Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
Machine Learning Approach for Structural Health Monitoring Using Decision Trees and XGBoost
Submission Author:
Heitor Rosa , SP
Co-Authors:
Heitor Rosa, Eloi Figueiredo, Samuel da Silva
Presenter: Heitor Rosa
doi://10.26678/ABCM.COBEM2023.COB2023-0216
Abstract
Structural Health Monitoring (SHM) is an essential strategy for detecting damage in aerospace, civil, or mechanical engineering infrastructure. This paper proposes a machine learning approach using XGBoost, a Decision Tree based supervised non-parametric algorithm for regression and classification tasks. An important aspect of the algorithm is its capability to extract Information Gain as a feature to accomplish the classification task, which we used to evaluate the sensors’ damage sensitivity. Our study was conducted on the Z24 bridge benchmark, considering two types of health conditions, a healthy and a damaged one. To measure the bridge response, eight sensors were attaches in different locations of the bridge. Feature Extraction was carried out considering statistical Time Domain Features and Natural Frequencies. We achieved satisfactory classification results with 0.943 Average Precision Score. A detailed feature importance analysis identifies specific sensor attributes, such as the second Natural Frequency and certain sensor tests, are significantly sensitive to damage detection. These findings open possibilities of selecting fundamental features to ensure faster and more robust data processing for damage classification
Keywords
Structural Health Monitoring (SHM), Decision Tree Learning, XGBoost, Feature Importance, Information Gain
