Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

This article was downloaded by: [Laurentian University]

On: 06 October 2014, At: 02:23


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Road Materials and Pavement Design


Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/trmp20

Influence of road profile obstacles on


road unevenness indicators
a
P. Múčka
a
Institute of Materials and Machine Mechanics, Slovak Academy of
Sciences, Račianska 75, 831 02 Bratislava 3, Slovak Republic
Published online: 21 Jun 2013.

To cite this article: P. Múčka (2013) Influence of road profile obstacles on road
unevenness indicators, Road Materials and Pavement Design, 14:3, 689-702, DOI:
10.1080/14680629.2013.811823

To link to this article: http://dx.doi.org/10.1080/14680629.2013.811823

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Road Materials and Pavement Design, 2013
Vol. 14, No. 3, 689–702, http://dx.doi.org/10.1080/14680629.2013.811823

Influence of road profile obstacles on road unevenness indicators


P. Múčka*

Institute of Materials and Machine Mechanics, Slovak Academy of Sciences, Račianska 75, 831 02
Bratislava 3, Slovak Republic

The study quantified the influence of the road profile obstacles on 27 different road uneven-
ness indicators (RUI). The influence of obstacle height, length and number of obstacles per
kilometre was assessed. The 2997 real longitudinal road-profile records from the Long-Term
Downloaded by [Laurentian University] at 02:23 06 October 2014

Pavement Performance program with considering asphalt concrete and cement concrete road
surfaces were processed. 5169 obstacles were identified by a median filtering approach and their
dimensions were quantified. The regression function based on linear coefficients for depen-
dence of the RUI on three obstacle parameters was derived. The marked dependence of the
road indicators on all three obstacle parameters was observed. The road indices describing the
short wavelength band proved to be the most sensitive for detecting obstacle presence in a
road profile.
Keywords: road surface; longitudinal profile; power spectral density; median filter; interna-
tional roughness index; ride number; obstacle; bump; pothole; straightedge

1. Introduction
1.1. Road unevenness indicators
The road surface quality, particularly its longitudinal unevenness, is an important factor for the
economy of motorised road transport. Various road indices are used for the description of lon-
gitudinal random road unevenness. A survey of contemporary road unevenness indicators (RUI)
used in Europe was given by Willet, Magnusson, and Ferne (2000). A comprehensive overview of
45 currently used unevenness indicators may be found in Boscaino and Praticò (2001) or Praticò
(2004). Survey of indicators used in the USA was provided by Sayers and Karamihas (1998).
The RUIs may tentatively be divided into two groups:

• direct: based on direct processing of the vertical road displacement data; for this group of
indicators, parameters relating to the power spectral density (PSD) of the road elevation
(ISO 8608; prEN 13036-5), statistics in particular waves bands (prEN 13036-5) or various
straightedge indices (Múčka, 2012b; Song and Hayhoe, 2006) may be used.
• indirect: based on translating the vibration response of a vehicle mechanical model (or a
measuring device) to the road roughness, taking into account the vibration response of the
road user or vehicle. From this group of indicators, the International Roughness Index (IRI)
proposed by Sayers (1995) and incorporated in technical standards (prEN 13036-5; ASTM E
1926-98) is well known. The IRI is based on summary statistics of the vibration response of

*Email: ummsmuc@savba.sk

© 2013 Taylor & Francis


690 P. Múčka

a quarter-car model (the relative velocity of suspension). The following indicators should
also be mentioned: Ride Number, Mays Ride Number, Mean Roughness Index, Profile
Index, Half-car Roughness Index, Average Rectified Velocity, Average Rectified Slope,
etc. (Sayers and Karamihas, 1996, 1998).

At present, the IRI (ASTM Standard E 1926-98) is the most popular single-number unevenness
indicator. Standard prEN 13036-5 provides various possible characterisations of the road profile
unevenness such as the IRI computation procedure, wave band analysis as well as the PSD
analysis. Standard ISO 8608 defines the processing and reporting of vertical road profile data by
using PSDs. The characteristics of the first distribution function, variance, skewness and excess
kurtosis as related to the road section’s Gaussianity were also investigated in this study.
Currently, the measured longitudinal road profile h(l) (m) is considered to be a particular
realisation of a random function H (l). Provided that this function is stationary, Gaussian, homo-
geneous and ergodic, it is fully described by its PSD function GH ()(m3 /rad), where  (rad/m)
Downloaded by [Laurentian University] at 02:23 06 October 2014

is the angular spatial frequency ( = 2π/L). The simplest form of the PSD may be expressed
as a line on a log–log chart, with the unevenness index C (radw−1 m3−w ) as the ordinate at the
reference point 0 (usually 0 = 1 rad/m) and the waviness w as the line slope. The simplest
approximation is of the form, GH () = C · −w (ISO 8608; prEN 13036-5). A comprehen-
sive overview of analytical models approximating measured PSDs was presented by Andrén
(2006).

1.2. Road obstacles


The properties or dimensions of obstacles in the real longitudinal road profiles are rarely reported
in the current literature. In Öijer and Edlund (2004), the frequency of potholes (sorted according
to depth) on US highways (2400 km) and Swedish roads (27 km) was quantified. Ferris (2001)
analysed about 106 obstacle events on road surfaces from highways to gravel roads to derive partial
components from which a global longitudinal profile encompassing obstacles may be derived.
Rouillard, Sek, and Bruscella (2001) analysed 415 km of Australian roads, proposing a Gaussian
distribution of transient phenomena.
The influence of obstacles on the indicators of the distribution function of the random road
elevation/acceleration was discussed in Rouillard and Sek (2001), Steinwolf, Giacomin, and
Staszewski (2002) or Steinwolf (2006). The application to the simulation of vehicle loading
processes was addressed in Rouillard and Sek (2001), Steinwolf and Connon (2005) or Bogsjö
and Rychlik (2009).
The simulation of the influence of obstacles on the road elevation PSD parameters (unevenness
index and waviness) was discussed in Kropáč and Múčka (2006). It was shown that the obstacles
changed the original straight-line form of the PSD in the log–log plot into two straight lines with
different waviness values. Only simulated sections with artificial obstacles were processed. The
considered obstacles were of analytical shapes (cosine and trapezoid) with approximated dimen-
sions. Ueckermann and Steinauer (2008) reflected the presence of obstacles in their somewhat
differently modified spectral-like representation.
Múčka and Kropáč (2009) analysed the statistical properties of the pure random component of
the longitudinal road profile based on 23,400 test sections from the LTPP database. Kropáč and
Múčka (2011a) presented the methodology for obstacle separation from longitudinal road profile
using median filtering.
This study deals with analysis of the real obstacle dimensions influence on the change of various
RUIs. This question has not been comprehensively addressed in the literature. The technique of
median filtering (Kropáč & Múčka, 2011a) was used for a detection of local obstacles and their
Road Materials and Pavement Design 691

dimensions. A small distortion effect of the median filter approach allows a clear separation of
the random part from the obstacle profile component. The study was intended to

(1) Quantify the relationships among the RUIs and the basic obstacle parameters – the height,
the length, and the number of obstacles per kilometre;
(2) Consider a sensitivity of the road indicators to obstacles;
(3) Consider the influence of particular obstacle dimensions on the road indicators;
(4) Consider the influence of road type (asphalt concrete (AC) vs. cement concrete (CC)) on
the road indicators.

2. Separation of obstacles from the longitudinal road profile by median filtering


For a random variable, the median is an alternative position indicator to the generally applied
mean value. Given a series {xj }, this sequence is ordered, i.e. it holds x(j) ≤ x(k) for j < k, j =
Downloaded by [Laurentian University] at 02:23 06 October 2014

0, 1, 2, . . . , 2N . Then, x(k) is the kth order statistics of the series {xj }. The mean value of the
ordered sequence {x(j) }, i.e. x(N ) is the estimate of its median at the jth position of the series {xj }.
When shifting the window along the original random sequence and ordering of the data in each
step, a median filter is obtained as the simplest case from a more general family of statistical
filters. It follows from the definition of the median filter that for the complete filtering of the
non-random component including m discrete data, the n ≥ 2m + 1 window length is needed. The
median filter works as a low-pass filter in a similar way as a linear filter.
There are two particularly important properties of median filters which are exploited when
trying to specify faults on longitudinal pavement profiles:

(1) Median filter removes large vertical irregularities (bumps, potholes, joints, vertical faults,
distresses, etc.) displayed along the track without affecting the random nature of the
surrounding parts of the road profile. This feature is a great advantage when compared
with the standard moving average filtering, where a distinct extreme value in the road
course is only partially reduced at the cost of affecting the surrounding parts of the random
profile part.
(2) Median filter, at the same time, preserves a single-vertical step to a new (quasi-) steady
level on the track (e.g. transition to a bridge) without affecting the random nature of the
surrounding parts of the road profile. On the contrary, the moving average filter partially
reduces the step height but simultaneously destroys the surrounding parts of the random
profile parts.

From the measured road profile h(lj ), the obstacle profile hO (lj ) was obtained in the
following form:
hO (lj ) = h(lj ) − hMF (lj ), (1)
where hMF (lj ) = medfilt1(h(lj ), n) is an estimate of the filtered random part, hR , (i.e. median-
filtered sequence using the nth order median filter). The program medfilt1 (Signal Processing
Toolbox® ) with the single parameter n as the filter window was applied.
Figure 1 shows the principle of generation of the median-filtered sequence hMF (lj ) from the
primarily given sequence h(lj ). Removing of both bump and pothole is indicated.
The window length for median filter was considered to be of order, n = 13 (nl = 13 ×
0.1524 m = 1.98 m), which approximately corresponds to the double of the estimated maximum
obstacle length (lOBmax ∼ 1 m) in the processed road sections. The minimal height of the detected
obstacle was considered to be hOBLIM = 0.3 cm and the uncertainty measure of obstacle profile
692 P. Múčka
Downloaded by [Laurentian University] at 02:23 06 October 2014

Figure 1. Principle of a median filter.

Figure 2. Scheme of the obstacle dimensions detection from the obstacle part hO of a median-filtered
profile.

hOBUNC = 0.05 cm was used (Figure 2). To indicate the obstacle dimensions more accurately, the
test sections sampled at 15.24 cm were re-sampled with a narrower sampling interval l = 1 cm,
while applying a linear interpolation. Three obstacle parameters were calculated – the height,
hOB , the length, lOB , and the number of obstacles per kilometre, NOB (Figure 2).
In total, 27 RUIs were estimated. The indicators defined in technical standards prEN 13036-
5 and ISO 8608 were preferred. The particular indicators were evaluated using the following
procedures:

• C (10−6 radw−1 m3−w ), w (–) – the parameters of the straight-line approximation of the PSD
were calculated in the wavelength band from 0.3534 to 90.909 m for 0 = 1 rad/m (ISO
8608; prEN 13036-5);
• DH (mm2 ) – variance of the road profile elevation was evaluated in the frequency domain
by integration of PSD in the wavelength band from 0.3534 to 90.909 m;
• RMSH (mm) – root mean square (rms) value of the road profile elevation was calculated
as a squared root of variance DH . This indicator is recommended in ISO 8608 for general
characterisation of the road profile;
Road Materials and Pavement Design 693

• IRI (mm/m) – International Roughness Index (IRI) was calculated according to the
procedure defined in Sayers (1995);
• RN (–) – Ride Number (RN) is based on the mathematical processing of longitudinal
profile to produce an estimate of subjective ride quality (Sayers & Karamihas, 1996). The
RN analysis method shares features with the IRI. It uses the same filtering method. The
RN is more sensitive to shorter wavelengths than the IRI. RN is defined as an exponential
transform of Profile Index (PI) according to the equation RN = 5exp(–160 PI);
• PI (mm/m) – Profile Index is the rms value of the filtered profile elevation and is transformed
to RN through the conversion equation (Sayers & Karamihas, 1996);
• 2AB , 2AF (mm) – the physical straightedge parameters also called double-amplitudes or
straightedge indices (Múčka, 2012b; Song & Hayhoe, 2006). The physical straightedge rests
on the two highest points of the pavement profile beneath the straightedge. The vertical
distance from the straight line representing the straightedge to each of the profile sample
points beneath the straightedge is computed. The maximum value of all the vertical distances
Downloaded by [Laurentian University] at 02:23 06 October 2014

is reported as the maximum deviation from the straightedge over its full length (2AF ) or
between the straightedge supports (2AB );
• 2ARSE (mm) – the rolling straightedge parameter (Song & Hayhoe, 2006); the maximum
deviation of the interior profile points from the straight line is found and reported as the
maximum deviation;
• MCD (mm) – the midcord deviation between the road profile and the rolling straightedge
(reference line) in the middle of the rolling straightedge (Farias and Souza, 2009).
• VA (mm/m2 ) – vertical acceleration represents the second derivative of the road profile
or the change of slope at a given point (Farias and Souza, 2009). The vertical acceleration
is equivalent to the MCD. The vertical acceleration equals an MCD multiplied by 2/b2
(b = base length, i.e. /2).

Three types of straightedge deviations (2AB , 2AF , 2ARSE ) are expressed for a processed road
section as the mean values per section length (denoted as < . >) and the MCD and VA variables
are expressed as the standard deviation, σMCD , and the root-mean-square value (RMSVA). The 3-m
lath ( = 3 m) shifted by 1.5 m (= /2) was used for all five straightedge indices calculation.

• Sk (–), Ex (–) – the routines (skewness, kurtosis) from the Matlab Statistical Toolbox© have
been used;
• CS2 , CL2 , wS2 , wL2 – were estimated by fitting a road elevation PSD in two sub-bands –
short (0.3534–6.28 m) and long wavelength bands (6.28–90.909 m) by two straightlines;
The unevenness indices were calculated for 0 = 1 rad/m;
• CS3 , CM3 , CL3 , wS3 , wM3 , wL3 – were estimated by fitting a road elevation PSD in three
sub-bands – short (0.78125–3.125 m), medium (3.125–12.5 m) and long (12.5–50 m),
respectively, which are defined in prEN 13036-5 for the wave bands indices; the unevenness
indices were calculated for 0 = 1 rad/m;
• LW, MW, SW (mm) – the particular rms values of the filtered profile in the short, medium,
and long waves bands were estimated in the spatial domain according to prEN 13036-5
Standard.

Thirteen of processed RUIs (DH , Sk, Ex, CL2 , wL2 , CS2 , wS2 , CL3 , wL3 , CM3 , wM3 , CS3 , and wS3 )
are not involved in various technical standards intended for road and airfield maintenance and
management purposes. The fitting of real road elevation spectrums revealed that approximation
by more than one straight line is better (Andrén, 2006; Kropáč & Múčka, 2008; Múčka, 2012a).
Thus, besides the single-waveband spectrum parameters (C, w), the two- and three- wavebands
694 P. Múčka

Table 1. Statistics of analysed road records from GPS experiments.

AC CC

Total number of analysed profile records 1888 1109


Number of profile records with at least one obstacle 1117 389
Mean profile record length (m) 145.66 145.60
Length of profile records with at least one obstacle (km) 162.703 56.647
Total number of obstacles (|hOB | > 3 mm) 4370 799
Number of obstacles per kilometre (km−1 ) 26.86 14.11
Number of obstacles per section 3.91 ± 3.14 2.05 ± 2.14
Obstacle height, |hOB | (cm) 0.470 ± 0.46 0.494 ± 0.624
Obstacle length, lOB (cm) 70.3 ± 12.7 74.5 ± 16.4
Distance of successive obstacles, dOB (m) 26.9 ± 23.1 34.0 ± 27.8
Downloaded by [Laurentian University] at 02:23 06 October 2014

parameters were calculated. Skewness and excess kurtosis were used as a measure of signal
Gaussianity. Road elevation variance (DH ) was estimated because the variance in defined spatial
range is sometimes used for road quality indication.
The road profile elevation PSD estimation was carried out by averaging 13 overlapped blocks
of road-profile section per 512 samples with the Matlab® function psd based on Welch’s mod-
ified, averaged periodogram method. Detrending on the mean value was used and the resulting
profile was windowed using a Cosine Digital Tapering Window (prEN 13036-5) of length 512
samples. The shifting of blocks with 32 samples was used because of a short signal record. The
obtained frequency resolution was Be = 0.08 rad/m. All computations were provided in Matlab®
environment.
Profile data for the left and right tracks from the various General Pavement Studies (GPS)
of the LTPP program was used for analysis (Karamihas, 2005). The test sections from GPS
experiments were pre-processed by a 30.48 cm moving average filter and stored with sampling
interval l = 0.1524 m. Several analysed sections (about 5%) were processed by a 30-cm moving
average and sub-sampled to 15 cm. The typical length of test sections is 145 m. The data from
years 1989 to 1998 were processed.
The statistics of analysed road records and detected local obstacles higher than 3 mm is given
in Table 1. In total, 2977 road records from left and right tracks were processed. The repeated
measurements at the same day were omitted. The total length of analysed sections was about
220 km. At least one obstacle higher than 3 mm was detected in about 60% of AC profiles and
in 35% of CC profiles. The similar mean height and length of obstacles were observed for both
types of surfaces. The substantial lower density of obstacles was observed for CC surfaces.

3. Processing of two test sections of different quality


First, two test sections of a different roughness quality were processed. The Section #1 (State:
California, Code 90012301, good quality, AC road surface, 145.5 m total length) was of good
quality (Figure 3(a)) with one pothole (hOB = 0.32 cm; lOB = 63 cm). In Section #2 (State Idaho,
Code 90072002, poor quality, AC road surface, 145.5 m total length), six potholes were detected
(Figure 3(b)). The mean obstacles dimensions were calculated as follows: hOBm = 0.47 cm;
lOBm = 63.5 cm, and dOBm = 21.9 m. In Figure 3, the original profile h (black line) and median-
filtered profile hMF (grey line) are plotted in upper row and an obstacle profile hO (Equation (1))
in lower row. The indicated obstacles are marked with black line.
Figure 4 shows the spectrum for a good quality section (Figure 4(a)) and a poor quality section
(Figure 4(b)). The spectrum fitting by straight lines in one-, two-, and three-wavelength bands is
Road Materials and Pavement Design 695

Figure 3. Detection of obstacles: (a) the good profile with one obstacle; (b) the poor profile with six
obstacles.
Downloaded by [Laurentian University] at 02:23 06 October 2014

Figure 4. Road elevation PSD of original profile with obstacles, h (black line) and of the pure random
component with removed obstacles, hR (gray line): (a) the good profile with one obstacle; (b) the poor profile
with six obstacles.

depicted in Figure 4 with dashed, solid, and dotted lines, respectively. The contribution of obstacles
to the total signal power in the short waves band ( > 1 rad/m, i.e. wavelengths, L < 6.28 m) is
apparent.
To estimate the RUIs change for the original profile with obstacles, h, in comparison with the
pure random part, hR , the variable δ (%) was introduced as follows:

RUI(h) − RUI(hR )
δ(%) = 100 × . (2)
RUI(hR )

The positive sign of variable δ presents the worse quality of an original profile, h, i.e. higher value
of particular RUI, in comparison with the pure random part, hR , and vice versa. Note that for
RN, the negative sign of δ presents the worse quality of an original profile. Higher value of RN
corresponds to a better state of the road. Similarly, it is, so for, the waviness in short waves band,
where lower waviness corresponds to worse quality in this spatial interval.
Table 2 presents the influence of the obstacles on particular RUIs. The changes in the PSD
indicators, C and w, are of small importance, because the single-line approximation of raw spectra
is not suitable for fitting as shown in Figure 4. The elevation variance, DH , and rms value, RMSH ,
are only slightly changed due to the major influence of the long waves on the total variance of
the road signal. The indices most influenced by obstacles were the short waves band parameters
696 P. Múčka

Table 2. The RUIs for two different sections (#1 – good quality (1 obstacle), #2 – pure quality (6 obstacles)).

Profile #1 Profile #2
# RUI hR h δ (%) hR h δ (%)

One-band indicators 1 C 0.23 0.52 126 0.15 1.01 573


2 w 2.83 2.67 −6 2.65 2.60 −2
3 DH 5.40 5.60 4 13.3 13.7 3
4 RMSH 2.32 2.36 2 3.64 3.71 2
5 IRI 0.675 0.958 42 0.819 1.316 61
6 RN 4.162 3.787 −9 4.444 3.515 −21
7 PI 1.146 1.736 51 0.737 2.203 199
8 < 2AB > 0.371 0.856 131 0.445 1.187 167
9 < 2AF > 0.689 1.199 74 0.690 1.650 139
10 < 2ARSE > 0.526 1.027 95 0.559 1.324 137
11 σMCD 0.761 1.084 42 0.48 0.996 108
Downloaded by [Laurentian University] at 02:23 06 October 2014

12 RMSVA 0.676 0.963 42 0.427 0.886 107


13 Sk −0.055 −0.162 195 0.333 0.314 −6
14 Ex 0.410 0.633 54 −0.271 −0.308 14
Two-band indicators 15 CL2 0.36 0.43 19 0.31 0.30 −3
16 CS2 0.28 0.92 229 0.08 2.02 2425
17 wL2 2.18 2.07 −5 2.89 2.92 1
18 wS2 2.91 2.94 1 2.34 2.91 24
Three-band indicators 19 CL3 0.36 0.39 8 0.65 0.62 −5
20 CM3 0.29 0.36 24 0.17 0.24 41
21 CS3 0.12 0.08 −33 0.04 0.62 1450
22 wL3 2.34 2.31 −1 2.58 2.61 1
23 wM3 2.61 2.40 −8 3.87 2.52 −35
24 wS3 2.33 1.09 −53 1.89 1.73 −8
25 LW 2.30 2.30 0 4.60 4.60 0
26 MW 0.83 0.91 10 0.61 0.65 7
27 SW 0.23 0.41 78 0.16 0.60 275

as unevenness indices CS2 , CS3 , or SW. The marked change in the RUIs was observed for all five
straightedge indices, which are sensitive to the short wavelength contents. The marked sensitivity
to obstacles was observed for PI, IRI, and RN too. The long wave band indicators as LW, CL2 ,
CL3 remained practically constant and the medium wave band indicators slightly increased.
Figure 5 presents the normalised obstacle shapes (hOBN = hOB / max(hOB ), lOBN =
lOB / max(lOB )) for six detected potholes in Section #2. The typical shapes of real road obstacles
were tackled in paper Kropáč and Múčka (2011b).

4. The RUIs dependence on the obstacle dimensions


The mean values of RUIs for the pure random part (hR ) and the original profile with obstacles
(h) for 1506 analysed road records with at least one obstacle in profile are presented in Table 3.
Mean percentage differences δ are added.
The impact on the RUIs mean values changes is similar as those presented for two analysed
profiles in Section 3. The largest impact on the RUIs was observed for short wave indicators
(CS2 , CS3 , SW) predominantly. The marked influence was calculated for IRI, PI, RN and all
straightedge indices. All those indices are markedly influenced by short wave contents. The long
wave band parameters remained practically unchanged. Skewness and excess kurtosis are only
slightly changed due to signal pre-processing by a moving average filter and longer sampling
interval. The tendencies in changes of RUIs are similar for both types of road surfaces.
Road Materials and Pavement Design 697
Downloaded by [Laurentian University] at 02:23 06 October 2014

Figure 5. Normalised shapes of 6 detected potholes in a poor quality surface – Section #2.

Table 3. Mean values of RUIs for 1506 real road records.

AC surfaces CC surfaces
# RUI hR h δ hR h δ
1 C 0.50 1.45 238 0.70 1.54 138
2 w 2.81 2.64 −6 2.78 2.64 −5
3 DH 46.95 47.74 3 25.49 26.44 6
4 RMSH 6.18 6.25 1 4.71 4.82 3
5 IRI 1.14 1.68 51 1.55 2.08 37
6 RN 4.12 3.45 −16 3.8 3.27 −14
7 PI 1.22 2.38 105 1.74 2.71 63
8 < 2AB > 0.66 1.41 120 0.90 1.63 87
9 < 2AF > 1.10 2.08 97 1.52 2.53 70
10 < 2ARSE > 0.90 1.69 96 1.28 2.03 63
11 σMCD 0.85 1.29 57 1.21 1.60 34
12 RMSVA 0.75 1.14 57 1.08 1.43 34
13 Sk −0.04 −0.05 −1 0.11 0.08 −2
14 Ex −0.46 −0.27 5 −0.21 0.09 39
15 CL2 0.70 0.80 16 1.46 1.68 16
16 CS2 0.45 2.82 691 0.74 2.81 310
17 wL2 3.01 2.92 −3 2.21 2.13 −4
18 wS2 2.68 2.80 4 2.77 2.83 2
19 CL3 1.44 1.50 5 2.81 3.02 6
20 CM3 0.57 0.77 43 1.04 1.36 34
21 CS3 0.47 1.67 229 0.57 1.34 164
22 wL3 2.90 2.88 −1 2.13 2.10 0
23 wM3 3.37 2.70 −21 2.97 2.47 −18
24 wS3 2.68 2.04 −25 2.63 2.27 −13
25 LW 0.92 1.03 13 1.33 1.47 12
26 MW 0.27 0.63 148 0.34 0.65 103
27 SW 0.50 1.45 238 0.70 1.54 138
698 P. Múčka

The analytical relationships between the mean obstacle dimensions (height, hOBm , length,
lOBm , and number of obstacles pre kilometre, NOBm ) and a percentage change of the RUIs for all
road records were also derived. The sensitivity to obstacle dimensions was evaluated through the
percentage change of the road indicator, δ (Equation (2)). The value of RUI(hR ) may be perceived
as an initial state indicator of road surface quality for profile without any local imperfections.
Preliminary data fitting revealed an appearance of linear terms and a negligible influence of
interaction and squared or cubic terms in the followed relationship. The general structure of the
expression for δ = f (hOBm , lOBm , NOBm ) should read as follows:

hOBm lOBm NOBm


δ = b1 + b2 + b3 . (3)
hOBmNOM lOBmNOM NOBmNOM

The obstacle parameters differ in dimensions, thus the weighting with nominal values was applied.
Downloaded by [Laurentian University] at 02:23 06 October 2014

As the nominal values, the mean values of detected obstacle dimensions in 1506 analysed records
were selected as follows: {hOBmNOM , lOBmNOM , NOBmNOM } = [0.47 cm, 70.3 cm, 26.86 km−1 ] for
1117 AC surfaces and [0.494 cm, 74.5 cm, 14.11 km−1 ] for 389 CC surfaces. The constants bi s
in Equation (3) correspond to the percentage change of RUI in case when particular obstacle
parameter mean value raises one time. This seems to be a suitable approach for the sensitivity
analysis of obstacle dimensions influence on the road indicators. For the section without obstacles,
variable δ should be equal to zero.

Table 4. Parameters of regression function (Equation (3)) for the change of the RUIs relating to the mean
obstacle dimensions – AC surfaces.

# RUI b1 std(b1 ) b2 std(b2 ) b3 std(b3 ) RMSE R2 δREF

1 C 30.9 4.2 72.7 5.1 109.5 2.2 94.3 0.535 213.1


2 w −14.4 0.4 8.7 0.4 −1 0.2 8.1 0.327 −6.7
3 DH 0.4 0.1 1.8 0.1 0.6 0.1 2.4 0.106 2.8
4 RMSH 0.2 0.1 0.9 0.1 0.3 0 1.2 0.106 1.4
5 IRI 8.1 0.9 16.8 1.1 24.7 0.5 21 0.540 49.6
6 RN −6.2 0.1 −3.1 0.2 −6.8 0.1 3.2 0.844 −16.1
7 PI 46.8 1.9 3.1 2.2 46.9 1 41.3 0.584 96.8
8 < 2AB > 24.8 2.3 44.4 2.7 42.5 1.2 50.1 0.364 111.7
9 < 2AF > 18.5 1.9 30.4 2.3 44.8 1 41.6 0.469 93.7
10 < 2ARSE > 21.2 1.7 21.7 2.1 49.9 0.9 38.9 0.573 92.8
11 σMCD 22.8 1.1 7.4 1.4 19.8 0.6 24.9 0.432 50
12 RMSVA 22.8 1.1 7.4 1.4 19.8 0.6 24.9 0.433 50
13 Sk −2.3 0.5 0.3 0.6 0.6 0.3 11.5 0.040 −1.4
14 Ex −3.2 1.8 0.7 2.2 −1 0.9 39.5 0.083 −3.4
15 CL2 2.1 0.5 11.6 0.6 −0.9 0.2 10.4 0.086 12.8
16 CS2 2 12.4 181.4 15 326.7 6.4 275.1 0.561 510.1
17 wL2 −0.4 0.1 −2.7 0.1 0.1 0.1 2.5 0.082 −3
18 wS2 −6.5 0.5 9.5 0.7 1.5 0.3 12 0.181 4.5
19 CL3 0.3 0.3 4.2 0.3 −0.4 0.1 5.6 0.041 4.1
20 CM3 8.1 0.7 21.1 0.9 7 0.4 16.3 0.222 36.2
21 CS3 −12 6.9 140.8 8.3 49.1 3.5 153.3 0.210 178
22 wL3 0 0.1 −0.9 0.1 0.1 0 1.5 0.047 −0.8
23 wM3 −4.2 0.4 −9.4 0.5 −6.5 0.2 8.4 0.417 −20.1
24 wS3 −28 0.9 13 1.1 −7.5 0.5 20.2 0.304 −22.6
25 LW 0.01 0.02 0.31 0.02 0.03 0.01 0.4 0.006 0.4
26 MW 3.1 0.3 6.9 0.4 1.2 0.2 6.6 0.143 11.2
27 SW 72 2.3 12 2.8 54.8 1.2 52.3 0.570 138.8
Road Materials and Pavement Design 699

The partial regression relationships were derived using robust regression function as imple-
mented in Matlab® (robustfit). The robustfit function uses reweighted least-squares algorithm
with the weights at each iteration calculated by applying the bisquare function to the residuals.
The identified constants of the regression function (Equation (3)) with standard deviations (std),
coefficient of determination (R2 ) and root mean squared error (RMSE) are presented in Tables 4
and 5. The weighted coefficients of determination were calculated according to the Pozzi, Di
Matteo, and Aste (2012). To better compare the RUIs changes, the reference δREF (Equation (3))
was issued in Tables 4 and 5, which were calculated for the mean obstacle dimensions.
The correlation between the empirical data and the fitting function is limited. The noisiness in
the calculated values of δ = f (hOBm , lOBm , NOBm ) is apparent in some cases. This noisiness may
be caused predominantly by:

• a variable sensitivity of RUIs to the different combinations of obstacle dimensions. The


different combinations of the mean obstacle dimensions should result to the same value of
Downloaded by [Laurentian University] at 02:23 06 October 2014

particular RUI;
• a different power ratio of the total road signal to the pure random part;
• the broad scale of detected obstacles shapes.

The identified parameters in Tables 4 and 5 show often the similar impact of the obstacle height,
length, and number of obstacles on δ. The obstacle height increased δ values except for waviness

Table 5. Parameters of regression function (Equation (3)) for the change of the RUIs relating to the mean
obstacle dimensions – CC surfaces.

# RUI b1 std(b1 ) b2 std(b2 ) b3 std(b3 ) RMSE R2 δREF

1 C 36 7.6 95.4 9.6 53.7 2.4 93.7 0.554 185.1


2 w −20.7 0.6 12.2 0.8 −0.1 0.2 7.9 0.477 −8.6
3 DH 0.2 0.2 2.6 0.3 0.3 0.1 3 0.080 3.1
4 RMSH 0.1 0.1 1.3 0.1 0.2 0 1.4 0.076 1.6
5 IRI 8.5 1.7 23 2.1 12.4 0.5 20.4 0.552 43.9
6 RN −6.6 0.3 −3.9 0.4 −3.4 0.1 3.5 0.835 −14
7 PI 54.6 3.4 9.1 4.3 23.2 1.1 41.5 0.617 86.9
8 < 2AB > 32.5 4.1 56 5.2 20.2 1.3 50.3 0.356 108.8
9 < 2AF > 44.5 3.3 22.9 4.1 22.2 1 40.2 0.550 89.6
10 < 2ARSE > 39.4 3 20.2 3.8 24.2 1 37.4 0.619 83.8
11 σMCD 36.6 2.1 1.5 2.7 10.3 0.7 26 0.533 48.5
12 RMSVA 36.6 2.1 1.5 2.7 10.3 0.7 26 0.534 48.4
13 Sk 2.2 2.1 −4.2 2.6 0.1 0.7 25.7 0.048 −1.9
14 Ex −0.2 3 −2.7 3.8 −0.8 1 37.4 0.052 −3.6
15 CL2 1.8 1 15.5 1.2 −2 0.3 12.1 0.145 15.4
16 CS2 12.7 24.8 216.6 31.4 187.3 7.9 306.3 0.613 416.6
17 wL2 −0.5 0.2 −3.5 0.3 0.4 0.1 2.8 0.142 −3.5
18 wS2 −6.7 1 9.5 1.2 1 0.3 11.8 0.186 3.8
19 CL3 −0.1 0.5 4.5 0.6 −0.1 0.2 6 0.027 4.4
20 CM3 11.9 1.3 21.9 1.7 4.3 0.4 16.2 0.373 38.2
21 CS3 −16.4 12.9 176.9 16.4 24.5 4.1 159.6 0.259 185
22 wL3 0.1 0.1 −0.9 0.2 0 0 1.5 0.027 −0.8
23 wM3 −2.9 0.7 −12.5 0.9 −3.6 0.2 8.5 0.434 −19
24 wS3 −33.5 1.7 15.1 2.1 −3 0.5 20.8 0.340 −21.3
25 LW −0.02 0.04 0.38 0.05 0.04 0.01 0.5 0.024 0.4
26 MW 3.7 0.5 8.4 0.7 −0.1 0.2 6.4 0.217 12
27 SW 78.1 4.1 24.3 5.2 28.1 1.3 50.8 0.618 130.5
700 P. Múčka

parameters. The waviness decreases predominantly in the short wave band, which is caused
by increase of signal power in the short wave band. The largest influence from three obstacle
parameters has the obstacle height for w, wS3 , σMCD , RMSVA. In some cases, the influence of
obstacle height is similar as the influence of obstacle length or number of obstacles. It was observed
for RN, PI, < 2AB >, < 2AF >, and < 2ARSE >.
The increased obstacle length increased all RUI values. The marked influence of obstacle length
was observed for all spectrum parameters in two- and three-wave bands and for MW and < 2AB >.
This was caused by modulation of spectrum in short wave band. The local contribution to the total
signal power in this frequency band depends on the mean obstacle length (Kropáč and Múčka,
2006). The obstacle length contributes to the total signal power at frequency  ∼ 2π/lOBm .
Number of obstacles had the marked influence on about half of analysed RUIs. The higher num-
ber of obstacles increased the RUIs predominantly for the short wave band indices (CS2 , CS3 ),
further for IRI, RN, PI, SW and for all straightedge indices. All straightedge indices are sensitive
to the short wavelength contents. Straightedge indices present the mean value of maximum devi-
Downloaded by [Laurentian University] at 02:23 06 October 2014

ations between straightedge and profile, thus the higher density of obstacles cause higher index
value. IRI and RN are most sensitive to the short wavelengths. These indicators correspond to the
accumulated relative motion of vehicle model suspension, which rises with number of obstacles.
The length of obstacle is an important factor for the impulse excitation of a motor vehicle. The
particular resonance frequencies (f ) of vehicle are connected with a wavelength by relationship,
f (Hz) = v (m/s)/L (m), where v is vehicle speed and L is wavelength. The number of obstacles
is important from the point of view of transient vibration events in the vehicle response due to
impulse excitation. The smaller sensitivity of some RUIs to those obstacle dimensions should
limit the RUIs correlation to the real vehicle vibration response markedly.
The calculated values of δREF for predefined constant mean obstacle parameters shows a marked
influence of obstacles presence on RUIs in short wave band (SW, CS2 , CS3 ). IRI, PI, and all five
straightedge indices have a similar sensitivity to the obstacles. Local obstacles partially influence
the medium wave band indicators. The PSD parameters in the long wave band remained practically
constant as would be expected. The values δREF for the single-band PSD indicators, C and w,
should be ignored because the one-line approximation of raw spectra is not suitable as shown in
Figure 4.
The highest correlation was observed for RN, IRI, PI, SW, and all five straightedge indices.
Seven RUIs (26%) for AC road surfaces and 10 RUIs (37%) for CC surfaces indicate R2 > 0.5,
which presents a moderate correlation between observed data and model according to Equation (3).

5. Conclusions
The main findings may be summarised as follows:

(1) This study is one of the first attempts to quantify the influence of the real obstacle
dimensions on the currently used RUIs.
(2) A three-parameter fitting relationship for linear dependence between the RUIs change
and three-obstacle parameters was derived. The acceptable correlation (R2 > 0.5) was
obtained for this model and for about 30% of estimated road indicators.
(3) The study often indicates a similar influence of the obstacle height, length, and density on
the RUIs. Not only obstacle height and density, but obstacle length also had the marked
impact on the RUIs change.
(4) The most sensitive unevenness indicators for the presence of obstacles were identified as
the road elevation rms value SW and the PSD parameters in the short waves band. All
Road Materials and Pavement Design 701

observed straightedge indices were markedly influenced by obstacles presence. The good
sensitivity was observed for the IRI, RN, and PI.
(5) RUIs that cover too broad wavelength interval (DH , RMSH , LW, MW, and spectrum indica-
tors in long- and medium-wave bands) are not possible to detect the presence of obstacles
in profile.
(6) The short wave PSD parameters have the advantage of having a closer relation to the
vehicle vibration response. The PSD parameters may be used directly to calculate the
vehicle vibration response or for generation of the artificial road sections.
(7) Only marginal differences between results for AC and CC road surfaces were observed.
This may be partially caused by applying a moving filter on the road data.
(8) The inverse results may also be used. From the change in a particular RUI value, the mean
obstacle dimensions may be estimated.

Limitation of the present results was the noisiness of the estimated dependences which was
Downloaded by [Laurentian University] at 02:23 06 October 2014

affected by the variability of processed test sections and obstacles and by the variable sensitivity
of RUIs to the different combinations of obstacles and their dimensions. Further, the results are
dependent on the sampling interval and elevation data pre-processing.

Acknowledgements
This work has been partially undertaken within the grant no. 2/0058/13 of the VEGA Grant Agency of
Slovak Academy of Sciences. The support of the agency is acknowledged. The author highly appreciates
the support of the LTPP Customer Service Team. The author is indebted to Dr Thomas P. Gunston for help
with English language correction.

References
Andrén, P. (2006). Power spectral density approximations of longitudinal road profiles. International Journal
of Vehicle Design, 40(1), 2–14.
ASTM Standard E 1926-98. (2003). Standard practice for computing International Roughness Index of
roads from longitudinal profile measurements. West Conshohocken, PA: ASTM International.
Bogsjö, K., & Rychlik, I. (2009). Vehicle fatigue damage caused by road irregularities. Fatigue and Fracture
of Engineering Materials and Structures, 32(5), 391–402.
Boscaino, G., & Praticò, F. G. (2001). A classification of surface texture indices of pavement surfaces.
Bulletin des Laboratoires des Ponts et Chaussees, 234, 17–34.
Farias, M. M., & Souza, R. O. (2009). Correlations and analyses of longitudinal roughness indices. Road
Materials and Pavement Design, 10(2), 399–415.
Ferris, J. B. (2001). Singular value decomposition of road events into characteristic shapes. Dynamic Systems
and Control Division DSC, ASME, 70, 121–126.
ISO 8608. (1995). Mechanical vibration – Road surface profiles – Reporting of measured data.
Karamihas S. M. (2005). Long-term pavement performance road profile data. Retrieved January 17, 2005
from http://www.umtri.umich.edu/erd/roughness/ltpp_erd.html
Kropáč, O., & Múčka P. (2006). Effect of obstacles on the indicators of road unevenness. Proceedings of
the IMechE, Part D: Journal of Automobile Engineering, 220(10), 1333–1343.
Kropáč, O., & Múčka, P. (2008). Indicators of longitudinal unevenness of roads in the USA. International
Journal of Vehicle Design, 46(4), 393–415.
Kropáč, O., & Múčka P. (2011a). Specification of obstacles in the longitudinal road profile by median
filtering. Journal of Transportation Engineering–ASCE, 137(3), 214–226.
Kropáč, O., & Múčka, P. (2011b). Shapes of obstacles in the longitudinal road profile. Shock and Vibration,
18(5), 671–682.
Múčka, P. (2012a). Longitudinal road profile spectrum approximation by split straight lines. Journal of
Transportation Engineering–ASCE, 138(2), 243–251.
Múčka, P. (2012b). Relationship between International Roughness Index and straightedge index. Journal of
Transportation Engineering–ASCE, 138(9), 1099–1112.
702 P. Múčka

Múčka, P., & Kropáč, O. (2009). Properties of random component of longitudinal road profile influenced by
local obstacles. International Journal of Vehicle Systems Modelling and Testing, 4(4), 256–276.
Öijer, F., & Edlund, S. (2004). Identification of transient road obstacle distributions and their impact on
vehicle durability and driver comfort. Vehicle System Dynamics, 41(Suppl.), 744–753.
Pozzi, F., Di Matteo, T., & Aste, T. (2012). Exponential smoothing weighted correlations. The European
Physical Journal B, 85(6), 21, Article no. 175.
Praticò, F. (2004). Nonstrictly-ergodic signals in road roughness analyses: A theoretical and experimental
study. Proceedings of 2nd international congress SIIV 2004, 27–29 October, University of Florence,
Florence.
prEN 13036-5. (2006). Road and airfield surface characteristics – Test methods – Part 5: Determination of
longitudinal unevenness indices. Brussels: CEN.
Rouillard, V., & Sek, M. A. (2001). Simulation of non-stationary vehicle vibrations. Proceedings of Institution
of Mechanical Engineers, Part D: Journal of Automobile Engineering, 215(10), 1069–1075.
Rouillard, V., Sek, M. A., & Bruscella, B. (2001). Simulation of road surface profiles. Journal of
Transportation Engineering–ASCE, 127(3), 247–253.
Sayers, M. W. (1995). On the calculation of IRI from longitudinal road profile. Paper No. 950842.
Downloaded by [Laurentian University] at 02:23 06 October 2014

Washington, DC: Transport Research Board, 24.


Sayers, M. W., & Karamihas, S. M. (1996). Interpretation of road roughness profile data (Report UMTRI-
96-101). University of Michigan, Transportation Research Institute.
Sayers, M. W., & Karamihas, S. M. (1998). The little book of profiling. Ann Arbor, MI: University of
Michigan, Transportation Research Institute.
Song, I., & Hayhoe, G. F. (2006). Airport pavement roughness index relationships using the Federal Aviation
Administration (FAA) profiling system. Proceedings of 2006 airfield and highway pavement specialty
conference (pp. 741–752). Reston, VA: ASCE.
Steinwolf, A. (2006). Closed-loop shaker simulation of non-Gaussian random vibrations – Part 1: Discussion
and methods. Test Engineering and Management, 68(3), 10–13.
Steinwolf, A., & Connon III, W. H. (2005). Limitations of the Fourier transform for describing test course
profiles. Sound and Vibration, 39(2), 12–17.
Steinwolf, A., Giacomin, J. A., & Staszewski, W. J. (2002). On the need for bump event correction in vibration
test profiles representing road excitations in automobiles. Proceedings of Institution of Mechanical
Engineers, Part D: Journal of Automobile Engineering, 216(4), 279–295.
Ueckermann, A., & Steinauer, B. (2008). The weighted longitudinal profile. A new method to evaluate the
longitudinal evenness of roads. Road Material and Pavement Design, 9(2), 135–157.
Willet, M., Magnusson, G., & Ferne, B. W. (2000). FILTER experiment – Theoretical study of indices.
FEHRL Tech. Note 2000/02, Crowthorne, Berkshire: Transport Research Laboratory.

You might also like