Crack width and spacing measurement using deep learning and damage assessment for reinforced concrete non-structural wall

Abstract

With technological advancements, the seismic resistance of reinforced concrete (RC) buildings has significantly increased. However, as the performance of structural members increases, damage to both structural and non-structural members becomes a severe challenge. To achieve sustainable development goals (SDGs) and a decarbonized society, damaged structures should be reused after large-scale earthquakes. In this study, deep learning (DL) was used for damage assessment in non-structural walls by investigating crack width and spacing. Images captured during previous loading tests for non-structural specimens were used as DL data. First, 12 DL models were designed using the parameters of a pre-trained model, a gradient descent method, and image extension, and these models were examined to determine the optimal model for crack detection. The model that used Xception and stochastic gradient descent method but not image extension performed better than the other models. Subsequently, a crack width measurement method was proposed, and the obtained values were similar to the crack widths measured using a crack gauge. However, the measured values were 1.67 times larger than those of the crack gauge. Additionally, a crack spacing measurement technique was proposed. The measured spacing was similar to that obtained through visual observation. However, when the distance between two cracks was small, the accuracy of the proposed technique decreased. Notwithstanding, the proposed technique was efficient when numerous images were used. Finally, the relationship between the maximum shear force and concrete damage was investigated. The crack width was more affected by the rebar ratio than by the rebar spacing. By contrast, the crack spacing was affected more by the rebar spacing than by the rebar ratio. Additionally, the crack spacing significantly affected the maximum shear force. Therefore, rebar spacing should be considered in shear strength estimation. These results will help in damage assessment and developing a new shear strength equation for RC structural and non-structural walls.

Keywords

• Crack width
• Crack spacing
• Deep learning
• Damage assessment
• Image analysis