PSI - Issue 46

Nithin Konda et al. / Procedia Structural Integrity 46 (2023) 87–93 Nithin Konda et al. / Structural Integrity Procedia 00 (2019) 000–000

92 6

3 . 1 Predictions using the RF and XG Models Since models are validated, predictions on the experimental fatigue data are made and comparison is made with the actual data visually. The actual versus predicted graphs are below in Figures 2(a-b)

Fig. 2. (a) Test values versus Predicted values plot for RF Algorithm; (b) Test values vs Predicted values plot for XGB Algorithm

It is clear from the graphs that the RF algorithm is over predicting most of the data and far away from the center line, whereas the data points in XGB algorithm are nearer to the center line and distributed uniformly. It is clear that XGB is superior.

Fig. 3. (a)Experimental vs Predicted S-N Data for RF Algorithm; (b) Experimental vs Predicted S-N Data for XGB Algorithm

To understand predictive accuracy of machine learning (ML) algorithms used in the present work, the actual S-N data is plotted along with predicted S-N data which are shown in the Fig.3 above. It is clear that patterns are well identified in XGB algorithm when compared to RF Algorithm. Hence, XGB is performing superior to RF for the Fatigue life prediction of Ti6Al4V alloy fabricated using L-PBF. 4. Conclusions Machine learning (ML) has been an effective tool for predictive analysis in the current times. Besides few ML models are not interpretable, such models could be viewed as statistical approximation techniques that substantiate underlying mechanism in structure property relationship. A novel attempt has been made to estimate fatigue life of L PBF fabricated Ti6Al4V alloy using Data Analytics in this work. By collecting the data pertaining to variations in manufacturing method and the testing condition, two ML models are used for predicting fatigue life of Ti-6Al4V and