PSI - Issue 68

Oleh Yasniy et al. / Procedia Structural Integrity 68 (2025) 132–138 O. Yasniy et al. / Structural Integrity Procedia 00 (2025) 000–000

136

5

a

b

Fig. 5. The predicted versus true (experimental) values of the strain, obtained using the ANN method for the period of loading (a) and unloading (b) of the SMA material.

As it can be seen from Figs. 1-5, the predicted values of material strain are very close to the experimental ones, which is also confirmed by the low value of the MAPE prediction error (Table 1). Table 1 shows that the best MAPE value was obtained for the ANN-based model. To further evaluate the performance of this model, we used the data of 127 load-unload cycles as test data. Fig. 6 shows the relationship between the experimental values of the material strain ! ! " $ #%! $ and the predicted values ! ! " $ #%! $ when testing the ANN performance on 127 loading and unloading cycles. Table 1. MAPE prediction error of different machine learning methods. MAPE (%) Machine learning method Boosted trees Random forests SVM KNN ANN UP direction 0.48 0.79 1.34 0.51 0.29 DOWN direction 0.63 0.91 2.22 0.88 0.38

a

b

Fig. 6. Predicted and true (experimental) values of the strain obtained using the ANN method for 127 cycles for the period of loading (a) and unloading (b) of the SMA material.