See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/236838812

A numerical method to calculate the Abbott parameters: A wear application

Article in Tribology International · September 2007

DOI: 10.1016/j.triboint.2006.12.007

CITATIONS READS

38 851

2 authors:

Maxence Bigerelle Iost Alain

Université Polytechnique Hauts-de-France Ecole Nationale Supérieure d'Arts et Métiers LILLE
206 PUBLICATIONS 5,967 CITATIONS 251 PUBLICATIONS 4,670 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Steel coils annealing View project

Roughness View project

All content following this page was uploaded by Iost Alain on 03 June 2019.

The user has requested enhancement of the downloaded file.

 ARTICLE IN PRESS

Tribology International 40 (2007) 1319–1334

www.elsevier.com/locate/triboint

Review

A numerical method to calculate the Abbott parameters:

A wear application
M. Bigerellea,b, A. Iostb,
a
Laboratoire Roberval, FRE 2833, UTC/CNRS, Centre de recherches de Royallieu BP20529, 60205 Compiègne, France
b
Laboratoire de Métallurgie Physique et Génie des Matériaux, Equipe CPP, ENSAM, USTL, CNRS, 8 Boulevard Louis XIV, 59046 Lille, France
Received 18 April 2002; received in revised form 16 November 2005; accepted 31 December 2006
Available online 2 May 2007

Abstract

A numerical technique was proposed to plot the Abbott curve and to compute its associated parameters defined by the DIN 4776 and
ISO 13565 norms. These parameters were then extended and applied to non-sigmoid Abbott curves. By studying the discretisation errors,
we show that a minimum of 200 intercepts, with parabolic interpolations between discretised data profiles, have to be taken into
consideration to calculate the parameters as accurately as possible. Experimental profiles were eroded by means of a numerical wear
model, and it was shown that the Abbott parameters correlate well with the wear model parameters. Our numerical estimations of
Abbott parameters were performed for electro-eroded, tool machined, polished, worn and sandblasted surfaces. Manual measures were
compared with our algorithmic method and it was shown that the difference is lower than 1% for Mr1 and Mr2 Abbott parameters, but
the numerical technique leads to a lower dispersion.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Roughness; Abbott; Algorithm; Wear; Profilometry; Numerical modelling

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1320
2. The Abbott curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1320
2.1. Asperities’ height distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1320
2.2. The Abbott curve parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1320
2.3. Remarks on the ISO 13565 normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1320
3. Materials, surface machining and roughness recording . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1321
3.1. Electro-erosion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1321
3.2. Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
3.3. Shot peening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
3.4. Wear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
3.5. Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4. Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4.1. Determination of parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4.1.1. Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4.1.2. The Abbott curves’ linear part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4.1.3. Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1322
4.1.4. Intercept calculation related to the Abbott curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1323

Corresponding author. Tel.: +33 3 2062 2233; fax: +33 3 2062 2957.
E-mail address: [email protected] (A. Iost).

0301-679X/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.triboint.2006.12.007
 ARTICLE IN PRESS
1320 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

4.1.5. t-Student computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324

4.1.6. Minimal percentage on the Abbott curves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324
4.1.7. Intercept numbers on the Abbott curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1326
4.2. The extended Abbott parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1326
4.3. Experimental profiles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1327
4.3.1. Electro-erosion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1327
4.3.2. Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1328
4.3.3. Shot peening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1328
4.3.4. Wear of hip prosthesis head . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1329
4.3.5. Polished surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1329
4.4. Simulated profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1330
4.5. Wear analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1330
4.5.1. Wear simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1330
4.5.2. Wear results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1331
4.6. Manual-computer comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1331
4.6.1. Methodology comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1331
4.6.2. Results from comparison. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1334
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1334
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1334
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1334

1. Introduction presented in Fig. 2(a) for an experimental electro-eroded

profile and Fig. 2(b) for its related Abbott curve.
The Abbott curves [1] are usually used to quantify wear
phenomena such as lubricant influence, bearing materials
2.2. The Abbott curve parameters
[2] or surface texture [3]. The roughness parameters
computed from these curves are defined by the DIN 4776
Parameters computed from the Abbott curves are
norm [4] superseded by ISO 13565 [5]. The present
represented in Fig. 2b. Let F(x) be the equation of the
development of methods (spectral, fractal, wavelet et al)
straight line that best adjusts the linear part of the Abbott
and their implantation led us to connect a computer to curve, and let us define c1 and c2 by c1 ¼ F(0) and
roughness profilometers and to compute numerical estima- c2 ¼ F(100). Finally, let us denote by A(x) the Abbott curve
tions of the roughness parameters. This way, we developed
equation and by A1(x) its reciprocal application. The
an algorithm to plot the Abbott curve and to compute the
Abbott parameters are defined in Table 1. All these
norm parameters. The discretisation of the Abbott curves
parameters depend on the F(x) equation that represents
is studied and the efficiency of the algorithm is tested on
the best-fitted line calculated in the ‘‘linear part’’ of A(x).
various kinds of profiles.
The main problem consists in finding the interval [a,b]
where A(x) could be considered as linear.
2. The Abbott curves

First, we briefly report on the Abbott roughness 2.3. Remarks on the ISO 13565 normalisation
parameters defined by the DIN 4776 and the ISO 13565
normalisation. The Abbott curves correspond exactly to 1f(x), where f
is the cumulative density function (CDF) of surface height,
2.1. Asperities’ height distribution provided the profiles are considered as a continuous
function. However, for a discretised profile, the results
Asperities’ height distribution (AHD) is the ratio (in could be different according to the method used to
percentage) between the length intercepted by the roughness calculate the length intercepted by the roughness profile
profile and the scan length L (Fig. 1) at a given height, called and the intercepted length c. It will be then judicious
c (cA[Rmin, Rmax] where Rmin is the minimal amplitude of the to use some parametric estimations (like maximal like-
profile and Rmax is the maximal one), from the reference line lihood estimation), if a probability density function (PDF)
which is the least square line of the profile: cannot be rejected, or a non-parametric one (like kernel
X density) to estimate the CDF in the other cases. The
AHDðcÞ ¼ 100 ðLi =LÞ, (1) norm specifies that a straight line must be fitted to the
i central line of the PDF corresponding to a 40% interval
where Li(c) are all the segments of the intersection of the length. No mathematical or statistical justification was
profile y ¼ f(x) with the intercept y ¼ c. The Abbott curve is given for this interval range. To obtain a linear Abbott
finally obtained by plotting c versus AHD(c). An example is curve in the whole interval meaning that A(x) ¼ F(x) ¼ a
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1321

15 4
Roughness amplitude (μm)

L9
L4 3
10
L7
L3 L5 L10 2

Amplitude height
L1
5 L2 L6 L11
L8 1
C
0 0

Density heights profile

4
3
2
-1 1
-5 0
-2 -1
-2
-3
-10 -3 -4
0 250 500 750 1000 1250 1500
0 100 200 300 400 Number of points
-4
Profile length (μm) 0 20000 40000 60000 80000 100000
Fig. 1. Roughness profile and description of the method used to plot the Number of points
Abbott curve from the DIN 4776 or ISO 13565 norms.
Fig. 3. Gaussian cumulative density function (CDF) of a simulated
Gaussian profile discretised in 100 000 points. The curve represents the
Gaussian theoretical CDF and the straight line the 40% linear part
calculated from ISO 13565 norm. The histogram represents the PDF of
15 the profile height.
Roughness amplitude (μm)

10
5
0 x+b 8xA[0, 100], the stationary signal has to be a
-5 triangular one mathematically speaking. The domain of
-10
the linear assumption can then be characterised as an
amplitude range where roughness can be approached by a
-15
triangular signal. If we suppose that the amplitude signal
-20 gets a Gaussian PDF (the most current PDF met in
0 2000 4000 6000 8000
tribology), then no linear part exists in the Abbott curve.
Profile length (μm)
However, the linear assumption can be a good approxima-
tion over a wide range. To visualise the results of the
15 approximation, a Gaussian profile (discretised in 100 000
Roughness amplitude (μm)

10 Rpk
points) is simulated, the CDF is calculated and then the
5
A1 C1 linear part is plotted according to the ISO 13565 norm. As
can be shown in Fig. 3, the central linear part is a good
0 Rk
approximation of the real PDF on the 40% interval length.
C2
-5 This approximation can be justified by the following
-10 Rvk remarks:
A2
-15
Mr1 Mr2
-20  A linear relation is no unreasonable approximation for
0 20 40 60 80 100 the CDF central parts.
Asperity height distribution (%)  The central part represents the maximal mass of the
material and a linear relation could involve mathema-
Fig. 2. An electro-eroded profile obtained by a tactile profilometer (a) and
tical simplifications of wear models (or more generally
the corresponding curve with definitions of the DIN 4776 or ISO 13565
norms parameters (b). physical models) based on the interaction of the surface
with a physical process [6,7].

Table 1 3. Materials, surface machining and roughness recording

Definition of Abbott parameters

Definition Formula
To test whether our algorithm holds or not, surfaces
were machined using various processes.
A1 Quantity of solid peaks R A1 ðc1 Þ (2)
A1 ¼ ðAðxÞ  c1 Þ dx
R0100
Quantity of solid valleys A2 ¼  Að100ÞÞ dx (3)
A1 ðc2 Þ ðAðxÞ
A2
3.1. Electro-erosion
Mr1 Threshold as the minimal AHD Mr1 ¼ A1 ðc1 Þ (4)
Mr2 Threshold as the maximal AHD Mr2 ¼ A1 ðc2 Þ (5)
Kernel or core roughness depth Rk ¼ c1c2 (6) 35NC16 steel sheets (0.35% C, 4% Ni, Cr) were cut by
Rk
‘‘Reduced’’ height peak amplitude Rpk ¼ A(0)c1 (7) electro erosion. All process parameters were unchanged
Rpk
‘‘Reduced’’ height valley amplitude Rvk ¼ c1A(100) (8) except for the pulse times, which are, respectively, 1.5 and
Rvk
2.5 m s for samples (el_a) and (el_b). Profiles were recorded
 ARTICLE IN PRESS
1322 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

using a mechanical stylus profilometer with 8 mm working 4.1.2. The Abbott curves’ linear part
length discretised in 8000 points. 4.1.2.1. Linear least-square methods (LLSM). Let n be
the number of points of the Abbott curve, xi the abscissa
3.2. Turning (in percentage) of the ith point and yi its ordinate (asperity
height). The coefficients a and b of the equation
Ten 35NCD4 steel sheets (0.35% C, 1% Ni, Cr, Mo) F(x) ¼ a+bx were calculated by the linear-least square
were machine-turned through the following process: five method with a residual variance s2 given by
with a feed rate of 0.1 mm/rev (U_a) and five with a feed X
n

rate of 0.35 mm/rev (U_b). Twenty profiles were recorded s2 ¼ ðyi  y0i Þ2 =n  2, (9)
i¼1
perpendicular to the grooves using a mechanical stylus
profilometer with 4 mm working length discretised in 8000 where y0i ¼ a þ bxi are the predicted values and the slope
points. variance estimator s2b is given by
X
n
3.3. Shot peening s2b ¼ s2 = ðxi  x̄Þ2 , (10)
i¼1
Tin base samples (0.95% Sn, 0.05% Sb) E5 were P
where x̄ ¼ n ðxi =nÞ.Following the Gauss–Markov hypoth-
sandblasted (53/106 mm) under two different environmental eses [8] and if the residual yi  y0i obeys a Gaussian PDF, the
conditions: E2 in dry environment and E3 in humid random value (bb)/sb (where b is the unknown value of the
environment. Thirty profiles chosen at random were slope and sb its standard deviation) obeys a Student PDF
recorded from each sample using a mechanical stylus with u ¼ n  2 degrees of freedom. The confidence interval
profilometer with 4 mm working length discretised in 8000 of the slope at the 1a level is defined by
points.
b  t1a=2;n2 sb pbpb þ t1a=2;n2 sb , (11)
3.4. Wear where tx,k is the fractile of the Student PDF with k degrees
of freedom and a the significant level of x.
Two regions of a Ti-6Al-4V hip prosthesis head,
explanted from a human body after 10 years, were named 4.1.2.2. Analysis of the LLMS. Eq. (11) is only theore-
TF (part of the prosthesis where friction is higher) and TB tically valid if the Gauss–Markov hypotheses are not
(lower friction area). Using a scanning confocal micro- violated and if the residual follows a Gaussian PDF.
scope. 280 profiles were recorded The 85 mm scaling length It can be shown that the Gauss–Markov hypotheses are
is discretised in 800 points. not respected, meaning that the LLSM method cannot
formally be applied. However, using the Bootstrap regres-
3.5. Grinding sion, it can be shown that these violations do not affect the
determination of the linear part of the Abbott curve. The
Ti-6Al-4V samples were polished using carborundum values of the coefficients a and b of the straight line are
disks 80 (P80), 1200 (P1200) or 4000 (P4000). The roughness unaffected; so are the Abbott parameters. Then, the LLSM
was recorded using a mechanical stylus profilometer with method can be applied without any restriction.
4 mm working length discretised in 8000 points.
4.1.3. Algorithm
4. Results and discussion
Let M be the number of intercepts used to calculate the
asperities’ height distribution. Let bxi ;xj with xioxj be the
4.1. Determination of parameters
slope obtained by regression analysis using the set of points
ððxi ; yi Þ; ðxiþ1 ; yiþ1 Þ; . . . ; ðxj ; yj ÞÞ with i 2 ½1; Mð1  wÞ and
4.1.1. Principle
j 2 ½i þ 1; Mw where w (wA[0..1]) represents the minimal
The algorithm used to calculate the linear part of the
ratio of adjacent points used to calculate the slope. In the
Abbott curve, F(x), could be summed up by five steps:
case of the ISO Norm, w ¼ 0.4 meaning that 40% of the
interpolated line will be used to compute F(x). (This ratio
(i) A minimal set of adjacent points of A(x) is chosen to
imposes a minimal critical length to avoid detection of
calculate F(x).
abnormal small linear part.) All the configurations bxi ;xj are
(ii) Linear least-square adjustment is computed.
computed with their confidence intervals and the ‘‘best’’
(iii) Slope precision is quantified by using an appropriate
slope is obtained when the range of confidence interval
statistical method (Student’s confidence interval).
(Eq. (11)) is minimised.
(iv) An adjacent point is further added and steps (ii), and
(iii) are reiterated. Remark. as an adjacent point is added or deleted during
All configurations are tested to find the set of points this algorithm, it is possible to optimise the computation
that leads to the best regression coefficient slope: time by using a recursive method to calculate the regression
(v) The Abbott parameters are calculated. equation.
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1323

4.1.4. Intercept calculation related to the Abbott curves (xi, yi) and (xi+1, yi+1), with yi1oyiocoyi+1, are
The intercepted length Ll can be defined by the following interpolated by a parabola. To calculate the intersec-
set of points: tion, we first evaluate the equation F(x) ¼ ax2+bx+d,
the two roots of the equation ax2+bx+d ¼ c are
Ll ¼ fðxl ; yl Þ; ðxk ; yk Þ; xk 4xl =yl ¼ c,
calculated and the solution xleft is the root included in
yk ¼ c; yi 4c; i 2 ½l þ 1::k  1; Ll ¼ xk  xl g. ð12Þ [xi, xi+1]. In the same way, the root xright included in
the interval [xj, xj+1] is calculated with [xj1, yj1],
[xj, yj] and [xj+1, yj+1] with yj14yj4c4yj+1.
4.1.4.1. Different methods. As the roughness profile is
discretised by the recording device, experimental points
 Fourth method (L3): In this method, (xi2, yi2),
(xi1, yi1), (xi, yi) and (xi+1, yi+1), where yi2oyi1o
with exactly the same amplitude as the intercept are rarely
coyioyi+1, are interpolated by a third-degree poly-
obtained. More precisely, it is unusual to get the conditions
nomial curve (Fig. 4d).
yl ¼ c, yk ¼ c and, consequently, an interpolation of
experimental points must be implemented. Four methods
4.1.4.2. Analysis of the intercepted peak length. At first,
can then be applied (Fig. 4) to interpolate intersection
peak length error has to be analysed. For this purpose,
between intercept and profile:
different intercepted peak lengths estimated by the four
methods are applied on the 800 experimental profiles
 First method (L0) considers the upper point (xl, yl) of
described in Section 3. As the sampling length depends on
the profile for the beginning of the intercept and the
the class of profiles, we shall normalise the error on the
lower point (xk, yk) for the end of the intercept (Fig. 4a).
length determination in sampling length unit. With this
In mathematical terms this relation is given by
( ) unscaled unit, the error is bound to 2. In the same
ðxl ; yl Þ; ðxk ; yk Þ; xk 4xl ; yl 4c; yl1 oc; yk 4c; way, to normalise the influence of the profile amplitude
Ll ¼ .
ykþ1 oc; yi 4c; i 2 ½l þ 1::k  1; Ll ¼ xk  xl on the peak length precision, the amplitude is divided by
the range roughness. Then, we calculate the mean error for
(13) the four methods. The differences between the length
obtained by method L3 (it is obvious that method L3
 Second method (L1): From the two adjacent points should be considered as the best one in terms of
(xl+1, yl+1) and (xl, yl) satisfying Eq. (13),the linear interpolation) and the others (L3L0, L3L1 and
equationD0 ¼ a0 x+b0 is calculated, xleftis defined by L3L2) allow us to analyse the error for 20 normalised
xleft ¼ ðc  b0 Þ=a0 . With the same method, xright is heights of intercept corresponding to all experimental
obtained with (xk, yk) and (xk+1, yk+1) which gives profiles. This statistical technique was applied to analyse
D00 ¼ a00 x+b00 with xright ¼ (cb00 )/a00 and finally Ll ¼ the bias on determining the length versus the height of
xrightxleft (Fig. 4b). intercept (in %). Then the error is plotted in sampling
 The third method (L2) could lead to better results. In length units versus the height of the intercept (Fig. 5).
method 2, to calculate the intersection between the As a matter of fact, the values obtained by method L3 are
profile and the value of the intercept, the Abbott curve higher than those obtained by applying the method L0
was interpolated by a straight line (Fig. 4c). A more for small intercept heights, and lower for higher intercepts.
precise intersection can be found when (xi1, yi1), The biggest difference reaches 4% of the sampling length:

(xl, yl) (xl, yl) (xk, yk)

C C xleft xright

(xr, yr) (xl−1, yl−1) (xk+1, yk+1)

L0 L1
(xi+1, yi+1) (xj−2, yj−2)
(xj−1, yj−1) (xi−1, yi−1)
(xi+1, yi+1) (xi, yi)
C C (xj, yj)
xleft xright
xright
(xi−yi) (xj, yj)
(xi−1, yi−1)
(xi−1, yi−1)

(xj+1, yj+1) (xi−2, yj−2) (xj+1, yj+1)

L2 L3

Fig. 4. Methods L0 (no profile interpolation), L1 (linear interpolation), L2 (parabolic interpolation) and L3 (third-degree polynomial interpolation) used
to calculate the intercepted length.
 ARTICLE IN PRESS
1324 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

8 4.1.5. t-Student computation

Error (%) of the sampling length

L3-L0 L3-L1 L3-L2

6 The tx,k calculation, used to estimate the confidence
4 interval (Eq. (11)), needs complex integral with root research
2 on integral functions. However, some approximate analy-
0 tical formulae exist to estimate the t PDF and tx,k are found
-2 using a large series function with cosine terms that require
-4
much computational time. As the algorithm requires the
computation of a high number of t PDF, an alternative
-6
approach consisting in approximating the t values, can be
-8
0 20 40 60 80 100 applied. As will be seen later, only one x value is required in
Profile height (%) our method. Consequently, tx,k was evaluated by a simple
formula for one x and for any k: t0.995,k is approximated by
Fig. 5. Mean distribution error of peak length versus the relative height of the original following equation (Fig. 6):
profile for all experimental profiles described in Section 3. Each peak
length for all the profiles is computed using the methods L0, L1, L2 and t0:995;k ¼ 2:57 þ 82:611e1:18k1:17 þ 5:43=ðk  0:898Þ. (14)
L3. Peak length is divided by the sampling length. Li3 is the error in
percent of the sampling length obtained by the L3 method minus the The high confidence level of 1a ¼ 0.995 is justified:
intercepted length obtained with method Li. For these statistics, 216 550 indeed when a high number of simulations is carried out, we
peak lengths are calculated. are certain (in five cases out of 1000) that the estimated slope
is included in the calculated confidence interval. If the usual
statistical interval 1a ¼ 0.95 was used, five cases out
when a peak, or a valley, is defined with fewer than 5 of 100 would not be included in the confidence interval. As
points, method L0 underestimates the intercepted length our algorithm leads to the calculation of over 1000 slopes,
for the valleys and on the contrary overestimates the it is then possible to find a high number of slopes that
intercepted length for the peaks. Method L1 gives would not be included in the confidence interval: conse-
worse precision than the L0 one. This surprising result quently for different values of 1a, different ranges of the
can be explained by the analysis of the second order linear part of the Abbott curve are obtained. The lower
derivative of the profile ðq2 y=q2 xÞ. In case of convexity the significant level 1a, the higher the range of the linear
ðq2 y=q2 x40Þ, method L1 overestimates the intercepted part. To quantify experimentally the influence of the
length. If the profile possesses an inflection point significant level on the determination of the Abbott
ðq2 y=q2 x ¼ 0Þ, the error is minimal and corresponds to parameters, the values of Mr1 and Mr2 are estimated
the value y ¼ 50% for a high number of experimental from 10 electro-eroded profiles for different values of 1a
profiles. In case of concavity ðq2 y=q2 xo0Þ, method L1 (Fig. 7). As could be shown, the values of Mr1 and Mr2
underestimates the intercepted length and method L2 (that affected all other Abbott parameters) are affected if
gives better results. The main problem of the L2 1a40.95. As a consequence, the majored value of
interpolation is that the parabola and the profile intersec- 1a ¼ 0.995 is retained for the determination of the linear
tion are not centred on the interpolation points. In part of the Abbott curves.
fact, L2 minimises xleft and xright. However, if we consider
that a peak can be statistically symmetrical to its maximal
4.1.6. Minimal percentage on the Abbott curves
value, the length L ¼ xleftxright should be centred on its
A small part of Abbott curves may present a good linear
true value. To verify this hypothesis, xleft and xright are
part that does not represent the central roughness
computed using both the L2 and L3 methods for 20 values
amplitude. To avoid this artefact, the DIN norm proposes
of c on each peak of the 800 profiles. An analysis of
to take exactly 40% of the points to plot the linear part of
variance shows that this difference does not depend on the
the Abbott curve. Is this value the best one? To answer this
intercept height. we obtain xL3 L2
left  xleft ¼ 0:0297  0:0003, question the Abbott parameters were calculated for all the
xL3
right  x L2
right ¼ 0:0301  0:0003 that gives a difference on
profiles using L3 method with 200 intercepts by imposing
the intercept length of LL3LL2 ¼ 0.000570.0003. This
the minimal range of AHD in the interval [5%, 95%]
difference is statistically insignificant, and therefore the
to estimate the precision on the amplitude coefficients
hypothesis (symmetrical peaks) is statistically confirmed.
(Rk, Rvk and Rpk). The precision on these coefficients
The conclusion from this analysis is that method L1 has to
depends on the precision in the determination of the slope
be proscribed and method L0 is appropriate if the
of the linear part of the Abbott curve. The confidence
optimisation of the computation time is searched rather
interval of F(x) at a point x0 is given by
than a better precision on the intercept length determina- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tion. Method L2 is more accurate when a better precision X n
in the intercept length calculation is wanted. Finally, a þ bx0  t S 1 þ 1=n þ ðx0  x̄Þ2 = ðxi  x̄Þ2 ,
1  a=2
method L3 is required when the co-ordinates of the i¼1
n2
interception between profile and intercept have to be
calculated for a particular roughness analysis. (15)
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1325

Fig. 6. Plot of t0.995,k, Student’s t value, versus degree of freedom (k). Full line corresponds to the non-linear regression given by Eq. (14).

11.79 90.71 p80 E3

el_a E5
0.1 el_b TB
11.77 90.69
p1200 TF
p4000
μ[σRk] μm

U_b
11.75 90.67
Mr1 (%)

Mr2 (%)

E2 U_a

11.73 90.65
MR1 (L) 0.01
MR2 (R)
11.71 90.63

11.69 90.61
0.9 0.92 0.94 0.96 0.98 1 0 20 40 60 80 100
Significant level Minimal length on the linear part (%)

Fig. 7. Variation of Mr1 and Mr2 versus the significant level. The points Fig. 8. Influence of the minimal length of the linear part on the mean for
reported on the graph correspond to the mean of the parameters the Rk standard deviation for each class of experimental profiles described
calculated from 10 electro-eroded profiles (Fig. 2a). in Section 3.

where x̄ is the mean of the n consecutive values of xi AHDmin (5–95% with an incremental step of 5%) and for
included in the linear part. each profile, sRk was calculated, and finally the mean
Consequently, Rvk, Rk and Rpk standard deviations are, m[sRk] and the standard deviation s[sRk] of sRk were
respectively, given by calculated for all profiles. It was found that the mean
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi increases exponentially when AHDmin increases (Fig. 8)
1 ðx̄Þ2 but is quite constant for AHDmin lower than 20%. As
sRpk ¼ s 1 þ þ Pn 2
, AHDmin increases, the probability that the straight line fits
n i¼1 ðxi  x̄Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi well with the Abbott curve decreases. Therefore, if the
1 ð100  x̄Þ2 linear assumption is not respected, the probability of
sRvk ¼ s 1 þ þ Pn 2
; and
n i¼1 ðxi  x̄Þ
finding points of the Abbott curves far from the linear part
sRvk þ sRpk rises, which will increase the standard deviation of
sR k  . ð16Þ sRk ; sRpk and sRvk . However, taking the shortest AHDmin
2
could lead to false results because of the presence of a short
These parameters can be considered as new roughness well-fitted linear part whose slope differs from the real
ones that characterise the disorder of the profiles. To linear Abbott domain. Consequently, AHDmin has also to
analyse the effect of the minimal range of AHD, (AHDmin), minimise s[sRk] for each category where s[sRk] is minimal.
profiles are normalised by dividing the profile height by the As regards (Fig. 9), the Abbott parameters get good
maximal range amplitude (Rt). Then for different values of stability by taking a minimal range AHDminE15%.
 ARTICLE IN PRESS
1326 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

4.1.7. Intercept numbers on the Abbott curves 12 88.4

The number of intercepts (N) used to plot the Abbott
11.5
curves influences the precision of the parameters since their 88.2
true values are ideally obtained for an infinite number of 11
88
intercepts. However, computational time increases with N

Mr1 (%)

Mr2 (%)
10.5
and its choice must be optimised. A1, A2, Mr1 and Mr2 87.8
MR1 (L)
precision depends on the discretisation step length between 10 MR2 (R) 87.6
two intercepts. For example, precision on Mr1 or Mr2 9.5 87.4
would not exceed 100/N.
9 87.2
To evaluate N, two methods can be applied:
8.5 87
-50 50 150 250 350 450 550
(i) Creating an iterative algorithm to find the minimal
Number of intercept lines
number of intercepts that gives the precision on the
parameters defining a norm and using a convergence Fig. 10. Mr1 and Mr2 versus the number of intercepts. Points are means
criterion. of Mr1 and Mr2 values for all experimental profiles described in Section 3.
(ii) Analysing convergence parameters on experimental
profiles and finding the pertinent N.
intercept lines and errors on A1 and A2 do not exceed 1%.
The first method is the more efficient since no condition For these reasons, all the Abbott curves will be plotted with
is imposed on the profile. However, increasing N increases 250 intercepts.
the number of points of the Abbott curve and leads to a
linear part with a high number of points. By a computation 4.2. The extended Abbott parameters
complexity study, it can be shown that an iterative
algorithm increases time by eight times more than the As in the DIN norm the Abbott parameters are defined for
deterministic one. For this reason, the iterative algorithm is sigmoid curves, some errors may appear when parameters
left aside. We have analysed the influence of the number of are computed on non-sigmoidal ones. Two possibilities can
intercepts used to plot the Abbott curves. All the profiles then be considered to avoid eventual artefacts. Firstly, when
are computed with a number of intercepts included in the computed parameters do not respect rigorously the DIN
the interval [20 y 500]. Then the means of the Abbott norm specification, the parameters will not be performed.
parameters for each group and the means of these means Secondly, the DIN norm is modified to calculate all the
are computed. As Mr1 and Mr2 parameters determine parameters for the non-sigmoidal curves, and to obtain the
totally the accuracy of the others, the variation of Mr1 and same values as the norm states for sigmoidal ones. For
Mr2 are plotted versus the number of intercept lines example, if we analyse the Abbott curve of a sinusoidal
(Fig. 10). It has to be noticed that after 250 points, the profile, no DIN parameter can be computed (Fig. 11). Let us
error on the determination of the parameters is lower than analyse the physical meaning of the Rpk: this parameter was
2% for Mr1 and lower than 0.1% for Mr2. The difference created to represent the mean height of peaks that passes
in accuracy between Mr1 and Mr2 is a consequence of the over the distribution of the ‘‘peaks of the interior profile’’ and
peak profiles’ variations being more important than the then to characterise some physical surface properties like the
valleys’. With the same analysis, it was proved that errors wear processes. For the sinusoidal profile, no peak exists over
on Rpk, Rk and Rvk are less than 0.6% by taking 250 the interior profile. Consequently, the DIN norm could
present this major defect when Abbott parameters are
E3
used to quantify the effect of a physical process on the
p80
el_a E5 modification of the surface topography: suppose that a wear
el_b TB phenomenon has to be characterised by the modification of
p1200 TF
U_b
the surface topography during wear cycles. On the initial
p4000
σ[σRk]μm

E2 U_a surface, Abbott parameters are computed respecting the

0.01
DIN norm. Then the wear process erodes the peaks at each
wear cycle that will decrease the RPK value. During the wear
process, it may happen that peaks are too eroded to be
characterised by the RPK with respect to the DIN norm. As a
0.001 consequence, the wear damage can only be characterised by
the Abbott parameters on a reduced time interval. This
0 20 40 60 80 100 example proved that the DIN norm has to be extended under
Minimal length on the linear part (%) the following assumptions:
Fig. 9. Influence of the minimal length of the linear part on the standard
deviation for the Rk standard deviation for each class of experimental (i) respect the DIN philosophy,
profiles described in Section 3. (ii) parameters can always be computed,
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1327

(iii) give exactly the same parameter values as the DIN be computed. However, let Mr1 be the value of the
norm if Abbott curves are sigmoidal, intersection between the highest point of the Abbott curve
(iv) the values of all parameters have to be numerically and the regression line, then A1 becomes the area between
continuous in the transition [calculable by the DIN the highest point and the Abbott curve. This area becomes
norm—non-calculable by the DIN norm], negative and characterises an excessive wear of the peak. It
(v) looking at Abbott parameters, experimenters must be becomes obvious that this parameter is continuous and no
able to conclude if the parameters tally with the DIN discontinuity appears in the transition A1 ¼ 0. The same
norm or are extended DIN parameters. method is applied for A2 computation.

For example, let us analyse the parameter Mr1 (Fig. 11) 4.3. Experimental profiles
of the Abbott curve applied to a noisy sinusoidal profile.
To be computed by the DIN norm, the Abbott curve must In this part, Abbott and extended Abbott parameters are
lie above the regression line. Mr1 equals zero if the computed to analyse the modification of surface topogra-
regression line and the Abbott curve are superposed. If the phy due to different physical processes.
Abbott curve is under the regression line, then Mr1 cannot
4.3.1. Electro-erosion
40 Does pulse time duration have an influence on surface
roughness? To answer this question, the Abbott parameters
Roughness amplitude (μm)

30
A1 Rpk were computed for different profiles (Fig. 12). By analysis
20
of variance, it can be remarked that Rpk, Rk, Rvk, A1 are
10 highly discriminating for the electro-erosion process. On
Rk
0 the contrary, Mr1, Mr2 and A2 do not allow us to
-10 discriminate it. To normalise these parameters, we decided
-20 to reduce each Abbott curve by dividing Rpk, Rk and Rvk by
-30 Rvk A2 Rt (with Rt ¼ Rpk+Rk+Rvk) to obtain the reduced
Mr1
Mr2 coefficients Rpk , Rk and Rvk (Table 2). A new analysis
-40
0 20 40 60 80 100 of variance led us to conclude that Rpkel_a ¼ Rpkel_b ,
Asperity length distribution Rkel_a ¼ Rkel_b and Rvkel_a ¼ Rvkel_b . That means that peaks
Fig. 11. Calculus of the modified Abbott parameters on a non-sigmoidal and valleys get the same morphology. As Mr1 and Mr2 do
Abbott curve. not depend on the scale unit and as it was proved by

Roughness (μ μm) versus The Abbott The Abbott

profile length (μm) curves Parameters
20 20 RPK 7.4 ±0.5
15 15
RK 7.6 ±0.4
10 10
5 5 RVK 5.1 ±0.6
0 0 MR1 11.7 ±0.6
-5 -5 MR2 90.7 ±0.7
-10 -10
A1 21.7 ±2.0
-15 -15
0 2000 4000 6000 8000 0 20 40 60 80 100 A2 -12.9 ±2.1

20 20
RPK 8.8 ±0.7
15 15
RK 9.3 ±0.7
10 10
5 5 RVK 6.3 ±1.4
0 0 MR1 11.9 ±1.5
-5 -5 MR2 91.5 ±1.7
-10 -10
A1 26.6 ±2.6
-15 -15
0 2000 4000 6000 8000 0 20 40 60 80 100 A2 -13.5 ±4.2

Fig. 12. Abbott parameter computation on two different electro-eroded profiles. Thirty profiles are measured and Abbott statistical parameters are
computed. Pulse time duration is 1.5 ms for sample A (a) and 2.5 ms for sample B (b).
 ARTICLE IN PRESS
1328 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

the statistical analysis that amplitude structure is scale standardised (Table 2). We can conclude that decreasing
independent, then electro-erosion profiles are statistically the feed rate decreases the peak amplitude due to friction
self-affine, describing a fractal structure. between the material and the turning tool. The material is
snatched away for small feed rates and is peeled off for
4.3.2. Turning higher ones. The values of Mr1 confirm this hypothesis
All the Abbott parameters characterise the feed rate (Mr1U_a ¼ 6.5, Mr1U_b ¼ 18.2) as the most important
parameters (Fig. 13). When the feed rate increases, volume of peaks for the profiles U_b (A1U_a ¼ 2.5,
all parameters that characterise peaks increase except A1U_b ¼ 81.9).
for Rvk. Increasing the feed rate from 0.1 mm/rpm
(Fig. 13a) to 0.35 mm/rpm (Fig. 13b) increases the 4.3.3. Shot peening
roughness amplitude (RpkU_a þ RkU_a þ RvkU_a ¼ 21:5 mm, By analysing variance, it can be shown that all the
RpkU_b þ RkU_b þ RvkU_b ¼ 28:8 mm). As shown earlier parameters discriminate among the different samples
(Section 4.3.1), the amplitude roughness parameters were except for Mr1 (Fig. 14). Those experimental profiles are

Table 2
Values of reduced Abbott amplitude parameters

Study Rpk (%) Rk Rvk

Electro-erosion el_a el_b el_a el_b el_a el_b

37 36 38 38 25 26
Tool machine milling U_a U_b U_a U_b U_a U_b
18 40 46 50 36 10

Shot penning E2 E3 E5 E2 E3 E5 E2 E3 E5

23 29 26 47 41 34 30 40 40

Wear of hip prothesis head TB TF TB TF TB TF

8 36 30 21 62 9

Roughness (μ μm) versus The Abbott The Abbott

profile length (μm) curves Parameters
15 15
RPK 3.9 ±0.7
10 10
RK 10.0 ±1
5 5
RVK 7.6 ±1.1
0 0
MR1 6.5 ±1.7
-5 -5
MR2 87.6 ±1.7
-10 -10
A1 2.5 ±13
-15 -15
0 1000 2000 3000 4000 0 20 40 60 80 100 A2 -32.8 ±6

15 15
RPK 11.5 ±1.0
10 10
RK 14.6 ±0.9
5 5
RVK 2.7 ±1.4
0 0
MR1 18.2 ±3
-5 -5
MR2 91.5 ±2.5
-10 -10
A1 81.9 ±20
-15 -15
0 1000 2000 3000 4000 0 20 40 60 80 100 A2 6.5 ±11

Fig. 13. The Abbott parameter computation on two different tool-machined profiles (2  30 profiles). (a) and (b) are machined, respectively, with an
advanced sample of 0.1 and 0.3 mm/rpm.
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1329

quite satisfactory as far as the fractal structure is (Table 2). This analysis is confirmed by the values
concerned. If we use the reduced amplitude Abbott of Mr1 and Mr2 (Mr1TB ¼ 7.5, Mr1TF ¼ 23.5 and
parameters, it can be concluded that profiles E5, before Mr2TB ¼ 80.7, Mr2TF ¼ 91.3). As regards these values,
shot peening, get the same parameters as the E3 ones profiles TB and TF seem to be inverted in amplitude. An
after shot peening in dry environment (Table 2). analysis of the wear mechanisms shows that the wear
It could statistically be proved by Student’s test that surface TF is damaged by a TA6V fibre between the hip
RpkE3  RpkE5 , RkE3  RkE5 , RvkE3  RvkE5 , Mr1E3  prosthesis and the cup. Biomedical and chemical investiga-
Mr1E5 and Mr2E3  Mr2E5 . This clearly means that shot tions show that fibre moves (when the human body does)
peening in dry environment does not affect the mass peak without constraint in the TB region and is immobilised in
distribution. However, profiles E5 present a more dis- the constraint zone (TF) during an undetermined period.
ordered aspect. This example proves that Abbott para- This hypothesis seems to be confirmed by this study of
meters do not characterise the frequency feature of the roughness parameters: immobilised fibre creates important
profile (like the usual roughness amplitude coefficient Ra). valleys and then the linear part of the Abbott curves is
displaced into the peak zone.
4.3.4. Wear of hip prosthesis head
Firstly, the worn surface of the prosthesis head gets 4.3.5. Polished surface
quite a higher amplitude than the unworn one (Rpk+ It may be logical to find a statistical self-affinity fractal
Rk+Rvk ¼ 5.49 mm against 6.78 mm) (Fig. 15). The main structure because the abrasive grains get the same morphol-
aspect is the difference between the percentage of valleys ogy (Fig. 16). By this fact, the same values of the reduced
for the more solicited part (TF) and the less solicited amplitude Abbott parameters are expected. However, this
one (TB). The central roughness amplitude is not large fact is not confirmed by the experimental values and could

Roughness (μ μm) versus The Abbott The Abbott

profile length (μm) curves Parameters
6 6
RPK 2.2 ±0.3
4 4
RK 4.5 ±0.4
2 2
RVK 2.9 ±0.9
E2

0 0
MR1 8.5 ±1.3
-2 -2
MR2 86.1 ±1.3
-4 -4
A1 6.1 ±1.5
-6 -6
0 1000 2000 3000 4000 0 20 40 60 80 100 A2 -20.4 ±2.4

3 3
2 2 RPK 1.3 ±0.1
1 1 RK 2.9 ±0.2
0 0
RVK 2.8 ±0.2
E3

-1 -1
-2 -2 MR1 7.2 ±0.9
-3 -3 MR2 88.4 ±1.3
-4 -4 A1 3.0 ±0.5
-5 -5
0 1000 2000 3000 4000 0 20 40 60 80 100 A2 -8.5 ±1.2

3 3
RPK 1.2 ±0.1
2 2
RK 1.6 ±0.1
1 1
RVK 1.8 ±0.3
E5

0 0
MR1 9.7 ±1.7
-1 -1
MR2 88.2 ±1.1
-2 -2
A1 3.1 ±0.7
-3 -3
0 1000 2000 3000 4000 0 20 40 60 80 100 A2 -5.0 ±0.8

Fig. 14. The computation of the Abbott parameters for two different shot peening surfaces (E2, E3) and the original one before shot peening (E5) (3  30
profiles).
 ARTICLE IN PRESS
1330 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

The Abbott
Roughness (μm) versus The Abbott Parameters
profile length (μm) curves 0.43 ±0.09
RPK
5 5
±0.11
4 4 RK 1.64
3 3 ±0.9
2 2 RVK 3.42
1 1 ±0.7
MR1 7.48
0 0 ±0.7
-1 -1 MR2 80.7
-2 -2 ±0.4
A1 1.45
-3 -3
-4 -4 ±1.1
A2 -21.8
-5 -5
0 20 40 60 80 100 0 20 40 60 80 100

5 ±0.2
RPK 4.76
5 4
4 ±0.1
3 RK 1.42
3 2 ±0.1
2 1 RVK 0.6
1 ±1.3
0 0 MR1 23.44
-1 -1 ±0.9
-2 MR2 91.3
-2
-3 -3 24.9 ±1.4
-4 A1
-4
-5 -5 ±0.9
A2 -2.7
0 20 40 60 80 100 0 20 40 60 80 100

Fig. 15. The computation of the Abbott parameters for two different parts of a TA6V prosthesis head (280 profiles). (a) Represent the less damaged
surface (TB) and (b) the more damaged one (TF).

be explained by the difficulty in evaluating the intercept profile, A1 ¼ A2 ¼ 0 and Abbott curves will be straight
length when the number of intercepts is decreasing. lines. In this case, A1 and A2 are a measure of the fractal
noise.
4.4. Simulated profiles For profiles (f) and (g), the slopes of the Abbott curves
are calculated from valley (f) and from peak (g) for two
To validate the efficiency of our algorithm, different reasons:
categories of profiles were simulated. As analytical func-
tions are not stochastic and lead to identical results (that (i) peaks or valleys get large amplitude when compared
do not allow us to validate our algorithm), a stochastic with the fractal amplitude function,
fractal profile was added to the original non-stochastic (ii) peaks or valleys get less variance in a statistical sense
ones to obtain global stochastic ones. To simulate than the fractal noise: points are aligned in the Abbott
consequent profiles that look like experimental ones, we graph in which regression is computed.
choose the Weierstrass–Mandelbrot function [9] (Fig. 17a).
Six categories of profiles were simulated on purpose In this case, Abbott curves formally have to be
to represent most of the profiles met in surface analyses decomposed in two parts that represent macro and micro
(Fig. 17b–g). One hundred profiles were simulated, the roughness, and the cut-off between these two stages could
related Abbott parameters were calculated by our algo- lead to new Abbott parameter definitions (micro and
rithm and finally descriptive statistics were computed. macro). Using these publication results, a new definition of
The profile (a) presents a sigmoidal Abbott curve on parameters characterising the two stages will emerge.
which the DIN norm is applied without any restriction. All
values are positive except for A2 meaning that the area is 4.5. Wear analysis
calculated under the straight line.
The Profile (b) shows an Abbott curve on which only The method developed before also has to be validated on
extended DIN parameters can be computed. As a result, A1, a dynamic physical process. In this part, we first simulated
A2, Rpk, Rvk get opposite signs because the Abbott curve is a wear model, applied it to the curves (a)–(f) of Fig. 17 and
under the straight line on the left and above on the right. then analysed the evolution of the wear process through the
Profiles (c) and (d) represent, respectively, a half- evolution of the Abbott parameters.
sigmoidal curve and inverted one, then Rpk and Rvk get
opposite signs. The range [Mr1, Mr2] is located on the 4.5.1. Wear simulation
upper profile height for (d) and in lower profile for (c). The following elementary model is used to simulate a
Profile (e) represents a sigmoidal profile with small A1 wear process. Let W(x), be the probability that a wear
and A2. If no fractal noise is introduced in this triangular phenomenon erodes an element with height dh during each
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1331

The Abbott
Roughness (μm) versus The Abbott Parameters
profile length (μm) curves
RPK 1.7 ±0.7
4 4
3 3 RK 2.8 ±0.6
2 2 RVK 1.7 ±0.6
1 1 MR1 12.1 ±4.7
0 0 MR2 85.1 ±7
-1 -1 A1 7.8 ±4
-2 -2 A2 -12.1 ±12
-3 -3
0 500 1000 1500 2000 0 20 40 60 80 100

RPK 0.48 ±0.16

0.60 0.60
RK 0.34 ±0.07
0.40 0.40
RVK 0.45 ±0.1
0.20 0.20
MR1 11.5 ±1.7
0.00 0.00
MR2 82 ±6.4
-0.20 -0.20
A1 1.34 ±0.7
-0.40 -0.40
A2 -2.8 ±1.8
-0.60 -0.60
0 500 1000 1500 2000 0 20 40 60 80 100

RPK 0.32 ±0.12

0.25 0.25
0.20 0.20 RK 0.14 ±0.01
0.15 0.15 RVK 0.12 ±0.01
0.10 0.10
0.05 0.05 MR1 9.12 ±1.5
0.00 0.00 MR2 88.8 ±1.8
-0.05 -0.05
-0.10 -0.10 A1 0.47 ±0.16
-0.15 -0.15 A2 -0.37 ±0.1
-0.20 -0.20
0 500 1000 1500 2000 0 20 40 60 80 100

Fig. 16. The computation of the Abbott parameters for three different polished surfaces (30 profiles). Surfaces are, respectively, polished with paper 80 (a),
1200 (b) and 4000 (c).

wear cycle (the higher the dh, the less wear resistant the 4.5.2. Wear results
material). Let Zmax be the maximal height of the profile Wear algorithm is applied to curves presented in Fig. 17.
and Zmin the minimal one. As the wear process does not With parameters l ¼ 0.1 and dh ¼ 0.01. For every 100
erode materials on a depth lower than Zmin, we defined the wear cycles, the Abbott parameters are computed up to
probability to erode an elementary volume of the material 2000 cycles. The profile and the associated Abbott curves
during every wear cycle by are represented at the end of the wear process in Fig. 18.
Z During the wear process A1 decreases, i.e. peaks are eroded
Zmax z
W ðxÞ ¼ k elx dx, (17) by the wear phenomenon. A1 is positive for fewer than 900
0 wear cycles and becomes negative after (Fig. 18d). As was
stated in section 4.2, the transition from the DIN
where k and l are real numbers. The higher the l, the higher parameters to the extended DIN ones respects continuity.
the wear phenomenae (intensity stress, corrosion and so on).
This trivial model can be applied only on regular profiles 4.6. Manual-computer comparison
without auto-correlation. Wear integral (Eq. (17)) is
R Z Z
normalised to obtain a PDF, then k 0 max min elx dx ¼ Generally, the Abbott parameters are manually calcu-
1 and the following expression can be deduced: lated and the main difficulty consists in drawing the slope
 1 of the Abbott curves and determining Mr1 and Mr2. In
k ¼ ð1  elðZmax zÞ Þ=l . (18) this part, we shall compare the manual and the computa-
tion methods.
The probability that an elementary material volume can
be eroded at a height z is finally given by 4.6.1. Methodology comparison
  Sixty profiles were recorded by a tactile profilometer on
W ðxÞ ¼ 1  k 1  elðZmax zÞ . (19) an electro-eroded surface discussed in Section 3.1 (Fig. 2a)
 ARTICLE IN PRESS
1332 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

N Roughness (µm) versus The Abbott The Abbott

μm)
profile length (μ curves Parameters
30 30 RPK 18.7 ±1.8
20 20
RK 21.7 ±1.3
10 10
0 0 RVK 19.1 ±1.5
-10 -10 MR1 9.6 ±0.8
-20 -20
MR2 90.2 ±0.7
-30 -30
-40 -40 A1 36.6 ±6.0
0 2000 4000 6000 0 20 40 60 80 100 A2 -37.0 ±4.9

40 40
30 30 RPK -6.47 ±0.4
20 20 RK 47.3 ±0.8
10 10
RVK -6.60 ±0.6
0 0
-10 -10 MR1 10.6 ±1.0
-20 -20 MR2 91.8 ±1.2
-30 -30
-40 -40 A1 -51.4 ±4.9
0 500 1000 1500 2000 0 20 40 60 80 100 A2 38.2 ±5.1

50 50
RPK -9.6 ±0.5
30 30
RK 44.9 ±0.3
10 10 RVK 5.2 ±0.7
-10 -10 MR1 25.0 ±1.6
-30 -30 MR2 98.4 ±0.5
-50 -50 A1 -161 ±13
0 500 1000 1500 0 20 40 60 80 100 A2 -2.3 ±1.1

40 40
30 30 RPK 4.3 ±0.8
20 20 RK4 5.4 ±0.3
10 10
0 0 RVK -9.6 ±0.5
-10 -10 MR1 1.6 ±0.5
-20 -20 MR2 75.7 ±1.6
-30 -30
-40 -40 A1 2.4 ±1.4
0 500 1000 1500 0 20 40 60 80 100 A2 156 ±16

30 30
RPK 4.4 ±0.6
20 20
RK 50.0 ±0.2
10 10
RVK 4.9 ±0.7
0 0
-10 -10 MR1 1.6 ±0.2
-20 -20 MR2 98.4 ±0.1
-30 -30 A1 1.9 ±0.3
0 500 1000 1500 0 20 40 60 80 100 A2 -1.8 ±0.3

20 20
10 10 RPK -9.5 ±0.5
0 0 RK 46.8 ±0.2
-10 -10 RVK 3.7 ±0.7
-20 -20 MR1 70.0 ±0.3
-30 -30
MR2 99.6 ±0.1
-40 -40
-50 -50 A1 -440 ±32
0 1000 2000 3000 0 20 40 60 80 100 A2 -0.4 ±0.1

50 50
RPK 3.2 ±0.7
40 40
30 30 RK 47.1 ±0.2
20 20 RVK -9.7 ±0.4
10 10 MR1 0.4 ±0.1
0 0
MR2 30.4 ±0.2
-10 -10
-20 -20 A1 0.5 ±0.1
0 1000 2000 3000 0 20 40 60 80 100 A2 456 ±25

Fig. 17. Abbott parameters computation on different kinds of profiles. Hundred Fractal functions are computed (a), reduced in amplitude and added to a
sinusoid (b), half-sinusoid (c and d), triangular function (e), ‘‘cracked’’ function (f) and ‘‘peaked’’ functions (g).

where the Abbott curves are sigmoidal. Then three retained only the Mr1 and Mr2 estimations since the
evaluations were carried out on all the profiles by different precision on the other Abbott parameters mainly depends
experimenters to determine the Abbott parameters. We on the determination of these two parameters.
 ARTICLE IN PRESS
M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334 1333

Fig. 18. Profiles and Abbott curves after 2000 wear cycles applied on original profiles of Fig. 17.
 ARTICLE IN PRESS
1334 M. Bigerelle, A. Iost / Tribology International 40 (2007) 1319–1334

Table 3
Comparison between manual and the computed methods to calculate Mr1 and Mr 2 parameters on 60 profiles of electro-eroded surface (Fig. 2)

Parameter Determination methods Mean Median Minimum Maximum Standard deviation

Mr1 Manual 11.98 11.95 7.06 16.30 2.24

Mr1 Algorithm 11.36 11.63 6.04 14.94 1.85
Mr2 Manual 89.82 89.94 83.42 96.19 2.37
Mr2 Algorithm 90.86 90.94 85.96 94.72 1.72

4.6.2. Results from comparison was applied for over 3 years in our laboratory and nothing
Statistical results are reported in Table 3. Both Mr1 PDF outrageous was found moreover it can be extended to 3D
(manual and automatic) obey a Gaussian law (critical surfaces [10] without any problem. Finally, some algo-
values of the Shapiro Wilks test on manual determination: rithms and results described in this paper will be helpful to
0.1497, automatic: 0.1352). Mr2 also obeys a Gaussian calculate other roughness parameters such as the number
PDF for both methods (manual: 0.97, automatic: 0.98). of peaks, the radius of curvature and so on.
Both automatic Mr1 and Mr2 determinations are higher
than manual ones: experimenters unconsciously minimise
Acknowledgement
the range of the linear part of the Abbott curves that lead
to a lower range for F(x) equation given by the algorithm.
We wish to thank Véronique Hague for her assistance in
The dispersion is higher if the parameters are evaluated
English.
manually without any computation. However, the differ-
ence between manual and automatic evaluations is less
than 1%. As a result our algorithm is reliable and could References
accurately be applied to experimental profiles.
[1] Abbott EJ, Firestone FA. Specifying surface quality. Mech Eng
1933;59:569–72.
5. Conclusion
[2] Torrance AA. A simple datum for measurement of the Abbott curve
of a profile and its first derivative. Tribol Int 1997;30(3):239–44.
An algorithm was performed in order to have an [3] Malburg MC, Raja J. Characterization of surface texture generated
automatic evaluation of the Abbott curve parameters. by plateau honing process. Ann CIRP 1993;42:637–9.
Execution time does not exceed 0.1 s on a personal [4] Deutsche Normen, Kenngrössen Rk, Rpk, Rvk, Mr1, Mr2 zur
computer. Discretisation artefacts are discussed by analys- Beschreibung des Materialanteils im Rauheitsprofil. Messung der
profiltiefe Pt von Oberflächen, DIN 4776, Berlin, Mai 1990.
ing experimental and theoretical profiles, and lead to the [5] NF ISO 13565. AFNOR, Paris, 1995.
conclusion that parameter estimations are accurate. As a [6] Lacey P, Torrance AA, Fitzpatrick JA. The relation between the
high number of different profiles were accurately analysed, friction of lubricated surfaces and apparent normal pressure. ASME J
we think that our algorithm can be applied to a high Tribol 1989;111:260–4.
number of experimental profiles. Like all numerical [7] Lacey P, Torrance AA. The calculation of wear coefficients for plastic
contacts. Wear 1991;145:367–83.
techniques, some particular profiles could lead to erro- [8] Gruber MHJ. Regression estimators, a comparative study, in
neous results (high isolated peak, or valley, non-stationary statistical modeling and decision science. Boston: Academic Press
profiles, micro and macro-roughness, etc.). However, the Inc.; 1990.
Abbott curves have no application in those typical cases [9] Ganti S, Bhushan B. Generalized fractal analysis and its applications
to engineering surfaces. Wear 1995;180:17–34.
and another parameters will be more appropriate to
[10] Dong WP, Sullivan PJ, Stout KJ. Comprehensive study of
characterise such types of profiles. It was shown that parameters for characterizing three dimensional surface topography.
fractal functions could be used to simulate a wear process III parameters for characterizing amplitude and some functional
or to validate the efficiency of the parameter. This method properties. Wear 1994;178:29–43.

View publication stats