Comparisons of Faulting-Based Pavement...

Research ArticleComparisons of Faulting-Based Pavement PerformancePrediction Models

WeinaWang1 Yu Qin2 Xiaofei Li3 Di Wang4 and Huiqiang Chen3

1State and Local Engineering Laboratory for Civil Engineering Material School of Civil Engineering Chongqing Jiaotong UniversityXuefu Avenue No 66 Nanrsquoan District Chongqing China2CREEC (Chongqing) Survey Design and Research Co Ltd Kunlun Avenue No 46 Liangjiang New Area Chongqing China3School of Civil Engineering Chongqing Jiaotong University Xuefu Avenue No 66 Nanrsquoan District Chongqing China4Pavement Engineering Centre Technical University of Braunschweig Raum 104 Beethovenstraszlige 51 b Braunschweig Germany

Correspondence should be addressed to Yu Qin qinyubridge163com

Received 15 April 2017 Revised 2 July 2017 Accepted 6 August 2017 Published 18 September 2017

Academic Editor Hainian Wang

Copyright copy 2017 Weina Wang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Faulting prediction is the core of concrete pavement maintenance and design Highway agencies are always faced with theproblem of lower accuracy for the prediction which causes costly maintenance Although many researchers have developed someperformance prediction models the accuracy of prediction has remained a challenge This paper reviews performance predictionmodels and JPCP faulting models that have been used in past research Then three models including multivariate nonlinearregression (MNLR) model artificial neural network (ANN)model andMarkov Chain (MC)model are tested and compared usinga set of actual pavement survey data taken on interstate highway with varying design features traffic and climate data It is foundthat MNLR model needs further recalibration while the ANN model needs more data for training the network MC model seemsa good tool for pavement performance prediction when the data is limited but it is based on visual inspections and not explicitlyrelated to quantitative physical parameters This paper then suggests that the further direction for developing the performanceprediction model is incorporating the advantages and disadvantages of different models to obtain better accuracy

1 Introduction

Transverse joint faulting is the common type of distressfor jointed concrete pavement which has negative effect ondriving safety and resulting costly rehabilitation [1] Pave-ment faulting prediction is essential for concrete pavementmanagement system and pavement design strategy Andpavement maintenance decision-making is based on currentand future conditions In the past decades many researchershave been focusing on the developing of pavement perfor-mance prediction and improving its accuracy However thechanging of pavement performance is a complex process-ing establishing an easy and accurate predicted model hasremained a challenge Therefore it is necessary to study theperformance of various pavement prediction models

The paper first presents a comprehensive literature reviewthat discusses previous work on the pavement performanceprediction and its classification Three different models

including MNLR model ANN model and MC model arebriefly introducedThese models are quantitatively evaluatedand compared using a set of concrete pavement surveyfaulting data with varying design features traffic and climatedata These survey faulting data are taken for evaluatingthe performance of the three models The results of theseprediction models are presented Then the strengths andweaknesses of these models are surveyed Finally the areas ofconcern in performance prediction and potential for futurework are addressed Suggestions for future research work areproposed by incorporating the advantages and disadvantagesof different models

2 Literature Survey

According to the prediction results the performance modelscan be divided into deterministic and probabilistic modelsFor the deterministic models future condition of pavement

HindawiAdvances in Materials Science and EngineeringVolume 2017 Article ID 6845215 9 pageshttpsdoiorg10115520176845215

2 Advances in Materials Science and Engineering

sectionwas predicted as the exact serviceability value or pave-ment condition index with the previous information of thepavement The probabilistic models predict the performanceof a pavement by giving the probability with which the pave-ment would fall into a particular condition state describingthe possible pavement conditions of the random process [2]Deterministic models included mechanistic model empiri-cal model and mechanistic-empirical models Probabilisticmodels included Markov models Bayesian approach andsurvival analysis

21 Deterministic Model Deterministic models are perhapsthe most common prediction method The main advantagesof using a deterministic model are easy to understandand develop Disadvantage of deterministic models is thatthe regression equation may express the deterioration of agroup of pavements well but do not predict the conditionof individual sections very well [3] Deterministic modelscan usually be categorized as mechanistic empirical andmechanistic-empirical models

Mechanistic models are based on the principles ofmechanics of materials and use the input of wheel loadsto predict the mechanistic responses such as stress strainand deflection The mechanistic models provide valuableinsights into the performance of the pavements There aresome researchers focused on mechanistic model Chua et alestablished the distress model that defined a performancefunction in which the level of distress may be determined asa function of the controlling structural response according tosome damage criterion [4] The continuous reinforced con-crete pavement computer program by National CooperativeHighway Research Program (NCHRP) has been modifiedseveral times to expand the ability of the mechanistic model[5] Al-Qadi et al also establish mechanistic pavement modelby finite element method and then in situ validation isalso taken [6] Since the deterioration process of pavementperformance is complex and not completely understoodpure mechanistic models developed so far cannot accuratelypredict the realistic pavement performance [7]Therefore thedevelopment of reliable and acceptable mechanistic modelsrequires a significant amount of time and effort for continu-ous studies

Empirical approach is widely used in the prediction areabut it suffers from the limitations associated with the scopeand range of available data As an empirical model themost important pavement performance model is created byAmerican Association of State Highways Officials and itis based on the results of actual road test [8] In a SHRPstudy conducted by Simpson et al based on early analysis ofLong-TermPavement Performance (LTPP) general pavementstudies data JPCP faulting models are developed whichare empirical models [9] Prozzi and Madanat use jointestimation techniques to combine experimental data andfielddata for modeling the pavement performance [10]

There are also some JPCP transverse joint faultingempirical models developed under previous research Yudevelops two separate JPCP faulting models for doweledand nondoweled pavements as part of FHWA RPPR projectThe development of these models identifies several pavement

design features and site conditions that significantly affecttransverse joint faulting Teng develops separatemechanistic-empirical JPCP faulting models for doweled and nondow-eled pavements for American Concrete Paving Association(ACPA) [11]

Mechanistic-empirical models are those in whichresponses predicted by mechanistic models were correlatedwith usage or environmental variable such as loadings orage to predict observed performance such as distress Mostmechanistic-empirical models are used for the project levelFew were used for the network level However as the speedand capacity of microcomputers increase and the cost ofcollecting more structural information decreases thesemechanistic-empirical models are used more and morein prediction models [12] Under the support of FederalHighway Administration (FHWA) Teng establishes themechanistic-empirical distress indicator prediction modelsand develops software Pave Spec 30 for JPCP [11] TheMechanistic-Empirical Pavement Design Guide (MEPDG2004) is also developed to replace the AASHTO 1993 Guidefor the Design of Pavement Structure [13] Most of distressprediction models in design guide are empirical-mechanisticmodels which causes wide public concern

In addition especially for faulting model there are alsosome research works on this Under the FHWA NationwidePavement Cost Model (NAPCOM) study Owusu-Antwidevelops the following mechanistic-empirical faulting modelfor doweled and nondoweled JPCP Titus Glove recalibratesNAPCOM JPCP transverse joint faulting model by LTPPdata But this model is recalibrated using LTPP data only[11] Ker et al also establish mechanistic-empirical faultingprediction models for rigid pavements using LTPP database[14] Jung and Zollinger present a mechanistic-empiricalfaulting model which is calibrated from the results of a newerosion test that involves the Hamburg wheel-tracking deviceand LTPP data [15]

22 Probabilistic Model Pavement performance is a stochas-tic process that varies widely with several factors many ofwhich are generally not captured by available dataThereforeprobabilistic models are often used to characterize perfor-mance The following list summarizes the major advantagesand disadvantages associated with probabilistic model [16]

The major advantages associated with probabilistic mod-eling approaches are as follows

(1) They provide a convenient way to incorporate fielddata into a prediction model

(2) They leave it to subjective inputs of experiencedagency personnel

(3) They provide a mathematical means for obtainingperformance predictions

(4) They provide a probabilistic distribution of theexpected condition value with time which will berequired to identify those sections performing signif-icantly differently than would be expected

(5) They reflect performance trends obtained from fieldobservations regardless of nonlinear trendswith time

Advances in Materials Science and Engineering 3

The major disadvantages associated with probabilisticmodels are listed

(1) They do not provide any guidance as to the physicalfactors that contribute to the change in condition

(2) They are time independent so that the probabilityof changing from one condition state to a lowercondition state is not influenced by the age of thepavement and the probabilities are constant over time

Probabilistic models include Markov models survivalanalysis and Bayesian approach

Markov Chains based on the concept of probabilisticcumulative damage are the most commonly used stochas-tic techniques for predicting the performance of variousinfrastructure facilities such as highways and bridges Lounisand Madanat combine the desired practicality of MarkovChain models and the accuracy of mechanistic models toimprove the effectiveness of bridge maintenance manage-ment systems [17] Golroo and Tighe apply a combinationof homogeneous and nonhomogeneous Markov Chain todevelop performance model [18] Pulugurta et al developeda Markov prediction model using the pavement conditiondatabase ofOhioDepartment of Transportation [19] Lethanhand Adey established exponential hidden Markov modelsfor roughness and texture depth indices [20] Abaza usedsimplified staged-homogenous Markov model for flexiblepavement performance prediction at the project level [21]

Survival analysis is generally defined as a set of methodsfor analyzing data where the outcome variable is the timeuntil the occurrence of an event of interest which hasbeen used in the performance prediction Survival meth-ods include parametric nonparametric and semiparametricapproaches Parametric methods assume that the underlyingdistribution of the survival times follows certain knownprobability distributions Weibull model is the popular oneMishalani et al develop a probabilistic model with Weibulldistribution function in different areas [22ndash24] Cox modelknown as the proportional hazard model is one of the mostpopular models in the semiparametric models [25] Mauchand Madanat use the Cox proportional hazards model tocreate a more descriptive model of deterioration without pre-specifying distributions for parameters relating independentvariables and deterioration [26] Nakat andMadanat also usea semiparametric Coxmodel to develop a pavement crackingmodel [27] As a nonparametric estimator of the survivalfunction theKaplan-Meiermethod iswidely used to estimateand graph survival probabilities as a function of time Chouet al used Kaplan-Meier method to estimate the mediansurvival time to the next treatment for pavement perfor-mance [28] Based on Kaplan-Meier method Pulugurta alsodevelops survival curves using available historical pavementdata [29]

Bayesian approach can be used with most approachesexcept the truly mechanistic models [10] Hong and Prozzidevelop the pavement deterioration forecasting model basedon the Bayesian approach and Markov model and useBayesian approach to obtain probabilistic parameter distri-butions through a combination of existing knowledge priorly

and information from the data collected [30] Morcousdevelops a performance prediction of bridge deck systemsusing Markov Chains and Bayesian approach [31] Gao et alpropose modeling the fatigue cracking of flexible pavementby means of survival model and adopt Bayesian approach byusing aMarkov ChainMonte Carlo (MCMC) algorithm [32]It is shown that various modifications to each of these typesof models can be used and it is in development

23 Other Models ANN models are varied in implemen-tation and interpretation An ANN is a mathematical rep-resentation of how mammalian brains were believed tofunction [8] Essentially an ANN model functions like aregression equation in that a number of parameters variablesare used to predict a dependent variable from a number ofindependent onesHowever unlike a regression equation thatdepended upon the ability of designer to comprehend theform of equation a priori neural networks use their internalmassively parallel structure to determine relationships withno input from the designer Thus a neural network is a toolwhich was used in most of the performance prediction area

The advantage of artificial neural networks is their abilityto be trained on previous situations Training is required tocontinuously adjust the connection weights until they reachvalues that allowedANN to predict outputs that are very closeto the actual outputs while being able to be generalized wellon new cases [33]

Some of the disadvantages for ANN are as follows [8]

(1) Prohibitively slow training times for large networks(2) Problems with previously unrepresented patterns in

supervised training(3) Ideal network architectures and training algorithms

remaining part of current research(4) Problems with local minima in training(5) Lack of ability to explain mechanisms in predictive

models

ANN can be used in the performance prediction indifferent areas Tack and Felker used the ANN methodto predict the performance and it is provided that thismethod performs well [8 34] Huang develops an applicationmodel based on ANN approach for estimating the futurecondition of bridges [35] Karwa and Donnell use ANNto predict pavement marking retroreflectivity by data fromNorth Carolina [36] Saghafi et al use ANN approach forpredicting faulting considering base condition and it isindicated that ANN approach can predict joint faulting injointed concrete pavements successfully [37] Recently ANNmodels are commonly used in pavement performancemodels[38 39]

24 Summary The overview of existing literature reflectssome prominent problems in the prediction area First thereare many types of prediction models The advantages anddisadvantages for each type are also introduced But basedon the advantages and disadvantages it is hard to estimateand compare the predicted performance for each model


Many models are able to perform well only in a dataset butnot in different dataset Hence calibrations are required toadjust the parameter inputs so that the models can performreasonably Second not all the prediction models can beused in the special case for they may lack some parametersthat cannot be acquired or there are not enough actual datato establish the model we are choosing So evaluating theeffectiveness of existing models on the same actual datasetthat had variable design features traffic and climate isessentially useful for researchers

For JPCP faulting models existing researches identifieda number of distinct relationships between faulting andtraffic age and various climatic site and pavement designvariables All of the models indicate that design featureshave a significant effect on faulting These models are almosteither empirical models or mechanistic-empirical modelswhich also suffers from the disadvantages of these types ofmodels But the review of these developed transverse jointfaulting models identified a number of variables that havebeen consistently found to significantly influence faulting

Hence in this paper we make a quantitative comparisonof three different prediction models with actual survey databeing conducted Two are the JPCP faulting predictionmodels while the other two are the prediction methodsthat can be used in this faulting prediction It is hoped thatthe comparison results will provide crucial information forresearchers and state DOTs on developing enhanced pave-ment performance models that can lead to a more accurateprediction for maintenance system and design system

3 Data Preparation

Actual pavement survey data used in the models are takenfrom interstate highway with varying design features trafficand climate data ([40] Web-1 httpwwwnoaagov) Thereare 9 sections with 143 records for this whole dataset These143 datasets are divided into two parts 107 records (approx-imately 75 of the whole dataset) are used for training 36records (approximately 25 of the whole dataset) are used forprediction These 36 records are the last 4 yearsrsquo records foreach section Training set is different from those predictionsets and it is greater than prediction set Faulting distributionis presented in Figure 1

4 Models Used for Comparison

Based on themodelingmethodology it is found that differenttypes of models have different characteristics It is importantto understand the feasibility of each prediction model bycomparing their results Since pavement performance dete-rioration process is complex and not completely understoodthe pure mechanistic models developed so far cannot accu-rately predict the realistic pavement performance Thereforemechanistic model is not chosen in the paper

MNLR model is a primal useful technique which hasbeen applied in all fields of engineering knowledge ANNmodel neither deterministic nor probabilistic can be used inall the performance predictions and performs well [37] nowbeing widely used MC model as a probabilistic model is the

3

18

1212

3 4

47

27

17

Training setPrediction set

0

10

20

30

40

50

The n

umbe

r of s

ampl

es

lt5 10ndash155ndash10 20ndash2515ndash20Faulting (mm)

Figure 1 Faulting distribution of training set and prediction set

most commonly used stochastic technique for predicting thevarious performances which is practical and relatively easyto develop [31]

Based on the previous study eight important factors thathave greater impacts on faulting are used in the modeling[41]They are ESAL age dowel diameter base type thicknessdrainage average annual rainfall and freeze-thaw cycletimes The eight important factors that affected faulting areused in multivariate nonlinear regression (MNLR) modeland artificial neural network (ANN) model MC model isjust associated with the faulting value and irrelevant to otherimportant factors

For the reasons listed above MNLR model ANNmodelandMCmodel are used for comparative studyThemodelingmethods were described as follows

41 MNLR Model Equation (1) is the form of a multivariatenonlinear regression (MNLR) model to predict faultingwhich performs well [42]

FAULTING = CESAL119886 [119887 + 119888 times AGE + 119889times DOWELDIA + 119890 times BASE + 119891 times THICKNESS

+ 119892 times DRAIN + ℎ times RAINFALL + 119894 times FTCYC] (1)

where FAULTING is the faulting values mm CESAL arethe accumulate equivalent single-axle loads AGE is thepavement age DOWELDIA is the dowel diameter in BASEis associated with erosion defined as 1 to 5 THICKNESS isthe slab thickness in DRAIN is the capability for drainagedefined as 0 to 1 RAINFALL is the average annual rainfallmm FTCYC are the freeze-thaw cycle times and a b c d ef g h i are the regression coefficients

42 ANN Model Artificial neural network (ANN) is math-ematical models and algorithms designed to mimic theinformation processing and knowledge acquisition that takesplace inside human brain ANNs are capable of learning


by example The back propagation neural network (BPNN)developed by Rumelhart et al is the most representativelearning model for the ANN BPNN is widely applied in avariety of scientific areas especially in applications involv-ing diagnosis and prediction [37] Back propagation is asystematic method that uses gradient descent based deltalearning rule also known as back propagation rule for trainingmultilayer feed forward artificial neural networks The backpropagation network design is a three-layer network withone of each input hidden and output layer After evaluatedcomputation the best results were obtained by an 8-8-1network structure It has 8 neurons in the input layer 8neurons in the hidden layer and 1 neuron in the output layer

43 Markov Chain Model Markov Chain (MC) model is aprobabilistic model widely spread in the world A MarkovChain is a special case of the Markov process whose devel-opment can be treated as a series of transitions betweencertain states A stochastic process is considered as first-orderThe probability of the future state in the Markov processdepends only on the present state [31] This property can beexpressed for a discrete parameter stochastic process (119883119905)with a discrete state space as

119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905 119883119905minus1 = 119894119905minus1 1198831 = 1198941 1198830= 1198940) = 119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905) (2)

where 119894119905 is state of the process at time 119905 and 119875 is conditionalprobability of any future event given the present and pastevents

Transition probabilities are obtained from the incrementof condition data to provide a better prediction [43] Transi-tion probabilities are represented by a matrix of order (119899 times 119899)called the transition probability matrix (P) where n is thenumber of possible condition states Each element (119875119894119895) inthis matrix represents the probability that the condition of afaulting increment component will change from state (119894) tostate (119895) during a certain time interval called the transitionperiod If the initial condition vector 119875(0) that describesthe present condition of a faulting increment component isknown the future condition vector 119875(119905) at any number oftransition periods (119905) can be obtained as follows

119875 (119905) = 119875 (0) times 119875119905

119875 = [[[[[[

11987511 11987512 sdot sdot sdot 119875111989911987521 11987522 sdot sdot sdot 1198752119899sdot sdot sdot sdot sdot sdot1198751198991 1198751198992 sdot sdot sdot 119875119899119899]]]]]] (3)

When the predicted value is gotten it is applied innext yearrsquos prediction Then the new transition probabilitymatrix with the predicted value is computed and next yearrsquospredicted value is obtained

5 Results

Three predictionmodels are used to evaluate the capability ofdifferent models for predicting the pavement performance

Here we present the results of three prediction models with36 records For comparing the capabilities of these proposedmodels measured faulting and predicted faulting are bothused A summary of experimental results is presented inFigure 2

Figure 2(a) represents the comparison of measured andpredicted faulting by MNLR model For this model mostpredicted values are smaller than the measured values Thedifference is nearly 5mm The ability of predicting futurevalue is not good It may result in the range of initial datausing for model regression As an empirical model thevarious data are really important for the predicted correctionThenmore data is needed for recalibrating theMNLRmodel

Figure 2(b) presents the relationship between measuredfaulting and predicted faulting computed by ANN model Itis observed that predicted values are greater than measuredvalue The difference is nearly 2mm Kumar and Minochapoint out that the number of weights required to be trained intheANNmodel will be very large [44]The training dataset of107 records used in this paper may be inadequate Thereforemore data are required for retraining ANNmodel Increasingthe number of training dataset is really good for gettingbetter predictions in further studying And only factors thatare statistically significant to the dependent variables can beincluded in a prediction model Testing the significance ofthose factors is also important

Figure 2(c) shows a plot of the measured and predictedfaulting using MC model It is revealed that the predictionsperformwell which are all near to the equality line Howeverthis model is just based on the condition data For instancethe prediction is only affected by the data quality and isnot related to the design feature or climate More dataattribute more accurate transition probability matrix Themodel without design features which is supposed to beapplied in pavement design is still difficultMoreoverMarkovmodel assumes that the future status is only determined bycurrent status based on the transition matrix which impliesthat previous status has no impact on future status So theimportant factors are not considered in Markov model andthe feasibility for prediction needs further study

Root mean squared error (RMSE) and mean absoluteerror (MAE) are used to quantify the prediction accuracy[37] The computation of root mean squared error and meanabsolute error is described as follows

RMSE = radic 1119899119899sum119894=1

(1199090 minus 119909119901)2MAE = 1119899

119899sum119894=1

100381610038161003816100381610038161199090 minus 11990911990110038161003816100381610038161003816 (4)

where 1199090 is the measured value 119909119901 is the predicted value and119899 is the total number of observationsRMSE and MAE of predicted and measured values for

the three prediction models are compared As shown inFigure 3 the RSME of MNLR model ANN model andMC model is 386 208 and 17 respectively The predictioncapability of MNLR model is worse than ANN model and


0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)

Section 1Section 2Section 3Section 4Section 5

Section 6Section 7Section 8Section 9Line of equality

5 10 15 20 250Measured value (mm)

(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model

Figure 2 Predicted faulting versus measured faulting

MC model ANN model and MC model perform well andthe difference for the prediction capability of these two isnot obvious The analysis of MAE that indicated a similarconclusion is presented Previous study shows ANN modelis more useful compared with World Bank developed modelthe ANN model is applicable to all types of distress [45]Also it is presented by other researchers that ANN also showsa higher capacity to predict joint faulting more accuratelycompared with MLR (multivariate linear regression) modeldeveloped with the same data [37] The prediction capacities

of ANNmodel and regression model concluded in this paperare similar to others

Figures 2 and 3 represent the best and worst model byusing actual data It is indicated that MC model can achievea best predicted value while MNLR model performs worstamong the three But for the limitation MCmodel seems notto be the bestmodel in the three In summary eachmodel hasits advantages and disadvantages their own applicability isdifferent It is necessary to choose the right prediction modelin the special case to get better performance Based on the


137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue

ANNMNLR MCModel type

Figure 3 Quantitative comparison of different models

results there does not exist a no-disadvantage model in therepresentative model in each type of models Developing amodel having more advantages and fewer disadvantages isindeed essential in further study

6 Conclusions

Although many researchers have developed pavement per-formance prediction models the accuracy of the model isstill a challenge It is difficult to effectively compare theperformance prediction models Most researchers just focuson the optimization of the prediction model but someresearches do some work on the comparison of differentmodels [37 45 46] The advantages and disadvantages ofeach predicted model are introduced in many papers Butit is hard to know which model performs well just basedon the summary of advantages and disadvantages for eachmodel Our research is motivated by this need to assess theperformance of various prediction models In this paperwe have conducted a comprehensive literature review of thevarious models used for performance prediction and JPCPfaulting prediction Three prediction models (MNLR modelANNmodel andMCmodel) are briefly introduced and theirperformance is quantitatively and objectively evaluated usingthe actual survey data

Based on the test results it is concluded that MNLRmodel performs the worst and MC model performs the bestANNmodel and MCmodel perform well and the differenceof the prediction capabilities of these two is not obviousMC model shows its promising performance compared withother models when data is limited It is concluded in ourcomparative study that MC model is a promising model inprediction but it is just based on its past condition andnot related to the design feature and other environmentfactorsThis characteristic makes it only applied in pavementmaintenance not in pavement design ANNmodel performsbetter thanMNLRmodel MNLRmodel for its low predicted

capacities needs more data to calibrate and ANN model alsoneeds more data for training the network to improve itsaccuracy

In the future more prediction models can be testedusing the actual survey data and compared with each othereffectively A bigger dataset that is composed ofmore complexsituation is also needed in the model comparison For dif-ferent models having different effectiveness and applicabilityit is important to find a developing and improving modelto predict the pavement performance Further direction fordeveloping the performance prediction model is incorporat-ing the advantages and disadvantages of different models toobtain better accuracy

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This paper is sponsored by National Natural ScienceFoundation of China [51508064 51408083] China Post-doctoral Science Foundation [2014M562287] ChongqingScience and Technology Commission [cstc2014jcyjA30018cstc2016jcyjA0128] Department of Human Resource andSocial Security of Chongqing [Xm2014094] and State andLocal Engineering Laboratory for Civil Engineering Materialof Chongqing Jiaotong University [LHSYS-2016-01]

References

[1] J S Jung Analytical Procedures for Evaluating Factors ThatAffect Portland Cement Concrete Pavement Performance [PhDthesis] Arizona State University Tempe USA 2004

[2] W D Paterson Road Deterioration and Maintenance EffectsModels for Planning andManagement JohnHopkins UniversityPress Baltimore USA 1987

[3] P L Durango Adaptive Optimization Models for InfrastructureManagement [PhD thesis] University of California BerkeleyUSA 2002

[4] K H Chua C L Monismith and K C Crandall ldquoMechanisticPerformanceModel for PavementManagementrdquo in Proceedingsof the Third International Conference on Managing PavementsAustin USA 1994

[5] S M Kim C M Won and F B McCullough ldquoCRCP-9Improved Computer Program forMechanistic Analysis of Con-tinuously Reinforced Concrete Pavementsrdquo Report No 0-1831-2 Center for Transportation Research Bureau of EngineeringResearch University of Teas at Austin Austin USA 2001

[6] I L Al-Qadi M A Elseifi and P J Yoo ldquoIn-Situ Validation ofMechanistic Pavement Finite Element Modelingrdquo in Proceed-ings of the 2nd International Conference onAccelerated PavementTesting Minneapolis USA 2004

[7] Y H Huang Pavement Analysis and Design Pearson PrenticeHall Pearson Education Inc Paramus USA 2nd edition 2004

[8] V Felker Characterizing the Roughness of KANSAS PCC andSuperpave Pavements [PhD thesis] Kansas State UniversityManhattan 2005


[9] A L Simpson J B Rauhut P R Jordahl E Owusu-Antwi IMDarter and R Ahmad ldquoEarly analysis of LTPP general pave-ment studies data volume III sensitivity analyses for selectedpavement distressesrdquo Report No SHRP-P-393 National Coop-erative Highway Research Program Transportation ResearchBoard Washington DC USA 1994

[10] J A Prozzi and S M Madanat ldquoDevelopment of pavementperformance models by combining experimental and fielddatardquo Journal of Infrastructure Systems vol 10 no 1 pp 9ndash222004

[11] T P Teng ldquoImproved Prediction Models for PCC PavementPerformance-Related Specificationsrdquo Volume Final ReportReport No FHWA-RD-00-130 National Cooperative HighwayResearch Program Transportation Research Board Washing-ton DC USA 2000

[12] M Nawaiseh Pavement Performance Prediction Using AdaptiveLogic Networks [PhDThesis] Florida International UniversityMiami Florida FL USA 2002

[13] ARA Inc Guide for Mechanistic-Empirical Design of New andRehabilitated Pavement Structures Final Report ndash NCHRPProject 1-37A National Cooperative Highway Research ProgramTransportationResearchBoardWashingtonWashUSA 2004

[14] H W Ker Y H Lee and C H Lin ldquoDevelopment of faultingprediction models for rigid pavements using LTPP databaserdquoInternational Journal of Pavement Research and Technology vol5 pp 658ndash666 2013

[15] Y Jung and D Zollinger ldquoNew laboratory-based mechanistic-empirical model for faulting in jointed concrete pavementrdquoTransportation Research Record no 2226 pp 60ndash70 2011

[16] R Haas W R Hudson and J Zaniewski Modem PavementManagement Publication No FHWAOH-2004016 Robert EKrieger Publishing Company Malabar USA 1994

[17] Z Lounis and SMMadanat ldquoIntegratingMechanistic and Sta-tistical DeteriorationModels for Effective BridgeManagementrdquoin Proceedings of the Seventh International Conference on Appli-cations of Advanced Technologies in Transportation (AATT) pp513ndash520 Boston Marriot Cambridge Massachusetts UnitedStates

[18] A Golroo and S Tighe ldquoUse of soft computing applications tomodel pervious concrete pavement condition in cold climatesrdquoJournal of Transportation Engineering vol 135 no 11 pp 791ndash800 2009

[19] H Pulugurta Q Shao and Y J Chou ldquoPavement conditionprediction using Markov processrdquo Journal of Statistics andManagement Systems vol 12 no 5 pp 853ndash871 2009

[20] N Lethanh and B T Adey ldquoUse of exponential hidden Markovmodels for modelling pavement deteriorationrdquo InternationalJournal of Pavement Engineering vol 14 no 7 pp 645ndash6542013

[21] K A Abaza ldquoSimplified staged-homogenousMarkovmodel forflexible pavement performance predictionrdquo RoadMaterials andPavement Design vol 17 no 2 pp 365ndash381 2016

[22] R G Mishalani and S M Madanat ldquoComputation of infras-tructure transition probabilities using stochastic durationmod-elsrdquo Journal of Infrastructure Systems vol 8 no 4 pp 139ndash1482002

[23] C Shin and S Madanat ldquoDevelopment of Stochastic Modelof Pavement Distress Initiation Applications of AdvancedTechnology in Transportationrdquo 744 pp 497ndash505 2003

[24] K Kobayashi K Kaito and N Lethanh ldquoDeterioration fore-casting model with multistage Weibull Hazard functionsrdquoJournal of Infrastructure Systems vol 16 no 4 pp 282ndash291 2010

[25] K D Kuhn Uncertainty in Infrastructure Deterioration Model-ing and Robust Maintenance Policies for [PhD thesis] Marko-vianManagement Systems University of California CaliforniaCA USA 2006

[26] MMauch and SMadanat ldquoSemiparametric hazard ratemodelsof reinforced concrete bridge deck deteriorationrdquo Journal ofInfrastructure Systems vol 7 no 2 pp 49ndash57 2001

[27] Z S Nakat and S M Madanat ldquoStochastic duration modelingof pavement overlay crack initiationrdquo Journal of InfrastructureSystems vol 14 no 3 pp 185ndash192 2008

[28] E Chou J Tack J Yu and J Wielinski ldquoEvaluation of theVariation in Pavement Performance between ODOT DistrictsrdquoReport No FHWAOH-2004016 Federal Highway Adminis-tration 2004

[29] H Pulugurta Development of Pavement Condition ForecastingModels [PhD thesis] University of Toledo Toledo Ohio OHUSA 2007

[30] F Hong and J A Prozzi ldquoEstimation of pavement performancedeterioration using Bayesian approachrdquo Journal of Infrastruc-ture Systems vol 12 no 2 pp 77ndash86 2006

[31] G Morcous ldquoPerformance prediction of bridge deck systemsusing markov chainsrdquo Journal of Performance of ConstructedFacilities vol 20 no 2 pp 146ndash155 2006

[32] L Gao J P Aguiar-Moya and Z Zhang ldquoBayesian analysisof heterogeneity in modeling of pavement fatigue crackingrdquoJournal of Computing in Civil Engineering vol 26 no 1 pp 37ndash43 2012

[33] MAttalla THegazy andRHaas ldquoReconstruction of the build-ing infrastructure two performance prediction modelsrdquo Jour-nal of Infrastructure Systems vol 9 no 1 pp 26ndash34 2003

[34] J N Tack Geostatistically Based Pavement Performance Pre-diction Using Universal Kriging and Artificial Neural Networks[PhD thesis] University of Toledo Toledo USA 2002

[35] Y-H Huang ldquoArtificial neural network model of bridge dete-riorationrdquo Journal of Performance of Constructed Facilities vol24 no 6 pp 597ndash602 2010

[36] V Karwa and E T Donnell ldquoPredicting pavement markingretroreflectivity using artificial neural networks Exploratoryanalysisrdquo Journal of Transportation Engineering vol 137 no 2pp 91ndash103 2010

[37] B Saghafi A Hassani R Noori and M G Bustos ldquoArtificialneural networks and regression analysis for predicting faultingin jointed concrete pavements considering base conditionrdquoInternational Journal of Pavement Research and Technology vol2 no 1 pp 20ndash25 2009

[38] S R Amin and L E Amador-Jimenez ldquoBackpropagation Neu-ral Network to estimate pavement performance dealing withmeasurement errorsrdquo RoadMaterials and Pavement Design pp1ndash21 2016

[39] H Ziari J Sobhani J Ayoubinejad and T Hartmann ldquoPre-diction of IRI in short and long terms for flexible pavementsANN and GMDHmethodsrdquo International Journal of PavementEngineering vol 17 no 9 pp 776ndash788 2015

[40] Y C Tsai Y CWu and CWang ldquoGeorgia Concrete PavementPerformance and Longevityrdquo Report No10-10 Georgia Depart-ment of Transportation Atlanta USA 2012

[41] W-NWang and Y-C J Tsai ldquoBack-propagation networkmod-eling for concrete pavement faulting using LTPP datardquo Inter-national Journal of Pavement Research and Technology vol 6no 5 pp 651ndash657 2013


[42] W N Wang Analysis of Design Factors Impacting Joint PlainConcrete Pavement Faulting and Development of ForecastingModel Using LTPP Data [PhD thesis] Changan UniversityXirsquoan Shaanxi China 2014

[43] Y B Lv andHM Peng ldquoForecasting the trend of oil productionof talimu oilfield in the near future by using markov chainrdquoJournal of Xian Shiyou University (Natural Science Edition) vol19 no 4 pp 77ndash79 2004

[44] A Kumar and V K Minocha ldquoRainfall runoff modeling usingartificial neural networksrdquo Journal of Hydrologic Engineeringvol 6 no 2 pp 176-177 2001

[45] D TThube ldquoArtificial Neural Network (ANN) based pavementdeterioration models for low volume roads in Indiardquo Interna-tional Journal of Pavement Research and Technology vol 5 no2 pp 115ndash120 2012

[46] C Chen and J Zhang ldquoComparisons of IRI-based pavementdeterioration prediction models using new mexico pavementdatardquo in Proceedings of the Geo-Frontiers 2011 Advances inGeotechnical Engineering pp 4594ndash4603 usa March 2011

Submit your manuscripts athttpswwwhindawicom

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

CorrosionInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Polymer ScienceInternational Journal of



CeramicsJournal of


CompositesJournal of

NanoparticlesJournal of



International Journal of

Biomaterials


NanoscienceJournal of

TextilesHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Journal of

NanotechnologyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

CrystallographyJournal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


CoatingsJournal of

Advances in

Materials Science and EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Smart Materials Research



MetallurgyJournal of


BioMed Research International

MaterialsJournal of



sectionwas predicted as the exact serviceability value or pave-ment condition index with the previous information of thepavement The probabilistic models predict the performanceof a pavement by giving the probability with which the pave-ment would fall into a particular condition state describingthe possible pavement conditions of the random process [2]Deterministic models included mechanistic model empiri-cal model and mechanistic-empirical models Probabilisticmodels included Markov models Bayesian approach andsurvival analysis

21 Deterministic Model Deterministic models are perhapsthe most common prediction method The main advantagesof using a deterministic model are easy to understandand develop Disadvantage of deterministic models is thatthe regression equation may express the deterioration of agroup of pavements well but do not predict the conditionof individual sections very well [3] Deterministic modelscan usually be categorized as mechanistic empirical andmechanistic-empirical models

Mechanistic models are based on the principles ofmechanics of materials and use the input of wheel loadsto predict the mechanistic responses such as stress strainand deflection The mechanistic models provide valuableinsights into the performance of the pavements There aresome researchers focused on mechanistic model Chua et alestablished the distress model that defined a performancefunction in which the level of distress may be determined asa function of the controlling structural response according tosome damage criterion [4] The continuous reinforced con-crete pavement computer program by National CooperativeHighway Research Program (NCHRP) has been modifiedseveral times to expand the ability of the mechanistic model[5] Al-Qadi et al also establish mechanistic pavement modelby finite element method and then in situ validation isalso taken [6] Since the deterioration process of pavementperformance is complex and not completely understoodpure mechanistic models developed so far cannot accuratelypredict the realistic pavement performance [7]Therefore thedevelopment of reliable and acceptable mechanistic modelsrequires a significant amount of time and effort for continu-ous studies

Empirical approach is widely used in the prediction areabut it suffers from the limitations associated with the scopeand range of available data As an empirical model themost important pavement performance model is created byAmerican Association of State Highways Officials and itis based on the results of actual road test [8] In a SHRPstudy conducted by Simpson et al based on early analysis ofLong-TermPavement Performance (LTPP) general pavementstudies data JPCP faulting models are developed whichare empirical models [9] Prozzi and Madanat use jointestimation techniques to combine experimental data andfielddata for modeling the pavement performance [10]

There are also some JPCP transverse joint faultingempirical models developed under previous research Yudevelops two separate JPCP faulting models for doweledand nondoweled pavements as part of FHWA RPPR projectThe development of these models identifies several pavement

design features and site conditions that significantly affecttransverse joint faulting Teng develops separatemechanistic-empirical JPCP faulting models for doweled and nondow-eled pavements for American Concrete Paving Association(ACPA) [11]

Mechanistic-empirical models are those in whichresponses predicted by mechanistic models were correlatedwith usage or environmental variable such as loadings orage to predict observed performance such as distress Mostmechanistic-empirical models are used for the project levelFew were used for the network level However as the speedand capacity of microcomputers increase and the cost ofcollecting more structural information decreases thesemechanistic-empirical models are used more and morein prediction models [12] Under the support of FederalHighway Administration (FHWA) Teng establishes themechanistic-empirical distress indicator prediction modelsand develops software Pave Spec 30 for JPCP [11] TheMechanistic-Empirical Pavement Design Guide (MEPDG2004) is also developed to replace the AASHTO 1993 Guidefor the Design of Pavement Structure [13] Most of distressprediction models in design guide are empirical-mechanisticmodels which causes wide public concern

In addition especially for faulting model there are alsosome research works on this Under the FHWA NationwidePavement Cost Model (NAPCOM) study Owusu-Antwidevelops the following mechanistic-empirical faulting modelfor doweled and nondoweled JPCP Titus Glove recalibratesNAPCOM JPCP transverse joint faulting model by LTPPdata But this model is recalibrated using LTPP data only[11] Ker et al also establish mechanistic-empirical faultingprediction models for rigid pavements using LTPP database[14] Jung and Zollinger present a mechanistic-empiricalfaulting model which is calibrated from the results of a newerosion test that involves the Hamburg wheel-tracking deviceand LTPP data [15]

22 Probabilistic Model Pavement performance is a stochas-tic process that varies widely with several factors many ofwhich are generally not captured by available dataThereforeprobabilistic models are often used to characterize perfor-mance The following list summarizes the major advantagesand disadvantages associated with probabilistic model [16]

The major advantages associated with probabilistic mod-eling approaches are as follows

(1) They provide a convenient way to incorporate fielddata into a prediction model

(2) They leave it to subjective inputs of experiencedagency personnel

(3) They provide a mathematical means for obtainingperformance predictions

(4) They provide a probabilistic distribution of theexpected condition value with time which will berequired to identify those sections performing signif-icantly differently than would be expected

(5) They reflect performance trends obtained from fieldobservations regardless of nonlinear trendswith time
















models







3 Data Preparation





3

18

1212

3 4

47

27

17


0

10

20

30

40

50

The n

umbe

r of s

ampl

es














119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905 119883119905minus1 = 119894119905minus1 1198831 = 1198941 1198830= 1198940) = 119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905) (2)



119875 (119905) = 119875 (0) times 119875119905

119875 = [[[[[[



5 Results







RMSE = radic 1119899119899sum119894=1

(1199090 minus 119909119901)2MAE = 1119899

119899sum119894=1

100381610038161003816100381610038161199090 minus 11990911990110038161003816100381610038161003816 (4)




0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)




(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model






137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of

















models







3 Data Preparation





3

18

1212

3 4

47

27

17


0

10

20

30

40

50

The n

umbe

r of s

ampl

es














119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905 119883119905minus1 = 119894119905minus1 1198831 = 1198941 1198830= 1198940) = 119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905) (2)



119875 (119905) = 119875 (0) times 119875119905

119875 = [[[[[[



5 Results







RMSE = radic 1119899119899sum119894=1

(1199090 minus 119909119901)2MAE = 1119899

119899sum119894=1

100381610038161003816100381610038161199090 minus 11990911990110038161003816100381610038161003816 (4)




0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)




(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model






137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of






3 Data Preparation





3

18

1212

3 4

47

27

17


0

10

20

30

40

50

The n

umbe

r of s

ampl

es














119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905 119883119905minus1 = 119894119905minus1 1198831 = 1198941 1198830= 1198940) = 119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905) (2)



119875 (119905) = 119875 (0) times 119875119905

119875 = [[[[[[



5 Results







RMSE = radic 1119899119899sum119894=1

(1199090 minus 119909119901)2MAE = 1119899

119899sum119894=1

100381610038161003816100381610038161199090 minus 11990911990110038161003816100381610038161003816 (4)




0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)




(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model






137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of





119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905 119883119905minus1 = 119894119905minus1 1198831 = 1198941 1198830= 1198940) = 119875 (119883119905+1 = 119894119894+1 | 119883119905 = 119894119905) (2)



119875 (119905) = 119875 (0) times 119875119905

119875 = [[[[[[



5 Results







RMSE = radic 1119899119899sum119894=1

(1199090 minus 119909119901)2MAE = 1119899

119899sum119894=1

100381610038161003816100381610038161199090 minus 11990911990110038161003816100381610038161003816 (4)




0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)




(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model






137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of



0

5

10

15

20

25Pr

edic

tion

valu

e (m

m)




(a) MNRL model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(b) ANN model

0

5

10

15

20

25

Pred

ictio

n va

lue (

mm

)




(c) MC model






137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of



137170169

208

334

386

RSMEMAE

0

1

2

3

4

5Va

lue




6 Conclusions







Acknowledgments


References
























































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of

















































CeramicsJournal of







Biomaterials




Journal of


Journal of





CoatingsJournal of

Advances in








MaterialsJournal of


Comparisons of Faulting-Based Pavement...

Documents

Transcript of Comparisons of Faulting-Based Pavement...