ارائه رابطه‌اي براي تخمين جابه‌جايي جانبي حاصل از روانگرايي با استفاده از بانك اطلاعاتي ژئوتكنيكي جديد

Developing a Model to Estimate the Liquefaction Induced Lateral Displacement Using a New Geotechnical Database

گسترش جانبي ناشي از روانگرايي يكي از عوامل اصلي به وجود آورنده خسارات عمده در طول زلزله به سازه‌هاي زميني و سازه‌هاي مدفون در خاك مي‌باشد. اين نوع از جابه‌جايي‌هاي دائمي سطح زمين، كه از چند سانتي‌متر تا 10 متر و بيشتر متغير مي‌باشد، خسارات اساسي را به تأسيسات زير بنايي و شالوده‌هاي شمعي سازه‌هاي بزرگ و پايه‌هاي پل‌ها در طول دهه‌هاي اخير تحميل كرده است. به همين علت، يك مدل جديد با توانايي بالا به‌منظور برآورد جابه‌جايي‌هاي جانبي ناشي از روانگرايي با استفاده از روش شبكه‌هاي عصبي نوع GMDH1 در تحقيق حاضر توسعه داده شده است. به اين منظور، يك بانك اطلاعاتي جامع شامل 526 الگو كه در طول 18 زلزله مهم ثبت شده است، به‌منظور مدل‌سازي و تحليل جمع‌آوري گرديد. به دليل اينكه مدل حاضر بر پايه تعداد زيادي از زلزله‌ها و نيز ساخت‌گاه‌هاي مختلف توسعه يافته است، مدل جامع‌تر و قابل‌اعتمادتري از مدل‌هاي قبلي ارائه مي‌كند. همچنين مقايسه عملكرد مدل توسعه داده شده در تحقيق حاضر با نتايج آزمايشگاهي موجود در متن بيانگر صحت پيش‌بيني مقادير توسط مدل جديد مي‌باشد.

The pore water pressure increasement in saturated sands is a consequence of cyclic shear stresses induced by earthquake loads. Following this change, the shear strength of soil rapidly decreases and liquefaction of soil may be occurred. Most types of failure associated with the liquefaction phenomena are: sand boil, flow failure of slopes, ground oscillation, loss of bearing capacity, ground settlement, and lateral spreading. Liquefaction-induced lateral spreading is one of the most important factors of major damage to Ground and underground structures during earthquakes. This type of permanent ground displacement, which has amplitudes ranging from a few centimeters to 10 meters and more, has caused substantial damages to lifelines and pile-foundations of buildings and bridge piers along the past earthquakes. Although the mechanism of soil liquefaction is well recognized, the prediction of liquefaction-induced horizontal displacement is associated with the complexity and difficulty, due to the involvement of multiplex parameters. Several researches have been done to develop techniques for lateral ground displacement prediction. These techniques can be divided into four classes, including simplified analytical, numerical, empirical, and artificial neural network methods. Since the simplified analytical methods consider the shear strength of soil unchanged during an earthquake, these methods may not provide an accurate estimate of lateral displacements caused by liquefaction of soil. Besides, due to the complexities related to the accurate modeling and the difficulties in measuring the in-situ parameters of soil layers, it is obvious that the consideration of some simplifications in numerical methods is required, which may reduce their capabilities. Due to the limitations related to analytical and numerical methods, many researchers developed empirical models based on ground displacement records. Empirical methods detect the relationship between in-situ displacements and various effective parameters by regression method. It is believed that ANN models compared to the conventional regression methods can predict complex problems, such as liquefaction-induced lateral spreading more accurately. In the present study, a new model with the ability to estimate the lateral displacement caused by liquefaction has been developed using the Group Method of Data Handling (GMDH) type neural networks. In this method, complicated relationships are developed according to their efficiency against a series of multi-input single-output data pairs. GMDH algorithms present a tool to find the appropriate relationship between data, recognize the optimal structure of the network, and improvement in accuracy of existing algorithms. In general, the GMDH type neural network includes certain advantages compared to other types of neural networks. In particular, it has the ability to find and select the most suitable input variables from a set of variables. By sorting different solutions, GMDH networks minimize the influence of the user on the structure and results of modeling. The computer automatically finds the optimal structure of the model and the laws acting on the system. In this study, a comprehensive database containing 526 case histories and recorded over 18 major earthquakes was utilized to correlate the liquefaction-induced lateral spreading with the most effective parameters. Since the presented model has been developed based on numerous earthquakes and site conditions, it is more general and reliable than previous models. The obtained results indicate that the GMDH model has the ability to predict the lateral spreading with a high degree of accuracy. In order to validate the new proposed model, the displacements obtained by 28 centrifuge tests were compared with the results of the GA-GMDH model. The comparison showed a high degree of accuracy of the new GA-GMDH model, indicating a good predictive capability of the model, even at small ground displacements. Moreover, comparing the performance of the model developed in the present study with experimental results in literature shows the accuracy of predicted values by the new model.