PSI - Issue 64

Eshwar Kumar Ramasetti et al. / Procedia Structural Integrity 64 (2024) 557–564 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

563

7

4.1. Classifier 1: Binary Classification (vehicle passed or not) Based on the acceleration data, the CNN model was evaluated to classify if there was any vehicle passing or not on the bridge. Fig. 7 (a) depicts the training and validation loss curve of the CNN model described in Section 3, which empathizes with its training performance.

(b)

(a)

Fig. 7. (a) Training and validation loss curves (b) Confusion matrix of true vs. predicted labels The training and validation losses diminish to a lower value as the number of epochs increases, eventually converge to a lower value. This decrement in losses shows that the number of epochs used was acceptable and that the model can perform effectively after training. Fig. 7 (b) shows the confusion matrix of the model’s performance on test data. The CNN classifier achieved an accuracy of approximately 98 %, which matches the accuracy obtained from Lawal et al. (2023) using the ANN classifier. 4.2. Classifier 2: Multi-label Classification (cars or trucks or large trucks) A CNN model with same hyperparameters was evaluated to further classify if the passing vehicle was a car or truck or large truck. Fig. 8 (a) depicts the training and validation loss curve for multi-label classification. As in the binary classification, the training and validation losses converge with an increase in the number of epochs. The model was evaluated on the test data sets, and an accuracy of 98 % was achieved.

(a)

(b)

Fig. 8. (a) Training and validation loss curves for multi-label classification (b) Confusion matrix of actual vs. predicted labels