Open Access Research on the Mechanical Strength of Emulsified Asphalt-Cementstabilized Macadam Based on Neural Network Algorithm

Size: px
Start display at page:

Download "Open Access Research on the Mechanical Strength of Emulsified Asphalt-Cementstabilized Macadam Based on Neural Network Algorithm"

Transcription

1 Send Orders for Reprints to The Open Civil Engineering Journal, 2015, 9, Open Access Research on the Mechanical Strength of Eulsified Asphalt-Ceentstabilized Macada Based on Neural Network Algorith Chenhao Guo 1,*, Xianpeng Cheng 2 and Xiaoing Zhang 3 1 School of Materials Science and Engineering, Chang' an University, Xi'an , China; 2 Xi'an Municipal Engineering & Research Institute Co., Ltd, Xi'an , China; 3 Shanxi Acadey of Building Research, Xi'an , China Abstract: In this paper, we researched the application of echanical strength experient on eulsified asphalt ceentstabilized acada based on neural network algorith. Based on the analysis of the early diseases of asphalt paveent, which uses sei-rigid base, solving the proble of sei-rigid base asphalt paveent's reflection crack, this paper proposes the research approach that uses a odified sei-rigid base, appropriate paveent structure as the point of starting. Cobining with the past eulsified asphalt-ceent-stabilized acada aterial ixing ethods, the construction technologies of ceent-stabilized acada and synchronous chip sealing, this paper has coe up with a new ethod for odification of sei-rigid base. The unconfined copressive strength and splitting strength of the ixtures would decrease as the ixing aount of eulsified asphalt increases, and the reduction rate of copressive strength decreases as the curing age grows. The ixing aount of eulsified asphalt has alost no influence on the flexural-tensile strength at noral teperature, so it can be ignored. The experiental result shows that the eulsified asphalt ceent-stabilized acada has good perforance than traditional ethods. Keywords: Eulsified asphalt-ceent-stabilized acada, echanical strength, neural network algorith, siulation. 1. INTRODUCTION With the social progress and constant developent of science and technology, there is no denying that road transportation plays a ore and ore iportant role in national econoy. The ore iportant role the road transportation has, the greater the deand for road quality. In the field of civil engineering, natural building aterials have been difficult to eet the requests of the road transportation, which is constantly being developed. Nobody could have failed to notice the fact that new aterials and odified aterials research are prevalent and pervasive issues with which we are all confronted. Due to any benefits of ceent-stabilized gavel, good intensity, the rigidity and integrity, good water stability and anti-frozen, wide source, it is widely used in the base. But for the understanding of ceent-stabilized gravel, base aterial intensity standard still has soe weakness. Botto tensile stress as the standard is stipulated in Specifications for Design of Highway Asphalt Paveent (JTG DSO-2006), Copressive strength is used as the standard in aterial coposition design and engineering quality. But unconfined copressive strength tested in laboratory nearly has no relation with tensile stress of sei-rigid base's botto [1]. Based on theoretical research and systeatically laboratory tests of ceent-stabilized gravel, the proper strength controlling standards and specification were put forward, and this ade the standards in design phase that can also be used in the construction phase. *Address correspondence to this author at the School of Materials Science and Engineering, Chang' an University, Xi'an , China; Tel: ; E-ail: guochenhao@sohu.co Mixing eulsified asphalt into ceent-stabilized acada can ake ixtures that have the characteristics of rigidity and flexibility [2]. It also can reduce drying shrinkage and increase ultiate strain of the ixtures so as to reduce its shrinkage cracking due to change of teperature and huidity as well as resulted in entire paveent cracks. The incorporation of eulsified asphalt would ake ixture have a certain teperature sensitivity, and rising experiental teperature would decrease its strength. The unconfined copressive strength and splitting strength of the ixtures would decrease as the ixing aount of eulsified asphalt increases and the reduction rate of copressive strength decreases as the curing age grows. The ixing aount of eulsified asphalt has alost no influence on the flexural tensile strength at noral teperature, so it can he ignored. The eulsified asphalt ceent-stabilized acada has good resistance to water daage, copressive strength after iersion has no variation and the loss of strength under drying curing condition is less. 2. THE FRAMEWORK AND NEURAL NET- WORK ALGORITHM After adding the eulsified asphalt to the fly-ash ceentstabilized soil and the ceent-stabilized soil, their unconfined copressive strength and indirect tensile strength has reduced, but it still eets the regulatory requireents. When the eulsified asphalt is between 2% and 4%, the eulsified asphalt-ceent powder-fly-ash stabilized powder eets the technical and regulatory requireents of highways and firstlevel roads base. The case is the sae, when the eulsified asphalt is 2%. Meanwhile, the copressive resilient odulus of both the fly-ash ceent-stabilized soil and the ceent / Bentha Open

2 930 The Open Civil Engineering Journal, 2015, Volue 9 Guo et al. stabilized soil is saller than that of the sei-rigid aterial of the sae age period. The water stability and freezing stability of fly-ash ceent-stabilized soil are better than that of ceent-stabilized soil by adding eulsified asphalt. When the content is 2%, the freezing resistance and abrasive resistance are the optiu. Eulsified asphalt can reduce the deforation of drying shrinkage and teperature shrinkage effectively. Coefficient of shrinkage of eulsified asphalt treated fly-ash ceent-stabilized soil and eulsified asphalt treated ceentstabilized soil is less than that of eulsified asphalt treated lie stabilized soil obviously. In the sae way, the forer two aterials have the siilar rule between their drying shrinkage and teperature shrinkage. The econoic analysis indicates that the sei-rigid base ixed with eulsified asphalt has iportant social and econoic benefits. Especially in the areas which lack highquality granules, it is very econoic and reasonable to ake use of the sei-rigid base ixed with eulsified asphalt. So it ay popularize. Reflective cracking of asphalt paveent based on seirigid type base is widely existed in our country s highway and becoes one of the difficult probles to deal with. Cracking in base aterials (norally ceent-stabilized stone) has been shown to be a aor source of this distress in roadways. Precise siulation of initiation and propagation of cracking in ceent stabilization stone under loading, can avoid tie-consuing and expensive indoors experients. Only by input loads, geoetry and aterial paraeters, the Indirect Tensile Test can be accurately siulated, and then useful echanical paraeters can be provided for preventing reflection cracking in the paveent design. The ceentstabilized stone is a coposite aterial coprised of ceent and stone. The cracking on acro scale is a coalescence of cracking on icro scale, whereas icro-cracking is on the ceent and aggregate particulate level. That is why how to link the icro echanical behavior with acro echanical behavior is the priary obective of the research work. A neural network for handwriting recognition is defined by a set of input neurons which ay be activated by the pixels of an input iage. After being weighted and transfored by a function (deterined by the network's designer), the activations of these neurons are then passed on to other neurons. This process is repeated until finally, an output neuron is activated. This deterines which character was read. Like other achine learning ethods, systes that learn fro data-neural networks have been used to solve a wide variety of tasks that are difficult to solve by using ordinary rule-based prograing, including coputer vision and speech recognition. Coputational devices have been created in CMOS, for both biophysical siulation and neuroorphic coputing. More recent efforts show proise for creating Nano-devices [3] for very large scale principal coponents analyses and convolution. If successful, these efforts could usher in a new era of neural coputing [4] that is a step beyond digital coputing, because it depends on learning rather than prograing, and because it is fundaentally analogous rather than digital even though the first instantiations ay in fact be with CMOS digital devices. Genetic algorith is a ethod of foring the research study established by John Holland and his colleagues at the University of Michgan Aerican student in the late 1960s and early 1970s [5, 6]. In this ethod, siulation echanis of biological evolution Model to construct artificial syste, has been widely used in recent years [7]. Fro Traditional BP neural network algorith s slow convergence and easy to fall into Local inia shortcoings, this paper is based on genetic algorith BP Network echaniss to iprove the convergence speed of the network, and then the iproved BP Neural networks is used for the evaluation of the level of the university library inforation. Each index can score fro reviewer s subective scoring ethod after obtaining. The data to be used in equation (1) is noralized. x i = x i! b i a i! b i (1) Where Xi and X respectively, are the i-th index and the actual value Standard value; ai, bi are the axiu, iniu, of the i-th index. Known evaluation indexes, n are the hidden layer nodes depending on probles and experiental data to deterine, you can also experience the value of the forula (2) for the decision. n = log 2 (2) Hidden node output is calculated as follows: = % i i $ & i= 1 h f ) w x! (3) ' ( Where θ is defined as the threshold value of hidden node. The output of the output node is calculated as follows: ( ( ) % i i $! & =! ' i= 1 ( f ) w x f f (4) Where in θ is an output node threshold. Equation (3) and Equation (4) in the transfer function is generally expressed as (0,1) interval of S-type function: ( ( ) % i i $! & =! ' i= 1 ( f ) w x f f (5) Genetic algorith to train the neural network can be used in the echanical strength experient on eulsified asphalt ceent-stabilized acada. The input layer connection weights can be denoted as w i, hidden layer and output layer connection weights can be denoted as t, hidden layer threshold can be denoted as θ and the output layer threshold can be denoted as θ. So connected together to for into a long string, the string corresponding to each position of a group of network weights and threshold value, constitutes an individual. It can generate an initial population of N individuals [8]. Guiding the evolutionary process toward the region of space ay contain the best individual conduct. Suitable function selection should have a great ipact on the training results. This paper uses a calculation error of the neural network (see equation 6), and the fitness function error. Always

3 Research on the Mechanical Strength The Open Civil Engineering Journal, 2015, Volue cutting along the neural network output error in fewer search direction, the convergence target is the iniu output error. T low and T high are the key paraeters to deterine whether the paraeter is serious load ibalance. If one node's load is higher than T high, and there is at least one node, whose load is lower than T low, then the request will be diverted to the lower one. If a node, whose load is two ties higher than T high, request will be diverted to the lowest node in syste, whether there has a node whose load is lower than T low or not. If all the nodes' load are two ties higher than T high, efficiency will be reduced, so the total nuber of effective connections in syste ust be liited, and the value of T high and T low ust be an appropriate set. Supposing the nuber of nodes is n, the largest nuber of the total effective connections is L, we can get:! ( t + n) = "! ( t) + #! (6) i i i where! evaporation is a coefficient such that (1-! ) represents the of trail between tie t and t+n i k # i (7) k = 1 "! = "! Then we have: L = ( N! 1) " T + T! 1 (8) high low Thus there always has a node, whose load is lower than T low. In order to iprove the throughput of syste, T low should be high enough. If the value of T low is T high inus, T low is very iportant to obtain, so the value of T high should be set as high as possible. Assuing the T low is known, the axiu of request delay inus the iniu of request delay is D, the average request delay is R, then we can get: T high ( Tlow + D / R) 2 = (9) 3. THE EXPERIMENT ON MECHANICAL STRENGTH OF EMULSIFIED ASPHALT- CEMENT-STABILIZED MACADAM BASED ON NNA Aggregate grade not only affects strength, stability and construction properties (separation of ixtures, etc.) of ceent-stabilized aterials, but also affects dry shrinkage properties. In this experient, it is the detailed aggregate grades taken for all of ceent-stabilized aterials into account. In order to analyze and copare, it also includes the low and high liit aggregates of the specifications. To overcoe the shortcoings of CCR, CCRDSS was eployed as the base course aterial in the paper. The principle of the structure is to increase the coarse aggregate content so as to ake the ixture for tight locked skeleton in order to reduce the shrinkage coefficient, enlarge tensile strength and iprove the anti-cracking perforance of the ixture. The void of the ixture is filled with ceent and fine aggregate, so the axiu density of the ixture can be obtained and the ixture's durability can be raised. The following sections present the evaluation ethod of anticracking perforance of CCR and the research process in laboratory and in field. Raw aterials in the research include as follows. Ceent: 425# Portland Ceent, its ain technical properties are listed in Table 1. Crushed Rock Aggregates: the size of each crushed rock aggregate can be classified into the grades of 20-40, and 5-10 etc. The test results of aggregate are presented in Table 2. River Sand: its fineness odulus is equal to 2.5, so it belongs to ediu sand. Suppose the aggregate is a sphere of the diaeter of D and is piled up according to the siple cube, that is to say the sphere of the diaeter of D is put into a cubic container of each side D. Under the condition, the aggregate is accuulated to be in the ost looseness. So percentage of void content of the container can be calculated as follows: Table 1.! 3 1" D % V = = 3 D Ceent properties. Index 425#Portland ceent Initial Setting (h:in) 2:51 Initial Setting (h:in) 3:23 Copression Strength (MPa) 27.1 Table 2. Bending Strength (MPa) 5.8 Ceent initial test results. Apparent Density (g/c3) 2.78 Loose Density (g/c3) 1.46 Percentage of Void (%) 45.2 Ratio of Water Absorption (%) 0.13 Aggregate Crushing Value (%) Content of Elongated Aggregate (%) 11.2 If the sphere is a diaeter of D/2, the container can hold 8 spheroids; If the spheroid is a diaeter of D/4, the container can hold 64 spheroids; the rest ay be deduced analogously, but the container's percentage of void content is always not changed, naely 47.6%. In the so-called Dense Skeleton Structure ade of axiu grain-size aggregates (20-40 crushed rock), are

4 932 The Open Civil Engineering Journal, 2015, Volue 9 Guo et al. used as the ain aggregates, whose quantity should not exceeded to that in the naturally loose-laid fraework. The quantity of the other grade crushed rock aggregates should be deterined by the filling of each grade, which should satisfy the filling of the previous grade void and not interfere with the Skeleton Structure of the previous grade. This kind of Aggregate Gradation with interlocking and filling ay be the best ixture in such aspects as friction, cohesive force and copactness. The detailed calculating procedure can be seen as follows. The apparent density of the ain aggregates (20-40 crushed rock) is usually about 2.68g/c3, whereas its axiu dry density is coonly 1.92 g/c3. Here, let it take 2.0 g/c3. Because aggregates are not really spheres, it is ipossible for the ain aggregates to array according to the ideal interlock principle. The percentage of void is usually a little larger than the ideal one, so let it take 50% when calculating. Therefore, weight percentage of the ain aggregates is calculated as follows. (1! 0.5) " 2.68 = 67% 2.0 In Fig. (1), "strain 1" eans the eulsified asphaltceent-stabilized acada grade A1, the rest are contrast sets and can be deduced accordingly. Maxial dry shrinkage strain of ixture A2 is about half that of A11 and A12. Due to containing ore coarse aggregates and less fine aggregates, strain of ixture A12 is greatest, which is about 2 ties of ixture A aggregate affects ore the dry shrinkage strain of sall bea. Dry shrinkage strain of ixture A2 is siilar to that of A3. This indicates that, when the total aount of 26.5 aggregate reaches 3%, even its aount is increasing, the effect of aeliorating sall bea's shrinkage isn't distinct. By controlling the pass ratio of aggregates of 4.75, 2.36 and 0.075, the effect of aeliorating sall bea's shrinkage strain is distinct. The shrinkage strain of ixture A4 or A5 is only 1/2 or 1/3 of ixture A2. And with tie going, this difference increases. Thus, during the highway construction, enhancing the control of aggregate grades and aking sure the ixture construction continuously and equably are needed. Especially controlling the fine aggregates of 4.75, 2.36 and and aking sure that aggregate grades lie with the ediu value of specifications are the urgent necessities. At the sae tie, haronizing the ixing, transportation, spreading and rolling of ixture and enhancing aintenance of ceent-stabilized bases are also required. Optiized by eulsified asphalt-ceent-stabilized acada, fatigue test is set to 500 ties the nuber for tension and copression, with the training Error of experts coented on the level of the tension and copression as a test saple. Saple root ean square error of 0.05, was Library level of tension and copression experient, and experts predicted values to score ore consistent result. Table 3 shows the test to validate the odel. Table 3. Test results for fatigue test. (A1-A5) Nuber of Predictive Intact rate saples value Error (A6-A10) Fig. (1). Mechanical strength experient on strain of sall bea with tie. In order to illustrate the actual effect of using eulsified asphalt-ceent-stabilized acada copared with the traditional ceent, we choose different loading nodes to do the test. The coparison before and after using eulsified asphalt-ceent-stabilized acada in fatigue test of tension and copression and can be seen fro Fig. (2) and Fig. (3). The result shows that during the sae experiental tie, the eulsified asphalt-ceent-stabilized acada can achieve better perforance in the tension and copression ability than the traditional one.

5 Research on the Mechanical Strength The Open Civil Engineering Journal, 2015, Volue rising experiental teperature would decrease its strength. The unconfined copressive strength and splitting strength of the ixtures would decrease as the ixing aount of eulsified asphalt increases, and the reduction rate of copressive strength decreases as the curing age grows. The ixing aount of eulsified asphalt has alost no influence on the flexural tensile strength at noral teperature, so it can be ignored. The eulsified asphalt ceent-stabilized acada has good resistance to water daage, copressive strength after iersion has no variation and the loss of strength under drying curing condition is less. CONFLICT OF INTEREST The authors confir that this article content has no conflict of interest. Fig. (2). The coparison before and after using stabilized acada in fatigue test of tension Fig. (3). The coparison before and after using eulsified asphaltceent-stabilized acada in fatigue test of copressing CONCLUSION In this paper, a new echanical strength experient on eulsified asphalt ceent-stabilized acada based on neural network algorith is presented and researched. Its architecture and features are introduced, and a siulation exaple of fatigue test of tension and copressing is deonstrated. Laboratory experients and coparative ethods are used to study the echanical laws of unconfined copressive strength, splitting strength and flexural tensile strength of the ixtures ixed with eulsified asphalt under different curing conditions such as age, teperature etc. The test results show that incorporation of eulsified asphalt would ake the ixture have a certain teperature sensitivity, and the ACKNOWLEDGEMENTS This work is supported by the Key Proect of Guangxi Social Sciences, China (No. gxsk201424), the Education Science fund of the Education Departent of Guangxi, China (No. 2014JGA268), and Guangxi Office for Education Sciences Planning, China (No. 2013C108). REFERENCES [1] J. Tai, W. Meleis, and J. Zhang, Adaptive Resource Allocation for Cloud Coputing Environents under Bursty Workloads, Northeastern University, Boston, USA, 2013, pp [2] J. M. Tirado, D. Higuero, and F. Isaila, Predictive data grouping and placeent ivor cloud-based elastic server infrastructures, 11 th IEEE/ACM International Syposiu on Cluster, Cloud and grid Coputing, IEEE, 2011, pp , [3] Y. Deng, and Z. Du, The cobination of artificial neural network and genetic algorith applied to forecast of oil and gas yield, Matheatics in Practice and Theory, vol. 38, no.15, pp , [4] J. Liu, and X. Gan, Optiu design of self-adaptive wavelet neural networks based on hybrid hierarchy genetic algorith, Fire Control and Coand Control, vol. 33, no. 11, pp , [5] B. Zhang, and G. Wu, Cooperation of artificial neural networks and iproved genetic algoriths for solving, Coputer Engineering and Applications, vol. 45, no. 34, pp.35-37, [6] Z. Chen, Application of genetic algorith and BP neural network in GDP forecasts, Coputer and Digital Engineering, vol.37, no.9, [7] L.H. Chen, Q.C. Chang, X.G. Chen, and Z.D. Hu, Using BP neural network to predict the water quality of Yellow River, Journal of Lanzhou University (Natural Sciences), vol. 39, no. 2, pp , (in Chinese). [8] Z.Y. Guo, Z.Y. Chen, L.Q. Li, B.P. Song, and Y. Lu, Artificial neural network and its application in regie prediction of groundwater quality, Journal of East China Noral University (Natural Sciences), vol. 1, pp , (in Chinese). Received: February 06, 2015 Revised: April 18, 2015 Accepted: May 28, 2015 Guo et al.; Licensee Bentha Open. This is an open access article licensed under the ters of the ( which perits unrestricted, noncoercial use, distribution and reproduction in any ediu, provided the work is properly cited.