You are on page 1of 9

Tribol Lett (2008) 32:1321

DOI 10.1007/s11249-008-9354-x

ORIGINAL PAPER

3-D Characterization, Optimization, and Classification


of Textured Surfaces
Gwidon Stachowiak Pawel Podsiadlo

Received: 11 May 2008 / Accepted: 6 August 2008 / Published online: 11 September 2008
 Springer Science+Business Media, LLC 2008

Abstract Surface texturing is a new technology aiming at


reducing friction in operating machinery. Different surface
features of varying shape and density are artificially introduced onto the existing surfaces. The methods currently
available for the 3-D characterization and description of
these surfaces are inadequate. Reliable surface description
is necessary for the optimization of those surfaces and
quality control during production. In this paper possible
ways of solving the problems associated with 3-D
description and optimization of textured surfaces are outlined and discussed. The problems associated with the
development of automated classification systems for textured surfaces are also presented.
Keywords Textured surfaces  Surface characterization 
Surface texture classification

1 Introduction
The market for advanced products from optics, telecommunication, airspace, automotive and machine industries,
etc. is growing at about 21% per annum [1]. Advanced
components manufactured by the modern industry use
surfaces of various forms and texture patterns. These surfaces are very different from the traditionally used surfaces
which usually could be characterized using standard surface roughness parameters. The characteristic feature of
these surfaces is that they use tessellated patterns, i.e.

G. Stachowiak (&)  P. Podsiadlo


Tribology Laboratory, School of Mechanical Engineering,
The University of Western Australia, 35 Stirling Highway,
Crawley, WA 6009, Australia
e-mail: gws@mech.uwa.edu.au

repeated structures over the surface. Surfaces with tessellated patterns are not new, as commercially available
abrasive surfaces, consisting of arrays of triangular based
micron-sized pyramids, are common. However, surface
features/patterns that are artificially introduced (like dimples, chevrons, etc.), of different density and orientation,
are new. During the past decade these surface patterns have
been studied as possible means to reduce friction and also
to increase wear resistance, hence components durability.
Friction reduction, in particular, is an important issue for
the transportation industry because of the increasing petrol
prices and global warming. Various initiatives are being
considered by governments across the world aiming at
reducing fuel consumption and exhaust gas emissions. The
transportation industry is one of the largest consumers of
petroleum and also one of the largest emitters of global
warming gases. It has been estimated that the surface
texturing, when optimized, can reduce the parasitic friction
loses by as much as 5% (S. Hsu, private communication).
As the global oil reserves are rapidly diminishing, reducing
petrol consumption becomes a critical issue to all of us.
The type of textures, i.e. shape, density, orientation, etc.,
of these surfaces are of interest because of their direct link
to the components functionality. It is thus important to be
able to characterize/describe these surface so that a consistent manufacturing result can be achieved. The problem
is that, currently, we do not have a method/standard that
could be used to reliably describe those surfaces at various
scales, i.e. micrometre, nanometre, etc. In this paper, possible ways of solving the problem of characterization of 3D surface topographies with tessellated patterns are
discussed.
There is a growing need for new surface characterization
methods to reliably characterize and classify the surfaces
with different tessellated patterns. Once this is achieved, in

123

14

combination with full understandings of the physics governing the lubrication and contact mechanics, it would be
possible to numerically optimize those surfaces for minimum friction and wear. In many instances improved
performance and durability is obtained by a try-and-see
approach. However, this is not very effective. Thus,
recently several works involving numerical simulation used
in the optimization of these surfaces have been published. In
these works the surface texture is described/simulated by
some mathematical functions [25] or parameters used for
deterministic surfaces [6, 7]. Some researchers even
attempted to provide analytical solutions for the effects of
surface textures on hydrodynamic lubrication [e.g. 8]. There
are limitations to this approach, however, as this tends to
work well with specific textures. Clearly, a new standard is
needed to reliably characterize all the textured surfaces.

2 Preparation of 3-D Surface Topography Data


for Further Processing
3-D surface topography can be acquired using either contact
or non-contact surface profilers. Instruments currently
available (e.g. Talysurf both contact and non-contact) provide excellent horizontal and vertical resolutions. 3-D
images of surface topography captured are then converted
into 2-D image Z (called a range image). Entries of this image
are outputs of a 2-D discrete image function z = f(x, y).
This function assigns a 3-D surface elevation data (encoded
into a brightness value) z = Lz Lz f1; 2; . . .; Nz g to a
point (encoded into a pixel) located on surface at x; y 2
Lx  Ly coordinates Lx f1; 2; . . .; Nx g and Ly f1; 2;
. . .; Ny g Nz is the number of grey scale levels, while Nx and
Ny are the number of pixels in the x and y directions,
respectively. As an example, shaded and range images of a
sandblasted surface are shown in Fig. 1. For the images the
brightest pixel value of 255 represents the highest elevation
point on the surface, while the darkest pixel value of 0
represents the lowest elevation point on the surface.
Fig. 1 Encoding surface
elevation data into the range
images of the sandblasted
titanium surface, Ra = 1.05 lm

123

Tribol Lett (2008) 32:1321

3 Fractal-Wavelet Modelling of 3-D Textured Surfaces


Surface characterization methods currently in use work
well with isotropic surfaces or surfaces with regular patterns. However, problems arise with the characterization of
anisotropic surfaces or surfaces with tessellated patterns,
e.g. abrasive surfaces. Surface texturing is a new technological development promising better friction and wear
control in working machine components.
It is known that surface topographies of manufactured
components are multi-scale objects (i.e. exhibiting different scales of surface features). Therefore, any new surface
characterization methods developed must be able to
describe these surfaces at various length scales (micrometre, nanometre, etc.). However, this is not an easy task.
A promising approach is the application of fractals [9
12]. 3-D surface topographies presented in the form of
range images exhibit a very useful feature, i.e. they exhibit
a self-transformability. Self-transformability means that
one part of the image can be transformed into another part
of the image reproducing itself almost exactly, as illustrated in Fig. 2. This way, the surface range image
(containing the 3-D surface data) is encoded into a fractal
model, called a Partition Iterated Function System (PIFS),
using a set of N contractive
affine transformations, i.e.

S
PIFS Nj1 fj DOMj [911]. Each transformation fi
converts a larger part of the surface (called domain) DOMj
into a smaller part (called range) RAN
elsewhere
 j located

on the same surface, i.e. fi DOMj RANj ; j
1; 2; . . .; N and
32 3 2 3
2 3 2
ej
aj bj 0
x
x
fj 4 y 5 4 cj dj 0 54 y 5 4 wj 5
0 0 sj
oj
z
z
where aj, bj, cj, dj, ej, wj are parameters used to rotate,
skew, scale and translate the DOMj domain and sj and oj
are scaling and offset parameters on the scale levels of the
surface image Z. The scale parameter controls the image
contrast while the offset parameter controls the image

Tribol Lett (2008) 32:1321

15

Fig. 2 Surface with marked


self-transformable part. A larger
part of the image transforms to a
smaller part of the image using
mathematical transformation
containing information about
the scale, translation, rotation,
etc. Set of these transformations
(PIFS) gives the fractal model
of the surface

Fig. 3 Example of textured


surface decoding from the
fractal model. At each iteration
an intermediate image or a
transition frame is generated

PIFS is constructed for the


textured surface image
transf 1
transf 4
transf 7
transf 10
transf 13
transf 16
transf 19
transf 22
transf 25
transf 28
transf 31
transf 34
transf 37
transf 40
transf 43

Textured surface image

..
.

etc.

transf 2
transf 5
transf 8
transf 11
transf 14
transf 17
transf 20
transf 23
transf 26
transf 29
transf 32
transf 35
transf 38
transf 41
transf 44

transf 3
transf 6
transf 9
transf 12
transf 15
transf 18
transf 21
transf 24
transf 27
transf 30
transf 33
transf 36
transf 39
transf 42
transf 45

PIFS

Two iterations

Initial image is
iterated through
PIFS

Initial surface image

brightness. When an arbitrary image is applied iteratively


to the PIFS, a sequence of decoded images (called
intermediate images or transition frames) converging to
attractor is obtained, i.e.
n
o
Z l PIFSl Zinit ; l 1; . . .; L
where PIFS is a PIFS constructed for the surface image Z;
Z(l) is the decoded surface image obtained after l iterations,
1
1
Zinit ; Z 2 PIFS2 Zinit PIFS1
i.e. Z1 PIFS

PIFS Zinit ; etc.; L is the total number of iterations and
Zinit is an initial image that is an arbitrarily chosen image.
The last decoded image (the attractor) is a very close
approximation of the original surface image Z, i.e. Z 
Z l  An example of the iterative decoding process of a
textured surface is shown in Fig. 3.

Three iterations

Ten iterations

This approach is different from traditionally used surface characterization methods as the entire surface
topography information is encapsulated in the set of
mathematical transformations, i.e. in the fractal model.
Therefore, if the fractal model is constructed for a textured
surface it will contain the detailed information about, if for
example dimples are used, dimple spacing, size, orientation, shape, depth and surface roughness between the
dimples. It is technically possible to obtain even the surface
roughness at the bottom of the individual dimple. This
process is schematically illustrated in Fig. 4.
Once the fractal model has been constructed it is also
possible to control/measure all the surface features. For
example, the roughness of surface features and their orientation can be controlled using the sj scale and aj, bj, cj, dj

123

16
Fig. 4 Schematic illustration of
the fractal model of a textured
surface

Tribol Lett (2008) 32:1321

Each parameter in this model


represents a particular
surface characteristic

Dimple
dimensions

Fractal model
Dimple spacing

transf 1
transf 4
transf 7
transf 10
transf 13
transf 16
transf 19
transf 22
transf 25
transf 28
transf 31
transf 34
transf 37
transf 40
transf 43

..
.

transf 2
transf 5
transf 8
transf 11
transf 14
transf 17
transf 20
transf 23
transf 26
transf 29
transf 32
transf 35
transf 38
transf 41
transf 44

transf 3
transf 6
transf 9
transf 12
transf 15
transf 18
transf 21
transf 24
transf 27
transf 30
transf 33
transf 36
transf 39
transf 42
transf 45

Dimple orientation

Dimple shape

etc.

Surface roughness
between dimples

Containing many parameters


representing all features for a
surface image

Dimple depth

etc.

Fig. 6 Controlling the dimple orientation using the fractal model

Fig. 5 Controlling the surface roughness using the fractal model.


Roughness (complexity) of individual surface features (dimples) can
also be controlled once they are identified

rotation parameters, as shown in Figs. 5 and 6. The depth


of surface features including dimples can be controlled
using the oj offset parameter (Fig. 7). In previous study the
scale parameter was used to calculate the surface fractal
dimension D, i.e.
N

P
log
jsi jpi
i1
D2
log2

123

Fig. 7 Controlling/measuring the depth of surface features

Tribol Lett (2008) 32:1321

17

carrying capacity is at maximum was found [16]. Recently,


the analytical optimal solution for the effect of uniform
periodic textures on the load carrying capacity in thrust
bearings has been obtained using homogenization technique and pertubation analysis [17].
In these studies, the optimization process of surface
textures was limited to selected features of textured surfaces such as, for example: the surface roughness, the area
fraction covered by dimples or the stiffness constant and
the average depth. It seems, therefore, that it would be
desirable to fully describe the textured surface with all its
features and use this description in optimization of its
performance. This has not been fully done yet as we still do
not know how to describe the textured surface. Once the
real surface data can be inputted into the simulation models
we could expect that the surface optimization of textured
surfaces for their optimal performance would follow. This
would be of major interest to industry.
The unique advantage of the fractal model is that while
applying iteratively some initial image to the numerical
model a number of intermediate images are generated.
There could be a number of final surfaces with different
features and corresponding fractal models as illustrated in
Fig. 9. The method proposed can thus be used for the
optimization of these surfaces with different textures. For
example, it is intuitively felt that the surface with dimples
might give a good performance but it is not known what
should be the optimal depth, size or density of these dimples. With fast computers the surface optimization process
would involve inputting the surface data from all the
intermediate surface images obtained from decoding of
PIFS to, for example, a governing Reynolds differential
equation of a thrust bearing. Then using a selected

where pi is the domain factor [12]. This suggests that the


scale parameter can provide a measure of the complexity of
textured surfaces.
To account for the multiscale nature of the surface
topographies a combination of fractals and wavelets has
been employed. Wavelets are used to decompose the surface image containing 3-D surface topography information
into different scale components, e.g. roughness, waviness,
form, etc. while fractals are used to characterize the surface
topography over the achievable range of scales. Detailed
description of this technique can be found in [13].

4 Textured Surface Optimization/Modelling


From the industrial viewpoint it would be desirable to
design the surface texture that would give the optimal
performance under specific operating conditions. Recently,
there have been numerous attempts to numerically model
the performance of textured surfaces. The effects of surface
textures on elastohydrodynamic films have been numerically modelled [25]. In these works the surface topography
was described using some mathematical function. The
output from this function was then worked iteratively with
the equation governing the minimum film eleastohydrodynamic thickness as schematically illustrated in Fig. 8.
The effects of different surface features on hydrodynamic and mixed lubrication have also been investigated
but mostly experimentally [e.g. 14, 15]. The numerical
analyses of these effects on hydrodynamic lubrication have
also been shown. For example, a partially textured thrust
bearing pad with square dimples was analysed and the
optimal area fraction of the textured pad for which the load
Fig. 8 Schematic illustration of
modelling the effects of surface
texture on elastohydrodynamic
film (adapted from [3])

Numerical shape of the microcavity

R(X,Y,T) = A cos
Non-dimensional film thickness

(c)

P(X',Y',T)dX'dY'
-

(b)

D/1.2R

2
2
H(X,Y,T) = H0 (T) + X + Y - R(X,Y,T) + 2
2
2
2

(a)

D e-2
21.2 R

(d)

(X - X') 2 + (Y - Y')2

(e)

123

18

Tribol Lett (2008) 32:1321

Fig. 9 Intermediate images


generated during the decoding
of fractal models of the surfaces

Fig. 10 Finding an optimal


surface for the case of static
conditions (intial and final
surface images adapted from
[5]), a, initial; b, 1st iteration; c,
2nd iteration; d, 3rd iteration; e,
4th iteration; f, final

criterion, e.g. the maximum load carrying capacity or


minimum friction, dimples of optimal depth or size could
be found for the best performance. This process is illustrated in Fig. 10. A PIFS is first constructed for the final
(extreme) image that contains deep large and small dimples
and rough surface (Fig. 10f). The PIFS constructed is then
decoded using the initial (extreme) image that contains
shallow large dimples and rough surface (Fig. 10a). This
decoding process results in intermediate images (Fig. 10bd) between these two extreme surfaces. Intermediate images are used and then substituted to the Reynolds equation.
Finally, the optimal surface, i.e. the surface that gives the
maximum load capacity or minimum friction, is found.
In dynamic systems surface features are continuously
changing their positions as the surfaces move relative to
each other. Thus, the number of features present in the
contact area at the instant can vary under those conditions.
This complicates the simulation calculations. One possible

123

solution is to produce a film of the surface frames taken at


N different time intervals, as shown in Fig. 11. For each
surface frame from the film a fractal model is generated
and input surface data are iteratively run through it
resulting in a number of intermediate images (i.e. the film
#1 at 1st iteration, the film #2 at 2nd iteration, etc.), as
illustrated in Fig. 12. The surface data from these images
are then substituted to the criterion equation, e.g. the
minimum film thickness, load or friction equation. One of
these images would be optimal, giving the best performance. This way, with the use of fast computers, the
surface textures can be optimized.
The optimization process can be performed in a rigorous
mathematical manner using the maximum principle of
Pontryagin [18]. In this case, the problem of finding the
maximum load capacity can be defined as a non-linear
optimal control problem. The load capacity becomes, for
example, the objective function that is subject to the

Tribol Lett (2008) 32:1321

19

Fig. 11 Film of surface frames


taken at different time intervals

Fig. 12 Schematic illustration


of the concept of optimization
of textured surfaces under
dynamic conditions (the film
equation is adapted from [2])

Reynolds equation, the inequality constraints of film


thickness and pressure and the boundary conditions of film
pressure. The film pressure, the constant of integration and
the load capacity become the state variables and the
objective function is maximized with respect to a film
profile (the control signal). The maximum principle
approach has been successfully used for the optimum
design of one-dimensional journal and slider bearings [19,
20]. However, there is limitation in the application of this
approach. The Pontryagins minimum principle provides
only necessary conditions for local minimum and it must
be proven first that the solution obtained is unique and
global. The optimization process of textured surfaces can

be also performed using a Bellmans optimality principle


[21]. However, the application of this pricinple is limited
by the fact that it provides only sufficient conditions for
global minimum. The difficult problem of finding a unique
and global solution for the surface optimization still needs
to be addressed.
It might be argued that the optimization process would
not work for textured surfaces that change shapes due to
contact, materials and operation conditions. However, the
shape changes can be added into the optimization process
of surfaces in the form of dynamically changing constraints. In this case, parameters of the fractal model will be
optimized subject to the constraints.

123

20

Tribol Lett (2008) 32:1321

Target class

(a)

Outlier
Similar image parts

of both wear particles and tribological surfaces [23, 24].


This classifier has the right complexity, does not require
the assumption of a normal distribution, can work with
small training sets of data and also has a learning ability. In
addition, the SVM classifier can be used for the surface
classification based on dissimilarity measures calculated
between an unclassified surface and a small number of
classified surfaces. If slack variables are used and nonlinear transforms are applied to input data the classifier also
works well with overlapping data [25].
Once the tasks described in this paper are fulfilled and
the problems solved the surfaces with textures giving
optimal performance can be characterized and produced.
Tools developed for the automated surface classification
would be vitally important during the production and
quality control.

6 Conclusions

(b)
Fig. 13 Examples of overlapping data problems, (a) overlapping data
on images of two surfaces belonging to different classes, (b)
overlapping classes [17]

From the work conducted the following conclusion can be


drawn:
A concept of characterization of textured surfaces has
been presented. Based on this concept new methods for
characterization of textured surfaces will be developed.

5 Classification of Textured Surfaces


Based on the combination of wavelets and fractals it might
also be possible to develop an automated classification
system for textured surfaces. This development would be
essential during the manufacturing to ensure the quality
and consistency of the finished product. The basic problems that need to be overcome first are following:

Overlapping data problems (overlapping classes) [22],


i.e. similar data can exist in images belonging to
different classes, as illustrated in Fig. 13,
Selection/development of optimal classifiers that can
separate between the overlapping classes of textured
surfaces [23, 24].

Various classifiers have already been compared and


tested on both computer generated and real tribological
surfaces, including the k-nearest neighbours, linear
discriminant, quadratic discriminant, Fisher linear discriminant function, linear programming machines and
support vector machines (SVM) [23]. It was established
that a support vector machine classifier gives the best
performance and is the most suitable for the classification

123

Fractal model, encapsulating complete surface topography information can be built based on surface
transformations,
To accommodate multi-scale nature textured surfaces
wavelets can be employed to decompose the surface
into individual scales,
Combination of fractals and wavelets allows for a full
description of any given surface, without use of any
parameters,
Surface characterization method developed could be
used in optimization of textured surfaces,
Based on this method a surface classification system
can be developed for the use in production and ensuring
the consistency of the produced components.

Acknowledgement The authors wish to thank the School of


Mechanical Engineering, University of Western Australia for its help
during preparation of the paper.

References
1. Willis, E.: Surface finish in relation to cylinder liners. Wear 109,
351366 (1986)
2. Wang, Q.J., Zhu, D.: Virtual texturing: Modelling the performance of lubricated contacts of engineered surfaces. Trans.
ASME 127, 722728 (2005)

Tribol Lett (2008) 32:1321


3. Mourier, L., Mazuyer, D., Lubrecht, A.A., Donnet, C.: Transient
increase of film thickness in micro-textured EHL contacts. Tribol.
Int. 39, 17451756 (2006)
4. Hu, Y.Z., Zhou, D.: A full numerical solution to the mixed
lubrication in point contacts. ASME J. Tribol. 122, 19 (2002)
5. Wang, X., Adachi, K., Otsuka, K., Kato, K.: Optimization of the
surface texture for silicon carbide sliding in water. Appl. Surf.
Sci. 253, 12821286 (2006)
6. Ren, N., Nanbu, T., Yasuda, Y., Zhou, D., Wang, Q.: Micro
textures in concentrated-conformal-contact lubrication: Effect of
distribution patterns. Tribol. Lett. 28, 275285 (2007)
7. Siripuram, R.B., Stephens, L.S.: Effect of deterministic asperity
geometry on hydrodynamic lubrication. ASME J. Tribol. 126,
527534 (2004)
8. Fowell, M., Olver, A.V., Gosman, A.D., Spikes, H.A., Pegg, I.:
Entrainment and inlet suction: Two mechanisms of hydrodynamic lubrication in textured bearings. Trans. ASME 129, 336
346 (2007)
9. Podsiadlo, P., Stachowiak, G.W.: Scale-invariant analysis of wear
particle morphologya preliminary study. Tribol. Int. 33, 289
295 (2000)
10. Podsiadlo, P., Stachowiak, G.W.: Scale-invariant analysis of wear
particle morphology. I: Theoretical background, computer
implementation and technique testing. Wear 242, 160179 (2000)
11. Stachowiak, G.W., Podsiadlo, P.: Surface characterization of
wear particles. Wear 225229, 11711185 (1999)
12. Podsiadlo, P., Stachowiak, G.W.: Scale-invariant analysis of wear
particle surface morphology. II: Fractal dimension. Wear 242,
180188 (2000)
13. Podsiadlo, P., Stachowiak, G.W.: Hybrid fractal-wavelet method
for characterization of tribological surfacesa preliminary study.
Tribol. Lett. 13, 241250 (2002)
14. Costa, H.L., Hutchings, I.M.: Hydrodynamic lubrication of textured steel surfaces under reciprocating sliding conditions. Tribol.
Int. 40, 12271238 (2007)

21
15. Johansson, S., Nilsson, P.H., Ohlsson, R., Anderberg, C., Rosen,
B.-G.: New cylinder liner surfaces for low oil consumption.
Tribol. Int. 41, 854859 (2008)
16. Marian, V.G., Kilian, M., Scholz, W.: Theoretical and experimental analysis of a partially textured thrust bearing with square
dimples. Proc. Inst. Mech. Eng. Part. J: J. Eng. Tribol. 221, 771
778 (2007)
17. Buscaglia, G.C., Ciuperca, I., Jai, M.: On the optimization of
surface textures for lubricated contacts. J. Math. Anal. Appl. 335,
13091317 (2007)
18. Pontryagin, L.S., Boltyanskii, V.G., Gamkrelidze, R.V., Mischenko, E.F.: The Mathematical Theory of Optimal Processes.
Wiley, New York (1962)
19. Maday, C.J.: The maximum principle approach to the optimum
one-dimensional journal bearing. Trans. ASME J. Lubr. Technol.
92, 482489 (1970)
20. Bourgin, P., Gay, B.: The optimum Rayleigh bearing in terms of
load capacity or friction for non-Newtonian lubricants. Trans.
ASME J. Tribol. 107, 157164 (1985)
21. Bellman, R.: Introduction to the Mathematical Theory of Control
Processes, vol. 1. Academic Press, New York (1967)
22. Stachowiak, G.W., Podsiadlo, P.: Towards the development of an
automated wear particle classification system. Tribol. Int. 39,
16151623 (2006)
23. Stachowiak, G.P., Podsiadlo, P., Stachowiak, G.W.: Comparison
of feature extraction methods for machine condition monitoring
and failure analysis. Tribol. Lett. 20, 133147 (2005)
24. Stachowiak, G.P., Podsiadlo, P., Stachowiak, G.W.: Automated
classification of wear particles based on their surface texture and
shape features. Tribol. Int. 41, 3443 (2008)
25. Podsiadlo, P., Stachowiak, G.W.: Classification of tribological
surfaces without surface parameters. Tribol. Lett. 16, 163171
(2004)

123

You might also like