PSI - Issue 17

Maria Apostolopoulou et al. / Procedia Structural Integrity 17 (2019) 914–923 Maria Apostolopoulou et al. / Structural Integrity Procedia 00 (2019) 000 – 000

919

6

twelfth case of trained and developed ANN model with two hidden layers, three encoding parameters, including the maximum diameter of aggregates in the input parameters and with the use of normalization technique. Furthermore, its transfer functions are the Log-sigmoid transfer function (logsig) for the first hidden layer, and the hyperbolic tangent sigmoid transfer function (tansig) for the second hidden layer and the the hyperbolic tangent sigmoid transfer function (tansig) for the output layer. In Table 2 the values of statistical indexes such as R, RMSE, MAPE and VAF are presented.

Table 2. Statistical Indexes of the optimum NN model

Statistical Indexes

Case

NN Architecture

Datasets

R

RMSE

MAPE

VAF

Training All Data Test Data

0.9803 0.9614 0.9455

0.5641 0.7632 0.7792

0.1281 0.2070 0.2071

96.1041 92.4218 89.3651

12

7-12-25-1

Figure 2 depicts the comparison of the exact experimental values with the predicted values of the optimum NN model for the case of test data. It is clearly shown that the proposed optimum 7-12-25-1 NN reliably predict the compressive strength of mortar materials.

Fig. 2. Experimental vs Predicted values of the mortar compressive strength for the case of Test datasets