State of the Art and a New Methodology Based on Multi-agent Fuzzy System for Concrete Crack Detection and Type Classification

• Published in: Archives of computational methods in engineering Vol. 28; no. 4; pp. 2509 - 2542
• Main Authors: Payab, Mahsa, Khanzadi, Mostafa
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.06.2021; Springer Nature B.V
• Subjects: Classification; Concrete; Engineering; Fuzzy systems; Image analysis; Image classification; Inspection; Intelligent agents; Mathematical and Computational Engineering; Modules; Multiagent systems; Original Paper; Particle swarm optimization
• ISSN: 1134-3060; 1886-1784
• DOI: 10.1007/s11831-020-09465-7

Routine inspection and automatic distress detection and classification are critical for civil infrastructures such as bridges. The main subject of this paper is to provide brief review a multi-agent fuzzy system (MAFs) based on image analysis for the detection and classification of, various types of cracking in concrete elements. For this purpose, the combination of fuzzy inference systems has been developed as an autonomous intelligent agent in the center of a multi-agent system (MAS) which communicate, and exchange information with each other. This work is presented in two main sections, (1) the crack recognition system and (2) the type detection system that both of them designed based on MAS fuzzy systems. The first module is a binary classification agent that made of 5 inputs, one output, 11 rules. This agent receives an image and classified it into two groups: crack and non-crack. The second module, which is made of 8 inputs, three output, 20 rules, and used for type classification (individual, pattern, and random). The input of this module is the images that were classified using the first module into the crack group. Particle swarm optimization has been used to find the optimal values of membership functions coefficients. The optimized results of multi-agent modules are compared with other methods. After an experimental characterization and optimization of modules, the MAFs are tested on various concrete distresses. The results show a high potential of MAFs for crack detection and classification. Analysis of the results showed that accuracy of detection, classification can be improved by 4% and 5%, respectively, with MAFs. This method enhances the speed, accuracy, and has higher precision, which indicates the satisfaction and reliability of the MAFs. Also, besides this system has high computational power to detect and classify complex cracking patterns in bridge components.