanical Engineering 15 (1) (2015) 11–18. Aref. A parametric study of optimum tall piers for railway [1] A. Bjork. DLS. Bekdas. algorithm. A.A. for automatic grouping of beam cross-section dimensions in reinforced concrete building structures. The conclusions are as follows: optimization: overview and conceptual comparison. F. [3] E. Regarding the slab.K. beams. Zavadskas. Medineckiene. Yepes.

design of frame structures using the Eagle Strategy with [31] M. VTOP. [23] C. multicriteria fuzzy optimization of a prestressed concrete [36] A. Design of retaining walls using big bang. 1975 https://mitpress. Journal of Structural Engineering Road Bridges. Camp. P. 1998. Yepes. F. Nurul Ghani. Journal of Structural Engineering 131 (3) (2005) 380– [37] J. J. K.psu. Fomento. Threshold accepting: a general purpose [25] S. Engineering Structures 48 (2013) 342–352. Ministerio Differential Evolution. 2008 (in Spanish).V. Rana. V. Talatahari.S. and [26] S.R. González-Vidosa. Yepes.A. Meccanica 45 (5) (2010) 693–704. Genetic algorithm.mit. T. prestressed concrete precast road bridges with hybrid [22] S. Deb. Code on Structural Concrete EHE-08. Latin American Journal of Solids and Structures 11 [30] J. A. stochastic hill climbing algorithms. Alcalá. Gandomi. IAP-98: Code on the Actions for the Design of big crunch optimization.B. Adaptation in Natural and Artificial Systems. X. Caltech Concurrent Computation Program [29] T. Akin. On Evolution. [33] I. Moscato. Holland. 1989 http://citeseerx. A. Marcos.27. Madrid. algorithm.V. Search. Informes de la [27] G. Gonzalez-Vidosa. Advances in Engineering Software 41 (3) [35] J. A.16). 30 (5) (2003) 841–849. Spain. Coastal. Fomento.V. V. [34] G. J. García-Segura. Cost optimum design of artificial-systems (accessed 10. de Fomento. García-Segura. González-Vidosa. J. Spain.11. simulated annealing. 138 (3) (2012) 438–448. archives of civil and mechanical engineering 17 (2017) 738–749 749 [21] T. Sirca. Madrid. A. [24] M. J. Ahsan. F.ist. Martí. Port. Engineering Structures 91 (2013) 16–25. open reinforced concrete abutments road bridges with hybrid Structural Optimization 15 (2) (1998) 132–140. bridges. Scheuer. Journal of Waterway. Loov. Hospitaler.16). Yang. Cost optimization of prestressed concrete Construcción 67 (540) (2015) e114. An approach to Ocean Engineering 127 (4) (2001) 213–221. Dissanayake. Alcalá. Yepes. S. Martí.1.H. Dueck. V. Canadian Journal of Civil Engineering by simulated annealing. Martí. S.1. Optimization concrete box-girder pedestrian bridges. Optimum simulated annealing.E. Design of (7) (2014) 1190–1205. concrete beams. Medina. Journal of Computational Physics 90 (1990) 161–175. posttensioned I-girder bridge using global optimization [38] P.9474 (accessed 09. S. Taniwaki.edu/books/adaptation-natural-and- [28] R.N. Algorithms.H. J. R. An optimization algorithm appearing superior to simulated improved software for design optimization of prestressed annealing. Payá-Zaforteza. Cost optimization of concrete On the Weibull cost estimation of building frames designed bridge infrastructure. Alcalá. [32] M. Design of bridge system considering cost and aesthetic feeling. Estimation of incident and reflected waves using (2010) 415–421. Engineering of concrete I-beams using a new hybrid glowworm swarm Structures 92 (2015) 112–122. Yepes. Luz. Pérez-López. Díaz. Optimization.F. Hybrid (Report 826). Fontan. E. Yepes. . Journal of Structural Engineering 138 (2) (2012) Algorithms and Martial Arts – Towards Memetic 273–284. 388.V. Hernández. D. V. Ohkubo.edu/viewdoc/ harmony search for sustainable design of post-tensioned summary?doi=10. H. F. V.11. Adeli. J. Hassanain.