Modeling Marshall Stability of Light Asphalt Concretes Fabricated Using Expanded Clay Aggregate with Artificial Neural Networks

Nihat MOROVA, Şebnem SARGIN, Serdal TERZİ, Mehmet SALTAN

Department of Electric and Electronical Engineering, Department of Civil Engineering,

Department of Structural Education Suleyman Demirel University,

32260 Isparta/Turkey nmorova@tef.sdu.edu.tr,

sebnemsargin@gmail.com, sterzi@sdu.edu.tr,mehmetsaltan@sdu.edu.tr

Sercan SERİNDepartment of Structural Education

Düzce University, Faculty of Technical Education

81620 Duzce/Turkey sercanserin@duzce.edu.tr

Abstract —In this study, an Artificial Neural Network (ANN) model has been developed to estimate Marshall Stability (MS) of lightweight asphalt concrete containing expanded clay. In the model, amount of bitumen (%), transition speed of ultrasound (µs), unit weight (gr/cm3)were used as inputs and Marshall Stability (kg) was used as output. Developed ANN model results and the experimental results were compared and good relationship was found.

Key Words: Lightweight asphalt concrete, expanded clay, Marshall Stability, Artificial Neural networks, Prediction

I. INTRODUCTION

Natural lightweight aggregate sources can be found in regions characterized by volcanic activity, where porous rocks (known as pumices) are available. Artificial lightweight aggregates (like the expanded clay obtained by thermal treatment of argillaceous materials) are produced in many countries, the raw materials being very common. They may exhibit higher resistance than natural lightweight aggregates, but this favorable result implies a greater productioncost [1].

Clays has formed a mass full of with gas bubbles when it is heated and called “expanded clay”. It has the highest compressive strength among lightweight building materials. They express volume increase during heating process. They produced granules when heating process reached between 1000-1300°C and contain homogeneous, secret and little gaps called porous ceramic has sintered hard shell structures [2, 3].

The use of artificial aggregates such as expanded clay in the production of asphalt concrete makes it possible to reduce both natural aggregate extraction and the use of nonrenewable raw resources, greatly benefiting the environment. Moreover, the expanded clay production process allows nondangerous waste

materials to be reclaimed and thereby avoids the necessity to dispose of them in a landfill or dump—this benefits the environment and also offers economic advantages [4].

Over the past two decades, there has been an increased interest in a new class of computational intelligence systems known as artificial neural networks (ANN). This type of networks (i.e., ANN) have been found to be powerful and versatile computational tools for organizing and correlating information in ways that have proved useful for solving certain types of problems too complex, too poorly understood, or too resource-intensive to tackle using more-traditional computational methods. ANN have been successfully used for many tasks including pattern recognition, function approximation, optimization, forecasting, data retrieval, and automatic control. This circular provides an introduction to ANN and their applications in the design and analysis of geomechanical and pavement systems. As ANN can be a useful complement to more-traditional numerical and statistical methods, their use merits continued investigation.

The main purpose of this paper is to develop an ANN methodology for estimating pavement Marshall Stability without any restrictive assumption by considering amount of bitumen (%), transition speed of ultrasound (µs) and unit weight (gr/cm3) as input variables and the Experimental Marshall stability (kg) results data as an output variable.

II. STABILITY OF FLEXIBLE PAVEMENTS

Stability of asphalt concrete determines the performance of the highway pavement. Low stability in asphalt concrete may lead to various types of distress in asphalt pavements [5, 6]. Cracking, especially fatigue cracking, due to repeated loading has been recognized as an important distress problem in asphalt concrete pavements. The stability of asphalt concrete pavements depends on the stiffness of the

������������������� ������������

mix, bitumen content, softening point of bitumen, viscosity of bitumen, grading of aggregate, construction practice, traffic, and climate conditions [7].

The Marshall mix design method and criteria were originally developed for airfield pavements, but were later also adopted for use in highway pavements. Due to its simplicity, the Marshall method of mix design was the most commonly used mix design method in the U.S. before the introduction of the Superpave design system, and it is still the most commonly-used mix design method throughout the world [8].

The Marshall stability is the maximum load the specimen can withstand before failure when tested in the Marshall stability test. The configuration of the Marshall stability test is close to that of the indirect tensile strength test, except for the confinement of the Marshall specimen imposed by the Marshall testing

head. Thus, the Marshall stability is related to the tensile strength of the asphalt mixture [8].

III. EXPERIMENTAL STUDY

In the experimental part of this study, Crushed Stone Aggregates (CSA) obtained from the proximity of the province of Düzce and the ECA used in this study was supplied by Germany Liapor Company. The study not only the range of 0-2 mm ECA was included in the Hot Mix Asphalt (HMA).Within the framework of this study, first of all, material tests were carried out based on American Codes (ASTM), in order to obtain the physical and mechanical characteristics of the materials to be used in the mixtures. The physical and mechanical characteristics of the aggregates used in the mixtures are given in Table 1.

TABLE I. PHYSICAL AND MECHANICAL CHARACTERISTICS OF CSA TO BE USED IN HMA

Sieve Diameters2,00-4,75 mm 4,75-9,5 mm 9,5-25 mm Codes

Water Absorption % * (3,54) 1,63 0,81 ASTM C 127Los Angeles % * * 23,804 ASTM C 131Fine Material % * (14,51) 1,27 0,45 ASTM C 117Organic Material Clear Clear Clear ASTM C 40Freeze-Thaw % * * 6,69 ASTM C 88Peeling Strength. % * More than %50 More than %50 HTS Part 403 App-AAverage Density (gr/cm3) 2,576 2,642 2,677 ASTM C 127Loose specific gravity (g/cm3) 1,61 1,40 1,41 ASTM C 29Compact specific gravity (g/cm3) 1,91 1,62 1,64 ASTM C 29

*Tests not required according to the technical specifications prepared by Highways Commission (Highways Technical Specifications-HTS)

TABLE II. PHYSICAL CHARACTERISTICS OF ECA TO BE USED IN HMA

Characteristics of ECA (0-2,00 mm)Test Name Average ValuesApparent density (g/cm3) 1.655Loose specific gravity (g/cm3) 0.82Compact specific gravity (g/cm3) 1.04Water absorption (%) 15.25Moisture content (%) 0.01

Figure 1. (a) CSA used in HMA (b) ECA used in HMA

Figure 1(a) and (b) illustrates CSA and ECA which were used as coarse and fine aggregate in the experimental study respectively.

In the experimental part of this study, AC 60/70 asphalt cement (AC), which is produced in Izmir Refinery of TÜPRAŞ (Turkish Petroleum Refineries Corporation), was used. The physical characteristics of the binder are given in Table 3.

TABLE III. BASIC PHYSICAL CHARACTERISTICS OF BITUMEN

Characteristics of BitumenTest Name Average ValuesPenetration (25 ºC) 60-70Flash Point 180ºCFire Point 230 ºCSoftening Point 45,5°C Ductility (5 cm/minute)

>100 cm

Specific Gravity 1,034 gr/cm³

For this aim, first of all, a series of tests were carried out in order to determine the optimum bitumen percentage. Empirical calculation methods were used to determine the pre-optimum bitumen percentages. Then, these values were altered by ±1%, and a total of 13 (%4.5, %5, %5.5, %6, %6.5, %7, %7.5, %8, %8.5, %9, %9.5, %10, %10.5) bitumen percentages were determined. Three samples were prepared for each bitumen percentage value, therefore a total of 39 asphalt samples were prepared and used for Marshall Stability test in order to determine optimum bitumen percentage value for the aggregate sample to be used. So that, the prepared this samples’ Marshall Stability (MS) and Flow rations were determined and then VMA, Vf, Vh, Dt, Dp and ultrasound values were determined too.

IV. ARTIFICIAL NEURAL NETWORK

Artificial neural networks (ANN) are modeling tools able to solve linear and non-linear multivariate regression problems. This methodology does not need the explicit expressions of the physical meaning of the system or process under study and considered to belong to the group of ‘‘black-box’’ models. Such models permit to study the relationship between the input variables and the target(s) or output(s) of the process using a limited number of experimental runs. Moreover, the ANN models can be easily developed by applying an adequate design of experiments [9].

Artificial neural networks (ANNs) are considered as one of the most widely reported data driven techniques in the last couple of decades. The performance of ANN model depends on quality of in-put data and network parameters. However, most of the published researches are exclusively focused on design and implementation of ANN models. The essential theoretical feature of ANNs is the ability to approximate continuous functions by learning from

observed data, under very general analytical conditions [10].

Artificial neural networks (ANN) try to mirror the brain functions in a computerized way by restoring the learning mechanism as the basis of human behavior. ANN can operate like a black box model, which requires no detailed information about the system or equipment. ANN can learn the relationship between input and output based on the training data [11].

V. DEVELOPED ANN MODEL STRUCTURE, PARAMETERS, AND RESULTS

The Artificial neural networks model developed in this research has three neurons (variables) in the input layer and one neuron in the output layer, as illustrated in Fig. 2. One hidden layer with three neurons was used in the architecture because of its minimumpercentage error values for training and testing sets. Some of the architectures with different numbers of neurons were studied here in hidden layers and their correlations with experimental results were investigated, while modeling amount of bitumen (%),transition speed of ultrasound (µs) and unit weight (gr/cm3) were used as inputs and Marshall Stability (kg) was used as an output. In this study, data sets were taken from experimental test results. For training sets, 32 samples (80% of all samples) were selected and the residual data (about 7–20% of all samples) were selected as a test set. The values of the training and test data were normalized between 0 and 1 using Eq. 1.

� � � � (1) FF/FFF minmaxmini ���

In this equation, F represents the normalized value, Fi represents i. Value of measured values and Fmax and Fmin represent maximum and minimum values of measured values.

A learning rate of 0.001 and momentum of 0.1, were fixed for the selected network after training and model selection was completed for the training set. The trained networks were used to run a set of test data. All of the developed networks (3,1,1-3,2,1-3,3,1-3,4,1-3,5,1-3,6,1-3,7,1- 3,8,1-3,9,1) were compared with experimental results.

Various combinations of network architecture to develop an optimum ANN model were examined. ANN (i, j, k) indicates a network architecture with i, j and k neurons in input, hidden and output layers, respectively. The ANN (3, 5, 1) appeared to be most optimal topology; the configuration is shown in Fig. 2.

Figure 2. The structure of the 3,5,1 model (3 inputs, 5 hidden and 1 output)

All of the results obtained from experimental studies and predicted by using the training and testing results of ANN (3,5,1) model is given in Fig. 3a and b, c, d, respectively. The linear least square fit line, its equation and the R2 values were shown in these

figures for the training and testing data. As it is visible in Fig. 3, the values obtained from the training and testing in ANN (3,5,1) model is very close to the experimental results. The result of testing phase in Fig. 3 shows that the ANN (3,5,1) model is capable of generalizing between input and output variables with reasonably good predictions.

The performance of the ANN (3,5,1) model is shown in Fig. 3. The best value of R2 is 94.30% for training set in the model. The minimum values of R2

are 93.30% for testing set in the model. All of R2

values show that the proposed ANN model is suitable and can predict Marshall Stability values very close to the experimental values.

Figure 3. The correlation of the measured and predicted Marshall Stability in a) training and b) testing phase for ANN (3,5,1) model

VI. CONCLUSIONS

In this presented study, an ANN model for predicting the Marshall Stability of lightweight asphalt concrete containing expanded clay and has various mix proportions has been developed.

As a result, Marshall Stability values of lightweight asphalt concrete containing expanded clay and has various mix proportions can be predicted using newly ANN model without any experiments.ANN method is useful artificial intelligent method for pavement engineering applications.

REFERENCES

[1] Cavaleri L, Miraglia N, Papia M (2008). Pumice concrete for structural wall panels. Engineering Structures. 25: 115-125.

[2] Gündüz L, Şapcı N, Bekar M (2006). Utilization of Expanded Clay As Lightweight Aggregate, J. Clay Sci. Technol. Kibited 1(2): 43-49.

[3] Subaşı S. (2009a). The Effects of Using Fly Ash on High Strength Lightweight Concrete Produced with Expanded Clay Aggregate, Scientific Research and Essay Vol. 4 (4) pp. 275-288.

[4] Losa, M. Leandri, P. and Bacci, R. (2008a). Mechanical and Performance-Related Properties of Asphalt Mixes Containing

Expanded Clay Aggregate, Transportation Research Record: Journal of the Transportation Research Board, No. 2051, Transportation Research Board of the National Academies, Washington, D.C., pp. 23–30.

[5] Kalyoncuoğlu, S. F., & Tığdemir, M. (2004). An alternative approach for modelling and simulation of traffic data: Artificial neural Networks. Simulation Modeling Practice and Theory, 12(5), 351–362.

[6] Tiğdemir, M., Karaşahin, M., & Şen., Z. (2002). Investigation of fatigue behaviour of asphalt concrete pavements with fuzzy logic approach. International Journal Fatigue, 24(8), 903–910.

[7] Özgan, E. (2011). Artificial Neural Network Based Modelling of the Marshall Stability of Asphalt Concrete. Expert Systems With Applications, 38, 6025-6030.

[8] Chen, W.F, & Richard Liew, J.Y., (2003). The Civil Engineering Handbook, New Directions in Civil Engineering, CRC Press.

[9] Khayet, M., & Cojocaru, C., Artificial neural network modeling and optimization of desalination by air gap membrane distillation. Separation and Purification Technology, In press.

[10] Sahoo, A.K., Zuo, M.J., & Tiwari, M.K. A data clustering algorithm for stratified data partitioning in artificial neural network. Expert Systems with Applications, In press.

[11] Mohanraj, M., Jayaraj, S., & Muraleedharan, C. (2012). Applications of artificial neural networks for refrigeration, air-conditioning and heat pump systems-A review. Renewable and Sustainable Energy Reviews, 16, 1340– 1358.