Defect Detection in Concrete Structures Based on Characteristics of Hammer Reaction Force and Apparent Stiffness of Concrete

Concrete structures play a vital role in civil engineering. However, their deterioration poses a significant safety risk, making regular inspection essential. The impact method detects defects by analyzing characteristic reaction forces generated when striking areas with internal cavities, offering the advantage of being unaffected by acoustic noise and not requiring direct sensor contact. Traditional approaches, relying solely on reaction force signals, have struggled with low defect detection accuracy due to the similarity of signals between healthy and defective concrete and limited data quantity. To address this challenge, we propose a novel method that enhances defect discrimination accuracy by integrating statistical processing and physical property analysis within a machine learning framework designed for force signal characteristics. Our method employs wavelet transformation to convert short-duration force signals into high-resolution time-frequency features, capturing their non-stationary behavior in detail. For dimensionality reduction, we use Uniform Manifold Approximation and Projection to accurately embed data clusters near decision boundaries in a low-dimensional space. The embedded data is then clustered using Fuzzy c-means, and defect cluster identification is automated based on the apparent stiffness of the concrete. Experimental validation through field and laboratory tests confirmed the effectiveness of our method, demonstrating a significant improvement in defect detection accuracy. By advancing the precision and automation of the impact method, this study contributes a valuable tool for enhancing the safety and maintenance of concrete structures.