Probabilistic performance assessment of eccentric braced frames using artificial neural networks combined with correlation latin hypercube sampling

Hysteretic behavior of shear links has a vital role in seismic responses of Eccentric Braced Frames (EBFs). So, their behavioral parameters have been suggested recently in order to predict their hysteresis behavior. For shear links, modeling variables have been defined through seven implicit parameters all of which having specific Coefficient of Variation (COV), and variation in these parameters may affect the structural responses. However, the effect of uncertainties in these parameters on the structural responses has not been yet addressed. In this paper, variations of seismic structural responses due to the variations in the behavioral parameters of shear links have been investigated while trying to maintain a balance between the accuracy of predicted responses and the required computational effort. For this purpose, a 4-story EBF system has been modeled and studied. Incremental Dynamic Analysis (IDA) has been employed in order to attain the seismic responses while the spectral acceleration corresponding to different performance levels (Sa(T1, 5%)) is used as the Intensity Measure (IM). A sensitivity analysis is carried out to rank the modeling parameters by their effect on seismic responses of the system in two performance levels. Then, Correlation Latin Hypercube Sampling (CLHS) method has been used to propagate the uncertainties of important modeling parameters into the structural responses of the frame. In the next step, Function Approximation (FA) with Artificial Neural Networks (ANNs) is used as a practical method to approximate the structural responses with lower computational effort. Screening design has been employed to consider the most optimum input data to train the network. ANNs in conjunction with CLHS provides a very efficient tool to address the effect of uncertainties in structural responses requiring lower computational effort. It has been found that web thickness and yielding strength of shear link play a key role in variation of seismic structural responses for the model studied. Moreover, incorporating modeling uncertainties may shift the median value of the probability distribution for structural responses.