Deep learning based crack damage detection technique for thin plate structures using guided lamb wave signals

Abstract

In this paper, convolutional neural network (CNN) is employed for crack damage detection in thin aluminum plates. Identifying a successful damage feature is a critical step in computationally efficient damage detection and characterization process; however, such feature identification process is generally time-consuming and often difficult to execute. The CNN model is a deep learning model that can be trained to represent high dimensional data for which traditional mathematical model is ill to describe. This study first formalizes the nondestructive evaluation (NDE) problem of notch-type crack damage detection in thin metal plates into an image classification problem in the machine learning domain, which is then solved with a deep CNN model trained using Lamb wave data converted images. Analytical formulas have been derived for generating Lamb wave signals subsequently used for training and validation test of the proposed crack damage detection technique for thin metal plates. Experimental test data from aluminum plate with simulated notch damage has also been used to independently validate the CNN model. The study results show its potential as a promising NDE tool for crack damage detection in thin plate structures.