Application of Artificial Neural Networks (ANN) to Predict Geomechanical Properties of Asmari Limestones
الموضوعات :Mahdi Razifard 1 , Mashallah Khamechiyan 2 , Mohammad Reza Amin‐Naseri 3
1 - Department of Engineering Geologyt,Tarbiat Modares University, P.O. Box 14115-175, Tehran, Iran
2 - Department of Engineering Geology, Faculty of Basic Sciences, Tarbiat Modares University, Tehran, Iran
3 - Department of Socio-economic Systems, School of Industrial and Systems Engineering, Tarbiat Modares University, Tehran, Iran
الکلمات المفتاحية: Asmari Formation, internal friction angle, Artificial Neural networks (AAN), Feedforward Backpropagation, Multivariate Linear Regression (MLR), Uniaxial Compressive Strength (USC), Modulus of Elasticity (E), Cohesive Strength (C),
ملخص المقالة :
A number of common laboratory rock mechanics tests are carried out in all geotechnical projects such as dams, to determine parameters such as porosity, density, water absorption, sonic velocity, Brazilian tensile strength, uniaxial compressive strength, and triaxial compressive strength. In this paper, data obtained from two dams in Asmari Formation including Khersan 1 and Karun 4 - both located in Chahar-Mahal Va Bakhtiari Province, Iran - have been subjected to a series of statistical analyses. Then, using Multivariate Linear Regression (MLR) and Artificial Neural Networks values of UCS, E, C, and φ were predicted using the input parameters including depth, compression ultrasonic velocity, porosity, density, and Brazilian tensile strength. The designed ANN in this research was a feedforward backpropagation network which is powerful tool to solve prediction problems. Designed network had two hidden layer (hidden layer 1: 18 neurons and hidden layer 2: 20 neurons). Via comparing designed MLR and ANN models, it was revealed that ANNs (R2 UCS= 0.91, R2 = 0.87, R2 =0.78, R2 EC phi = 0.61) are more efficient than MLR models (R2 UCS= 0.69, R = 0.69, R = 0.66, and R2 22 EC phi = 0.50) in predicting strength and shear parameters of the intact rock. Also, to enhance the credibility of this study, some extra tests were carried out to evaluate the efficiency of network designed for prediction of strength parameters. The results obtained from this network were as: R2 UCS= 0.85, R2E = 0.81.
