PSI - Issue 60

S. Mahesh et al. / Procedia Structural Integrity 60 (2024) 382–389 Mahesh et al. / Structural Integrity Procedia 00 (2019) 000 – 000

387

6

are continuously tweaked with multiple trial runs before arriving at the working model. The model was trained with data for one hundred and fifty epochs/iterations. 2.3. Testing and saving the model Testing the algorithm for its effectiveness was carried out using the standard “train -test- split” approach. From the entire input data (as shown in Table 1), certain percentage of data was randomly chosen and used for training the algorithm. Subsequently the rest of the left-over data was used to test the trained model. In the present work, as the training data-set is very small, the ratio of splitting the data was 60% for training and 40% for testing. More on the “train -test- split” approach can be found in the work of Brownlee (2016). Predictions were carried out for the same material for varying R values: 0.2, 0.4, 0.6 and 0.8. The results of the prediction are discussed in the following section.

3. Results and Discussion

(b)

(a)

Fig. 4. Comparison of predicted (P) and experimental (E) Paris law constants (C and m)

(b)

(a)

Fig. 5. Comparison of predicted (P) and experimental (E) FCGR plots