Académique Documents
Professionnel Documents
Culture Documents
CHAPTER-1
INTRODUCTION
1.1. Introduction
quality and performance of the product with lower cost in short span
have been implemented during the past decades. CNC machine tools
radial depth of cut, Nose radius that control the cutting operation can
both tool and work piece are uncontrolled [HUYNTH and FAN, 1992].
hands on data tables that are furnished in data hand books. Lin
(1994) suggested that are trial and error approach could be followed in
parameters such as Nose radius, spindle speed, feed rate, axial depth
of cut and radial depth of cut for keeping the desired surface finish
expects [DEMING].
4. Quality is what the customer says, it is [FEIGENBAUM].
5. Quality is the minimum loss imparted by the product to the
TAGUCHI].
4
presented below.
Modeling Techniques
A. Conventional modeling techniques
1. Full factorial design.
2. Response surface methodology.
3. Regression modeling.
4. Bond graph.
5. FEM.
of P20 mould steel on CNC end milling machine with coated carbide
5
The project work has been carried out on CNC End milling
Chapter-1: Introduction.
networks.
network.
CHAPTER-2
LITERATURE SURVEY
2.1. Introduction
metal at a very fast rate. The machine can hold one or more number
7
The first milling machine came into existence in about 1770 and
milling machine was designed by, Eli Whitney in the year 1818.
as the work piece material, the cutting tool material, the rigidity of the
machine, the rigidity of the work piece, cutting speed, feed, axial
described below:
2.3.2. Feed
which the work piece advances under the cutter. The Units are
methods.
The feed per tooth is defined by the distance the work advances
expressed in mm/tooth.
The feed per cutter revolution is the distance the work advances
in the time when the cutter turns through one complete revolution. It
The feed per tooth, the feed per cutter revolution, and the feed
expressed in mm.
The depth of the tool along its radius in the work piece as it
makes a cut. If the radial depth of cut is less than the tool radius, the
radial depth of cut is equal to the tool diameter, the cutting tool is
Surface integrity is the sum of all elements that describes all the
describes the roughness, lay, or texture of this outer most layer of the
work piece i.e., its interface with the environment. The second is
below the surface with respect to the base of matrix material. This
11
that created the surface. The deviation also depends on the properties,
2.6.1. Roughness
2.6.2. Waviness
2.6.3. Lay
geometry used.
and valleys in the direction that the tool was drawn across the
surface. The shape of the lay can take one of several forms as shown
2.6.4. Flaws
of a part surface.
2.7.1. Profile
waviness and form error. Optionally, the roughness may also exclude
16
error) that determines its friction in contact with another surface. The
for coating or sealing. For moving parts the roughness determines how
the surface will wear, how well it will retain lubricant, and how well it
the sum of the squares of the deviations of the profile from the mean
"level" the raw traced profile. The traced profile is relative to the
tilted with respect to the measured surface. A least squares line fit
(usually a sample length) of the profile such that the total areas
between the line and the profile are the same above and below the
above the mean line and negative below the mean line.
A profile peak is a region of the profile that lies above the mean
line and intersects the mean line at each end. In the fig.2.7, each
profile that lies below the mean line and intersects it at each end. The
depth of a valley is the depth of the lowest point within the valley.
consideration.
Fig.1 Fig.2
21
Fig.3 Fig.4
c = Sampling length,
d = Direction of lay,
e = Machining allowance,
50 N12 ~
25 N11
12.5 N10
6.3 N9
3.2 N8
1.6 N7
0.8 N6
22
0.4 N5
0.2 N4
0.1 N3
0.05 N2
0.025 N1
Roughness average(Ra)
roughness. Therefore,
Ra = ----------------------- (2.3.)
Where Ra= the arithmetic average deviation from the mean line,
L= sampling length,
Rq = --------------------- (2.4)
This is the distance between two lines parallel to the mean line that
contacts the extreme upper and lower points on the profile within the
not reveal true depth pattern. Most comparative methods require skill
and judgment.
following.
1. Straight edge.
2. Straight gauge.
3. Optical flat.
5. Profilometer.
6. Profilograph.
7. Talysurf.
8. Surtronic machine.
machining of modified AISI P20 tool steel in end milling process. They
25
concluded that, higher the feed rates, larger the cutting forces. They
also developed the genetic network model to predict the cutting forces.
increase of feed rate but decreases with increase of insert nose radius.
cutting speed, feed and depth of cut on cutting forces with different
comes to high speed milling, where small cut sections are employed.
This can cause uneven wear of the tool tips due to uneven chip loads.
the tool life in end milling of hardened steel AISI D2. This technique
the cutting parameters (cutting speed, feed and depth of cut) on the
tool life.
machining of hardened AISI H13 and D2 moulds, dies and press tools.
The primary objective was to assess the drilling and tapping of AISI
26
D2 and H13 with carbide cutting tools, in terms of tool life, work piece
quality, productivity and costs. The secondary aim was to assess the
rate (MRR) of AL6061 in CNC face milling operation. The objective was
removal rate. According to them, in rough operation, the feed rate and
depth of cut are the most significant effect parameters on R a and MRR
milling operation of modied AISI P20 tool steel using the response
predict the tool life in end milling of hardened steel AISI H13 hardened
tool steel. This technique used central composite design in the design
the P20 mould steel (357 HB) in dry and wet milling conditions. They
considered a range of 75125 m/min for the cutting speed and a feed
ranging between 0.3 and 0.7 mm/tooth: they found the cutting forces
Liao and Lin[15] (2007) studied the milling process of P20 steel
and the feed between 0.1-0.2mm/tooth. The authors found that the
tool life is higher with MQL, due to an oxide layer formed on the tool
CHAPTER-3
NETWORKS
3.1. Introduction
Other Definitions
1. Factorial Design.
3. Mixture Design.
4. Taguchi Design.
project work.
the engineers experiment passes (Ealey, 1998). The array allows the
engineer to vary multiple variables at one time and obtain the effects
design process, such that the quality is build into the product during
approach.
---------- (3.1)
experiments.
Analysis is easy.
exactly.
means.
English Statistician Sir Ronald A. Fisher during the first part of this
century. Much of the early work in this area dealt with agricultural
response.
(Box and Draper, 1987) and then migrated into modeling of numerical
al., 1990, Toropov et al., 1996). In RSM the errors are assumed to be
random.
3. D-optimal designs.
6. Audze-Eglais approach.
E(y) = Y = E [f (x1, x2, -------xn)] + E () = f (x1, x2, -------xn); ---------- (3.3)
The variables x1, x2, -------xn equation (3.3) are usually called the
terms of the coded variables the response function (3.3) will be written
as
For the case of three variables, the first order model in terms of
coded variables is
called a main effects model, because it includes only the main effects
of the true variables x1, x2, and x3. If there is an interaction between
Often the curvature in the true response surface is strong enough that
37
situations. For the case of three variables the second order model is
written as
several reasons.
order model. The method of least squares can be used for this
purpose.
problems.
-------------------- (3.9)
order greater than two are used. The general motivation for a
the Taylor series expansion around the point X10, X20-------------, Xk0.
Finally, lets note that there is a close connection between RSM and
analysis.
Def: when ever two variables are so related that the changes in one
along X-axis and other along the Y-axis. The resulting showing a
points in this scatter diagram seem to line near a line the correlation
is called linear. If all points seem to line near some curve the
correlated.
40
Correlation Coefficient
SS error
R 2 1 ---------------- (3.11)
SS tota l
Regression
neurons.
and has survived at least one major setback and several eras.
the technology available at that time did not allow them to do too
much.
patterns and detect trends that are too complex to be noticed by either
given to analyze.
time.
structure called a synapse converts the activity from the axon into
electrical effects that inhibit or excite activity from the axon into
Dendrites
Axon hillock
Axon
Synapse
Bias
bk
I1
Wk1
k Output
Wk2
I2
Summing Activation /
Ok
Wk3 Junction Transfer function
I3
.
.
Wkm
.
Fig.3.2. Artificial Neuron
Im
Inputs
Weights NN.
1. Feed-Forward
2. Recurrent NN.
4. Self-organizing Map.
6. Others.
inputs x1, x2, -------xn and output y . Each arrow in the figure
45
layers. The input layer consists of just the inputs to the network. Then
Fig.3.3. A feed forward network with one hidden layer and one
output.
---------------- (3.12)
Where the weights {wj, bj} are symbolized with the arrows feeding into
the neuron.
the output is called the output layer. In Figure 3.3 there is only one
46
the perceptron, with the exception that their activation functions can
------------------(3.13.)
the parameter vector . The size of the input and output layers are
-----------(3.14.)
the squared errors between the prediction and the experimental data.
The weights and biases are changed incrementally after each iteration.
neurons are organized in layers, and send their signals forward, and
inputs by neurons in the input layer, and the output of the network is
and then the error (difference between actual and expected results) is
48
this error, until the ANN learns the training data. The training begins
with random weights, and the goal is to adjust them so that the error
will be minimal.
--------------- (3.15)
inputs to the jth neuron in the hidden layer. W ij is the weight of the
connection from the ith neuron in the forward layer to the j th neuron in
the hidden layer. A neuron in the network produces its output (Out j)
1
Out f(net )
j j netj ------------------- (3.16)
1 e
output when certain inputs are given. Since the error is the difference
between the actual and the desired output, the error depends on the
error. We can define the error function for the output of each neuron:
49
-------------- (3.17)
output units of the network, and the sums run over time and over the
output units. When the mean square error is minimized, the power of
the error (i.e. the power of the difference between the desired and the
We take the square of the difference between the output and the
greater if the difference is big, and lesser if the difference is small. The
error of the network will simply be the sum of the errors of all the
depends on the output, inputs, and weights. After we find this, we can
new old
Wij = Wij + Wij -------------------- (3.18)
-------------------- (3.19)
Where E is the mean square error and out j is the jth neuron output.
Telecommunication,
Signal processing.
Prediction.
Process control.
CHAPTER-4
DATA COLLECTION
4.1. Introduction
tool removes the metal from flat work piece. During the operation,
high temperature and forces affect the life of the cutting tool. If cutting
tool fails, it will lead to poor surface finish. So, the surface roughness
cutting tool. The five cutting parameters selected for the present work
is nose radius (r, mm), cutting speed (V, m/min), feed rate (f,
mm/min), axial depth of cut (d, mm) and radial depth of cut (rd, mm).
51
outcome effects are not linearly related, it has been decided to use five
level tests for cutting speed, feed rate, axial depth of cut, radial depth
of cut and two levels for nose radius. The tool holder used for milling
milling machine and its specifications are given in Fig.4.1 and Table-
4.1 respectively.
DESCRIPTION SPECIFICATION
Y-axis 510 mm
Z-axis 510mm
Power 13 KW
Positioning 0.005mm
52
Repeatability 0.0025 mm
Make HARDINGE(TAIWAN)
Baudrate 9600
the work material is given in the Table 4.2. The different coated
carbide cutting tool inserts (TN450) used for the present project work
Carbon 0.35-0.45
Silicon 0.2-0.4
Manganese 1.3-1.6
Chromium 1.8-2.1
molybdenum 0.15-0.25
The machining parameters used and their levels chosen are given in
Table 4.3.
Levels
Control
Units symbol Level Level Level Level Level
parameters
1 2 3 4 5
Cutting
m/min V 75 80 85 90 95
speed
Axial depth
mm d 0.5 0.75 1 1.25 1.5
of cut
Radial
depth of mm rd 0.3 0.4 0.5 0.6 0.7
cut
55
this experiment. Because, nose radius with two levels and cutting
speed, cutting feed, axial depth of cut and radial depth of cut with five
L50(2^1*5^11)
S. NO Nose Cutting Cutting Axial depth Radial depth
Radius Speed Feed of cut of cut
1 1 1 1 1 1
2 1 1 2 2 2
3 1 1 3 3 3
4 1 1 4 4 4
5 1 1 5 5 5
6 1 2 1 2 3
7 1 2 2 3 4
8 1 2 3 4 5
9 1 2 4 5 1
10 1 2 5 1 2
11 1 3 1 3 5
12 1 3 2 4 1
13 1 3 3 5 2
14 1 3 4 1 3
15 1 3 5 2 4
16 1 4 1 4 2
17 1 4 2 5 3
18 1 4 3 1 4
19 1 4 4 2 5
20 1 4 5 3 1
21 1 5 1 5 4
22 1 5 2 1 5
23 1 5 3 2 1
56
24 1 5 4 3 2
25 1 5 5 4 3
26 2 1 1 1 4
27 2 1 2 2 5
28 2 1 3 3 1
29 2 1 4 4 2
30 2 1 5 5 3
31 2 2 1 2 1
32 2 2 2 3 2
33 2 2 3 4 3
34 2 2 4 5 4
35 2 2 5 1 5
36 2 3 1 3 3
37 2 3 2 4 4
38 2 3 3 5 5
S. NO Nose Cutting Cutting Axial depth Radial depth
Radius Speed Feed of cut of cut
39 2 3 4 1 1
40 2 3 5 2 2
41 2 4 1 4 5
42 2 4 2 5 1
43 2 4 3 1 2
44 2 4 4 2 3
45 2 4 5 3 4
46 2 5 1 5 2
47 2 5 2 1 3
48 2 5 3 2 4
49 2 5 4 3 5
50 2 5 5 4 1
(mm)
Axial
Nose Cutting Feed Radial
depth of
S.No Radius Speed depth of
cut
(mm) (mm/min) (mm/tooth) cut (mm)
(mm)
Axial
Nose Cutting Feed Radial
depth of
S.No Radius Speed depth of
cut
(mm) (mm/min) (mm/tooth) cut (mm)
(mm)
Table -4.6
machine.
Parameter Description
Measurement metric/inch.
Ra MRR MRR
S.NO S.NO R Ra (m)
(m) (cm3/min) (cm3/min)
1 0.94 23.39 26 0.98 46
2 1.16 58.73 27 1.2 97.4
3 1.12 116.6 28 1.68 69.06
4 0.86 197.8 29 1.06 136.3
5 0.66 298.8 30 0.52 230.8
6 0.82 63.13 31 1.14 37.07
7 1.44 121.7 32 2.48 83.54
8 0.7 206.6 33 1.74 154.9
9 0.92 127.7 34 1.48 253.4
10 1.28 66.4 35 1.72 109.2
11 1.08 117.9 36 0.52 88.5
12 1.3 81.54 37 0.92 161.3
13 1.48 158.4 38 0.76 257.7
14 1.44 76.1 39 0.64 45.33
15 1.54 152.1 40 0.96 105.3
16 0.56 94.13 41 0.8 156.3
17 0.46 174 42 0.5 103.8
18 0.42 81.3 43 1.54 56.18
19 0.58 161.3 44 1.27 120.6
20 0.5 109.1 45 1.32 214.1
21 0.46 173.4 46 0.87 119.3
22 1.1 81.97 47 1.1 61.35
61
Chapter-5
METHODOLOGY
5.1. Methodology
Response surface model and ANN model methodology for the data
presented in chapter-4.
The method of finding the equation of the line which the best fits for a
Ra = 0+1(R)+2(V)+3(f)+4(d)+5(rd)------------------------------------(5.1)
Ra*d = 0d+1Rd+2Vd+3Fd+4d2+5drd-------------------(5.6)
Ra*rd= 0rd+1Rrd+2Vrd+3Frd+4drd+5rd2-------------(5.7)
regression coefficients.
MRR = 0+1(R)+2(V)+3(f)+4(d)+5(rd)--------------------------------(5.8)
MRR*R = 0R+1R2+2RV+3Rf+4Rd+5Rrd-------------(5.10)
MRR*d =0d+1Rd+2Vd+3Fd+4d2+5drd---------------(5.13)
MRR*rd= 0rd+1Rrd+2Vrd+3Frd+4drd+5rd2--------(5.14)
63
regression coefficients.
The data collected from the experiments was used to build the
(mm/tooth), axial depth of cut (mm), radial depth of cut (mm) and the
and MRR.
axial depth of cut (mm) and radial depth of cut (mm). The relationship
expressed as follows.
Ra = 0+1(R)+2(V)+3(f)+4(d)+5(rd)----------------------------------(5.15)
64
found to be 0.217. This shows that first order model can explain the
Ra = 2.28840+0.509R-0.01866v+0.836f-0.2744d-0.089rd.--------(5.16)
depth of cut (mm) and radial depth of cut (mm). The relationship
expressed as follows
Ra=0+1(R)+2(V)+3(f)+4(d)+5(rd)+6R2+7v2+8f2+9d2+10rd2+11Rv
was found to be 0.304. This shows that second order model can
Ra = -11.7967+0.509R+0.262569v+25.2817f+0.831886d-0.089rd-
0.00165429V*V-81.4857f*f-0.553143d*d. ------------------(5.18)
are presented in the Table 5.1. This analysis is carried out for a level
Total 49 8.82046
calculated value is greater than F table value (F 0.05, 8, 41= 2.176). Hence
cut (mm) and radial depth of cut (mm). The relationship between the
MRR = 0+1(R)+2(V)+3(f)+4(d)+5(rd)-------------------------------(5.19)
found to be 0.93. This shows that first order model can explain the
MRR = -275.586+2.08R+0.514940v+784.560f+123.377d-233.079rd
-----(5.20)
(mm), cutting speed (m/min), feed (mm/tooth), axial depth of cut (mm)
and radial depth of cut (mm). The relationship between the MRR and
MRR = 0+1(R)+2(V)+3(f)+4(d)+5(rd)+6R2+7v2+8f2+9d2+10rd2+
11Rv+12Rf+13Rr+14Rrd+15vf+16vd+17vrd+ 18rrd---(5.21)
was found to be 0.998. This shows that second order model can
MRR=1135.67-21.1811v-4915.39f+166.731d-226.271rd +0.09638v*v
+3000.75f*f+35.0434d*d-166.407rd*rd+47.5774v*f-2.51327v*d
in the table 5.2. This analysis is carried out for a level of significance
Total 49 202850
adequate.
68
are obtained from the training process applied to the given neural
error between the calculated outputs and observed outputs from the
model. These networks are used for a generalization of the MLP (Multi-
layer perceptron) such that connections can jump over one or more
layers. The network has five inputs of Nose radius, cutting speed,
feed, axial depth of cut and radial depth of cut and outputs of surface
roughness and MRR. The size of hidden layer is two of the most
layer feed forward network. Two hidden layer was adopted for the
hidden layer and two neurons in the output layer. Using a neural
networks is checked with the means square error and best network is
selected which has the lowest mean square error among the different
networks. In this study 5-8-8-2 network was selected which has the
this study, was designed using NeuroSolutions 4.0 and shown in Fig
5.5. The network consists of one input, two hidden and one output
layer. The input layer has five neurons, two hidden layer has eight
neurons and output layer has two neurons respectively. Since surface
Feed, axial depth of cut and radial depth of cut was the main interest
Hidden layrer-1 Hidden layrer-2
in this research. Neurons in the input layer corresponding to the Nose
radius, cutting
Input layer speed, feed, axial depth of cut and radial depth of cut,
Output layer
the output layer corresponds to surface finish and MRR. The input
Nose radius
Surface
layer, hidden and output layer will apply a sigmoidal roughness
activation
Cutting speed
function.
Outj=
Feed
f (netj) = 1/ (1+e-netj)
Axial depth of cut
MRR
data sets is utilized. The new data set consists of 50 analysis results
data set used for testing of the developed model which were not used
predicted by artificial neural networks for train and test data are given
in Table 5.3 and Table 5.4 and percentage deviation is calculated and
2
%Deviation Absolute % Deviation
1.5
n 1
0.5
tio
via
0
De
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
% -0.5
-1
-1.5
-2
Experiment No.
2.5
2 % Deviation Absolute % Deviation
1.5
1
0.5
0
-0.5 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-1
-1.5
% Deviation for MRR
-2
-2.5
Experiment No.
error (MSE) between the network prediction and training data set as
shown below:
new old
W W W -------------- (5.23)
ij ij ij
k E
W k -t out ------------ (5.24)
ij t 1 W j
ij
Where E is the MSE and outj is the jth neuron output. is the learning
rate of convergence of the network. The learning rate [step size 1.0,
2.41E-06. Fig.5.4 and Fig.5.5, depicts the average MSE with standard
output for training data sets are presented in Table 5.3. In Order to
judge the ability and efficiency of the model to predict the surface
Experimental Predicted
i 100%
Experimental
(4.10)
74
n
i
i1
n
Exp ANN
Exp MRR ANN MRR
S.No Ra Ra
(cm3/min) (cm3/min)
(m) (m)
1. 1.96 73 1.992 56.34
2. 1.96 73 1.992 56.34
3. 0.6 203.5 0.728 277.9
4. 0.78 200.5 0.789 272.7
5. 0.64 134 0.641 134.6
6. 1.02 154 1.075 148.7
7. 1.16 107 1.01 141.8
8. 1.56 92 1.512 92.05
9. 0.68 179.2 0.687 221.5
10. 1.24 92 1.321 122.3
11. 0.94 104 0.955 102.9
12. 1.18 118 1.182 136.2
13. 1.28 84 1.3 91.33
14. 0.62 197 0.64 212.7
75
experimental results and the network can be used for testing of the
network. Hence the testing data sets are applied which were never
respectively.
50
% Deviation Absolute % deviation
40
30
20
10
0
-10 1 2 3 4 5 6 7 8 9 10 11 12 13 14
-20
-30
% Deviation for MRR
-40
-50
Experiment No.
CHAPTER-6
6.1. Introduction
(mm/tooth), axial depth of cut (mm) and radial depth of cut (mm) on
Fig 6.1. to Fig. 6.8 as contours. These response contours can help in
experimental domain.
78
Fig.6.2. shows that at lower levels of nose radius and feed rates
removal rate increases. The increase of feed rate increased the heat
the surface roughness increases while axial depth of cut increases the
cut the material cant be removed fully which leads to high surface
Fig.6.4. shows that when nose radius and radial depth of cut
high and radial depth of cut increases the MRR increases due to this
roughness increases.
From the Fig.6.7, it is observed that when cutting and radial depth of
model:
the data are spread roughly along the straight line. Hence it is
considered that the data are normally distributed Fig.6.10 shows, the
that the residuals are distributed evenly in both positive and negative
99
95
(response is Ra)
90
80
70
60
Percent
50
40
30
20
10
1
-1.0 -0.5 0.0 0.5 1.0
Residual
(response is Ra)
1.0
0.5
Residual
0.0
-0.5
1 5 10 15 20 25 30 35 40 45 50
Observation Order
0.5
Residual
0.0
-0.5
The results predicted by the RSM model were compared with the
200
% deviation Ra absolute % deviation Ra
150
100
50
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-50
% Deviation of Ra
-100
-150
-200
Exeriment No.
Exp Ra RSM Ra
S.No R V F d rd
(m) (m)
1 0.8 78 0.14 0.8 0.4 1.96 1.24442
2 0.8 78 0.14 0.8 0.4 1.96 1.24442
3 0.8 78 0.16 1.35 0.6 0.6 1.04679
4 0.8 87 0.14 1.35 0.6 0.78 0.93657
5 0.8 87 0.16 0.8 0.6 0.64 1.14984
6 0.8 87 0.16 1.35 0.4 1.02 0.97109
7 1.2 78 0.16 1.35 0.4 1.16 1.26819
8 1.2 78 0.16 0.8 0.6 1.56 1.44694
9 1.2 78 0.14 1.35 0.6 0.68 1.23367
10 1.2 87 0.16 0.8 0.4 1.24 1.37124
11 1.2 87 0.14 1.35 0.4 0.94 1.15797
12 1.2 87 0.14 0.8 0.6 1.18 1.33672
13 0.8 90 0.1 1.5 0.3 1.28 0.53186
14 0.8 95 0.1 1.5 0.7 0.62 0.27889
87
Fig 6.14. to Fig. 6.19 as contours. These response contours can help
cut increases the MRR also increases. Generally, as the depth cut
depth of cut increases the material removal rate increases. When feed
axial depth of cut, radial depth of cut and cutting feed increases the
cut the material is being clear of from the surface, hence MRR
increases.
91
in Fig.6.20. The data are spread roughly along the straight line. Hence
Shows the residuals against the observation order. Fig 6.21 is used to
show that the residuals are distributed evenly in both positive and
negative along the run. Hence the data can be said to be independent.
99
(response is MRR)
95
90
80
70
Percent
60
50
40
30
20
10
1
-7.5 -5.0 -2.5 0.0 2.5 5.0
Residual
(response is MRR)
5.0
2.5
Residual
0.0
-2.5
-5.0
-7.5
1 5 10 15 20 25 30 35 40 45 50
Observation Order
percentage deviation has been calculated for MRR and shown in Fig.
20
% deviation MRR absolute % deviation MRR
15
10
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-5
% deviation MRR
-10
-15
-20
Experiment No.
for MRR.
nose radius, cutting speed, feed, axial depth of cut, radial depth of cut
observed that 5-8-8-2 network structure has the lowest MSE and it is
Performance Ra MRR
MSE 2.00597E-06 0.189657695
NMSE 1.13711E-05 4.67483E-05
MAE 0.000871394 0.308762985
Min Abs Error 8.69751E-06 7.98035E-05
Max Abs Error 0.00669662 1.491375732
R 0.999994394 0.999977327
From the Table 6.5, the MSE value for surface roughness is
training and testing data set are presented in Table.6.6 and Table 6.7.
increases the mean square error is constant. The mean square error
at 5-8-8-2 is 2.13126E-05.
predicted values by the ANN model and the measured values resulted
Sensitivity Ra MRR
R 1.611631274 51.30732727
V 0.023935925 3.643531799
f 10.01920319 1249.439209
d 0.3577438 142.9523315
rd 0.633553028 439.4009094
101
n /
train data.
ANN ANN RSM RSM
Ra MRR
Exp.No Ra MRR MRR Ra
(m) (Cm3/min)
(m) (Cm3/min) (Cm3/min) (m)
106
1 0.94 23.39 0.9395 23.704 27.4 0.962
2 1.16 58.73 1.1608 59.075 61.185 1.162
3 1.12 116.55 1.12 116.47 118.139 1.191
4 0.86 197.78 0.8599 198.12 198.262 1.049
5 0.66 298.8 0.6617 297.31 301.553 0.736
6 0.82 63.13 0.8212 62.538 63.461 1.01
7 1.44 121.65 1.4397 121.9 122.95 1.141
8 0.7 206.61 0.7006 206.62 205.608 1.101
9 0.92 127.66 0.9203 127.87 131.223 0.934
10 1.28 66.4 1.2805 65.519 62.071 1.068
11 1.08 117.92 1.0804 117.86 116.36 0.906
12 1.3 81.54 1.2999 81.645 79.127 1.013
13 1.48 158.39 1.48 158.23 158.216 0.903
14 1.44 76.1 1.4397 76.962 78.999 1.139
15 1.54 152.13 1.5406 151.65 154.216 1.033
16 0.56 94.13 0.5586 94.07 92.183 0.695
17 0.46 174.01 0.4619 174.51 173.807 0.688
18 0.42 81.3 0.4203 81.159 84.526 1.025
19 0.58 161.29 0.579 161.3 162.278 1.021
20 0.5 109.05 0.4979 108.87 110.523 0.891
21 0.46 173.41 0.4667 173.52 177.997 0.288
22 1.1 81.97 1.0997 81.962 78.652 0.727
23 0.86 65.33 0.8602 65.287 64.048 0.87
24 0.48 137.55 0.4777 137.58 138.231 0.797
25 0.74 241.38 0.7394 241.55 235.583 0.553
26 0.98 46 0.9791 45.993 42.44 1.139
27 1.2 97.4 1.1998 97.5 93.789 1.339
ANN ANN RSM RSM
Ra MRR
Exp.No Ra MRR MRR Ra
(m) (Cm3/min)
(m) (Cm3/min) (Cm3/min) (m)
28 1.68 69.06 1.6799 68.824 68.053 1.412
29 1.06 136.31 1.0607 135.77 136.466 1.27
30 0.52 230.77 0.5199 230.9 228.049 0.957
31 1.14 37.07 1.1401 37.07 36.489 1.232
32 2.48 83.54 2.4794 83.622 84.269 1.362
33 1.74 154.89 1.7393 155.46 155.217 1.322
34 1.48 253.38 1.4794 252.87 249.335 1.111
35 1.72 109.17 1.7191 109.3 112.825 1.244
36 0.52 88.5 0.5183 87.755 89.084 1.128
37 0.92 161.29 0.9205 161.08 162.567 1.189
38 0.76 257.73 0.759 258.24 259.22 1.08
39 0.64 45.33 0.6387 46.357 39.927 1.36
40 0.96 105.34 0.9601 105.44 103.435 1.254
41 0.8 156.25 0.7999 156.51 158.516 0.872
42 0.5 103.81 0.5034 104.16 101.235 0.909
43 1.54 56.18 1.5408 55.708 56.859 1.246
44 1.27 120.58 1.2709 120.81 122.902 1.243
45 1.32 214.13 1.3202 214.07 212.113 1.068
46 0.87 119.33 0.867 118.69 116.83 0.509
47 1.1 61.35 1.1009 62 62.39 0.948
48 0.78 128.64 0.7801 128.3 130.968 1.046
107
ANN model for training data. From the Fig.6.40, it is concluded that,
difficult to distinguish the experimental and the fit curve because the
fit is so good.
108
Test data.
roughness, MRR by the RSM and ANN model for testing data. It is
experimental tests.
Chapter-7
surface roughness in the end milling of P20 mould steel has been
experimentation.
2. The feed is most dominant parameter for the surface roughness
and MRR. The cutting speed shows the minimal effect on the
P20 mould steel, the second order model has been modeled using
mould steel.
10. The ANN model could predict the surface roughness and MRR
11. The ANN model could predict the surface roughness and MRR
3. Further study could also consider the tool wear that would affect
Appendix-1
Table-1: Performance of Artificial Neural Network
Appendix-2
input and outputs lie between 0-1. For each training pair, Assume
form.
between l<m<2l.
= [random weights]
= [random weights]
= =0
Step-4: Further training data, present one set of inputs and outputs.
Present the pattern to the input larger [I] I as inputs to the input layer.
By using linear activation function, the outputs of the input layer may
be evaluated as = .
115
Step-6: Let the hidden layer units evaluate the output using the
sigmoidal function as
Step-8: Let the output layer units evaluate the output using sigmoidal
function as
Step-9: Calculate the error and the difference between the network
output and the desired output as for the ith training set as
116
Step-10: Find d as
Step-12: Find
Step-13: Find
Step-14: Find .
Step-15: Find .
Step-17: Repeat steps 4-16 until the convergence in the error rate is
Nose radius
Hidden layer
Cutting speed
Surface roughness
Feed
Axial depth of cut
MRR
Step-1:
Step-2:
Step-3:
118
Step-4:
Step-5:
step-6:
Step-7: Error =
d=
step-9:
Step-10:
Step-11: =
119
Step-12:
Step-13:
Step-14:
Step-1:
120
Step-2:
Step-3:
Step-4:
Step-5:
Step-6:
Step-7: Error =
Step-8:
d=
Step-9:
121
Step-10:
Step-11: =
Step-12: =
Step-13:
Step-14:
122
REFFERENCES
1. Dr. Mike S. Lou, Dr. Joseph C. Chen & Dr. Caleb M. Li surface
15, Number1-November1998-January1999).
2. Mr. John L. Yang & Dr. Joseph C. Chen A systematic approach
2008).
4. B. Sidda Reddy, K. Thirupathi Reddy, Syed Altaf Hussian
Number2-November2007-January 2008).
1-14, doi:10.1155/2007/92717.
123
Australia, pp.1-7..
21(2008) 4, 592-602.
AISI H13 and D2 moulds, dies and press tools, J. Mat. Proc.
10.1007/s11740-010-0232-7.
241247.
12. A.M. Khalid Hafiz, AKM Nurul Amin, ANM Mustafizul Karim and
inserts during the high speed end milling of AISI H13 Hardened
547.
http://www.neurosolutions.com/downloads/documentation.htm.
Richard A. Johnson
Choudhury,
125