Abstract

The ability of an ultrasonic testing method to distinguish flaws in close proximity relative to the wavelength is limited by the theoretical resolution limit. Ultimately the reflections become superposed in the same wave packet. In many ultrasonic testing scenarios, the maximum useable frequency is limited by attenuation, and it may therefore become difficult to detect flaws in close proximity and impossible to increase the frequency. The purpose of this work is to use a convolutional neural network in order to separate and identify the time of arrival overlapping echoes. The machine learning algorithm was trained using finite element simulations and was then tested on experimental measurements. The convolutional neural network was able to distinguish shallow flat bottom hole in an aluminum block with a depth corresponding to only 0.5?.

How to Cite:

Chapon, A. . & Bélanger, P. ., (2019) “Detection of flaws in close proximity using convolutional neural networks”, Review of Progress in Quantitative Nondestructive Evaluation .