PSI - Issue 64

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

562

6

Fig. 6. The classification of vehicles passed on the bridge based on acceleration data (30 sec max. value).

3.2. Neural Network Architecture The convolutional neural network (CNN) model for the classification approach was chosen to study the classification problem of the vehicle movement on the bridge. A CNN model commonly consists of three layers: a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer bears the major chunk of the computation, pooling layer curtails the spatial size of the representation and shortens the number of computations; and the fully connected layer is connected to both layers. The reason for using CNN is to simplify the accurate classification of events that occur on the road bridges. To use the CNN architecture, a set of inputs needs to be fed into the network. These inputs are the five mentioned features extracted from the acceleration data caused by the vibration of the bridge from the vehicle ’s movement. The dataset was randomly split into 80 % training data and 20 % testing data, and the developed model was used on the test dataset. Table 2 gives a summary of the hyperparameters used in training the CNN model.

Table 2: CNN model hyperparameters

Parameter No. of inputs Kernel size

Value

5 3

Activation function

Relu

Loss function

Binary cross-entropy

Optimizer Adam Output layer activation function Sigmoid

4. Results and Discussion In this section, the results using the CNN model to classify the vehicle movement on the bridge is explained. In the Classifier 1, the model was used to classify whether any vehicle passed or not based on the acceleration data. The model was extended in Classifier 2 to classify further whether the passed vehicle was a car, truck, or large truck.