Mustapha, Samir; American University of Beirut Faculty of Engineering and Architecture; Lebanon
Fakih, M.A.; American University of Beirut; Lebanon
Islam, M.S.; Hamad Bin Khalifa University; Qatar
Khoa, N.L.D.; CSIRO; Australia
Mustapha, S.A.; American University of Beirut; Lebanon
The study focuses on the development of a robust algorithm that can assist in the detection and classification of cracking in steel reinforced concrete structures using vibration data. The study was performed on a jack arch that emulates one of the major components on the Sydney Harbour Bridge (Figure 1). Data were collected using 10 single axis accelerometers for five different states of the structure covering the intact case and four damage cases (Figure 2).
Different time domain and frequency domain features were extracted based on several approaches including statistical analysis, symbolic dynamics (SD), and frequency response function (FRF). The study proposes different methods for multi-source data fusion, as well as fusion of the extracted features depending on their nature. A series of multi-class ensemble classifier algorithms (Decision Tree, Random Forest, Extra Trees, Ada Boost, and Gradient Boosting) were employed for the assessment of the five different health states of the structure, and the models were evaluated according to their accuracy.
The accuracy of the classification was significantly improved with the use of the Extra Trees classifier when compared to the other investigated ensemble classifiers. While low accuracies (below 57%) were obtained when using the SD-based features with all the employed classifiers, statistical-based features performed much better leading to accuracies ranging between 85 and 97%. Moreover, the use of only two features from the FRF data led to very high accuracies of above 98% except for the Ada Boost classifier (Table 1).