تجزیه و تحلیل حساسیت از انتشار ترک در ساختار لایه بندی شده قیر آسفالت با استفاده از یک شبکه های عصبی مصنوعی یکپارچه سازی سیستم ترکیبی و روش اجزاء محدود

The paper presents results of sensitivity analysis to crack propagation of a pavement bituminous layered structure using the Finite Element Method (FEM) and Artificial Neural Networks (ANN). The performed preliminary study determined stresses and displacements in the pavement layer system. The pavement structure consisting of asphalt layers, a base layer, subbase and subgrade layers was analyzed as a 2D finite element model using the ABAQUS computer software. The second method, i.e. the extend Finite Element Method was applied, to simulate cracking process of the bituminous layer of a road surface. The pavement model was subjected to static load. Both linear and non-linear material properties of the pavement layers were considered to discuss crack propagation sensitivity in the pavement layers. A hybrid system integrating Artificial Neural Networks (ANN) and FEM was considered to model the material of asphalt layers in flexible pavements. The tests for Radial Basis Function (RBF) and Multi-Layer Perceptron (MLP) networks were carried out. In the formulated model the ANN requires inputs such as: layers thickness, load value and the Young’s moduli of each layer creating the pavement. The data for the ANN were obtained from Finite Element Method analysis. The aim of the network learning process as non-destructive testing was to evaluate the pavement material behavior and estimation of the crack propagation sensitivity. The main conclusion is that cracking considerably increases with a decrease in the thickness of bituminous layer B2. The thickness of the asphalt layer B1 has much less considerable effect on the cracking of the subgrade layer.

Asphalt pavement cracking poses a serious problem for the development of road infrastructure. The phenomenon concerns worn-out roads as well as relatively new roads that are used in accordance with contractor specifications. The occurrence of cracking in multilayered road pavements results from numerous factors, such as strength parameters of particular layers, their thickness, load and production technology applied. The present study is an attempt at the application of an FEM–ANN system to predict cracking sensitivity in a subgrade layer (Fig. 1). The system is based on MLP and RBF networks. The applied input variables included asphalt pavement layer parameters and different load modes. The expected output network response would predict a crack occurrence. The Abaqus-calculated numerical analysis results were used as learning data.

The presented test method is an attempt at examining the application of a hybrid ANN-FEM system to evaluate road pavement behavior. The conducted preliminary analyses confirm the possibility of its effective application to investigate cracking in road pavements. Based on these studies, it seems that the presented system could also be applied to evaluate other phenomena occurring in similar layered materials or rheological behaviors. It should be emphasized that the method applied offers possibilities of estimating the effect of particular parameters of a road pavement on cracking process sensitivity. Cracking considerably increases with a decrease in the thickness of bituminous layer B2. The thickness of layer B1 has much less considerable effect on the cracking of the subgrade layer. In further studies, the effect of other road pavement parameters on both cracking and other strength parameters of an already constructed road pavement will be analyzed. The temperature effect on the pavement cracking will be also investigated.