Detection of multidamage to reinforced concrete using support vector machine‐based clustering from digital images

Abstract

Damage to reinforced concrete (RC) can be hazardous to structures and threaten users' health and work efficiency; it is an indicator of the deterioration level of a facility. Developing automatic detection methods for facility damage has always been a goal that competent authorities of facility management strive to achieve, and such methods can provide engineers with references when evaluating facility damage and maintenance. This study introduced a hybrid machine learning (HML) that combined cluster analyses and a support vector machine (SVM) to create SVM-based clustering. The method was proposed to detect four types of RC damage: rebar exposure, spalling, efflorescence, and cracking. First, grouping was implemented according to damage features in images using cluster analysis, and the results were used as the standard for SVM classification. Second, the detection efficacy of the two types of machine learning, namely, SVM-based clustering and SVM, were compared. The functions of the suggested HML were evaluated with the six indicators based on image data with human annotations. In the classification experiment of three models, HML was superior to the single supervised machine learning. SVM-based clustering had the highest detection efficacy for cracks (accuracy = 99.3%), followed by the detection of the ternary damage of rebar exposure, spalling, and efflorescence (accuracy = 94.9%). HML can quickly extract damage features from digital images and classify multiple types of RC damage in a certain time. The proposed method is effective for facilitating the damage evaluation of facilities and is useful for facility competent authorities and user safety.