PSI - Issue 64

Manish Prasad et al. / Procedia Structural Integrity 64 (2024) 1524–1531 Manish Prasad / Structural Integrity Procedia 00 (2019) 000 – 000

1528

5

3. Implementation of ML on the prediction of CCS failure 3.1. The experimental database

Experimental data were collected from 45 experimental campaigns, giving a total of 140 beams with CCS failure mode (Al-Ghrery et al., 2021; Smith & Teng, 2002b). Collected data contains all relevant geometrical, mechanical, and material properties of all beam specimens. The selected variables which include cross-sectional geometric properties (beam width b , depth h , effective depth d , FRP thickness t f , FRP width b f , and tensile steel reinforcement area A s ), positional parameters ( a , a f ) as shown in Fig. 2, material properties (elastic modulus of steel, FRP and concrete, E s , E f , and E c , respectively), yield strength of steel reinforcement ( f y ), tensile strength of FRP ( f f ), compressive and tensile strength of concrete ( f c , and f ct , respectively) and the maximum load ( P ), acting as an output or target variable.

Fig. 2. Geometrical properties of the beam.

In this preliminary study, which explores the possibilities that ML tools offer for the prediction of the load capacity in FRP strengthened RC elements. All variables have been transformed to dimensionless features, except for the concrete compressive strength (fc). The features used in this study are the ratios of the beam width to height ( b/h ), the shear span to depth ratio ( a/d ), the shear span to unbonded length ( a/a f ), the internal and external reinforcement ratios ( r s , r f ), and the modular ratios ( E s / E c , Ef/E c ). From the total amount of collected data (140 beams), data is then split into 80:20 proportion for training and testing the ML models, respectively. An outline of the statistical measurements of the experimental data is presented in Table 1.

Table 1. Statistical measurement of the experimental data

f c (MPa)

P exp

b/h

a/d

a/a f

E s /E c

E f /E c

f f /f y

 s

 f

Minimum Maximum

22.00 80.00 44.98 45.95

0.43 1.33 0.73 0.70

2.28 6.25 3.60 3.53

1.21

5.20

1.34

2.31 9.15 6.03 6.02

0.18 3.49 0.50 0.36

0.06 1.04 0.39 0.39

16.50

28.00

65.15

10.24

140.20

Mean

9.63 8.82

8.27 6.90

6.32 7.21

81.45 77.50

Median

Standard Deviation

13.35

0.24

0.98

7.15

8.93

2.38

1.65

0.43

0.22

35.48

3.2. Fitting the ML algorithm A typical ML training procedure is shown in Fig. 3. When the data is manually split into 80:20, the last 20% of the data may contain important relationships that are not being used during training the model. Therefore, it is important