ANALYSIS OF PAVEMENT ROUGHNESS FOR THE AASHTO DESIGN METHOD IN PART OF BAGHDAD CITY

7/31/2019 ANALYSIS OF PAVEMENT ROUGHNESS FOR THE AASHTO DESIGN METHOD IN PART OF BAGHDAD CITY

1/20

(Manuscript No: I12528-05)

April 15, 2012/Accepted: April 21, 2012

1

ANALYSIS OF PAVEMENT ROUGHNESS

FOR THE AASHTO DESIGN METHOD IN

PART OF BAGHDAD CITY

Siham E. Salih *

Lecturer (Dep.of H.W.Y.s &Transportation Eng. - University of Al-Mustansiriyah)

(Email: [email protected])

Abstract - Road roughness is a major factor in evaluating the condition of a highway pavement section because

of its effects on ride quality for road users and vehicle operating costs. The objective of the present study is todevelop the prediction model for international roughness index (IRI) for flexible pavement in part of Baghdad

city. The measures to predict model were used serviceability (Present Serviceability Index, PSI), that include

pavement deteriorations, a (150) selected pavement sections in many sites in the study area .The pavementswere rated to measure the required data for (IRI) model building requirements. These data include: Present

Serviceability Index (PSI), cracking, patching, and rutting and slope variance. Serviceability is an indicator that

represents the level of service a pavement provides to the users. This subjective opinion is closely related toobjective aspects, which can be measured on the pavements surface. This research aims specifically at relating

serviceability results obtained by a 9-member evaluation panel, representing the general public as closely aspossible, to parameters (particularly of roughness) measured with instruments on 50, and 100 road sections of

asphalt concrete and Portland cement concrete, respectively. Results show that prediction of serviceability isquite accurate based on roughness evaluation, while also revealing that, by comparison to studies in more

developed countries, this study are seemingly more tolerant, it is assign a somewhat higher rating to ride

quality. Furthermore, visible distress does not have a significant influence on serviceability values in this study .

Keywords - International Roughness Index, Flexible pavement, Rigid pavement, Present Serviceability Index ;

Longitudinal and Transverse Cracking , Rut Depth , Patching, prediction of pavement condition.

INTROUDECTION

Highway agencies use pavement roughness to monitor the condition and performance of their road networks dueto its effects con ride quality and vehicle operation costs. Pavement roughness can be defined as irregularities in

the pavement surface that adversely affect the ride equality of a vehicle"(Kasibati and Al-Mahmood, 2002)". In

its broadest sense, road roughness has been defined as "the deviations of a surface from a true planer surface

with characteristic dimensions that affect vehicle dynamics, ride quality, dynamics loads, and drainage"[Sayers,1985]. Despite this broad description, the practice today is to limit the measurement of roughness qualities to

those related to the longitudinal profile of the road surface which cause vibrations in road-using vehicles. Roadroughness can also be defined as "the distortion of the road surface that imparts undesirable vertical

accelerations and forces to the vehicle or to the riders and thus contributes to an undesirable, uneconomical,unsafe, or uncomfortable ride" (Hudson, 1981).

In general, road roughness can be caused by any of the following factors (Yoder and Hampton, 1958):

i. Construction techniques which allow some variation from the design profile.ii. Repeated loads, particularly in channelized areas, that can cause pavement distortion by plastic

deformation in one or more of the pavement components.

iii. Frost heave and volume changes due to shrinkage and swell of the subgrade.iv. Nonuniform initial compaction.

During the last three decades, several studies pointed out the major penalties of roughness to the user. In 1960,Carey and Irick (1960) showed that the driver's opinion of the quality of serviceability provided by a pavement

surface is primarily influenced by roughness. Between 1971 and 1982, the World Bank supported several

research activities in Brazil, Kenya, the Caribbean, and India. The main purpose of these studies was toinvestigate the relationship between road roughness and user costs. In 1980.


2/20

International Journal Of Structronics & Mechatronics

2

Rizenbergs (1980) pointed to the following penalties associated with roughness: rider no acceptance anddiscomfort, less safety, increased energy consumption, road-tire loading and damage, and vehicle deterioration.

Gillespie and Sayers (1981) examined the relationship between road roughness and vehicle ride to illustrate the

mechanisms involved and to reveal those aspects of road roughness that play the major role in determining the

public's perception of road serviceability. It has been widely suspected that the initial roughness of a pavementsection will affect its long-term performance. Recently, a study conducted by Janoff (1990) suggestedthat initial

pavement roughness measurements are highly correlated with roughness measurements made 8-10 years afterconstruction.

Due to the importance of pavement roughness, most highway agencies have established smoothnessspecifications for new pavement construction. Smoothness specifications are normally written for the use of

profilographs. About half of the states require that a specific limit of smoothness be met, whereas the remainder

of the states are using a variable scale with pay adjustments, depending on the degree of the smoothness

achieved (Wood strom, 1990). These pay adjustment factors are made based on the assumption that lower initialpavement roughness will result in better pavement performance.

SERVICEABILITY AND ROUGHNESS INDICES

A.

SERVICEABILITY INDEX

Pavement Serviceability represents the level of services that pavement structures offer users. This indicator first

appeared as a rating made by users with respect to the state of the road, particularly the roads surface. Thisrating is represented by a subjective index called Present Serviceability Rating (PSR) and may be replaced by

an objective index called Present Serviceability Index (PSI). The latter index is determined on a strictlyobjective basis by applying the users rating scale to sections of roads featuring different states of distress. This

scale enables users to rate the pavements state in terms of its service quality. The scale rates pavements from 0to 5, from an extreme state of distress to a new or almost new pavement [Jorge alberto,2001]. Thus, a

quantitative relationship is established between this Serviceability rating and certain parameters that measurephysical distress of pavement surface.

Roughness Index

Roughness is defined as irregularities in pavement surface that adversely affect ride quality, safety, and vehicle

maintenance and operating costs. Roughness is the factor that most influences users evaluation when rating ride

quality. One of the problems faced by technicians when rating ride quality and comfort for vehicle users andcomparing experiences among countries is the great diversity of techniques, equipment, and indicators available

in each country.

Consequently, there arose an international interest in developing a single and common index as reference. This

index had to be independent from equipment or techniques used to obtai n the profiles geometry, and at thesame time had to represent the full range of users perceptions when driving an average vehicle at an averagespeed. The need for this index originated in the mid-eighties, giving rise to the concept, definition, and method

for calculating the International Roughness Index (IRI) [7, 8].

IRI is a statistical indicator of surface irregularity in road pavements. The real profile of a newlybuilt roadrepresents a state defined by its IRI with an approximate range of 1.02.5 (m/km). After the road is constructed,

pavement roughness varies as a function of traffic, gradually increasing the pavement IRI values (greaterirregularities).

Categories or Classes of Equipment for Measuring Roughness

The different evaluation methods available to measure surface roughness were grouped into four categories,

classified according to how directly their measurements came close to the IRI [7, 9]. These methods may be

summed up as follows: Class 1, Precision Profiles (which require the longitudinal profile of a rut to be measured

in a precise manner); Class 2, Other Methods for Profile Measuring (calculation of IRI is based onmeasurements of the longitudinal profile, but is not as accurate as Class 1 measurement method); Class 3,


3/20

Siham E. Salih

3

Estimations of IRI through Correlation (systems for measuring roughness by Profilometer, Rod &Level ); Class4, Subjective Ratings and Un calibrated Measurements (devices with an un calibrated response, sensations of

comfort and safety which a person experiences when driving on a road).

METHODOLOGY OF THE EXPERIMENT

In order to achieve the objectives proposed in this paper, it was first necessary to select a sufficient number of

pavement sections for study in Baghdad city, covering the range of possible conditions (good, fair, and poor).

Next, roughness of these sections had to be measured, first using Profilometer and then Rod &Level (Machine

for Evaluating Roughness using Low-cost Instrumentation). Also, surface integrity had to be established usingcondition survey of the pavement. The last stage of data collection would involve evaluating serviceability by a

panel of people representative of habitual vehicle users. Figure 1 shows the principal stages of the methodology

of the experiment.

Figure 1- Methodology of the Experiment

Selection of Pavement Sections

Selection of pavement sections for the study had to be conducted by an objective process that would allow

discrimination among the different pavements to be studied. Therefore certain requirements, based mainly on

the feasibility of evaluating roughness and serviceability, were established [10]. These requirements were:length, safety (number of lanes, vehicle flows and visibility), accessibility, possibility of measuring with

equipment. Pavement sections that met conditions as defined by the evaluation panel were selected fromdifferent municipal districts.

Relative Importance of each Requirement

To be able to discriminate when selecting the sections, each requirement had a different importance in the final

weighting. The percentage assigned to each weight depended mainly on the possibility of measuring the

sections roughness with the available equipment and of the traffic modal composition, leaving safety at asecondary level.


4/20


4

The previously selected sections were then divided into three levels of serviceability, according to the scoresobtained by the work group: good, fair and bad. A weighting with the previous criteria was applied to each

section, and a list of sections arranged by each serviceability level and pavement type was obtained.

Procedures Used to Measure Roughness

A. ROD &LEVEL INSTRUMENTMeasures of surface elevation are obtained at constant intervals along a line on a traveled surface to define alongitudinal profile. The line used for the profile is called a wheel track, a path followed by the tire of a road

vehicle. The measured numbers are recorded and entered into a computer for graphical display and analysis. Theprofile points are used as input to a computational algorithm that produces a summary roughness index.

This method describes the use of conventional survey equipment comprising an optical level and graduated rod,but it may also be applied to automated techniques (for example, laser-based systems) with appropriate

adjustments. At a minimum, two persons are required; one to locate and hold the rod (the rod-man), and a

second to read relative heights through the leveling instrument and record the readings. For better efficiency, it

is recommended that a third person record the readings to allow the instrument operator to concentrate on

adjusting and reading the instrument. When maximum measuring speed is desired, a fourth crew member isrecommended to act as relief.

B. LASER PROFILERA Two Laser Profiler (TLP) was used to measure the cross-section profile and calculate roughness (IRI) of the

projects sections [12]. It is a Class 1 type of equipment as it is able to obtain the profile with great precision,which then allows the calculation.

To calculate IRI, the Laser Profilers computer program has a profile processing module, which is independentfrom the measurement and can be performed at any time after the profile has been measured. Only the

processing distance is needed, that is, the distance from which the program is to report the IRI. A distance of 10meters was established as reasonable, because it allows one to recognize singularities and to obtain a sufficient

amount of IRI values from the sections 400 meters.

The profile processing yields a file text which may then be easily worked on with spreadsheets. Interesting

results that can be seen on the file, in the different columns, are: the distance traveled from the beginning of the

section, the IRI value of the left rut and the IRI value of the right rut [15].

C. DISTRESS SURVEY METHODOLOGYExisting levels of distress are a very important measurement of pavement sections requirements. Thisinformation is added to roughness data measured with Rod &Level and the Profilometer. There are different

types of deterioration and each type has different degrees of severity.

Every distress condition is the result of one or more factors, which when known give a very good diagnosis ofthe pavements weaknesses. Thus, a detailed distress survey of the pavement is one of the steps necessary t o

establish pavement condition. In this research, the condition survey of the pavement consisted of detecting,recording and quantifying the distress conditions that each section had at the moment of conducting the study.

There are several distress survey procedures [9], and it is felt that the most complete one, supported by years ofstudy and experience, is the procedure proposed by the Strategic Highway Research Program [13]. This is the

methodology used in this study.

Evaluation of Serviceability

This section describes those planning aspects which are relevant for serviceability rating by the evaluationpanel.


5/20

Siham E. Salih

5

A. COMPOSITION OF THE PANELThe people that made up the evaluation panel was one of the most important aspects in this study: (i) they had to

represent the public which generally circulates on the countrys streets and roads; (ii) they needed a broad range

of experience both as drivers and passengers in cars, as well as passengers in public transportation buses; and(iii) they should not have any kind of bias or prejudice regarding trips in cars and buses.

The size of the evaluation panel had to be defined so that it was administratively manageable while permittingan adequate precision. The number of people needed to obtain a certain degree of certainty in the PSR, at a

given level of confidence, had been tabulated in previous studies [14].

As the group had to represent the general public as closely as possible, the panel was finally made up by 11 men

with different activities, obtaining 90% of level of confidence and an error of 0.5 in the PSR value (14).

Design of the Rating Form

In this study, we adopted the widely used AASHTO scale. It consists of reporting in words the levels of quality,in addition to a line where the person performing the rating makes a mark. The other evaluation category thatwas used is the acceptance criteria. In it, the evaluator is asked to judge if ride quality on the section seems

acceptable so as to include it in: (a) expressways, and (b) initial streets .

The responses to this segment of the form provide a measure of the minimum acceptance threshold of functional

quality of pavements.It was important for the form to be simple, so that it enabled the evaluator to rapidly judgeand decide the serviceability rating as well as his position regarding the acceptance or not of that ride quality.

Figure 2 shows this form.

Figure 2: Rating Form used by the Evaluation Panel

Training of the Members of the Evaluation Panel

Training of the evaluation panel and the instructions they would be given were very important aspects in the

process of subjective rating. Studies that show that a team rating a subjective variable without receiving any


6/20


6

instructions obtains results that are significantly different to another team who has received instructions [15].Besides, the variance in the team that receives instructions is much lower than in the other one [16].

Considering the above, careful instructions were developed for this studys evaluators. The instructions weredesigned to be as simple as possible, but at the same time to have the sufficient level of detail to prevent any

kind of confusion as to procedures and definitions.The rating procedure was explained to all evaluators in a

training session. They were given instructions in writing which were discussed by the research team; also, allpanel members questions were answered.

After the training session, evaluators were taken for a ride on some pavement sections featuring a broad range of

roughness. During the ride, evaluators were motivated to discuss the procedure both among themselves as wellas with those in charge. The purpose of this was to orient the evaluation panel so that they could perceive the

differences and acquire confidence with the procedure.

Evaluation Sessions

In order to prevent results from being influenced by changes in pavement characteristics (e.g. new distress or

possible rehabilitation), section evaluation by the panel in different vehicles must carried out over a brief period

of time. It is also important that evaluations in the same vehicle are not made very far apart in time, so thatvariations in the mechanical response of the vehicles body does not alter the results.

It must also be borne in mind that panel members weariness and fatigue may alter their rating of a section. A

suggestion made in a prior study to have breaks every 1.5 or 2 hours was adopted in this study [15]. Based onthe above, daily evaluation sessions were from 9:00 a.m. to 5:30 p.m., with time off for lunch between 12:30

p.m. and 2:00 p.m., and rest periods halfway through the morning and afternoon. In each session, sections of

asphalt and concrete were included, so that each time the team covered a wide spectrum of roughness in circuits

that optimized driving time, while simultaneously avoiding any effects due to the evaluations sequence. Thesections were evaluated at a constant speed of approximately 50 km/hour.

RESULTS OBTAINED

A. RELATIONSHIP BETWEEN ROD &LEVEL AND THE PROFILOMETER Different adjustment curves were tested, using data obtained for IRI from measurements with Rod &Leveland

from those obtained with the Profilometer. Considering the good adjustments obtained, the larger sample size

and for simplicity of future general treatment, the use of the linear equation (Equation 1) obtained by using data

for all types of pavements (concrete and asphalt ,) is recommended.

Concrete:

1. IRI PRO = 0.0171*( R&L) + 1.8227 R2 = 0.9169

Figure 3: Relationship Proposed and Existing Relationships For Concrete Pavement

R2

= 0.9169

1.5

1.6

1.7

1.8

1.9

2

2.1

2.2

2.3

2.4

0 5 10 15 20 25 30 35

IRI Rod&Level

IRIProfilomete


7/20

Siham E. Salih

7

Asphalt

2. IRI PRO =0.4072* Rod&L + 7.7176 R2= 0.9297

Figure 4: Relationship Proposed and Existing Relationships For Flexible Pavement

Rating Serviceability

The evaluation panel had to drive on all road sections under study using passenger car types of vehicles. Allevaluations were performed by the same group of 11 evaluators, under the direction of the personnel responsible

for the study. The evaluation panel members were subjected to prior training and were asked to drive on sometest sections so they could be in a position to compare their opinions. Subsequently, during the evaluation

sessions, they rated their perception of the pavements on an individual and secret basis. Finally, it is worthmentioning that all evaluations were performed within a time frame no longer than two weeks. The evaluators

were asked to indicate possible conditions of comfort and ride quality from very bad to very good.

This subjective rating was converted into a numerical value, assigning a score to each road section which could

range from 0 to 5. The average of the individual scores assigned by each evaluator for the same length of road isthe PSR of the road section. Thus, the panel evaluated a total of 50 road sections for rigid pavement and 108

road sections for flexible pavement. The results of the evaluation panel are shown in Table 1

Relationship between Roughness and Serviceability

The regression between the panel values for PSR and IRI is called PSIROUGH.. Serviceability ratings are

available for two types of asphalts (AC, ACC) used in the study, and the vehicle used in mention is made of theServiceability relationship for cars, because it is the one most commonly used and the one established by the

AASHTO Test.

The best adjustments were obtained with square root and exponential models (nonlinear equations). In order to

use regression analysis to calibrate these equations, some transformations (such as log transformations) weremade to change the nonlinear relationship between PSIROUGHand IRI into a linear relationship. For flexible

pavements, Equations 3 and 4 were obtained and Figure 5 shows the representative graph of the regression forthese pavements.

Concrete:

3. PSIROUGH= -6.785Ln( IRI ) + 10.336 R2 = 0.882

R2

= 0.9297

0

5

10

15

20

25

30

35

40

45

0 10 20 30 40 50 60

IRI ROd& Level

IRIProfilometer


8/20


8

4. PSIROUGH= 82.962e-1.1669 IRI R2 = 0.9232

Figure5: PSIROUGH Regression in Concrete Pavements

In an analogous manner, Equations 5 and 6 were obtained for Asphalt pavements:

Asphalt:

5. PSIROUGH= -6.8101 Ln IRI + 10.737 R2 = 0.885

6. PSIROUGH= 144.31 e-1.2897 IRI R2 = 0.8921In Figure 6, the regressions shown previously may be observed

Figure 6: PSIROUGH Regression in Flexible Pavements

R2

= 0.8823 R2

= 0.9231

-2

-1

0

1

2

3

4

5

6

0 2 4 6

IRI m/km

P,Serviceability

Series1

Log. (Series1)

Expon. (Series1)

R

2

= 0.8921

R2

= 0.885

-2

-1

0

1

2

3

4

5

6

0 2 4 6

Series1

Expon. (Series1)

Log. (Series1)


9/20

Siham E. Salih

9

Although the exponential and LOG equations are very similar, in this study we have preferred LOG regressionsto predict the values of PSR for two reasons: (a) they have a higher coefficient of determination; and (b) for low

IRI values they predict a higher Serviceability. The position of the adjustment curves between the types of

pavement is set forth in Figure 7. It can be seen that for the same IRI value, a concrete pavement is rated better

than an asphalt pavement and that the difference between them is greater as roughness increases.

Figure 7: Comparison of PSIROUGH Regressions between Asphalt and Concrete

Effects of Other Pavement Distress on Serviceability

The equations of serviceability developed by AASHTO to predict the PSR [6] include slope variance and other

pavement distresses like surface rutting, cracking, and patching. All these distresses had been measured in this

research on a condition survey of the pavement. In order to determine if some types of distress had an effect onServiceability for this study , it was necessary to consider the results of the condition survey incorporating thosedistresses that could contribute to the regression, and finally, prove its significance in the model. Tables 2 and 3

show the different models, the explanatory variables, and the t-statistic. Interestingly, surface rutting (RD),cracks (C), and patching (P) are not significant in determining Serviceability. However, the IRI is always

significant. This means that predict model for this study the surface distresses are not significant in determining

Serviceability compared to IRI.

Table 2: Effects of Other Pavement Distress on Serviceability in Concrete Pavements.

IRI: International Roughness Index (m/km); C: cracks (m2/1000m2)P: patch (m2/1000m2)

R2 = 0.8153

R2 = 0.8831

0

1

2

3

4

5

6

2.4 3.4 4.4 5.4 6.4

IRI m/km

Serviceability

,

Asphalt

Concrete

Model Variable Regressor R2

SEM T-Statistic T Critical

(95%)

a + b * IRI a

b

5.189560

-0.92057

0.6044 0.76074 29.1426

-13.5991

a + b IRI +c*

PC

A

b

c

6.65957

-1.08183

-0.15178

0.45 0.41362 5.39982

-1.25120

-6.41522


10/20


10

Table 3: Effects of Other Pavement Distress on Serviceability in Asphalt Pavements.

RD: rutting (inches);IRI: International Roughness Index (m/km); C: cracks (m/1000m2)

P: patch (m2/1000m2)

Application of the AASHTO Method of Design

Serviceability ratings performed by the evaluation panel of Iraqian users were much higher than in other similar

studies. These higher ratings generate the problem that final Serviceability is also higher, whereby the loss in

Serviceability is lower. Therefore, these Serviceability results according to Iraqian users are not recommended

for use in the AASHTO design method. In order to solve this problem, the PSI was calculated by resorting to

procedures recommended by AASHTO, using information such as slope variance, rutting, cracked surface andpotholed surface. Applying these procedures to data obtained during the study, the PSIAASHTO was then

calculated for each one of the road sections analyzed and compared to the corresponding IRI for each roadsection. The equations of the IRI-p relationships obtained with the above information may be used in designing

pavements according to AASHTO (Equations 7 and 8):

Concrete:

7. IRI PSIAASHTO = -1.1406Ln(IRI) + 3.9946 R2 = 0.943Asphalt:

8. IRI PSIAASHTO = -1.343Ln(IRI) + 4.1807 R2 = 0.9622

Model Variable Regressor R2

SEM T-StatisticTCrit(95

%)

a + b *RD2 a

b

4.81641

-2.70457

0.666 0.80430 27.1561

-14.5627

a + b IRI +c*RD2A

b

c

7.13651

-0.87602

-2.19867

0.865 0.51179 32.7820

-12.4630

-17.5961

a + b IRI +c* PC A

b

c

6.94556

-1.28535

-0.00710

0.61 0.87644 18.5390

-11.2899

-6.0859

a + IRI +c*RD2+d PC

A

b

C

d

6.55689

-1.16816

-0.00337

-0.00337

0.5477 0.9354 17.1449

-10.0822

-1.1842

-1.1842


11/20

Siham E. Salih

11

The comparison of these new design curves developed by other studies [25] is more favorable, because the

Serviceability value which they predict is quite similar. Figure 12 shows this comparison in the case of asphalt,

and Figure 8 for concrete.

Final Serviceability values recommended by this study originate from the evaluationpanels results, but must beread in terms of the PSIAASHTO curve, because it is this curve which generates the values to be used for design

purposes.

Figure 8 - Comparison of IRI-PSIAASHTO curve with previous studies for Concrete

Figure 9 - Comparison of IRI-PSIAASHTO curve with previous studies for Asphalt

R2

= 0.943

R2

= 0.9019

0

1

2

3

4

5

6

0 2 4 6 8 10

IRI

Serviceability

;P

IRI PSI AASHTO

IRI_PSI

Log. (IRI PSI

AASHTO)

Log. (IRI_PSI)

R

2

= 0.9622

R2

= 0.8852

-2

-1

0

1

2

3

4

5

6

7

0 10 20 30

IRI

Serv

iceability;

IRI PSI AASHTO

IRI _PSI

Log. (IRI PSI

AASHTO)

Log. (IRI _PSI)


12/20


12

DEVELOPMENT OF MODEL FOR THE PREDECTION OF INTERNATIONALROUGHNESS INDEX

The statistical techniques used for the models development required for evaluation of the pavement

serviceability and performance of the selected roads in the study area. A suitable amount of data representingmany variables is presented in this investigation. For the purpose of model development of the present

serviceability index, the data include; patching, cracking, slope variance and rutting. The choice of sample sizeis presented in the following paragraph.

1. Selecting Sample SizeThe following formula is used to determine the required sample size. [17],

E =V t / (n) 0.5

V = S / X

Where

E = Error of the mean,

V = Coefficient of Variation,

T = tstatistics,

n = Sample Size,

S = Standard deviation, and;

X = Sample Mean.

N = 80

For confidence level = 95%, df = 79, then t = 1.99

E = (1.107029* 1.99) / (80) 0.5

E = 0.2485

Then, the sample size is accepted with this percent of error.

2. THE MODELS DEVELOPMENT PROCESSThe following steps, which are recommended and presented by many statisticians and researchers,[17] are

followed in this study;

i. Identifying the dependent variables.ii. Listing potential predictors.iii. Gathering the required observations for the potential models.iv. Identifying several possible models.v. Using statistical software to estimate the models.vi. Determining whether the required conditions are satisfied.vii. Using the engineering judgment and the statistical output to select the best models.

3. REGRESSION MODELRegression analysis is a statistical method that uses the relationships between two or more quantities variables

to generate a model that may predict one variable from the other(s). The term multiple linear regression (MLR)


13/20

Siham E. Salih

13

is employed when a model is a function of more than one predictor variable. The objective behind (MLR) is toobtain adequate models, at a selected confidence level, using the variable data while at the same time satisfying

the basic assumptions of regression analysis.

The main assumptions of regression include:

severe multicollinearity does not exist among predictor variable Influential observation or outliers do not exist in the data. The distribution of error is normal. The mean of error distribution is zero.

The objective is accomplished by selecting the model, which provides the highest adjusted coefficient of

determination (R) and lowest mean square error (MSE), for a given data [17]. The same variables and criteriaused to perform AASHO Road Test (IRI) model are used to develop a (IRI) of the present study. Accordingly,

multiple linear regressions are used for the development process of this model.

Outliers

If one or more of observations is different significantly from all others, it is called outlier . The cause of afaulty observation may be a mistake. Outliers and influential observations are checked by usingChauvinist's criterion [17]. The results of this test can be found in Tables (4).

Table (4): Results of Chauvenet' Test for Outliers of PSI Database

Sample Size: 80 , X m = value of outlier. x= sample mean. s = standard deviation.

(xmx

/s)tabulated= 2.74 > all calculated values. Thus the outliers are not rejected

Multicollinearity

It is a condition that exists when the independent variables are correlated with another one. The adverse effect of

multicollinearity is that the estimated regression coefficient (b1, b2, etc.) tends to have large sampling

variability. By using STATISTICA software the correlation coefficients between all of the variables werecalculated and the correlation matrix was setup.

Developed Model

Scatter plot was carried out between the dependent and independent variables for the requirements of IRI model

building process. From the plots, the nature of relation between these variables can be expected and the best

relations are selected, the Scatter plots for selected function are illustrated in (Figures 10, 11, and 12) forpatching, cracking, and rut depth.

Variables Mean Minim MaximumStandard

Deviation (s )

xmx /s

X m = min

xmx /s

X m=maxi

PSI 4.074748 2.61206 4.852703 0.535917 2.729 1.45

SlopeVariance

0.0262481 0.00263 0.1005665 0.0293210 0.857 2.534


14/20


14

y = 0.0035x - 0.0063

R2

= 0.9034

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

050100150200250300350

Paching

SV

Figure 10 - Patching vs SV value

y = 0.8285x - 0.2928

R2

= 0.8109

-0.5

0

0.5

1

1.5

2

00.511.52

RD

SV

Figure 12 - RD vs. SV value

y = 0.0018x - 0.014

R2

= 0.8502

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0100200300400500600

Cracking

SV

Figure 11 - Cracking vs. SV value


15/20

Siham E. Salih

15

The multiple linear regression technique that is used for the purpose of IRI model development results in thefollowing model form;

12. IRI=0.244759-.155070 LN (PSI)Where:

IRI = International Roughness Index

PSI = Present Serviceability Index

The summary of the multiple linear regressions, and several possible developed models can be seen in Tables

(5) and (6).

Table (5): Regression Summary for IRI Model

Table (6): Several Possible models From the Multiple Linear Regression Analysis for the IRI Model.

Results of the Analysis

The multiple linear regressions, using STATSTICA software has served its purpose in drawing attention todeveloping a model by using the same independent variables as those used previously by Observed AASHO

model. The model developed is shown at the end of the previous section as IRI model (equation 2). Theindependent variables ;Serviceability (PSI) that represent in slope variance , rut depth, cracking and patching

and those used in the model development process show that, the value of IRI is strongly affected by these

mentioned variables . The model indicates that the value of IRI decreases with the increase in Serviceability .

Regression Summary for Dependent Variable: SVR= .93524029 R= .87467440 Adjusted R= .87306766F(1,78)=544.38 p


16/20


16

Discussion of Results

Referring to the IRI model; one variable was found to be common to the general picture of the model

development, but this variable is content from many criteria slope variance, rut depth, cracking, and patching.The coefficient of determination was found to be 0.94 that means; 94 percent of the IRI prediction can be

explained by this model.

Models Limitation

As with all regression models, the model is only valid within the ranges of the variables they were developed

from .Some additional limitations may be related to the study area. Specific specifications can be listed as

follows;

i. The select of sections is randomly in the study area and is uniform of a (1200 ft) each is used forthe purpose of IRI model development.

ii. The range of data for IRI model can be seen in Tables (7).The intention of the limitation is not tosuggest that the modeling effort has not been successful. It merely serves to alert of the limitations

of the data.

Table (7): Ranges of Data in IRI Model Database

VALIDATION OF THE DEVELOPED MODEL

The final step in the model building process is validation of the developed models. The objective is to assess the

ability of pavement condition index prediction model to accurately predict amount of IRI in the field. A reviewof the statistical researches suggested the following methods for validation of a regression model [Ahmed,

Namir G 2002].

check on model predictions and coefficients collection new data comparison with previously developed Models data splitting prediction sum of squares

SELECTION OF VALIDATION METHODS

The literature suggests that all available methods of validation could be used. However, in this case, it is notpossible to use all the methods of validation .Therefore, the applicability of each method in terms of the

validation of the IRI model will be discussed and the most appropriate methods of validation will be selected.

The third method (Comparison with Previously Developed Models).The results of a newly developed model are compared with the previously developed model or with a theoretical

model. AASHO (IRI) Model is used to be compared with the developed IRI model.

Variable Mean Minimums Maximum

IRI 0.026214 0.002634 0.100567

PSI 4.144817 2.529651 4.949475


17/20

Siham E. Salih

17

Due to the time constraints and the difficulties of collection additional data to maximize the sample size, theabovementioned third method (Comparison with Previously Developed Models) is proposed to be used in the

validation process of the developed IRI model.

Validation Results

As previously mentioned, AASHO _IRI model was used in the validation process of the new IRI_ model .The

values of the IRI estimated by use of AASHO model are plotted against those obtained by the application of thenew developed model. This plot can be seen in Figure (13).

The relation between observed and estimated IRI can be found in the following form in eq. 12;

12. (IRI Observed) =-4.1632* LN (Developed IRI) + 0.557These findings seem to be in good agreement with the relation y= x. The results of checking the goodness of fit

for the relation between observed and estimated IRI model by using Chi-square test t- test and the distributionof errors ,these testing can be seen in the following paragraphs .

Goodness of Fit

To checking the goodness of fit for the predicted models. t test and Chi- square test were carried out and thefollowing results are expressed;

T-test :

n= 80 4, df = 159 confidence level = 95%

There is no reason to reject the null hypothesis.

Thus the difference is not significant.

X2test

n =80, d f = 79, confidence level = 95%

y = -4.1632Ln(x) + 0.557

R2 = 0.7752

-0.5

0

0.51

1.5

2

2.5

3

3.5

4

00.20.40.60.811.21.4

IRI Estimation

IRIObserve

Figure (13) Observed IRI Model versus Estimation IRI


18/20


18

For x2


19/20

Siham E. Salih

19

REFERENCES

1. W. N. Carey and P. E. Irick, The Pavement Serviceability Performance Concept.Highway ResearchBoard. Record250, 1960.

2. K. T. Hall and C. E. Correa, Estimation of Present Serviceability Index from International RoughnessIndex, Transportation Research Record, 1655 (1999), pp. 9399.

3. B. Al-Omari and M. I. Darter, Relationships Between International Roughness Index and PresentServiceability Rating,Transport Research Record, 1435 (1994), pp. 130136.

4. W. D. O. Paterson,Road Deterioration and Maintenance Effects. Washington, D.C.; The World Bank,1986. de Solminihac, Salsilli, Khler, and Bengoa 160 The Arabian Journal for Science and

Engineering, Volume 28, Number 2B. October 2003

5. AASHO, The AASHO Road Test;American Association of State Highway Officials Report 5 PavementResearch. Highway Research Board, 1962.

6. M. W. Sayers, T. D. Guillespie and W. D. Paterson, Directrices para Realizar y Calibrar Medicionesde Rugosidad de Carreteras, Technical Report of the World Bank46, 1986.7. I. Snchez, H. de Solminihac, El IRI: Un Indicador de la Regularidad Superficia l, Revista de

Ingeniera de Construccin,Chile, N6, 1989, pp. 117.

8. H. de Solminihac, Gestin de la Infraestructura Vial. Santiago; Ediciones Universidad Catlica deChile, 1998.

9. E. Khler, Diagnstico de Serviciabilidad de Pavimentos Urbanos en Chile, Tesis de Magister,Escuela de Ingeniera,Pontificia Universidad Catlica de Chile, Santiago, 1998.

10. M. Cundill, The MERLIN Low-Cost Road Roughness Measuring Machine, Transportation andRoad ResearchLaboratory, Research Report 301, 1991.

11. ARRB, User Manual for Two Laser Profiler. Melbourne, Australia, ARRB Transport Research LTD,1996.

12. SHRP, Distress Identification Manual for the Long-Term Pavement Performance, National ResearchCouncil, SHRP-P-338,Washington, DC, 1993.

13. T. F. Fwa and K. T. Gan, Bus-Ride Panel Rating of Pavement Serviceability,Journal TransportationEngineering, 115 (1989), pp. 176191.

14. B. G. Wildman, M. T. Erickson and R. N. Kent, The Effect of Two Training Procedures on ObserverAgreement and Variability of Behavior Ratings, Child Development, 46 (1975), p. 520.

15. S. W. Nair, R. Hudson, and C. E. Lee, Realistic Pavement Serviceability Equations Using The 690DSurface Dynamics Profiler, Austin, Texas; Center for Transportation Research. The University of

Texas at Austin, 1985.

16. Ahmed, Namir G. ,[2002] , " The Development of Models for the prediction of Thermal Cracking inFlexible Pavements ," Ph. D Thesis, Civil Engineering Department , College of Engineering ,

University of Baghdad , Iraq .

17. Kennedy , J. B.; & Neville A.M [1975],"Basic Statistical Methods for Engineers andScientists , 2

ndEdition, Thomas Y., Crowell Company, In, USA .

18. Keller, G.; & Warrack , B. [ 1997 ], " STATISTICS for Management and Economies , 4th Edition ,USA .


20/20


20

19. Ogden, K.W. Safer roadsA guide to road safety engineering.Avebury Technical, Aldershot, 1996.20. Rigden P.J. Skid resistance of roads and streets. CSIR Special Report Pad 64,Pretoria, 1988.

ANALYSIS OF PAVEMENT ROUGHNESS FOR THE AASHTO DESIGN METHOD IN PART OF BAGHDAD CITY

Documents