Deep learning‐based evaluation for mechanical property degradation of seismically damaged RC columns

Self Cite

In the aftermath of an earthquake, damage detection and performance evaluation of structural components are imperative for assessing the residual seismic capacity of a building. In this study, an integrated image processing and deep learning approach was developed to evaluate the degradation in strength and stiffness (i.e., strength reduction and stiffness reduction) of reinforced concrete (RC) shear walls. The approach comprised two main tasks: detecting and localizing visible seismic damage from photographs and evaluating strength and stiffness degradation based on this information. The semantic segmentation network, Damage‐Net, was used for damage detection and localization. A novel crack morphological processing layer and a patch feature extraction layer were developed for damage feature extraction and compression. A lightweight deep convolutional neural network named DegradeEval‐Net_v2, featuring the upgraded dilated and separable convolution block and multi‐layer perception, was developed to link the damage feature with strength and stiffness degradation. A database comprising test data and photographs of 14 RC shear wall specimens with a flexural‐dominated behavior mode and high to intermediate ductility was constructed to train and test the DegradeEval‐Net_v2 network. The results indicate that DegradeEval‐Net_v2 substantially improved the performance assessment accuracy of damaged RC shear walls, with a 35% smaller root mean square error (RMSE) for stiffness degradation evaluation and 75% smaller RMSE for strength degradation evaluation, compared with the provisions specified in JBDPA and FEMA guidelines. Moreover, evaluation results on test sets demonstrate that introducing the damage feature extraction and compression layers effectively preserved local crack information and improved the accuracy with which stiffness reduction was evaluated. In addition, DegradeEval‐Net_v2 outperformed ResNet18 and MobileNet V3 in terms of balanced efficiency and accuracy. Interpretability analysis demonstrates that the model learned the distinct contribution patterns of various visible damage indexes to stiffness and strength degradation across different loading levels.

Section: Feature Extraction and Compressionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Evaluation of strength and stiffness degradation of RC shear walls: An integrated image processing and deep learning approach

Ji,

Yu,

Gao

et al. 2024

Self Cite

“…Except for the aforementioned damage state‐based estimation, the authors recently developed a deep CNN model to directly correlate the visible damage appearance to the mechanical property degradation of RC columns, 22 which could achieve refined estimates of stiffness and strength reduction factors directly from damage images. Nevertheless, the deep CNN model had not yet been extended to estimations of RC beams and walls.…”

Section: Framework For Vision‐based Postearthquake Residual Capacity ...mentioning

confidence: 99%

Vision‐based seismic damage detection and residual capacity assessment for an RC shaking table test structure

Zhuang

Miao

et al. 2022

Self Cite

In this paper, the existing postearthquake performance assessment framework for reinforced concrete (RC) building structures is improved by adding a new feature of the computer vision‐based damage detection. In this framework, visible seismic damage is classified and quantified from photographs of damaged RC components using the developed deep convolutional network (CNN) Damage‐Net of semantic segmentation, and then the mechanical property degradation factors of components determined from the detected damage states are used to update the numerical model. Pushover analysis of the updated model assesses the residual capacity of the damaged structure. Large‐scale shaking table tests of a three‐story RC building structure, which was heavily instrumented with sensors and recorded with a large volume of photographs, were used as a case study to demonstrate the improved postearthquake performance assessment framework. The vision‐based approach accurately detected multicategory seismic damage of the test structure and effectively estimated the residual crack widths and angles under various lighting, image acquisition, and surface conditions. The updated model, which incorporated the mechanical property degradation of the damaged components, provided accurate estimate on the fundamental vibrational frequencies of the damaged structure after various levels of seismic motion shaking, which matched well with the system identification results. Using the mechanical property reduction factor values recommended by FEMA 306 & Chiu et al., pushover analysis of the updated models provided residual capacity curves that reasonably captured the measured hysteretic responses of the structure. In addition, the damage states of components as estimated by the vision‐based methods were also compared with the measured plastic hinge rotation data. The successful implementation of the vision‐based assessment in this test case indicates its potential for application in the postearthquake evaluation of buildings.

“…The study also correlated the cracking length and crushing areas to the post-earthquake damage states of beam-column subassemblies using FEMA P-58 and HAZUS guidelines. In another study, Miao et al 22 developed a deep Convolutional Neural Network (CNN) model to relate visible seismic damage features to the stiffness and strength degradation of flexural-dominated RC columns. Based on the presented literature, machine learning is a state-of-the-art computer-based technique capable of assessing structural integrity and has been shown to have remarkable performance in detecting and extracting visible damage features.…”

Section: Introductionmentioning

confidence: 99%

“…The study also correlated the cracking length and crushing areas to the post‐earthquake damage states of beam‐column subassemblies using FEMA P‐58 and HAZUS guidelines. In another study, Miao et al 22 . developed a deep Convolutional Neural Network (CNN) model to relate visible seismic damage features to the stiffness and strength degradation of flexural‐dominated RC columns.…”

Section: Introductionmentioning

confidence: 99%

Three‐dimensional fragility surface for reinforced concrete shear walls using image‐based damage features

Asjodi

Dolatshahi

Burton

2023

This paper proposes a new data-driven method to generate three-dimensional fragility surfaces for post-earthquake damage assessment of reinforced concrete (RC) shear walls (SWs) using image-based damage features. A research database comprised of 212 images corresponding to 66 damaged reinforced concrete shear walls tested under quasi-static cyclic loads is utilized. The walls are categorized into three damage states defined based on different load points along the backbone curve. Convolutional kernel-based filters are then employed to measure the crack patterns and crushed areas from images of the damaged walls. A set of 360,000 sub-cracks from the 212 images is analyzed using Gaussian mixture modeling to distinguish between shear and flexural cracking. These two types of cracking, along with crushing, are the prominent characteristics used to determine the damage states of the RCSWs. The extracted features of shear and flexural cracking are also compressed into a unique cracking index using the Principal Component Analysis theory for dimensionality reduction. Using the compressed feature of cracking and crushing, a new methodology for generating three-dimensional fragility surfaces, entitled the Box-counting method, is introduced, and the damage surfaces are developed based on visual damage features. The damage surfaces are finally formulated using statistical models or machine learning regression learners. The proposed fragility surface can be used for postearthquake damage state identification, risk assessment, and loss estimation of damaged RCSWs.