Effect of sampling rates on the accuracy of acoustic-laser technique in defect detection and upsampling using machine learning

As a non-destructive testing (NDT) method, the acoustic-laser technique has demonstrated its effectiveness in defect detection of composite systems. The technique is particularly useful for the detection of near-surface defects in fiber reinforced polymer bonded concrete by vibrating the material with an acoustic excitation and measuring the vibration signals with a laser beam. However, same as the other vibration-based measurement methods, the accuracy of acoustic-laser technique is sensitive towards sampling rate during the measurements. The sampling rate of acoustic-laser measurement adopted in previous studies is as high as 50000 Hz to assure the accuracy of measurement. However, such high sampling rate cannot always be guaranteed due to the limitation of data acquisition system, or any missing data generated during the measurement. In this study, the effect of sampling rates on the accuracy of acoustic-laser technique is investigated through the experimental study. Six sampling rates, i.e. 10 Hz, 100 Hz, 1000 Hz, 10000 Hz, 25000 Hz, and 50000 Hz, are adopted to measure the FRP bonded system with an interfacial defect, in order to study the relationship between the sampling rate and measurement accuracy. Moreover, an upsampling method using machine learning is proposed in this study, in order to reconstruct the missing data to the target sampling rate, so that the accuracy of the detection can be improved from low sampling rate measurement with missing data.