Figure 1. Experimental setup for the proposed MLR-IPSRR system.

99
The Journal of Technology Studies

The hardware included:

A Fadal vertical CNC milling machine
with multiple tool changing and a 15 HP
spindle.
A Kistler 9257B type dynamometer sensor, which provided dynamic measurement of the three orthogonal components
of a force signal (Fx, Fy, and Fz).
A Micro Switch 922 series 3-wire DC
proximity sensor, used to collect the
signal for counting the rotations of the
spindle as the tool was cutting.
A power supplier, used to amplify the
signals from the proximity and the
dynamometer sensors. This amplified
signal was then sent to the A/D board.
An omega CIO-DAS-1602/12 A/D
converter, used to convert both the
dynamometer and proximity sensor data
from analog to digital signals.
A P5 133 personal computer, which was
connected to collect data from the A/D
converter output via an I/O interface.

A 6061 aluminum workpiece with dimensions of 1.00" x 1 .00" x 1.00", which was
cut in the end milling operations.
In order to control end milling operations
and analyze the spindle revolution and cutting
force signals, the following software was
required: (a) Basic CNC codes, which were
applied to conduct cutting operations, and (b)
A/D converter software, which was used to convert data (proximity and cutting forces) from
analog signals to digital values. Using the
hardware and software setups, tests of cut were
performed. Figure 2 shows the data obtained
from this experimental run using spindle speed
(S = 2500 rpm), feed rate (F = 8 ipm), and
depth of cut (D = 0.08 in.).
The Cutting Forces Analysis
From Figure 2, the cutting force data were
collected from the dynamometer sensor; these
three forces (Fx, Fy, and Fz) cannot individually
represent the actual force affecting surface rough-

Figure 2. Proximity and cutting force digital data using cutting condition
S = 2500 rpm, F = 8 ipm, and D = 0.08 in.

The Journal of Technology Studies

100

Figure 3. Individual resultant cutting force Fr_xy and four peak forces
in one revolution at cutting condition of F = 20 ipm, S = 2000 rpm, D = 0.08 in.

ness. Four cutting force magnitudes

r_xy_peak,
z, and
r_xyz) were consid( r_xy,
ered as possible candidates for an input factor for
the MLR-IPSRR system. They are defined as:

ting tool teeth. In this study, r = 4.

1. Average resultant force of the x and y

directions per revolution ( r_xy).

The third type of force analyzed in this

study was the average cutting force in the z
direction per revolution ( z) and could be
given as:

By using the following equation, one could

find the individual resultant force (Fr_xy) from
the x and y directions as shown in Figure 3.

, (4)
where i = 1, 2, . . . m and m is the total data
points in one revolution.

(1)
where i is the data point in one revolution.
Then, the average resultant force in
one revolution ( r_xy) could be
given as:

4. Average resultant force of x, y, and z

directions per revolution ( r_xyz).

, (2)
where i = 1, 2, . . . m and m is the total data
points in one revolution.
2. Average resultant peak force (

3. Average z direction cutting force per

revolution ( z).

The researcher also wanted to analyze the

average resultant force of the x, y, and z directions in one revolution ( r_xyz). The force is
given as:
, (5)

).

r_xy_peak

By using the data shown in Figure 3, one

could also identify the peak force (Fr_xy_peak) from
the average resultant forces of the x and y directions ( r_xy) in the cut period of each tooth.
Then, the average resultant peak force in each
revolution ( r_xy_peak) could be given as:

, (3)
where i = 1, 2,. . . r and r is the number of cut-

where i is the data point in one revolution.

Then, the average resultant force in one
revolution ( r_xyz) could be given as:
, (6)
where i = 1, 2, 3. . . m and m is the total data
points in one revolution.
After the above-mentioned cutting forces
were formed, we examined the correlation coefficient between these cutting forces and surface
roughness. Equation 7 was used to compute the

, (7)

where is the correlation coefficient between the

average resultant cutting force (
) and
surface roughness, Rai is the ith surface roughness, i = 1, 2, . . . n (n is total data sets; here
n = 384), and is the ith average resultant cutting
force of the x and y directions, i = 1, 2, . . . n ,
(n is total data sets; here n = 384).

Similarly, the,, and were calculated. The

largest value of correlation coefficients between
the above-mentioned cutting forces and surface
roughness represented the most significant cutting force, which was then used in the development of the MLR-IPSRR system.
Experimental Design
In order to identify the most significant
cutting force for the MLR-IPSRR system, an
experimental design matrix was used to run and
collect the training data. The experimental
design matrix, including eight levels of feed rate
(6, 8, 10, 12, 14, 16, 18, and 20 ipm), four levels
of spindle speed (1750, 2000, 2250, and 2500

rpm), and three levels of depth of cut (0.04,

0.06, and 0.08 in.), was designed for the experiments with two replicates of each experiment.
Two end milling tools (1/2 in. with four teeth)
were used to cut the workpiece. Therefore, a
total of 8*4*3*2*2 = 384 sets of training data
were collected. Cutting forces (Fx, Fy, and Fz)
were collected using a dynamometer, as shown
in Figure 1. The average resultant force of the x
and y directions
, average resultant
peak force
, average cutting force
of the z direction
, and average resultant
force of the x, y, and z directions
were calculated using Equations 2, 3, 4, and 5.
The 384 specimens were measured offline
with a Pocket Surf stylus type profilometer (produced by Federal Products Co.) to obtain surface
roughness (Ra) in this study. A JMP (a product
of the SAS Institute) statistical software package
was used to calculate the correlation coefficient
between surface roughness and cutting forces.
The results were
therefore, the average resultant peak force of the
x and y directions
had the highest
correlation coefficient with surface roughness
and was selected as the input parameter for the
MLR-IPSRR system.
The Proposed MLR-IPSRR System
After the most significant force was identified, the MLR-IPSRR system shown in Figure 4
was proposed. From Figure 4, one can see the

Figure 4. The structure of the proposed MRL-IPSRR system.

101
The Journal of Technology Studies

correlation coefficient between surface roughness (Ra) and the average resultant force of the
x and y directions ( r_xy).

The Journal of Technology Studies

102

input parameters (average resultant peak force

[
], spindle speed [S], feed rate [F],
and depth of cut [D]) and the output parameter
(surface roughness [Ra]) used to generate the
MLR-IPSRR system. The proposed MLRIPSRR system is given as:

where are coefficients of the regression model,

Rai is the surface roughness, Fi is the feed rate,
Si is the spindle speed, Di is the depth of cut,
is the average resultant peak force of
the x and y directions, and i ~ N(0, 2), where
i is the number of data sets. To obtain data for
the development of a multiple regression prediction model, a total of 384 experimental runs
have taken place using the cutting combination
indicted in the experimental design section.
Therefore, in this study, i = 1, 2, 3,K384.
Analysis and Results of the System
After utilizing the JMP software package,
the results of the surface roughness MLR model
were generated as follows:
Ra (predicted) = 57.066-0.024*S + 4.142*F 0.001*(S - 2125)*(F-13) + 491.056*D +
0.630*(S - 2125)*(D - 0.06) + 41.820*(F 13)*(D - 0.06) - 0.351*
-0.0007*
- 75.787) + 0.015*
(S - 2125)*(
(F - 13)*(
- 75.787) - 0.326*
(D - 0.06)*(
- 5.787) + 0.099*
(S - 2125)*(F - 13)*(D - 0.06) - 0.0001*(S 2125)*(F - 13)*(
- 75.787) + 0.01*
(S - 2125)*(D - 0.06)*(
- 5.787) +
0.722*(F - 13)*(D - 0.06)* (
- 75.787)
-0.0016*(S - 2125)*(F - 13)*(D - 0.06)*
(75.787)
The effect of tests showed that the feed rate,
spindle speed, average resultant peak force, and
depth of cut influenced the surface roughness
significantly since the p values of each main
effect (feed rate, spindle speed, average resultant
peak force, and depth of cut) were less than a =

0.01 significant level). That is, the surface

roughness was mainly determined by the feed
rate, spindle speed, average resultant peak force,
and depth of cut in end milling operations. The
MLR model was also significant with the p
value less than a = 0.01. Therefore, the MLR
model can be effectively used in this research.
Once the MLR model had been established,
the MLR recognition model was tested using 20
sets of cutting conditions that were different
from the cutting conditions of experimental
designs. The MLR-IPSRR model was implemented in the prediction of the surface roughness while the machining process was taking
place. The results of the predicted surface
roughness (RaiMR) were compared with the finished parts (Rai-m) that were measured by using
a Pocket Surf portable surface roughness gauge.
Then, the individual precision _ iMLR of each
experimental run (i) was evaluated based on the
following equations.

(8)

(9)
where iMLR is the precision of ith testing run,
MLR
and is the average precision of the 20 testing data, i = 1...20.
The results showed that the capability of the
surface roughness prediction was about 86% for
the testing experimental data in this study.
Therefore, one can see that the surface roughness (Ra) can be predicted effectively by the
above-mentioned MLR-IPSRR system.
What We Learned
The purpose of this study was to analyze
cutting forces to find out the most significant
cutting force magnitude that affected surface
roughness and to evaluate whether a MLR
approach for surface roughness recognition could
be used for prediction in the IPSRR system. Our
main conclusions are summarized as follows:
)
The average resultant peak force (
was identified to be the most significant force
to affect surface roughness in this study.

tool material, tool radius, workpiece material,

and lubricants in the system and (b) using different methodologies, such as fuzzy logic,
neural networks, and fuzzy nets, to provide the
IPSRR system with a learning capability. With
this capability, the system could be adopted to
different machines produced from different
CNC manufacturers.

This research assumed that the CNC

milling machine was effective and stable to
conduct all experiments under each cutting condition using an HSS end mill to cut 6061 aluminum material. We believe that the proposed
IPSRR system could eventually be implemented
in the new age of CNC machines. This would
be more likely if additional research and testing
could be done such as (a) including different

Dr. Lieh-Dai (John) Yang is an associate

professor in the Department of Industrial
Management at Nan-Kai College, Nantou,
Taiwan.
Dr. Joseph Chen is a professor in the
Department of Industrial Education and
Technology at Iowa State University. He is a
member of Alpha Xi Chapter of Epsilon Pi Tau.

References
Chen, J. C., & Lou, M. S. (2000). Fuzzy-nets based approach to using an accelerometer for an in-process
surface roughness prediction system in milling operations. International Journal of Computer
Integrated Manufacturing, 13(4), 358-368.
Chen, J. C., & Lou, S-J. (1999). Statistical and fuzzy-logic approaches in on-line surface roughness
recognition system for end-milling operations. International Journal of Flexible Automation and
Integrated Manufacturing, 6(1&2), 53-78.
Chen, J. C., & Savage, M. (2001). A fuzzy-net-based multiple in-process surface roughness recognition
system in milling operations. International Journal of Advanced Manufacturing Technology, 17, 670-676.
Coker, S. A., & Shin, Y. C. (1996). In-process control of surface roughness due to tool wear using a new
ultrasonic system. International Journal of Machine Tools & Manufacture, 36(3), 411-422.
Degarmo, E. P., Black, J. T., & Kohser, R. A. (1999). Material processes in manufacturing. Upper Saddle
River, NJ: Prentice Hall.
Fuh, K-H., & Wu, C-F. (1995). A proposed statistical model for surface quality prediction in end-milling
of Al alloy. International Journal of Machine Tools & Manufacture, 35(8), 1187-1200.
Jang, D-Y., Choi, Y-G., Kim, H-G., & Hsiao, A. (1996). Study of the correlation between surface
roughness and cutting vibrations to develop an on-line roughness measuring technique in hard
turning. International Journal of Machine Tools & Manufacture, 36(4), 453-464.
Lee, T. S., & Lin, Y. J. (2000). A 3D predictive cutting force model for end milling of parts having
sculptured surfaces. International Journal of Advanced Manufacturing Technology, 16, 773-783.
Montgomery, D. C. (1997). Design and analysis of experiments. New York:Wiley.
Susic, E., & Grabec, I. (1995). Application of a neural network to the estimation of surface roughness
from AE signals generated by friction process. International Journal of Machine Tools &
Manufacture, 35(8), 1077-1086.
Tsai, Y., Chen, J. C., & Lou, S. (1999). An in-process surface recognition system based on neural
networks in end milling cutting operations. International Journal of Machine Tools & Manufacture,
39, 583-605.

103
The Journal of Technology Studies

Spindle speed, feed rate, average peak cutting

force, and depth of cut are significant in
affecting surface roughness in end milling
operations and the determination of the
coefficient is R2 of 0.62 in the MLR model.
The MLR-IPSRR system is approximately
86% accurate in predicting surface roughness
while the machining process is taking place.