| Title |
Machine Learning Based Damage Prediction of Seismically Vulnerable Unreinforced Masonry Walls |
| Authors |
신동현(Shin, Dong-Hyeon) ; 음영채(Eum, Yeong Chae) ; 이수민(Lee, Su Min) |
| DOI |
https://doi.org/10.5659/JAIK.2025.41.1.263 |
| Keywords |
Seismic vulnerable structure; Unreinforced masonry; Rapidly damage prediction; Machine learning model |
| Abstract |
The seismic responses of vulnerable unreinforced masonry buildings are strongly dependent on the damage or failure modes of unreinforced
masonry walls. The main purpose of this study is to develop machine learning based damage prediction models of seismically vulnerable
unreinforced masonry walls. To do this, the damage or failure modes of unreinforced masonry walls are classified into rocking, diagonal
tension, bed-joint sliding, and toe-crushing. Dataset including geometrical information, material properties, and damage states was established
from the experimental results of reference studies. In order to training machine learning based classification models, deep neural network
(DNN), K-nearest neighbor (KNN), random forest (RF), and support vector machine (SVM) were utilized and input variables were categorized
as two groups. The estimating performance of machine learning models were evaluated by comparing performance measurement indices,
accuracy, precision, recall, F1-score, AUC values which can be calculated from the confusion matrix and ROC curve. From the observation,
DNN model has produced largest performance measurement indices among considered 8 machine learning models and is also presented
reasonable classification performance for diagonal tension and bed-joint sliding modes which can be regarded as critical damage or failure
modes of unreinforced masonry walls. |