PSI - Issue 64

5

Nikhil Holsamudrkar et al. / Procedia Structural Integrity 64 (2024) 580–587 Holsamudrkar Nikhil et al./ Structural Integrity Procedia 00 (2019) 000 – 000

584

Fig. 3. Dataset count across four classes as an input to the CNN model

Table 2. CNN model parameters

Geometric features Filter Size – 3 x 3 Pool Size – 2 x 2 Stride – (2, 2) Batch size – 30 Epochs – 15 Dropout – 0.5

Functional features

Hyperparameters

Activation – RELU Optimizer – Adam Loss – Categorical cross-entropy Metrics – Accuracy Cross-Validation – Yes Classification Function - SoftMax

Learning rate – 0.0001 Beta 1 – 0.9 Beta 2 – 0.99 Epsilon – 1e-7

Fig. 4. VGG16 CNN model architecture adopted from (Simonyan & Zisserman, 2014)