Predicting the Seismic Response of Embankments and Earth Dams with Artificial Neural Network

The artificial neural network (ANN), one of the most powerful tools of artificial intelligence, was used to control the seismic responses of the cross section, and to predict earth dam seismic responses rather than utilizing dynamic time-based analyses. In terms of the artificial neural network, apart from the aforementioned ANN-based applications in various engineering problems, an increasing number of articles have been published over the last decade where the efficient implementation of ANNs in geotechnical earthquake engineering is presented. Most of these studies focus on liquefaction potential under seismic excitations, which is an extremely computationally intensive issue and therefore suitable for ANNs. In some of the studies in this field, the applicability of ANNs in soil dynamic analysis was examined. In addition to the generation of spectrum compatible accelerograms, prediction of earthquake parameters, wave propagation approximations, estimation of peak ground acceleration using microtremors, and the detection of earthquake electric field patterns. Not much has been done into predicting the seismic responses of embankments and earth dams using artificial neural networks tools. The first study on the domain was carried out by Tsompanakis et al. [1] in 2009 in which using the artificial neural network, a seismic response (maximum horizontal acceleration) of homogeneous and symmetric embankment in different locations was predicted. Although being a case study on a simple embankment, earthquake records (accelerograms) were used, which contained a variety of frequency concepts, simplicity or complexity, and various PGA levels, in addition to a general neural network that could be used in all PGA levels the seismic response was not accurately predicted. The objective of the present study was to investigate the ability of the artificial neural network to reach the dynamic response of the embankment and earth dams under seismic loads and different levels of intensity. In other words, it was planned to reduce the heavy cost of calculating the applicable issue in seismic geotechnical engineering. To fulfill the aim, a comprehensive study was carried out and various parameters were evaluated. According to the results, the artificial neural network could reach a good approximation of the seismic response of embankments and earth dams. Although comparing the results of artificial neural networks segmented from generic artificial neural network showed the accuracy of the estimated artificial neural network structured for each PGA level, especially in the linear region, the study was successful in eliminating the defects of previous research, making a perfect arrangement for a generic artificial neural network to be used for this type of neural network to achieve a favorable estimation of all PGA levels. Finally, it can be concluded that considering the complexities of the problem and the performance of the artificial neural networks in solving them, as well as the history of successful performance reports on this tool in other similar applications, the application of Artificial Neural Networks proved useful for estimating and predicting seismic responses of embankments and earth dams. Using artificial neural networks to meet this purpose would reduce the cost of calculating the seismic response assessment of embankments and earth dams. Therefore, applying this tool in the complicated field of seismic geotechnical engineering may be highly promising. Overall, future developments in this research field and promotions in artificial neural network training in other situations by changing the dimensions and properties of materials and load conditions would realize this idea and create a more general neural network for this application.