Title: A Neural Network to Detect Damage Intensity and Location

Authors: Christopher M. Basic, Nicholas A. Smith, James E. Steck

DOI: 10.33599/nasampe/s.22.0761

Abstract: Reliable structural health monitoring can inform more in-depth inspection, but it is difficult to get reliable results for real structures and a practical set of sensors. In this work the effectiveness of a neural network using only acceleration is investigated. Inputs for the neural network are generated from a finite element model of a wingbox. Accelerations from discrete locations due to an applied load at controlled frequencies, with cracks of varying length and location are used to train a neural network. This is meant to simulate accelerometers which could be embedded in a structure.

Initial results show an ability to predict crack length and location independently, future work is required to predict both using one unified model. Future work could also be done to optimize the number and placement of sensor nodes (accelerometers).

References: 1. S. Chandrasekaran, Structural Health Monitoring with Application to Offshore Structures (WORLD SCIENTIFIC, 2019). 2. A. Dourado and F. A. C. Viana, ""Physics-Informed Neural Networks for Corrosion-Fatigue Prognosis,"" PHM_CONF 11, (2019). 3. M. T. I. Khan, Structural Health Monitoring by Acoustic Emission Technique (IntechOpen, 2018). 4. C.-M. Chang, T.-K. Lin, and C.-W. Chang, ""Applications of neural network models for structural health monitoring based on derived modal properties,"" Measurement 129, 457–470 (2018). 5. A. Krizhevsky, I. Sutskever, and G. E. Hinton, ""ImageNet classification with deep convolutional neural networks,"" Commun. ACM 60, 84–90 (2017). 6. T. Developers, TensorFlow (Zenodo, 2021). 7. S. Grondel, J. Assaad, C. Delebarre, and E. Moulin, ""Health monitoring of a composite wingbox structure,"" Ultrasonics 42, 819–824 (2004).

Conference: SAMPE 2022

Publication Date: 2022/05/23

SKU: TP22-0000000761

Pages: 9

Price: $18.00