Impact Detection on Composite Plates Based on Convolution Neural Network

Abstract

This paper presents a novel Convolutional Neural Network (CNN) based metamodel for impact detection and characterization for a Structural Health Monitoring (SHM) application. The signals recorded by PZT sensors during various impact events on a composite plate is used as inputs to CNN to detect and locate impact events. The input of the metamodel consists of 2D images, constructed from the signals recorded from a network of sensors. The developed meta-model was then developed and tested on a composite plate. The results show that the CNN-based metamodel is capable of detecting impacts with more than 98% accuracy. In addition, the network was capable of detecting impacts in the other regions of the panel, which was not trained with but had similar geometric configuration. The accuracy in this case was also above 98%, showing the scalability of this method for large complex structures of repeating zones such as composite stiffened panel.

Cross-structures Deep Transfer Learning through Kantorovich potentials for Lamb Waves based Structural Health Monitoring.

“…Deep Learning (DL) great successes in image processing [16] or sound analysis [17] encourage its use to achieve the four SHM tasks mentioned previously. The damage detection based on DL has been successfully proposed by many papers [18][19][20][21][22][23][24][25][26]. Damage classification is more difficult because of the large diversity of damages that can be encountered in practice in composite structures, however anomaly classification was achieved successfully in [19,[24][25][26][27][28].…”

“…Most of the previous papers trains their models with experimental data obtained on simple case studies such as plates [20,21,26,27,[30][31][32], gearboxes [24,25], or storey structures [18,28,33]. Those approaches are trained on large dataset impossible to acquire practically in an industrial context because of their cost and the lack of information on the damage.…”

“…Numerical dataset are easier to generate [19,23,29] but there is still an existing gap between the numerical models and the experimental reality that prevents their use in practice as the simuated numerical models do not exactly correspond to the actual reality. As input data format, some papers use the temporal signals directly [20,30], others transform the data in 2D images (for example using time-frequency transforms) [26,34], and others compute specific SHM features such as Damage Indexes (DIs) [21], Time Varying Damage Indexes [32] or localization figures [31]. The architecture of the DL networks strongly influence the accuracy of the model.…”

“…It must be adapted to the data format and the dataset size. Most of the papers uses Convolutional Neural Networks (CNNs) [18][19][20][21][22][23][24][25][26][27][29][30][31] but other types of architecture present good results in [29] or [18] . This brief overview shows that despite the difficulties inherent to data acquisition in LWSHM, CNN and others DL architectures obtain interesting results and deserve to be deployed to solve the SHM tasks.…”

Acoustic Emission Source Localization Methodology for Composite Materials based on Artificial Intelligence

“…Nowadays, various features other than arrival time are used to localize the AE sources. Approaches based on features form the AE signal can be divided into features from time domain [8,9], and features from frequency domain [10,11]. Time-domain feature-based approaches are possible to reflect the time continuity of the signal.…”