DEVELOPMENT OF COMPUTER VISION FOR THE CLASSIFICATION SYSTEM OF RAILWAY FASTENING COMPLETENESS USING A CONVOLUTIONAL NEURAL NETWORK APPROACH

• Main Author: Ekaputra, Ariansyah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83618
• Institution: Institut Teknologi Bandung

Rail fastener is one of the most important components in railway system. It functions to secure rail into sleepers. It has responsibility to minimize rail vibration caused by train movement. However, rail fastener often go missing due to excessive usage and lack of maintenance which prompt them to be broken and missing. The other fact said this component often stolen. Consequently, track inspection has to be carried out periodically. Currently, the inspection depends on inspector capability to identify rail fastener condition. This traditional method lead to time consuming, require a large human resource, and tend to be inaccurate. Computer vision use camera to imitate human’s eye function to capture rail fastener. The images of rail fastener are processed by computer or single board processor to check whether rail fastener is complete or not. Convolutional Neural Networks (CNN) consist of feature extraction and classification algorithms. The features of image are extracted by convolution and dimension reduction process. Then, the features will be classified by Multi-Layer Perceptron (MLP) to yield the prediction output. In this research, CNN capable to achieve high value in all of evaluation metrics. The model yields 98,6% accuracy; 97,8 precision; 98,80% f1-score; 99,27% specificity; and 95,76% negative predictive value. The architecture of model consisting of four convolutional layer, four max pooling layer, and three hidden layer. Keywords: railway system, computer vision, rail fastener, convolutional neural network ?