Date of Award
2024
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Computer Science
Committee Chair
Huaming Zhang
Committee Member
Gang Wang
Committee Member
Deepak Acharya
Research Advisor
Huaming Zhang
Subject(s)
Deep learning (Machine learning), Lamb waves, Cantilevers--Nondestructive testing, Structural health monitoring
Abstract
Guided Lamb waves offer a promising solution for the early detection of internal damages in structures due to their high sensitivity to small damages. However, noise can adversely impact the development of data-driven models for damage detection, leading to inaccurate monitoring systems. This thesis explores deep learning techniques to robustly predict the location and severity of damage in cantilevered beams using noisy guided wave responses. Initially, Multi-Layer Perceptron (MLP) is trained with frequency domain features to achieve robust performance against noisy data. Further performance improvement is achieved using end-to-end learning models, which include autoencoder and one-dimensional Convolutional Neural Network (CNN). The autoencoder demonstrates better dimensionality reduction compared to frequency-based feature extraction while also exhibiting better performance. The one-dimensional CNN model outperforms other techniques, achieving an R2 score of 0.9908 in the highest noise level settings. These results facilitate the development of robust structural health monitoring using deep learning techniques.
Recommended Citation
Gopali, Sameer, "Deep learning approach for robust structural health monitoring using guided Lamb wave responses" (2024). Theses. 665.
https://louis.uah.edu/uah-theses/665