1、 GEOTECHNICAL SPECIAL PUBLICATION NO. 195 PERFORMANCE MODELING AND EVALUATION OF PAVEMENT SYSTEMS AND MATERIALS SELECTED PAPERS FROM THE 2009 GEOHUNAN INTERNATIONAL CONFERENCEAugust 36, 2009 Changsha, Hunan, China HOSTED BY Changsha University of Science and Technology, China CO-SPONSORED BY ASCE Ge
2、o-Institute, USA Asphalt Institute, USA Central South University, China Chinese Society of Pavement Engineering, Taiwan Chongqing Jiaotong University, China Deep Foundation Institute, USA Federal Highway Administration, USA Hunan University, China International Society for Asphalt Pavements, USA Jia
3、ngsu Transportation Research Institute, China Korea Institute of Construction Technology, Korea Korean Society of Road Engineers, Korea Texas Department of Transportation, USA Texas Transportation Institute, USA Transportation Research Board (TRB), USA EDITED BY Halil Ceylan, Ph.D. Xueyan Liu, Ph.D.
4、 Kasthurirangan Gopalakrishnan, Ph.D. Likui Huang Published by the American Society of Civil Engineers Library of Congress Cataloging-in-Publication Data Performance modeling and evaluation of pavement systems and materials : selected papers from the 2009 GeoHunan International Conference, August 3-
5、6, 2009, Changsha, Hunan, China / hosted by Changsha University of Science and Technology, China ; co-sponsored by ASCE Geo-Institute, USA et al. ; edited by Halil Ceylan et al. p. cm. - (Geotechnical special publication ; no. 195) Includes bibliographical references and indexes. ISBN 978-0-7844-104
6、7-9 1. Pavements, Asphalt concrete-Design and construction-Evaluation-Congresses. 2. Pavements, Concrete-Design and construction-Evaluation-Congresses. I. Ceylan, Halil. II. Changsha li gong da xue. III. American Society of Civil Engineers. Geo-Institute. IV. GeoHunan International Conference on Cha
7、llenges and Recent Advancements in Pavement Technologies and Transportation Geotechnics (2009 : Changsha, Hunan Sheng, China) TE278.P447 2009 625.8-dc22 2009022665 American Society of Civil Engineers 1801 Alexander Bell Drive Reston, Virginia, 20191-4400 www.pubs.asce.org Any statements expressed in
8、 these materials are those of the individual authors and do not necessarily represent the views of ASCE, which takes no responsibility for any statement made herein. No reference made in this publication to any specific method, product, process, or service constitutes or implies an endorsement, reco
9、mmendation, or warranty thereof by ASCE. The materials are for general information only and do not represent a standard of ASCE, nor are they intended as a reference in purchase specifications, contracts, regulations, statutes, or any other legal document. ASCE makes no representation or warranty of
10、 any kind, whether express or implied, concerning the accuracy, completeness, suitability, or utility of any information, apparatus, product, or process discussed in this publication, and assumes no liability therefore. This information should not be used without first securing competent advice with
11、 respect to its suitability for any general or specific application. Anyone utilizing this information assumes all liability arising from such use, including but not limited to infringement of any patent or patents. ASCE and American Society of Civil EngineersRegistered in U.S. Patent and Trademark
12、Office. Photocopies and reprints. You can obtain instant permission to photocopy ASCE publications by using ASCEs online permission service (http:/pubs.asce.org/permissions/requests/). Requests for 100 copies or more should be submitted to the Reprints Department, Publications Division, ASCE, (addre
13、ss above); email: permissionsasce.org. A reprint order form can be found at http:/pubs.asce.org/support/reprints/. Copyright 2009 by the American Society of Civil Engineers. All Rights Reserved. ISBN 978-0-7844-1047-9 Manufactured in the United States of America. Geotechnical Special Publications 1
14、Terzaghi Lectures 2 Geotechnical Aspects of Stiff and Hard Clays 3 Landslide Dams: Processes, Risk, and Mitigation 7 Timber Bulkheads 9 Foundations 2Director, Pengfang Paving Engineering Research Co. Ltd., 10 Qiaonan Village, Banan District, Chongqing 400054, China; 3Pengfang Paving Engineering Re
15、search Co. Ltd., 10 Qiaonan Village, Banan District, Chongqing 400054, China; 4Xinjiang University, 21 Youhao Road, Urumqi 830008, China; ABSTRACT: The conventional methods of pavement performance assessment indices were established by statistical analyses based on single item and multiple linear
16、regression techniques. These regression models have many deficiencies and are not able to truly reflect the inherent complex nonlinear relationships among the performance indices. However, the Back-Propagation (BP) neural network method with ca comprehensive nonlinear dynamic system is able to addre
17、ss some of these weaknesses. In this paper, the International Roughness Index (IRI), Damage Rate (DR), Structure Strength Index (SSI), Sideway Force Coefficient (SFC), and Rutting Depth (RD) were selected as the five index variables. These variables are considered as some of the most significant fac
18、tors that affect pavement performance. Additionally, these indices were easily classified as non-dimensional quantities and became input data units in the application of the BP neural network. In the study, Pavement Management Index (PMI) was accordingly sub-divided into five groups representing fiv
19、e grades; namely (1) excellent, (2) good, (3) medium, (4) subordinated, and (5) inferior. In this paper, pavement performance assessment based on the BP neural network method and PMI is presented along with a practical application example; followed by a summary of findings and recommendations 1. INT
20、RODUCTION The assessment of asphalt pavement performance is the basis of a series of 1pavement management works (such as prediction of performance, making a plan of maintenance or rebuilding, rational allocation of maintenance funds, etc.). Consequently, the authenticity of the assessment results co
21、nstitutes a key factor in determining the success or failure of the whole pavement management decision. Additionally, it also influences the carryover effect of pavement management system directly. Therefore, it appears particularly important to research the assessment methodology for asphalt paveme
22、nt performance management. Currently, combining subject and object matters is a common way to establish a comprehensive assessment index system for pavement workability, which is achieved through statistics of single item and multiple linear regression techniques. This subsequently allows for the es
23、tablishment of a connection between subjective scoring and objective measured data such as Present Serviceability Index (PSI) in AASHTO, Riding Comfort Index (RCI) in Canada, Maintenance Control Index in Japan, MCI, tc. (Li N. et al., 1997). This methodology has a certain application value. But the
24、regression technique has its own deficiency; so it is difficult to truly reflect the internal complex nonlinear relationships through the specific regression relation established by this methodology. Furthermore, its adaptability is subject to certain restrictions. Based on the foregoing, this paper
25、 explores the design of Back-Propagation (BP) neural network and the selection of pavement performance assessment indices as basis of assessing pavement performance. The paper also provides an example of applying the BP neural network for pavement performance assessment. 2. BACK-PROPAGATION NEURAL N
26、ETWORK In the 80s of 20th century, headed by Rumelhart and McClelland, experts put forward the BP algorithm of Multilayer Feed forward Neural Networks (MFNN). It is a study process with supervision and also an application of Gradient Descent in MFNN (XU Li-na, 2003). BP networks excel at data modeli
27、ng because of their superior function approximation capabilities (Meier and Tutumluer, 1998). Artificial neural network (ANN), as a highly complex nonlinear dynamical system, has high-dimensionality. In recent successful applications, the use of ANNs was introduced for the analysis of jointed concre
28、te pavement responses under dual-wheel and tri-tandem type aircraft gear loadings (Ceylan et al., 1998 and 2000). An ANN model was verified and validated with the results of the ILLI-SLAB finite element solutions, which were intended to enable pavement engineers to easily incorporate current sophist
29、icated finite element methodology into routine practical design. As a result of this verification analysis, a simplified HMA |E*| prediction model was accordingly developed based on the ANN methodology (Gopalakrishnan et al., 2008). GEOTECHNICAL SPECIAL PUBLICATION NO. 19523. BP NEURAL NETWORK DESIG
30、N FOR PAVEMENT PERFORMANCE ASSESSMENT 3.1. The Selection of Pavement Performance Assessment Index Pavement Performance is a synthesis concept, which characterizes the changing trends of pavement behavior and service function under traffic loading, environmental changes (i.e., moisture variations, te
31、mperature fluctuations, etc), and other influencing factors. In essence performance reflects the different degrees of pavement behavior meeting or adapting to the driving requirements including functional performance and structural performance. Based on the current and ever-growing heavy traffic spe
32、ctrum in China including the traffic channelization and overloaded trucks, this paper will analytical determine the subentry indices that are considered more influential on the asphalt pavement assessment. These indices include International Roughness Index (IRI), Damage Rate (DR), Structure Strengt
33、h Index (SSI), Sideway Force Coefficient (SFC), and Rutting Depth (RD). 3.2. Dimensionless Processing of Pavement Performance Assessment Indices Based on the theoretical discussions above, it is clearly evident that comprehensive evaluation of asphalt pavement performance is affected by many factors
34、. Due to the different dimensions of the subentry indices and different types of dimensions, there is no general or common characteristic of these indices and it is practically difficult to compare them directly when analyzing them. For this reason, it is often recommended to normalize these subentr
35、y indices to some similar dimensionless interval with a common utility function so as to enable comprehensive analysis and obtain accurate results. In this paper, the linear dimensionless method-extremum method was selected and utilized to calculate the pavement performance indices. Comparing to PMI
36、, SSI and SFC belong to the direct index grouping, which means that a larger index in terms of magnitude is better. The IRI, CR, and RD on the other hand belong to the inverse group of indices, which means that the smaller index in magnitude is the better the result in terms of pavement performance.
37、 On this basis, different models were adapted to conduct the dimensionless processing in this paper (JIN Cong, 2001). 3.3. Design of BP Neural Network The three-layer BP neural network designed in this paper is as follow: Input layer is the input variable of the BP network. For comprehensive assessm
38、ent of freeway asphalt pavement performance, the designed number of input cells was five, GEOTECHNICAL SPECIAL PUBLICATION NO. 195 3which corresponded to the five factors affecting comprehensive pavement performance assessment (i.e., IRI, CR, RD, SSI, and SFC). These are shown in Figure 1. For accur
39、ate results, each neural network input index should be processed with a dimensionless method. In this paper, the neural network contains only one hidden layer. The number of hidden layer cells is determined by Experiential Formula (Wang Ai-min et al., 2006). The number of hidden layer cells in BP ne
40、ural network designed in this paper is shown in Table 1. In this table, the corresponding suggested values are put forward. The output layer is the output variable of the neural network. The index for pavement performance comprehensive assessment referred in this paper is the Pavement Management Ind
41、ex (PMI). In accordance with the degrees of pavement conditions, five grades representing five pavement comprehensive performance conditions were utilized (i.e., excellent, good, medium, subordinated, and inferior). Therefore, in the design of the BP neural network, the desired output of output laye
42、r is indicated by five output variables, namely excellent, good, medium, subordinated, or inferior, which correspond to the orthogonal vectors: =1 0 0 0 0=0 1 0 0 0=0 0 1 0 0=0 0 0 1 0 =0 0 0 0 1. The BP neural network structure for asphalt pavement performance comprehensive evaluation is accordingl
43、y shown in Figure 1. FIG. 1. BP neural network structure diagram for pavement performance assessment Table 1. The Construction of Neural Network GEOTECHNICAL SPECIAL PUBLICATION NO. 1954BP neural network number of input layer cells “m” number of hidden layer cells “p” suggested values “p” Number of
44、output layer cells “n” asphalt pavement 5 3-12 6 5 Learning rate is a variable indicator of the quantitative power value produced in every circuit of the BP network in terms of training time. Larger learning rate may cause unsteadiness of the system; smaller learning rate may cause longer training t
45、ime and slower the convergence pace. However, a smaller learning rate tends to minimize network errors due to trough confinement. As a result, a smaller learning rate would generally be desired to ensure system steadiness and error minimization. Thus, a learning rate of 0.01-0.8 was recommended for
46、this study. 4. THE REALIZATION OF PAVEMENT PERFORMANCE ASSESSMENT BASED ON BP NEURAL NETWORK Here in this paper, “Microsoft Visual Basic” Visual was adapted as the programming language for the BP neural network modes for pavement performance comprehensive assessment. Figures 2 and 3 show specific le
47、arning training parameters interface and assessment application interface, respectively. FIG. 2. Learning training interface FIG. 3. Assessment and application of BP neural network interface of BP neural network GEOTECHNICAL SPECIAL PUBLICATION NO. 195 55. THE APPLICATION OF PAVEMENT PERFORMANCE COM
48、PREHENSIVE ASSESSMENT METHODOLOGY BASED ON BP NEURAL NETWORK Pavement performance comprehensive index-Pavement Management Index is mathematically obtained by assigning scores based on the assessed pavement performance. The specific method applied in this study uses a 5-point approach and assigns exc
49、ellent, good, medium, subordinated, and inferior grades to characterize the different pavement performances; with 1 point for each grade. Specific assessment grades and interrelated maintenance management measures are shown in Table 2. Table 2. Maintenance Criteria of PMI Index Assessment Grade inferior subordinated medium good excellent PMI (0,1 (1,2 (2,3 (3,4 (4,5 Maintenanc
