A near realtime decisiontreebased algorithm for identifying the

A near real-time decision-tree-based algorithm for identifying the extent of structural damage in braced-frame buildings Mojtaba Salkhordeh, Masoud Mirtaheri, Siavash Soroushian Abstract Rapid health assessment of essential buildings such as hospitals, fire stations, and large residential complexes is crucial after damaging earthquakes. The use of advanced technologies such as wireless sensors, learning algorithms, and signal processing methods became more attractive in such fast applications due to their higher reliabilities and efficiencies compared to conventional visual inspection methods. This study presents a robust post-earthquake damage detection framework for predicting the extent and location of damage occurrence in the braced frame structures after an earthquake. Keywords: Rapid assessment, Health monitoring, earthquake, damage detection Introduction Recent experiences from previous earthquakes illustrate that severity of damage can significantly vary in steel braced-frame buildings [1]. Some of these damages could be either obvious due to the large deformation of the brace elements or be hidden because of small deflection in the brace components. For instance, in 1994 Northridge earthquake, the initial assessment of a four-story braced frame building indicated that the structure had not endured much damage. However, after removing the nonstructural components, the site engineers reported that the brace elements were fractured during the earthquake [2]. the present study proposed a data-driven structural health monitoring framework to identify the performance level of braced frame buildings after an earthquake. To determine the damage occurrence, damageextent, and damage location of the building under the earthquake, acceptance criteria recommended by ASCE-41 [3] are employed. This paper first describes the numerical methods used as the case studies in this paper. Then, the proposed features as well as their derivation steps from raw acceleration data are explained. In subsequent, the pattern recognition method along with the optimization procedure used to classify the dataset are presented. Case studies (a) One-story one-bay chevron braced frame (b) Three-story X-braced frame A 2 D single story braced frame is used to evaluate the ability of proposed framework in detecting the damage induced to the brace elements during an earthquake. This structure has one bay with the length of 30 ft and one story with a height of 15 ft. The second building model is a three-story steel-frame structure designed based on the modern building code. Special concentric braced frame of the building was selected as the primary lateral resisting system of this building. Figure 2 illustrates detailed information about the elements of this 2 D frame. Figure 1: A general view of the one-story one-bay SCBF structure. Figure 2: A general view of the three-story SCBF structure. Implementation The proposed framework for the purpose of structural damage detection in braced frame buildings is shown in Figure 3. Figure 4: Result of severity detection models in the ideal condition, a) Confusion matrix; b) ROC curve (b) Three-story X-braced frame Results Figure 5 shows the confusion matrix of the test samples as well as a comparative ROC curve between different possible outcomes of the model. The accuracy of the cross validation process is equal to 96. 5%. It is clear that the model has a high capability in diagnosing the braces that experienced damage during an earthquake. Also, all the possible outcomes of the model has a minimum ROC curve near to the upper left corner of the plot and the AUC value is equal to 0. 97. This section discusses the results obtained from classification procedure. In order to present the result in a perceptible method, confusion matrix and receiver operating characteristic curve (ROC) are illustrated in addition to the cross validation accuracy for each of the classification procedures. A confusion matrix is a specific table layout where each row of the matrix represents the instances in an actual class while each column represents the instance in an predicted class (or vice versa). Moreover, a ROC curve is defined by plotting the true positive rate (TPR) versus the false positive rate (FPR) in different thresholds. Figure 5: Result of location detection model in the ideal condition, a) Confusion matrix; b) ROC curve (a) One-story one-bay chevron braced frame Figure 4 shows the performance of model on detection of the extent of damage when the structural response does not contain noise. The accuracy of this model is equal to 99. 6%, which indicates performance level of the model. Moreover, the ROC curves of this model for different damage intensities are shown in figure 4 -b, where the AUC values of all possible decisions are approximately equal to 1, which indicates high accuracy of model in an ideal condition. References 1. 2. 3. Zhao B, Taucer F, Rossetto T. Field investigation on the performance of building structures during the 12 May 2008 Wenchuan earthquake in China. Engineering Structures 2009; 31(8): 1707– 1723. Watanabe E, Sugiura K, Nagata K, Kitane Y. Performances and damages to steel structures during the 1995 Hyogoken- Nanbu earthquake. Engineering Structures 1998; 20(4 -6): 282– 290. ASCE. Seismic Evaluation and Retrofit of Existing Buildings (ASCE/SEI 41 -17). In: American Society of Civil Engineers. ; 2017.