ABSTRACT
A decrease in skilled infrastructure inspectors and the cost of maintenance are big issues in Japan. Thus, an effective, automated inspection system is much needed. In this study, local vibration testing, a non-destructive testing method, coupled with machine learning, was utilized. Local vibration testing has the benefit of detecting damage that exists in deep layers because the waves travel through-thickness of concrete members. Using a portable vibrator, local vibration testing was conducted on reinforced concrete (RC) beams to detect the location of damages within. The RC beam specimens had two types of damages: flexural cracks and shear cracks. Results showed that the damaged conditions’ frequency responses varied from that of the intact condition. Additionally, high abnormality degrees were detected around the crack region. This implies that the location of damages can be identified from the data produced by local vibration testing. Though this method alone yields results, introducing automation would further increase the diagnostic efficiency. Therefore, machine learning was introduced. A non-supervised learning method was applied. Using this method, the damages were identified with a very high accuracy rate. This indicates that the combination of local vibration testing and non-supervised learning can produce an effective, automated inspection system for the integrity diagnosis of RC beams.
