Sie sind auf Seite 1von 8

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

journal homepage: www.elsevier.com/locate/jmatprotec

Prediction of surface roughness proles for milled


surfaces using an articial neural network and
fractal geometry approach
I.A. El-Sonbaty, U.A. Khashaba , A.I. Selmy, A.I. Ali
Mechanical Design and Production Engineering Department, Faculty of Engineering, Zagazig University,
P.O. Box 44519, Zagazig, Egypt

a r t i c l e

i n f o

a b s t r a c t

Article history:

Articial neural networks (ANNs) models were developed for the analysis and prediction of

Received 21 March 2007

the relationship between the cutting conditions and the corresponding fractal parameters

Received in revised form

of machined surfaces in face milling operation. These models can help manufacturers to

19 August 2007

determine the appropriate cutting conditions, in order to achieve specic surface rough-

Accepted 3 September 2007

ness prole geometry, and hence achieve the desired tribological performance (e.g. friction
and wear) between the contacting surfaces. The input parameters of the ANNs models are
the cutting parameters: rotational speed, feed, depth of cut, pre-tool ank wear and vibra-

Keywords:

tion level. The output parameters of the model are the corresponding calculated fractal

Surface roughness prole

parameters: fractal dimension D and vertical scaling parameter G. The model consists

Neural network

of three-layered feed-forward back-propagation neural network. ANNs models were utilized

Fractal geometry approach

successfully for modeling and predicting the fractal parameters D and G in face milling

Milling operation

operations. Moreover, WM fractal function was integrated with the developed ANNs models

Speed

in order to generate an articially fractal predicted proles at different cutting conditions.

Feed

The predicted proles were found statistically similar to the actual measured proles of test

Depth of cut

specimens.
2007 Elsevier B.V. All rights reserved.

1.

Introduction

Surface roughness is a commonly encountered problem in


machined surfaces. It is dened as the ner irregularities
of surface texture, which results from the inherent action
of the production process. Consequently, surface roughness
has a great inuence on product quality, and the part functional properties such as lubricant retentivity, void volume,
load bearing area, and frictional properties. Furthermore a
good-quality machined surface signicantly improves fatigue
strength, corrosion resistance, and creep life (Stark and Moon,
1999). Surface roughness is consisting of a multitude of appar-

Corresponding author.
E-mail address: khashabu@zu.edu.eg (U.A. Khashaba).
0924-0136/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2007.09.006

ently random peaks and valleys. When two rough surfaces


are brought to be in contact, it is occurred in smaller area,
which is called the real area of contact. This area is not only
a function of the surface topography but also on the study of
interfacial phenomena, such as friction and wears (Bhushan,
1999). Lee and Ren (1996) explained that surface roughness
plays an important role in affecting friction, wear, and lubrication of contacting bodies. Lundberg (1995) has investigated
the effect of surface roughness on the lubricant lm characteristics under conditions of combined normal and sliding
motion. Xiao Li et al. (2003) showed that surface roughness is
one of the parameters that greatly inuence the friction under

272

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

certain running conditions. Mahrenholtz et al. (2005) stated


that surface roughness of the contacting surfaces inuences
the frictional properties of those surfaces during the forming processes. It is clear now that surface roughness geometry
strongly inuences the manner in which the contacting surfaces are interacting. Furthermore, it is well known that the
nal geometry of surface roughness is inuenced by various
machining conditions such as spindle speed, feed, depth of
cut, tool ank wear, and vibration level (Selmy et al., 1989).
Many investigators used the neural networks for prediction
surface roughness parameters (Abburi and Dixit, 2006; Priya
and Ramamoorthy, 2007; Erzurumlu and Oktem, 2007), on the
other hand many future research works are needed to assessment the prediction of surface roughness proles that having
great impact on machine development.
The objective of the present work is to develop ANNs
models that can correlate the relationship between the cutting conditions and the corresponding surface roughness
proles characteristics. Fractal geometry approach will be
utilized for characterizing each machined surface by two
scale-independent parameters D and G. These parameters
are employed for simulating each measured roughness prole
by using WeierstrassMandelbrot (WM) fractal function. The
input patterns of the ANNs models are the cutting conditions,
and the output patterns are the fractal parameters that measured for each machined surface. The training process was
implemented several times in order to achieve the best ANNs
models. The actual and predicted proles will be compared by
plotting the probability density functions for both proles.

2.

Fractal geometry

Euclidian objects like points, curves, surfaces and cubes are


characterized by their dimensions as 0, 1, 2 and 3, respectively. Whereas, a multitude of objects found in nature (e.g.
rough surfaces) are disordered and irregular for which the
measures of length, area, and volume are scaledependent.
Fractal objects are dened as rough or fragmented geometric
shapes that can be subdivided in parts. Each part is nearly a
reduced copy of the whole object (Hammad and Issa, 1994).
Fractal objects have self-similarity or self-afnity property
so that if a small part of the object is enlarged sufciently
it appears statistically very similar to the whole object as
shown in Fig. 1. Moreover fractal objects are characterized
by non-integer dimension called fractal dimension D which
takes fractional values (1 < D < 2) (Bhushan, 2001). In order to

Fig. 2 Qualitative description of statistical self-afnity for


a surface prole (Zhu et al., 2003).

understand the non-integer dimension of fractal objects, it is


important to review some forms of classical fractals.
The Von Koch curve shown in Fig. 1, is constructed by
removing the middle third of a line segment of a unit length
(E0 ), and replacing it by two segments of equal lengths (E1 ).
In the subsequent stages each straight segment is broken into
three parts and the middle portion of each segment is replaced
by two parts (E2 , E3 , Ei ). If this process is repeated innite times
(E ), the Von Koch curve is obtained. The fractal dimension D
of Koch curve can be calculated from Eq. (1).
D=

log N
log m

(1)

where N is the number of equal parts (N = 4), and m is the


magnication value (m = 3). Fractal objects have the following properties: rst, they are continuous but not differentiable
anywhere. Secondly, the curve is exactly self-similar. Thirdly,
although the curve contains roughness at a large number of
scales, the dimension of the curve remains constant at all
scales. Again from Eq. (1) the fractal dimension of Koch curve
equals to D = 1.26. This scale-invariance of the dimension is
an important property, which will be utilized to characterize rough surfaces shown in Fig. 2. However, because rough
surfaces are self-afne objects (not self-similar), Eq. (1) could
not be used for characterizing the fractal dimension of rough
surfaces.

2.1.
Fractal characterization of surface roughness
proles
The proles of machined surfaces appear random, multi-scale,
and disordered. The mathematical properties of such proles are continuous everywhere but non-differentiable at all
points. Such proles are also known to be self-afne in roughness structure (Zhu et al., 2003). In addition, it was found that
the WM fractal function satises these properties of continuity, non-differentiability, and self-afnity. Therefore the WM
fractal function can be used to characterize and simulate such
proles (Majumdar and Bhushan, 1990; Majumdar and Tien,
1990). The modied WM function is given by:
max
 G D1 n
cos(2 n x/L)

z(x) = L

Fig. 1 Construction of the Von Koch curve at different


iterations Ei .

n=0

 (2D)n

(2)

where G is the vertical scaling parameter, D the fractal dimension (1 < D < 2), L the sample length of measured prole, x the
coordinates of x-axis for the measured sample length,  is chosen to be equal to 1.5 providing both the phase randomization

273

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

and high spectral density, and n is the frequency modes, which


corresponding to the reciprocal of the wavelength as  n = 1/n
and  > 1 (Komvopoulos, 2000; Sahoo et al., 2002). The structure function of the WM fractal function follows a power law
and can be expressed in the form (Lundberg, 1995):

S() =

S() =

1
L

[z(xss + ) z(x)] dx

(3)

1
(N )/x

[z(xi + ) z(xi )]

(4)

i=1

S() = G2(D1)  2(2D)

Wt (%)

C
Mn
Cr
S
Mo
Cu

0.07216
1.02
0.045
0.2798
0.0235
0.02935

Table 2 The machining trails that employed in


producing the rst and second groups
Speed (rpm)

(5)

The fractal parameters D and G can be determined


from the slope and interception point, respectively, of a leastsquares line that t the structure function data plotted on a
loglog graph. Because D and G are not functions of the
correlation spatial distance , they represent intrinsic scaleindependent surface roughness parameters. Hence a rough
surface can be generated numerically for contact mechanics
analysis by introducing the measured fractal parameters D
and G, sample length L, and scaling parameter  into Eq. (2)
(Komvopoulos, 2000).

Experimental work

The experimental work is planned to obtain the inputoutput


patterns for developing the ANNs models that can simulate
the machining process in face milling operations. The input
patterns to the developed ANNs models were; the cutting conditions (speed, feed, and depth of cut), pre-machining wear
values (the tool wear at the beginning of the machining process) and the machine tool vibrations (as an indication of
the machined tool rigidity). The corresponding output patterns are the fractal parameters of the measured proles of
all machined specimens.

3.1.

Element

(N)/x

where  is the spatial distance, N the number of the measured


prole data points, x the distance between each two adjacent
points of the prole heights and S() is the structure function of
the prole data points over the spatial distance . When the
plot of structure function S() of a measured surface prole
follows a power law with , a surface prole is said to be
fractal, and hence the fractal parameters D and G can be
calculated from Eq. (5) as follows (Bhushan, 1999):

3.

Table 1 Chemical composition of workpieces

Collection of input data sets

Machining tests were conducted on a 7.1 hp vertical milling


machine tool with speed range from 56 to 1800 rpm, and feed
range from 11.2 to 90 mm/min. An inserted cutter of 50 mm
outer diameter with tips in place, and three tips of P30
throwaway-cemented carbide material were used. All the test
specimens were prepared with dimensions; 80 mm length,
45 mm width, and 20 mm thickness. The chemical composition of the work-pieces is given in Table 1.

560
900
1400

Feed rate
(mm/min)

Depth of cut
(mm)

11.2
18
28

0.5
0.75

The total number of machined specimens was 41 specimens. They were divided into two groups; the rst group was
consisting of 36 specimens (that used as the training and
validation sets for developing the ANNs). The second group
was consisting of ve specimens (that used to test the performance of the trained ANNs models). The machining trials
that were employed for producing the rst and second groups
are illustrated in Tables 2 and 3, respectively. The machining
operations, which indicated in Table 2, were repeated with
different pre-machining tool wear to investigate how the tool
wears affecting the nal product surface quality. Before conducting the machining trials, up to 3 mm thickness of the
top surface of each specimen was cleaned in order to eliminate any skin defects that can adversely affect the machining
results. During the machining operations, cutting forces are
exerted on the cutting tool tips, which in turn, cause vibration during the machining process. Thus for each machining
operation the vibration level was determined by using an integration vibration meter, type 2513. The observed values of
vibration level are used as one of the patterns input signal
to train the developed ANNs models. Furthermore, since the
tool wear inuences the quality of the machined surfaces, the
dimensional accuracy, and consequently the economics of the
cutting operations, the cutting edge ank wear of inserted tips
were measured for each machining condition, and used as one
of the patterns input signal to train the ANNs models. The
measurements of tool ank wear were implemented by using
an optical microscope of type 2158 with a venire resolution of
0.01 mm.

Table 3 The machining trails that employed in


producing the third group
Test sample no.
1
2
3
4
5

Speed (rpm)
450
710
1120
1400
1800

Feed rate
(mm/min)
35.5
35.5
14
45
22.4

Depth of
cut (mm)
0.8
0.5
0.5
0.75
0.35

274

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

Fig. 3 The structure of ANN model for prediction of fractal parameters D and G.

3.2.

Collection of output data sets

All the machined surfaces were characterized by measuring the roughness proles at different six positions over
each machined surface, which gives 216 total measurements.
The measurements were carried out using Taylor Hobson
Surtronic 3+ surface roughness measuring instrument, taking a traversing length of 4.0 mm and a cut-off length of
0.8 mm.
The roughness prole was acquired in a data le through
personal computer (every micron is represented by two data
points) that connected with the roughness instrument. This
could be performed by using the surface analysis software
program ST3PL.EXE which associated with the measuring
instrument. The fractal parameters D and G were calculated and used as the output patterns to train the developed
ANNs models.

4.

Results and discussions

4.1.
Prediction of fractal parameters using ANNs
models
A neural network is a massively parallel-distributed processor made up of simple processing units, which have a natural
propensity for storing experimental knowledge and making it available for use. Modeling the manufacturing process
using the ANN model could be denoted as system identication problem, where the set of actual experiments, which
applied on the machine tool (unknown system), can be utilized as training examples for training and developing ANN
model. The difference between the actual output of manufacturing process and the corresponding response of the
ANN model provides an error signal. This error signal in
turn is used in adjusting the free parameters of the neural
network to minimize the squared difference between the output of the unknown system and the neural network (Haykin,
1999).
The networks are selected to be feed-forward backpropagation neural networks. The networks consist of three
layers: the input, hidden and output layer. The input layer
consists of ve input nodes, and the output layer has only

one node. Thus, in order to reach the best structure of the


neural networks models, it is required to change the number of neurons in the hidden layer for each network. The
network structure, such as number of neurons and layers
are very important factors that determine the functionality
and generalization capability of the network (Ezugwu et al.,
2005). Thus, for prediction purposes of fractal parameters D
and G by using ANNs models, the experimental data are
extracted and used as the inputoutput pairs to train and validate the developed ANNs models as depicted in Fig. 3. In
order to determine the best structure of ANNs models, four
different networks with different number of neurons in the
hidden layer were designed and tested. For each designed ANN
model, the training process is performed six times to involve
the possible variations in the measured fractal parameters for
each machined specimen. In addition, because the outcome
of the training process greatly depends on the initialization
of the weights and biases that is done randomly according
to the NguyenWidrow technique, each network structure
was employed three times. The networks were trained with
LevenbergMarquardt algorithm. This training algorithm was
chosen due to its high accuracy for function approximation
purposes. The whole application was developed using Neural Network Toolbox for use with MATLAB R.6.5 (Demuth and
Beal, 2000).
The criterion of selecting the best network structure is considered to be the minimum average absolute percent error
(APE) between the actual calculated fractal parameters D
and G and the corresponding predicted values obtained from
the trained network. Moreover the percent errors APE, which
resulted by the different structures of ANNs for both fractal
parameters D and G, are recorded in Table 4. Thus, it is
obvious from this table that the ANNs models, which is consisting of 5-5-1 structures, are selected to be the best neural
networks structures, since they have the minimum APE errors
(1.97 and 6.33% for D and G, respectively).
After obtaining the best structures of ANNs models, all cutting conditions (input pattern) were applied to them in order
to get the corresponding output pattern, which represented
the predicted fractal parameters D and G. These results
were recorded and used later in computing the arithmetic
mean of both predicted and experimental measured fractal
parameters. The differences between the actual and predicted

Table 4 The different structures of ANNs models for prediction the fractal parameters D and G

6.33a
5-5-1
5-5-1
5-5-1

2.33
2.55
2.27

1.34
1.73
1.59

2.24
1.98
2.19

1.38
1.60
1.93

2.19
1.98
2.00

2.21
0.91
3.06

1.97a

7.02
9.71
7.12

7.68
3.41
8.54

5.91
6.09
7.96

4.15
7.17
6.17

8.00
3.72
4.99

5.93
6.34
4.10
5-10-1
5-10-1
5-10-1

2.68
2.58
3.17

1.43
1.43
1.29

3.15
3.87
2.64

1.18
1.54
1.80

1.66
3.01
2.04

0.82
2.72
2.12

2.18

7.31
11.08
7.92

6.54
10.90
8.07

8.56
11.46
6.22

7.27
8.75
8.55

7.38
7.13
8.67

10.03
4.87
10.45

8.4

5-15-1
5-15-1
5-15-1

2.64
2.13
2.32

3.04
2.63
2.29

2.88
2.35
3.32

1.87
2.14
3.38

2.40
1.65
3.05

1.77
2.27
2.67

2.15

6.44
11.07
10.64

7.90
10.98
10.94

10.90
11.45
5.69

9.77
12.66
11.72

9.64
11.16
5.99

13.39
4.84
5.71

9.5

5-20-1
5-20-1
5-20-1

2.75
3.48
2.95

2.39
3.59
2.26

3.96
2.97
3.87

2.23
2.01
1.94

3.27
1.69
2.30

2.68
2.26
4.95

2.87

12.16
8.13
11.36

13.18
6.81
10.40

6.94
10.46
9.59

12.77
12.91
10.14

6.01
11.23
7.54

8.34
12.89
11.64

10.14

275

It is obvious from the above gures that there is a close


agreement between the actual measured and predicted fractal
parameters for training, validation and test data sets. However, they are not the same. Therefore, in order to decide
whether these differences are statistically signicant or not,
one-way analysis of variance (ANOVA) technique was conducted to both actual and predicted fractal parameters in
Table 5.
It is clear that all calculated p-values are greater than
0.05 (the estimated signicance level), which prove the goodness of the developed ANNs models. The performance of

4.2.
Performance assessment of the developed ANNs
models

fractal parameters for training, validation and test data sets


are depicted in Figs. 4 and 5.

Fig. 5 The actual and predicted D values for (a) training


and validation sets, (b) for test set. () Mean values of actual
D, () mean values of predicted D.

Fig. 4 The actual and predicted D values for (a) training


and validation sets, (b) for test set. () Mean values of actual
D, () mean values of predicted D.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

refers to the minimum values.


a

Mean
value (%)
Average values of APE at six different
positions for vertical scaling parameter (G)
Mean
value (%)
Average values of APE at six different
positions for fractal dimension (D)
Structure of
ANNs models

276

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

Table 5 ANOVA test results


The type of investigated parameter
Fractal dimension parameter (D)
Vertical scaling parameter (G)

p-Values for training and validation data sets


0.26
0.90

p-Values for test data set


0.76
0.98

Fig. 6 Performance of trained ANNs models, using regression analysis for prediction of (a) fractal parameter D, (b) fractal
parameter G.

Fig. 7 Comparison between the actual and predicted proles for (a) specimen no. 1, (b) specimen no. 2.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

the developed ANNs models was investigated by applying


the regression analysis between the actual measured and
the corresponding predicted fractal parameters D and G,
respectively as shown in Fig. 6. The correlation coefcient of
approximately 0.9 and 0.93 were obtained, which give strong
indication that the developed ANNs models can be employed
effectively to model and predict the fractal parameters D and
G.

4.3.

Prediction of surface roughness proles

For prediction purposes of surface roughness proles, the best


selected ANNs models were introduced by the new data sets
of cutting conditions, Table 3, as an input pattern. Accordingly the corresponding pairs of fractal parameters D and
G are obtained, which in turn substituted into the WM
fractal function to obtain the corresponding fractal generated roughness proles. Fig. 7(a and b) shows the actual
and the corresponding predicted prole for two test specimens (where each test specimen is represented by only one
prole). From these gures, it is obvious that, although the
actual and the predicted proles are not identical, they have
approximately the same probability distribution function PDF.
These differences between the actual and the predicted proles may be attributed to the following reasons: rst, due
to the randomness of surface roughness proles measuring
process themselves (the geometry of measured roughness
proles depends mainly on the size of stylus-tip). Secondly,
the predicted proles have more details than the actual measured roughness proles, because the WM fractal predicted
proles are mathematically generated proles. And hence,
they are not limited by the instrument resolutions. Consequently, it is sensible to accept the differences between
the actual and predicted proles that are shown at PDF
plots.

5.

Conclusions

In this work, the articial neural networks (ANNs) technique was utilized for the prediction of fractal parameters
D and G. The predicted fractal parameters were used in
the WM fractal function in order to generate an articially
fractal predicted proles. These models can help manufacturers to determine the appropriate cutting conditions, in
order to achieve specic surface roughness prole geometry, and hence achieve the desired tribological performance
(e.g. friction and wear) between the contacting surfaces. Several attempts were made to reach the best structure of
these networks. The obtained results can be listed as the
following:

1. The fractal geometry approach was successfully used in


characterizing the machined surfaces by two parameters
D and G, which are unique and scale-invariant.
2. The ANNs models of structure 5-5-1 were found to be the
best neural networks that can predict the fractal parameters D and G with about 98 and 94% accuracies,
respectively.

277

3. The performance of developed ANNs models, were statistically demonstrated by applying the following techniques:
rst, ANOVA technique which gave p-values of 0.76 and
0.98. Secondly, the regression analysis method which
resulted in the correlation coefcients R of 0.895 and 0.925
for prediction of fractal parameters D and G, respectively.
4. The trained ANNs models for prediction of fractal parameters D and G was integrated with the WM fractal
prole simulation function, in order to predict the surface roughness proles of corresponding applied cutting
conditions. The predicted proles were found to be statistically similar to the actual measured proles, based
on the plotted probability density functions. Moreover,
the predicted proles exhibited more details than the
actual measured roughness proles, so that they could
be used effectively for establishing different contact
models.

references

Abburi, N.R., Dixit, U.S., 2006. A knowledge-based system for the


prediction of surface roughness in turning process. Rob.
Comp. Integr. Manuf. 22, 363372.
Bhushan, B., 1999. Handbook of Micro-Nano Tribology. CRC Press.
Bhushan, B., 2001. Modern Tribology Handbook, vol. 1. CRC Press,
LIC.
Demuth, H., Beal, M., 2000. Neural Network Toolbox Users Guide
version 4 (Release 12). The Mathworks Inc.
Erzurumlu, T., Oktem, H., 2007. Comparison of response surface
model with neural network in determining the surface quality
of moulded parts. Mater. Des. 28, 459465.
Ezugwu, E.O., Fadare, D.A., Bonney, J., Da Silva, R.B., Sales, W.F.,
2005. Modelling the correlation between cutting and process
parameters in high-speed machining of Inconel 718 alloy
using an articial neural network. Int. J. Machine Tools Manuf.
Vol. 45, 13751385.
Hammad, A.M., Issa, A.M., 1994. Fractal dimension as a measure
of roughness of concrete fracture trajectories. J. Adv. Cem.
Based Mater. 1, 169177.
Haykin, S., 1999. Neural Networks a Comprehensive Foundation.
Prentice-Hall Inc.
Komvopoulos, K., 2000. Headdisk interface contact mechanics
for ultrahigh density magnetic recording. J. Wear 238, 1
11.
Lee, S.C., Ren, N., 1996. Behavior of elasticplastic rough surface
contacts as affected by surface topography, load, and material
hardness. Tribol. Trans. 39 (1), 6774.
Lundberg, J., 1995. Inuence of surface roughness on
normal-sliding lubrication. Tribol. Int. 28 (5), 317322.
Mahrenholtz, O., Bontcheva, N., Iankov, R., 2005. Inuence of
surface roughness on friction during metal forming
processes. J. Mater. Process. Technol. 159, 916.
Majumdar, A., Bhushan, B., 1990. Role of fractal geometry in
roughness characterization and contact mechanics of
surfaces. J. Tribol. 112, 205216.
Majumdar, A., Tien, C.L., 1990. Fractal characterization and
simulation of rough surfaces. Wear J. 136, 313327.
Priya, P., Ramamoorthy, B., 2007. The inuence of component
inclination on surface nish evaluation using digital image
processing. Int. J. Machine Tools Manuf. 47, 570579.
Sahoo, P., Chowdhury, Roy, S.K., 2002. A fractal analysis of
adhesive wear at the contact between rough solids. J. Wear
253, 924934.

278

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 0 ( 2 0 0 8 ) 271278

Selmy, A.I., El-Sonbaty, I., Shehata, F., Khashaba, U.A., 1989. Some
factors affecting the accuracy of turned parts. Scientic
Bulletin of the Faculty of Engineering, vol. 24 (2). Ain Shams
University, Egypt, pp. 356368.
Stark, G.A., Moon, K.S., 1999. Modeling of surface texture in the
peripheral milling process, using neural network, spline, and
fractal methods with evidence of Chaos, Trans. of the ASME. J.
Manuf. Sci. Eng. 121, 251256.

Xiao Li, B., Rosen, G., Naser Amini, Nilsson Per, H., 2003. A study
on the effect of surface topography on rough friction in roller
contact. J. Wear 254, 11621169.
Zhu, H., Ge, S., Hung, X., Zhang, D., Liu, J., 2003. Experimental
study on the characterization of worn surface topography
with characteristic roughness parameter. J. Wear 255,
309314.

Das könnte Ihnen auch gefallen