Professional Documents
Culture Documents
Nomenclature
aw weld thickness
bfb beam flange width
bfc column flange width
db bolt diameter
dh horizontal distance between bolts
fub bolt ultimate stress
fyb beam yield stress
fyc column yield stress
fyep endplate yield stress Fig. 2. Characterization of the beam-column joints components.
fyfb beam flange yield stress
fyfc column flange yield stress
h1 first bolt row height
h2 second bolt row height
hb beam height
hc column height
lep distance from the beam top flange to the endplate
free edge
Lta top and seat angle length
Lw a web angle length Fig. 3. Bilinear approximation of the moment-rotation curve [1].
tep endplate thickness spring model depicted in Fig. 2 can be simplified by altering
tfb beam flange thickness each series of springs with an equivalent elasto-plastic spring,
tfc column flange thickness keeping their related characteristics. By this procedure, a non-
tta angle thickness linear equivalent model for the analysis of beam-column joints can
twc column web thickness also be obtained [11]. Upon defining the equivalent elastic model,
Mj,Rd joint flexural resistance the design process continues with a postbuckling stability analysis
Mj,Rd,LGP joint flexural resistance predicted by linear genetic utilizing an energy-based formulation [33]. Since these procedures
programming are still to be evaluated, this study uses the bilinear approximation
Mj,Rd,exp experimental joint flexural resistance of the moment-rotation curve suggested in Eurocode 3, see Fig. 3.
Mj,Rd,EC 3 Eurocode 3 joint flexural resistance
Sj,ini joint initial rotation stiffness 3. Genetic programming (GP)
Sj,ini,LGP joint initial rotation stiffness predicted by linear
genetic programming Genetic programming (GP) creates computer programs to solve
Sj,ini,exp experimental joint initial rotation stiffness problems using the principle of Darwinian natural selection.
Sj,ini,EC 3 Eurocode 3 joint initial rotation stiffness GP was introduced by Koza as an extension of the genetic
algorithms in which programs are represented as tree structures
and expressed in the functional programming language LISP [17]. A
comprehensive description of GP is beyond the scope of this paper
and can be found in [17,18].
Linear Genetic Programming (LGP) is a recently emerged
subset of GP. Comparing LGP to the traditional Koza tree-based
GP, there are some main differences such as the graph-based
functional structure of LGP, evolvement of these programs in an
imperative programming language (like c/c ++) [34], and machine
code [35] rather than in expressions of a functional programming
language (like LISP) and the coexistence of structurally ineffective
codes with effective codes in LGP. Ineffective code in genetic
programs, referred to as ‘‘intron’’ represents instructions without
any influence on the program behavior. Structural introns act as
a protection that reduces the effect of variation on the effective
Fig. 1. Classification of joints by initial rotational stiffness. code and also allow variations to remain neutral in terms of
fitness change. Because of the imperative program structure in
is performed. A reliable database including previously published LGP, these ineffective instructions can be identified efficiently. This
flexural resistance and initial rotational stiffness of beam-column allows the corresponding effective instructions to be extracted
steel joint test results is utilized to develop the proposed models. from a program during runtime. Since only effective programs
are executed when testing fitness cases, evaluation is significantly
2. Overview of Eurocode 3 model accelerated (see Fig. 4).
The instructions from imperative languages are restricted to
The European Committee appraises the behavior of beam- operations that accept a minimum number of constants or memory
column joints by the component method. Joint components variables, called registers (r ), and assign the result to a destination
have been introduced via a simple mechanical model to predict register, e.g., r0 := r1 + 1.
beam-column moment against rotation curves [32]. Fig. 2 shows Automatic Induction of Machine code by Genetic Programming
an endplate beam-column joint together with its associate (AIMGP) is a particular form of LGP. AIMGP induces binary machine
mechanical model. This model is composed of springs and rigid code directly without any interpreting steps that results in a
links representing relevant joint components. A comprehensive significant speedup in execution compared to interpreting GP
description of the joint components is presented in [33]. The systems. This LGP approach searches for the computer program
1740 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
hc
tfc
bfc
tep
lep tfb
dh
h1 hb
h2 fub db
twc bfb
Table 1
Input data for endplate joints.
Test bfc tfc hc fyc bfb tfb hb fyb tep lep fyep db fub h1 h2 dh
T110001 300 13.5 296 475 220 15.5 543 300 25 105 416 24 1323 580.3 474.8 30
T110002 300 13.5 296 482 220 15.5 543 309 25 105 410 24 1323 580.3 474.8 30
T110005 300 12.5 296 532 220 15.5 547 290 25 105 404 24 1323 584.2 478.8 30
T109003 180 14.1 179 300 151 11.2 300 323 30 70 273 20 1000 334.4 244.4 46
T109004 180 14 180 306 192 14 454 285 41 84 323 24 1000 496 381 65
T109005 240 16.4 240 276 192 14 454 285 41 84 323 24 1000 496 381 65
T109006 240 17 240 275 220 18.6 597 288 41 82 325 24 1000 634.7 514.7 62
T101004 160 12.6 163 280 99.7 8.4 198.8 351 15 60 315.5 16 1000 229.6 149.6 25
T101007 163 12.6 160 280 99.7 8.4 198.8 351 15 60 315.5 16 1000 229.6 149.6 25
T101010 160 12.6 163 280 150.9 10.8 298.9 303 20 70 291.5 20 1000 333.5 243.5 30
T101013 120 9.7 241 310 99.7 8.4 198.8 351 18 60 330.5 16 1000 229.6 149.6 24
T101014 151 10.8 299 303 99.7 8.4 198.8 351 15 60 327 16 1000 229.6 149.6 25
T839 240 12 230 256 180 13.5 400 235 12 80 214 20 785 433.3 343.3 35
T8310 200 15 200 243 180 13.5 400 235 14 80 312 20 980 433.3 343.3 35
T8311 300 14 290 417 300 23 490 235 14 113 312 24 785 546.5 426.5 60
T911 301 14.4 302 317 181 14.6 401 323 30 90 266 24 980 443.7 338.7 61
T912 301 14.3 302 317 181 14.4 401 317 30 90 261 24 980 443.8 338.8 61
T913 301 14.5 301 279 181 14.4 401 279 30 90 239.5 24 980 443.8 338.8 61
TC5 204 10.9 206 283 133.3 7.6 205.4 283 16 111.2 250 16 633 268.8 138.8 30
TC6 204 10.9 206 283 133.3 7.6 205.4 283 16 111.2 250 20 980 268.8 138.8 30
TC7 204 10.9 206 283 133.3 7.6 205.4 283 20 111.2 250 16 633 268.8 138.8 30
TC8 204 10.9 206 283 133.3 7.6 205.4 283 20 111.2 250 20 980 268.8 138.8 30
TC9 204 10.9 206 283 133.3 7.6 205.4 283 16 111.2 250 16 633 268.8 138.8 30
TC10 204 10.9 206 283 133.3 7.6 205.4 283 16 111.2 250 20 980 268.8 138.8 30
TC11 204 10.9 206 283 133.3 7.6 205.4 283 20 111.2 250 16 633 268.8 138.8 30
TC12 204 10.9 206 283 133.3 7.6 205.4 283 20 111.2 250 20 980 268.8 138.8 30
Geometrical characteristics are in mm and mechanical properties are in MPa.
Table 2
Input data for welded joints.
Test bfc tfc tw c hc fyfc bfb tfb hb fyfb aw
T107001 178 8.9 6.2 173 334.6 119 10.2 238 389.8 6.8
T107002 179 8.8 6.2 175 342.5 149 10 300 306.2 5.8
T107003 242 10.7 9.4 233 354.2 149 10 300 304.8 9.1
T107004 242 10.8 11.3 233 343.9 190 13.2 452 287 11.3
T105002 160 13.3 7.6 159 260 162 11.4 328 286 7.5
T105003 179 13.5 9 183 288 149 10 298 334 7
T105004 179 13.4 8.2 183 277 180 12.3 398 284 9.3
T105005 200 13.9 8.4 202 273 171 10.7 361 271 7.2
T105006 239 16 9.7 239 276 223 18.1 604 306 10.3
T105007 299 18.6 11.7 300 296 181 13.2 400 309 9.1
T105008 301 19.3 12.4 297 292 300 17.8 301 357 13
T105009 139 11.5 7.7 139 300 110 10 218 312 6.2
T105010 138 11.6 7.8 146 281 111 9.1 221 361 6.2
T105011 140 12 7.5 142 298 151 11.1 302 304 6.4
T105014 178 13.7 8.1 177 292 170 12.3 359 290 7.5
T105015 179 13.5 9.4 178 275 180 12.5 397 290 10.2
T105016 200 14.6 9.4 200 279 171 12 361 273 8.9
T105018 240 16.1 9.7 239 269 189 14.1 451 284 10.1
T105019 239 16.3 10.1 240 274 224 18.7 605 322 8.6
T105020 299 18.3 10.8 298 301 210 17.2 551 268 11.9
T105021 299 18.9 12.3 303 266 220 19.4 600 268 11.6
T105023 301 18.7 12 302 276 299 22.9 402 281 12.1
T105024 300 18 10.6 298 271 301 27.6 500 271 12.7
T105025 301 21.3 11.9 361 276 224 18.2 604 316 11.4
T106001 145 21.4 13.2 159 283 149 11 295 358 11.8
T106002 186 22.8 14.5 200 265 177 13.5 398 358 12.1
T106003 204 25.4 15.9 221 268 180 12.6 401 265 11
T106004 204 25.6 15.9 222 267 199 15.1 498 248 14
T106005 204 24.5 16 222 280 224 18.7 600 277 12.7
T106006 222 25 16.5 241 278 209 18 552 361 13.5
T106007 308 37 21.2 340 237 300 28 603 262 14.2
T108032 199 15 9 199 370 186 13.1 450 386 4
T108038 202 12 8.7 209 291 168 12.5 361 307 5.5
T108042 255 18.5 12.2 265 267 200 15.5 498 288 5.7
Geometrical characteristics are in mm and mechanical properties are in MPa.
steel joints, two separate models with a single output were program size. The parameter selection will affect the model
developed; one for flexural resistance and one for initial rotational generalization capability of LGP. They were selected, after a trial
stiffness. The various parameters involved in the LGP models are: and error approach, upon some previously suggested values [37,
population size, mutation rate and its different types, crossover 38]. The parameter settings are shown in Table 4. In the analysis,
rate, homologous crossover, function set, number of demes and the data sets were divided into training, validation and testing
1742 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
tfc hc
bfc
twc
tfb
hb
aw
twc bfb
hc
tta
lta
db
lwa hb
tfc
Table 3 Table 4
Input data for bolted joints with angles (in mm). Parameter settings for LGP.
Test tfc hb tta Lta Lw a db Parameter Settings
(a) LGP Single Solution-Endplate joints. (b) LGP Team Solution-Endplate joints. (c) LGP Single Solution-Welded joints.
(d) LGP Team Solution-Welded joints. (e) LGP Single Solution-Bolted joints with angles. (f) LGP Team Solution-Bolted joints with angles.
Fig. 8. Comparison of predicted vs experimental flexural resistance using LGP single and team solutions.
obtained by LGP single and team solutions. A comparison of the performance of models. Considering the R, MAPE and RMSE values
ratio between predicted flexural resistance values by the LGP, for flexural resistance, given in Table 5, it can be seen that the best
as well as those obtained from the European design code [1], performance for bolted endplate joints is obtained by LGP single
ANNs [16], GP/SA [27,28] and experimental values are shown solution (R = 0.9986, MAPE = 8.55%, RMSE = 11.12%). For the
in Figs. 9–11. As mentioned previously, R, MAPE and RMSE prediction of flexural resistance of welded joints, ANNs model
are selected as the target statistical parameters to evaluate the (R = 0.996, MAPE = 10.90%, RMSE = 14.87%) outperforms
1744 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
Fig. 11. Relative comparison of predicted flexural resistance of bolted joints with angles.
Table 5
Performance statistics of models for flexural resistance prediction for all element test data.
Type of joint LGP (single solution) LGP (team solution) ANN EC3 GP/SA
R MAPE RMSE R MAPE RMSE R MAPE RMSE R MAPE RMSE R MAPE RMSE
Bolted endplate 0.9986 8.5496 11.1212 0.9975 5.9564 13.5602 0.9967 9.808 15.3734 0.9604 29.5156 54.9477 0.9793 15.4511 45.1124
Welded 0.9819 22.0597 32.7646 0.988 18.5485 25.9763 0.996 10.897 14.8681 0.9169 54.0939 81.4623 0.9761 20.7471 36.5184
Bolted with angles 0.9918 3.2328 5.0813 0.9964 2.6267 3.5665 0.9921 4.1056 5.0685 0.964 11.85 13.9173 0.9846 4.5621 6.8337
the other models. Considering the bolted joints with angles, it team solutions and likewise those obtained from European design
is observed that the best team solution obtained by LGP with R, code, ANNs [16], GP/SA [27,28], Kishi et al. [3] and experimental
MAPE and RMSE values equal to 0.9964, 2.63% and 3.57% produces values are illustrated in Figs. 13–15. Performance statistics of
more appropriate results than other models. Based on the values of models for initial rotational stiffness prediction are presented in
performance measures in Table 5, it is shown that Eurocode 3 does Table 6. This table indicates that for bolted endplate joints the
not perform as well as others. best performance is obtained using LGP team solution with R =
The results of LGP for the prediction of initial rotational stiffness 0.9985, MAPE = 2.04% and RMSE = 5.41%. The same result is
are shown in Fig. 12(a-f). A comparison of the ratio between the obtained by the LGP team solutions for welded joints and bolted
predicted initial rotation stiffness values by the LGP single and joints with angles with R, MAPE and RMSE values equal to 0.9735,
A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750 1745
(a) LGP Single Solution-Endplate joints (b) LGP Team Solution-Endplate joints
(c) LGP Single Solution-Welded joints (d) LGP Team Solution-Welded joints
(e) LGP Single Solution-Bolted joints (f) LGP Team Solution-Bolted joints with
with angles angles
Horizontal axis represents predicted initial stiffness (MN.m)
Fig. 12. Comparison of predicted vs. experimental initial rotational stiffness using LGP single and team solutions.
9.35%, 12.85% and 0.9923, 1.73%, 2.22%, respectively. Table 7of the of the best thirty programs evolved by LGP. According to Fig. 16,
appendix shows a comparative analysis of results. for endplate joints, flexural resistance is more sensitive to column
In order to evaluate the number of times each input appears in flange width (bfc ) and bolt ultimate stress (fub ); while for initial
the contribution of the fitness of the LGP programs (importance rotation stiffness they are endplate yield stress (fyep ) and second
of input parameters), frequency [37] values of input parameters bolt row height (h2 ). Considering the welded joints, it can be seen
of the flexural resistance and initial rotational stiffness predictive from Fig. 17 that beam height (hb ) and column height (hc ) are the
models are obtained and presented in Figs. 16–18. A value of 1.00 most influencing parameters on flexural resistance; and those for
in the figures indicates that the input variable appeared in 100% initial rotation stiffness are beam height (hb ) and weld thickness
1746 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
Fig. 13. Relative comparison of predicted initial rotation stiffness of endplate joints.
Fig. 14. Relative comparison of predicted initial rotation stiffness of welded joints.
Table 6
Performance statistics of models for initial rotation stiffness prediction for all element test data.
Type of joint LGP (single solution) LGP (team solution) ANN EC3 GP/SA
R MAPE RMSE R MAPE RMSE R MAPE RMSE R MAPE RMSE R MAPE RMSE
Bolted endplate 0.9969 3.2788 7.6883 0.9985 2.0408 5.4111 0.9973 2.578 7.0930 0.9778 17.0412 25.7569 0.9836 3.6232 17.7062
Welded 0.9734 9.9467 14.1824 0.9735 9.347 12.8468 0.8546 21.2279 28.0037 0.9455 32.3606 48.3830 0.9490 10.2478 17.7226
Bolted with angles 0.9901 1.88 2.3452 0.9923 1.7272 2.2249 0.9923 1.7433 2.0616 0.9271a 6.1606a 7.9737a 0.9784 1.3071 3.4205
a
Kishi et al. model.
(aw ). As shown in Fig. 18, flexural resistance of bolted joints with prediction of semi-rigid joints, and its performance comparisons
angles is more sensitive to angle thickness (tta ) and top and seat were presented. The LGP-based models were developed upon
angle length (Lta ) while those for initial rotational stiffness are available experimental results. The geometrical characteristics and
angle thickness (tta ) and web angle length (Lwa ) compared to the mechanical properties of joints were utilized as inputs. The LGP
other inputs. results were compared to the experimental results, as well as
Eurocode 3 and other existing models in the literature. In order to
6. Conclusions determine the contribution of each input in the models, frequency
values of different input parameters of the flexural resistance and
The first application of a particular subset of GP, namely initial rotational stiffness predictive models were also evaluated.
LGP, to the flexural resistance and initial rotational stiffness Based on the values of performance measures, it was observed that
A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750 1747
Fig. 15. Relative comparison of predicted initial rotation stiffness of bolted joints with angles.
1.0
0.9 Flexural Resistance
0.8 Initial Rotational Stiffness
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
bfc tfc hc fyc bfb tfb hb fyb tep lep fyep db fub h1 h2 dh
Fig. 16. Frequency values of input parameters of endplate joints predictive models.
1.0
0.6
0.5
0.4
0.3
0.2
0.1
0.0
bfc tfc twc hc fyfc bfb tfb hb fyfb aw
Fig. 17. Frequency values of input parameters of welded joints predictive models.
the proposed LGP models give very reliable estimates of flexural flexural resistance of bolted endplate joints. It was found that the
resistance and initial rotational stiffness of beam-column steel LGP based models yielded better results for the prediction of both
joints. The results roughly demonstrate that the best single and flexural resistance and initial rotational stiffness of bolted endplate
team solutions created by LGP were more accurate than ANN joints than those of bolted joints with angles and welded joints.
models and far more accurate than Eurocode 3 and the other Because of the high precision of the models developed by the LGP
models. Furthermore, the LGP team solutions produced slightly approach, an excessive number of experiments and computations
better results than the single solutions except for the prediction of can be avoided, which leads to reduction of the costs of product
1748 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
Table 7
Results of LGP models vs. experimental and other models results.
Test Ref. Mj,RdEXP Mj,Rd,LGP /Mj,Rd,EXP Mj,Rd,ANN / Mj,Rd,EC 3 / Mj,Rd,GP /SA / Sj,RdEXP Sj,Rd,LGP /Sj,Rd,EXP Sj,RdANN / Sj,RdEC 3 / Sj,RdGP /SA /
(kN m) Mj,Rd,EXP Mj,Rd,EXP Mj,Rd,EXP (kN m) Sj,Rd,EXP Sj,Rd,EXP Sj,Rd,EXP
Single Team Single Team
Table 7 (continued)
Test Ref. Mj,RdEXP Mj,Rd,LGP /Mj,Rd,EXP Mj,Rd,ANN / Mj,Rd,EC 3 / Mj,Rd,GP /SA / Sj,RdEXP Sj,Rd,LGP /Sj,Rd,EXP Sj,RdANN / Sj,RdEC 3 / Sj,RdGP /SA /
(kN m) Mj,Rd,EXP Mj,Rd,EXP Mj,Rd,EXP (kN m) Sj,Rd,EXP Sj,Rd,EXP Sj,Rd,EXP
Single Team Single Team
14S1 [23,24] 77.70 1.02 1.13 1.10 0.79 1.06 22.03 1.10 1.10 1.10 0.65 1.01
14S2 [23,24] 107.00 1.01 1.00 1.00 1.21 1.03 33.33 0.99 0.97 0.95 1.12 1.20
14S3 [23,24] 73.90 0.84 1.06 0.91 0.70 0.86 13.09 1.32 1.37 1.16 1.10 1.06
14S4 [23,24] 92.90 0.85 0.94 0.92 0.94 0.89 25.07 0.97 0.97 0.97 0.57 0.89
14S5 [23,24] 86.20 1.00 1.02 0.98 0.75 1.13 27.90 0.88 0.96 0.98 0.61 0.80
14S6 [23,24] 119.00 0.98 1.00 1.02 0.87 1.10 32.30 0.99 1.01 1.02 1.39 1.24
14S8 [23,24] 176.40 1.00 0.99 0.99 0.85 0.93 65.40 0.99 0.99 1.01 1.22 0.96
14S9 [23,24] 115.70 1.01 1.02 1.05 0.90 1.03 29.20 1.10 1.11 1.12 1.54 1.14
a
Kishi et al. model; Bold sets are test sets.
Fig. 18. Frequency values of input parameters of bolted joints with angles predictive models.
development. In addition, LGP unlike most models is a white-box [17] Koza JR. Genetic programming: On the programming of computers by means
model that provides transparent programs of an imperative or of natural selection. Cambridge (MA): MIT Press; 1992.
[18] Banzhaf W, Nordin P, Keller R, Francone F. Genetic programming – an
machine language [41]. introduction. In: On the automatic evolution of computer programs and its
application. Heidelberg/San Francisco: dpunkt/Morgan Kaufmann; 1998.
References [19] Bäck T. Evolutionary algorithms in theory and practice: Evolution strategies,
evolutionary programming, genetic algorithms. USA: Oxford University Press;
1996.
[1] CEN, Eurocode 3. May 2003 EN1993-1-8: Design of Joints. CEN, European
[20] Ashour AF, Alvarez LF, Toropov VV. Empirical modelling of shear strength of RC
Committee for Standardization Brussels, Brussels, p. 124.
deep beams by genetic programming. Computers & Structures 2003;81(55):
[2] Zoetemeijer P, Munter H. Extended endplate with disappointing rotation
331–8.
capacity: Test results and analysis. Report [6-75-20 KV-4]. University of
[21] Alavi AH, Gandomi AH, Gandomi M, Sahab MG. A data mining approach
Technology, Delft. 1983.
to compressive strength of CFRP confined concrete cylinders. Computers &
[3] Kishi N, Chen WF, Matsuoka KG. Moment-rotation relation of top-and-seat
Structures, 2008 [in press].
angle with double web angle joints. In: Joints in steel structures. 1987.
[22] Javadi AA, Rezani M, Mousavi Nezhad M. Evaluation of liquefaction induced
p. 121–134.
lateral displacements using genetic programming Computers and Geotechnics
[4] Janss J, Jaspart JP, Maquoi R. Experimental study of non-linear behaviour of
2006;33(4–5):222–33.
beam-to-column bolted joints. Joints in steel structures: Behaviour, strength
[23] Folino G, Pizzuti C, Spezzano G. Genetic Programming and simulated
and design. In: Proceedings of the 1st international workshop on joints. 1997.
[5] Azizinamini A, Bradburn JH, Radziminski J. Initial stiffness of semi-rigid steel Annealing: A hybrid method to evolve decision trees. In: Poli R, Banzhaf W,
beam-to-column joints. Journal of Constructional Steel Research 1987;8: Langdon WB, Miller JF, Nordin P, Fogarty TC, editors. Genetic programming,
71–90. proceedings of EuroGP’2000, vol. 1802. Edinburgh: Springer-Verlag; 2000.
[6] Azizinamini A, Radziminski J. Static and cyclic performance of semi-rigid p. 294–303.
steel beam-to-column joints. Journal of Structural Engineering 1989;115(12): [24] Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E. Equation of
2979–99. state calculations by fast computing mechanics. Journal of Chemical Physics
[7] Simek I, Wald F. Test results on endplate beam-to-column joints. G-1121 1953;21(6):1087–92.
Report. Czech Technical University, Prague, 1991. [25] Kirkpatrick Jr S, Gelatt CD, Vecchi MP. Optimization by simulated annealing.
[8] Aggarwal AK. Comparative tests on end plate beam-to-column joints. Journal Science 1983;220(4598):671–80.
of Constructional Steel Research 1994;30(2):151–75. [26] Cerny V. Thermodynamical approach to the traveling salesman problem:
[9] Faella C, Piluso V, Rizzano G. Structural steel semi-rigid joints - theory, design an efficient simulation algorithm. Journal of Optimization Theory and
and software. CRC Press; 1999. p. 505. Applications 1985;45:41–52.
[10] Hummer C, Tschemmernegg T. A non-linear joint model for the design of [27] Gandomi AH, Sahab MG, Heshmati AA, Alavi AH, Gandomi M, Arjmandi P.
structural steel frames. Costruzioni Metalliche, No. 1. 1988. Application of a coupled simulated annealing-genetic programming algorithm
[11] Simões R, Simões da Silva L, Cruz P. Behaviour of end- plate beam-to-column to the prediction of bolted joints behavior. American-Eurasian Journal of
composite joints under cyclic loading. Steel Composite Structure 2001;1(3): Scientific Research 2008;3(2):153–62.
355–76. [28] Gandomi AH, Alavi AH, Sahab MG, Gandomi M, Safari Gorji M. Empirical
[12] Haykins S. Neural networks – A comprehensive foundation. 2nd ed. Englewood models for Prediction of flexural resistance and initial stiffness of welded
Cliffs (NJ): Prentice Hall International Inc.; 1999. beam-column joints. In: 11th East Asia-Pacific conference on structural
[13] Abdala KM. Stavroulakis, back propagation neural network model for semi- engineering & construction, 2008.
rigid joints. Microcomputers in Engineering 1995;10(2):77–87. [29] Brameier M, Banzhaf W. Linear Genetic Programming. Springer Science +
[14] Stavroulakis GE, Abdala KM. A systematic neural network classificator in Business Media, LLC; 2007.
mechanics. application in semi-rigid steel joints. International Journal of [30] Oltean M, Grosan C. A comparison of several linear genetic programming
Engineering Analysis and Design 1994;1:279–92. techniques. Complex Systems 2003;14(4):1–29.
[15] Anderson D, Hines EL, Arthur SJ, Eiap EL. Application of artificial neural [31] Baykasoğlu A, Güllü H, Çanakçı H, Özbakır L. Prediction of compressive and
networks to the prediction of minor axis joints. Computers & Structures 1997; tensile strength of limestone via genetic programming. Expert Systems with
63(4):685–92. Application 2008;35(1-2):111–23.
[16] de Lima LRO, da S, Vellasco PCG, de Andrade SAL, da Silva JGS, Vellasco MMBR. [32] Jaspart JP. Recent advances in the field of steel joints-column bases and
Neural networks assessment of beam-to-column joints. Journal of the further configurations for beam-to-column joints and beam splices. Chercheur
Brazilian Society of Mechanical Sciences 2005;27(3):314–24. qualifié du F.N.R.S., Universite de Liege, Faculte des Sciences Appliques 1997.
1750 A.H. Gandomi et al. / Journal of Constructional Steel Research 65 (2009) 1738–1750
[33] Da Silva LS, Coelho AMG. Ductility model for steel joints. Journal of [38] Feldt R, Nordin P. Using factorial experiments to evaluate the effect of genetic
Constructional Steel Research 2001;57(1):45–70. programming parameters. In: EuroGP 2000. LNCS, vol. 1802. 2000. p. 271–82.
[34] Brameier M, Kantschik W, Dittrich P, Banzhaf W. SYSGP – A C++ library of [39] Brameier M, Banzhaf W. Evolving teams of predictors with linear genetic
different GP variants. Technical Report [CI-98/48]; Germany: Collaborative programming. Genetic Programming and Evolvable Machines 2001;2(4):
Research Ceter 531 University of Dortmund, 1998. 381–407.
[35] Nordin PJ. A compiling genetic programming system that directly manipulates
[40] Francone FD, Deschaine LM. Extending the boundaries of design optimization
the machine code. In: Kenneth E, Kinnear JR, editors. Advances in Genetic
Programming. USA: MIT Press; 1994. p. 311–31. (Chapter 14). by integrating fast optimization techniques with machine-code-based linear
[36] Deschaine LM. Tackling real-world environmental challenges with linear genetic programming. Information Sciences 2004;161(3–4):99–120.
genetic programming. PCAI magazine 2000;15(5):35–7. [41] Weise T. Global optimization algorithms — Theory and application. On-
[37] Francone F. DiscipulusTM owner’s manual. In: Version 3.0, Register Machine line E-book, GNU free documentation license (FDL). 2007. Available at:
Learning Technologies. 2000. http://www.it-weise.de/projects/book.pdf.