97 / 2021-03-30 13:55:02

Machine learning applied to rotary hammering sound test with the optimum hammering conditions

Concrete structure,Defect detection,Rotary hammering,Self-organizing map,Hammering condition

A hammering sound test is widely used for inspection of internal defects in concrete structures. In this paper, we use a rotary hammer to improve the inspection efficiency and the hammering force variation. The rotary hammering device is comprised of a rotary hammering part and a spring for applying a pressing load to the part. The hammering sounds are measured with a microphone moving with the rotary hammering device. We propose a method using the self-organizing map (SOM) for automatically detecting the defective parts of concrete structures. We examine the hammering conditions (the pressing force and the spring constant of the rotary hammering device) when the difference between the hammering sounds at the defective parts and healthy parts is remarkable. The hammering sound test experiments of concrete specimens with artificial defects are conducted. Then, the SOM training is preformed where the frequency spectra of hammering sounds measured at the impact locations are used as input data. The results show that there are suitable conditions for the pressing force and the spring constant that maximize the ratio of the average value of the sound pressures at the defective parts to the average value at the healthy parts. The SOM results partition the impact locations into several groups. By removing the impact locations belonging to the group with the smallest average overall value, the impact locations corresponding to the defective parts are extracted. The extracted impact locations identify the exact locations and widths of the defects.