Image-based structural performance prediction for reinforced concrete exterior beam-column joints

• Main Author: Teo, Wilson
• Other Authors: Li Bing
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Civil engineering
• Online Access: https://hdl.handle.net/10356/172533
• Institution: Nanyang Technological University

This report detailed the Final Year Project – “Image-Based Structural Performance Prediction for Reinforced Concrete Exterior Beam-Column Joints.” In the past, it was commonly believed that the strength of reinforced concrete beams-columns joints was primarily determined by the member with the lowest strength within the connection. However, extensive research has been conducted that it is untrue as the beam-column joints are critical region that experienced high normal and bond stress. Joints are often the weakest links in a structural system and are often the with different aspect ratios. The loading-induced cracks affects and degrade the structural strength, ductility, and stiffness in a seismic behaviour. Therefore, rationally assessing the level of exterior beam and column joint is valuable for understanding structural condition. This study integrates advanced image processing and machine learning approaches, commonly referred to as computer vision techniques, to identify structural damage and estimate the degraded structural performance of exterior beam-and-column joints. The study has successfully linked surface observations to the quantitative estimation of damage levels based on stiffness, load, and displacement. For this project, 141 shear-critical specimens were gathered from earlier experimental studies, containing a total of 500 images and datasets. Data on the experiment's setup details, applied loads, displacement levels, and geometrical characteristics were gathered. Subsequently, python programming is used to visualise and analyse the collected datasets to weed out anomalies or outliers. Using a programming and numeric computing platform - MATLAB, the images are segmented into beam, column and joint components which are converted into a consistent binary format. The employed image segmentation techniques that are utilised include thresholding and edge detection, which are aimed at extracting textural and geometric attributes of surface crack patterns and converting them into quantifiable information. Machine learning regression techniques are utilised to generate estimation models from the surface crack patterns. These models are then utilised to predict the damage states based on the identified crack patterns. Within the regression prediction models, numerous pairs of predictors and responses are identified and utilised. Subsequently, statistical measures of errors were used to assess the performance of the regression models. The results clearly showcase the effective performance of the estimated models across diverse geometric configurations, loading patterns, types of loads, concrete strengths, and reinforcement specifications. In conclusion, the regression analysis results exhibit a prominent level of predictive accuracy and strong correlations, indicating the feasibility of employing image-based damage recognition of exterior beam-column joints for conducting visual structural inspections.