Computer Vision Based Inspection Approach to Predict Damage State and Load Level for RC Members

Abstract

This paper focuses on a computer vision based inspection approach capable of relating surface damage observations to quantitative damage and load levels in structures. The approach is based on image processing and machine learning techniques which are used to build predictive models capable of estimating internal loads (i.e., shear and moment) and damage states in RC beams and slabs derived from surface crack pattern images. The presented predictive models have been developed and tested using image data sets obtained from 17 different earlier published studies, which provided about 900 crack pattern images captured from 130 RC beams and slabs with and without web reinforcement across a range of load and damage levels. These earlier studies focused on investigating different phenomena and parameters such as size effect, aggregate size, concrete strength, longitudinal and transverse steel ratio on shear strength of reinforced concrete beams. The collected data sets were divided into two major categories: (i) beams with web reinforcement; and (ii) beams without web reinforcement (i.e., beams intended to be shear-critical). Working with these existing image data sets, various textural and geometric attributes of surface crack patterns have been defined and evaluated with respect to their effectiveness in building useful regression predictive models. Rela-tively simple crack representations have been used, consistent with the varying nature of the images available in the earlier studies, but also with an eye toward potential field ap-plications in which image capture and segmentation quality could be limited. In general terms, the results support the potential utility of image-based estimations of load history and damage states of the components, with future applications in field inspection and post-disaster assessments