A convolutional neural network method for damage location based on raw guided Lamb wave technique

Abstract

This paper studies the convolutional neural network (CNN) for damage localization based on raw Lamb waves. Locating damage is a critical step in structural health monitoring (SHM), while it is generally time-consuming and often difficult to implement. The CNN model is a deep learning model that can be trained to represent the high-dimensional data, which the traditional mathematical model is challenging to describe. Using CNN to detect damage faces two difficulties: the lack of enough damage samples to train the model, and the complex pre-processing. This paper introduces the numerical simulation approach to provides an alternative solution for this problem. Three different frequency signals are used to generate multi-channel images, which are then used as the input of the neural network to predict the damage location. The results indicate that the detection accuracy of the CNN trained with the simulation data reaches 95%.