DAMAGE IDENTIFICATION AND LOCALIZATION OF BRIDGE STRUCTURE USING PROBABILISTIC PRINCIPAL COMPONENT ANALYSIS

• Main Author: Farhan Fathurrahman, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70027
• Institution: Institut Teknologi Bandung

The main problem of infrastructure management in Indonesia in general is the lack of historical data on infrastructure conditions. The Structural Health Monitoring (SHM) system is one of the technologies to solve infrastructure management problems. However, the problem of missing data on the SHM system can interfere with the performance of the SHM system. In this research, damage identification and localization of structure is developed which is a crucial part of the SHM system. The developed system utilizes the vibration data of structure measured by the accelerometer in ambient conditions. The system was tested using the Z24 data bridge by comparing the vibration data of the structure in a healthy condition and in the event of damage. Vibration data of the structure in healthy condition will be modeled using principal component analysis (PCA) and probabilistic principal component analysis (PPCA). PPCA model parameter values are estimated using the expectation-maximization algorithm so that vibration data containing missing data can be modeled. By using vibration data in unknown condition and structure model in healthy condition, structural damage identification is performed by calculating the T 2 -statistic which can detect changes in operational or environmental conditions and the Q-statistic which can detect damage to the structure. In addition, localization of structural damage was also carried out by clustering the bridge structure into clusters that have a large correlation. The PCA and PPCA models built have normalized root mean square error (NRMSE) values of 6.45% and 6.45%, respectively. In this study, the PPCA model was also constructed using incomplete data with a ratio of missing data of 10% and achieved similar accuracy with NRMSE value of 6.52%. The results of structural damage identification using the PCA and PPCA models succeeded in detecting all cases of damage. The results of structural damage localization show that the PPCA model has a more accurate location than PCA. The results of this study also show that the PPCA model with incomplete data has the same results as the PPCA model with complete data.