Monitoring the health of structures and detecting damage has been an
interdisciplinary field of interest across various engineering domains. Nowadays,
structural health monitoring systems aim to enhance the performance of structures,
ensuring human safety, reducing the need for frequent inspections, and cutting
maintenance costs. Early and accurate damage detection has always been a primary
goal of structural health monitoring programs. In this regard, significant efforts
have been dedicated to vibration-based methods, utilizing structural vibration
responses to assess their conditions and identify damages.
In this study, two methods are employed for structural damage detection. Initially,
a subspace-based approach is used for damage detection in the structure, followed
by a combination of one-dimensional Convolutional Neural Networks (CNN) and
Dynamic Fuzzy Neural Networks (DFNN) for more accurate damage detection and
localization. Among various methods for modal parameter identification, subspacebased techniques are considered as one of the most favored and effective choices.
These methods exhibit remarkable capabilities in detecting changes in the
structure's behavior and specific characteristics, managing both input and output
data, and accurately extracting modal parameters.
Traditional systems for structural damage detection typically consist of two stages:
feature extraction and classification. The performance of such classical systems
heavily relies on the selection of features and the dependent classifier. While
manually crafted features might be unsuitable for specific structures, they often
demand high computational power, leading to unstable performance in
classification. Addressing these challenges, this research presents a data-driven
solution using deep learning techniques to tackle structural damage detection
problems without the need for human-defined physical features. In this approach,
abstract features are extracted from the data, embedding profo