Analysis of impact position based on deep learning CNN algorithm

Recently, various studies have attempted to detect the locations of impacts on metal surfaces. The most generalized method uses the time difference between the generation of the stress wave during impact and its arrival at the surface. However, accurate measurement of the arrival time of a stress wave is fraught with difficulties. To address this issue, in this paper, we attempt to identify the signal characteristics of a stress wave corresponding to the impact location by learning the image of the stress wave signal, as measured by a Piezoelectric sensor, via deep learning without directly extracting information about the wave's arrival time. To improve the accuracy of the measurement used in deep learning, we appropriately optimize the epoch variable and present a deep learning-based algorithm to measure the location of the impact on an aluminum plate of dimensions 1 m × 1 m. As a result, the proposed image-based NDT is successfully verified.