Wave-Based Neural Network with Attention Mechanism for Damage Localization in Materials

Cracks are omnipresent in materials and lead to billions of dollars in losses annually due to catastrophic and spectacular failures. Nondestructive wave-based methods are used to identify cracks, but these methods are cumbersome and require experts, leading to limited investigation. This research pr...

Full description

Saved in:
Bibliographic Details
Published in:Proceedings (IEEE International Conference on Emerging Technologies and Factory Automation) pp. 122 - 129
Main Authors: Moreh, Fatahlla, Hasan, Yusuf, Rizvi, Zarghaam Haider, Wuttke, Frank, Tomforde, Sven
Format: Conference Proceeding
Language:English
Published: IEEE 18.12.2024
Subjects:
Accuracy
attention mechanism
Attention mechanisms
Complexity theory
Data models
Deep learning
Dense-Structure
Location awareness
Neural networks
Numerical models
Optimization
Predictive models
seismic wave
Training
ISSN:1946-0759
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Cracks are omnipresent in materials and lead to billions of dollars in losses annually due to catastrophic and spectacular failures. Nondestructive wave-based methods are used to identify cracks, but these methods are cumbersome and require experts, leading to limited investigation. This research propose MicroCracksAttNet50E model that leverages numerical data to detect and localize damage in materials and structures, with a particular focus on microcracks that are imperceptible to the naked eye or conventional imaging methods but have the potential to develop into larger, hazardous fissures. The paper also includes a comparative analysis between the current study and the previous work, specifically evaluating the model that performed best in the prior paper (1D-DenseNet-Resize&Conv). Despite having approximately eight times fewer layers and over 200,000 fewer trainable parameters than DENSE variants, MicroCracksAttNet50E achieves similar or even better performance, with an accuracy of 0.860 and a precision of 0.881, compared to the best-performing DENSE model with an accuracy of 0.836 and a precision of 0.875. This improvement primarily highlights the effectiveness of the attention mechanism in MicroCracksAttNet50E, which focuses on critical areas to detect smaller cracks more accurately.
ISSN:1946-0759
DOI:10.1109/ICMLA61862.2024.00023