Abstract
This paper presents a Convolutional Neural Network (CNN) based strategy targeting regression and classification tasks based on post-processed Guided Wave Imaging (GWI) images issued from a Structural Health Monitoring (SHM) configuration. The studied use-case is a network of piezo-electric sensors permanently integrated on a structure to inspect. A GWI process is applied to the propagated guided wavepackets between every pair of sensor to generate a picture representing the health of the inspected region. If such image provides to an trained operator both detection and localization by a quick look, automated detection and diagnosis is a challenge, especially if the collected data are noisy. Moreover, GWI does not directly provide information regarding the defect size. The paper presents the use of a CNN to automate the detection, localization and sizing of a defect. More specifically, to train the CNN, data are generated using a numerical finite element solver, then the theoretical performances of the process are quantified on numerical data. Finally, the model built by the CNN is used to conduct the inversion on real experimental data and excellent detection, localization and sizing are obtained.
How to Cite:
Kulakovskyi, A., Miorelli, R., Mesnil, O., Chapuis, B. & D’Almeida, O., (2019) “Convolutional neural network for automated diagnostic from guided wave imaging in a structural health monitoring context”, Review of Progress in Quantitative Nondestructive Evaluation .
Downloads:
Download PDF