PREDICTION OF BRIDGE LOAD RATING FACTOR FOR THE UPPER STRUCTURE OF STEEL FRAME BRIDGES USING GAUSSIAN PROCESS REGRESSION.

• Main Author: Fauzi, Hafiz
• Format: Theses
• Language: Indonesia
• Subjects: Teknik sipil
• Online Access: https://digilib.itb.ac.id/gdl/view/80463
• Institution: Institut Teknologi Bandung

One of the methods for bridge capacity assessment is the Load & Resistance Factor Rating (LRFR) method, which produces a Rating Factor (RF). The RF is the ratio of the structural element's capacity to the forces generated by loading. The current analysis method for calculating bridge capacity is lengthy and time-consuming, necessitating a method to assess bridge load rating factors accurately, quickly, effectively, and efficiently. In this research, a machine learning approach, specifically the Gaussian Process Regression (GPR) method, is employed to calculate the rating factors for typical bridges. The study focused on the structural design of a standard Type A steel frame bridge with a span of 50 m, referenced to the current standard drawing. The analysis utilized the latest loading criteria, and the rating factors were calculated for both inventory and operating loading conditions. The research involves modeling bridge deterioration, namely corrosion in the main bridge elements, to assess its impact on the rating factor. The corrosion was varied based on the depth of penetration in the flange and web of the top chord, bottom chord, diagonal chord, and cross beam elements. The maximum modeled corrosion penetration depth was 600 ?m. A total of 60 training data points were used in the GPR model, obtained from Finite Element Analysis modeling. Based on the analysis, the top chord element was identified as the most critical, exhibiting the lowest rating factor values under inventory and operating conditions. Even in undamaged conditions, the rating factor was found to be below the acceptance criterion of RF. To assess the performance of the GPR model, predictions for new data points were compared with FEA modeling. The maximum difference obtained was 0.527% under inventory and operating loading conditions. The predictions from the GPR model, obtained through simulation, demonstrate its ability to assess the rating factor rapidly and accurately, thus providing a representation of the upper structure's condition in the bridge.