MACHINE LEARNING APPROACHES FOR BRIDGE STRUCTURAL DAMAGE DETECTION AND LOCALIZATION

• Main Author: Aulia Akbar Kustiana, Willy
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/57234
• Institution: Institut Teknologi Bandung

Most of the bridge structure monitoring systems are done by manual data processing. Monitoring is carried out only for a certain period, some of which are for a short period so the historical data is not properly collected. In addition, there is no early warning system to notify indications of damage. To provide information on the condition of the bridge, this study improves the existing system by utilizing certain features to detect damage and localization using a machine learning approach. The proposed system utilizes the vibrations obtained by the accelerometer-based wireless sensor nodes. The bridge acceleration data collected by the accelerometer wireless sensor node is transformed using the Fast Fourier Transform (FFT) to obtain the dominant frequencies and the appropriate magnitude as a machine learning feature. In this study, comparisons and selection of machine learning methods are also carried out to determine the most suitable method for detecting and localizing damage to bridges. The observed machine learning method is Multilayer Perceptron (MLP) classification method with the Adam optimizer and gradient descent, and the Support Vector Machine (SVM) classification method with a linear kernel and Radial Basis Function (RBF). The results were successfully tested in a bridge model, where the SVM method with the RBF kernel had the best performance with 97% accuracy and 96% precision, 97% recall rate, and 96% f1-score in terms of classifying and detecting damage and location detection.