Professional Documents
Culture Documents
Materials
Processing
Technology
Journal of Materials Processing Technology 62 (1996) 440-447
Abstract
Machine tool characteristics such as stiffness and vibration have a large influence on cutting processes. This paper presents an
evaluation approach of machine tool characteristics with adaptive prediction. Adaptive prediction can predict tool wear and surface
roughness by analysis and neural network, and adapt the parameters used for the prediction to practical processes. Machine tool
characteristics of two machine tools can be evaluated by the processes that are predicted using the parameters acquired in the same
operations. The effects of machine tool characteristics on the optimum cutting conditions and machining scheduling are shown with
the results of adaptive prediction.
Key words: tool wear, surface rough7ze.c.r,flzachiile tool, neural network, optimization, operr2tionplannirl2g, scheduling
(relative intensity of
units, 1 output unit)
I _ 1 Chip discontinuity
.I {chip strain) .
I>“”
r Built-up edge
ttmg temperature * (average temperature
around cutting edge)
-i
Table 2
Results of energy approach
Chip flow angle % 20.13 deg
Effective rake angle -6.40 deg
Effective shear angle ; 20.60 deg
Friction angle B 35.85 deg
Tool-chip contact length I, 1.588 mm
Shear velocity Ys 223 m/min
Chip velocity Vc 79.0 m/min
Principal component FH 507N
Feed component Fv 349N
Radial component FT 277N
Cutting conditions: material cut, 0.45% carbon steel;
tool, carbide P20 (-5,-5,5,5,15,15,0.8); cutting
speed, 200 m/min; depth of cut, 1.0 mm; feed rate,
0.2 mm/rev; lubrication. dry.
Table 1
Orthogonal cutting data and thermal constants
Work: 0.45% carbon steel, Tool: carbide P20
4 = e?cp( 0.0731/+ 1.458 X 103r1 +0.78Oa- 1.471) rad
/? = enp (-0.04OV - 1.522 X 103,1 +0597n f0.037) rad
zs = exp(O.O091/-2.011 X 102tI +0.414o-i-4.045)X9.8 MPa
I,= tzexp (- 0.096P 5.255 X 103c1 ~ 0.504crS 2.797) mm
Work Tool
Thermal conductivity W/(m . K) 46.05 67.0
Specific heat J@s . K) 502 398
Density kg/m 3 Fig. 4 Cutting temperature distribution calculated by FDM
784 1120 Cutting conditions are same as in Table 2. The width of
4, shear angle; B, friction angle on tool face; zs shear stress on lank wear land is 0.2mm. Thermal constants in Table 1 are
shear plane; and $, tool-chip contact length; V, cutting speed; used.
tl undeformed chtp thickness
of the width of wear land can be calculated by the following
estimating heat generation. 4 is assumed to be equal to equations:
flank wear rate for any cutting conditions and tool geometry. The network for predicting surface roughness has the
Initial wear offset VBOvaries with cutting conditions, tool, and following input information:
workpiece. Initial wear offsets, therefore, can be stored in the (a) Cutting speed
form of neural network, in which cutting conditions and tool (b) Affinity between tool and workpiece, which can be given
geometry are input information and V,, is that of output. The by relative intensity of Fe-K, on rake face measured with
network is built up for the combination of workpiece and tool EPMA [7].
material. (c) Chip discontinuity, which can be evaluated by chip strain.
Chip strain can be calculated in the prediction of cutting
2.2. Adaptive prediction of surface roughness force.
(d) Built-up edge formation, which can be evaluated by average
In the network prediction, neural networks shown in Figure temperature around cutting edge.
2 associate surface roughness and grooving wear in a given (e) Width of flank wear land, which can be predicted as
operation with machining results obtained before. mentioned in the previous section. Contact condition
In the neural network, the information propagates from between flank face and workpiece can be evaluated.
input layer (1st layer) to output layer (3rd layer), through (f) Theoretical roughness R,, considering tool wear. The
hidden layer (2nd layer). The input to i th unit in the k th layer, equation, presented by Solaja [13], is approximated by the
ni k, is given as: following equation:
‘ll,k= tC “z,~,k x "j,k - 11 + t~,k (5) R,, = f2 i 8H + (V, ~ Vi) tan y. (f < 2H sin C,) (9)
J
where wijkI I is the weight betweenj th unit in (k-l) th layer and wheref is feed rate; R, Ye, and C, are nose radius, end relief
i th unit in the k th layer; ujkml (u,,,J is the output of j (i )th unit angle, and end cutting angle of tool; Vs and VB’ are size of
in the k-l (k )th layer; and tik is the threshold value associated grooving wear and that of flank wear land on front cutting
with the i th unit in the k th layer. The output of a given unit is edge respectively. Being assumed to be equal to flank wear
assumed to be a sigmoid function of the input and can be on side cutting edge, VB’ can be predicted as mentioned in
expressed as: the previous section. Grooving wear, however, cannot be
predicted analytically. Another neural network, therefore,
u
‘.k = f@l.k) = { 1 + ex;( - ll,,k)j can be used for the prediction as shown in Figure 2. The
information of input layer units is cutting conditions, tool
Corresponding to the output (0.0 <f&J < 1.0) of output layer geometry, and grooving wear size (Vs)r at the time T
unit, surface roughness and grooving wear rate are related to considered; and that of output layer unit is grooving wear
the range from 0.0 to 1.0. rate dV#T. Grooving wear size (Vs)l.+dr at T+dT can be
The parameters to be adapted are the weight parameters
obtained by following equation:
between each unit connected in the networks. In adaptive
process, they are trained by back propagation [12]. For I th (V,),+dT=(VS)T+(dVsidT)‘dT (10)
data, predicted data Zpredr is compared to the practical data
Grooving wear size (VslrtdT is used for the next prediction
z prnc , I and then the error E of m data is calculated as follows:
of wear rate after time increment dT.
Surface roughness is the time-dependent process because
input information of the network includes the width of flank
Back propagation adjusts the weights and the threshold in wear land and size of grooving wear. Surface roughness,
Equation (5) using gradient information to minimize the error therefore, can be predicted for every machining time after the
E as follows: prediction of flank wear and grooving wear.
Table 3 0.20 mm
Machine tool data
Lathe A Lithe B
Allowance
Maximum machining diameter cpl80 @lO
Maximum machining length 280 600
Spindle
Power AC5.5 kw AC%5 kw
Revolution 112~4OOrpm 20-3600rpm 0.10
100 200 300 100 200 300
Cutting speed m/min Cutting speed m/min
Weight 2300kgf 3200kgf
(a) Lathe A (b) Lathe B
Fig. 6 Initial wear offset adapted
Cutting conditions are same as in Table 2, where cutting
speed and feed rate are variable.
mm
.20
i 8 Lathe A Measured
5 ) :: Lathe B Measured
Y -.-. - Lathe A Predicted
5 - Lathe B Predicted
ii I
0 _.
0 0
0 5 10 15 100 200 300
Cutting time min Cutting speed m/min Cutting speed m/min
(a) Lathe A (b) Lathe B
Fig. 5 Prediction of flank wear
Cutting conditions are same as in Table 2. Fig. 7 Prediction of flank wear at 10 minutes machining
Cutting conditions are same as in Table 2, where cutting
speed and feed rate are variable.
get the parameters used for the prediction. Flank wear processes predicted on two machine tools after
(2) The system shown in Figure 2 predicts cutting processes the adaptation of the parameters are shown in Figure 5.
using the parameters adapted. The predicted processes Predicted processes agree with practical processes, and those
show us the effects of machine tool characteristics on results confirm that the parameters can be adapted very well. In
cutting processes. the results, machine tool characteristics have a large influence
(3) The effects of machine tool characteristics on machining on initial flank wear. On the other hand, they have little
operations can be evaluated by operation planning and influence on flank wear rate. Figure 6 shows initial wear
machining scheduling using the results of adaptive offsets adapted on each machine tool with cutting speed and
prediction. Though operation planning and machining feed rate. There is a large difference in initial wear offset with
scheduling vary with planning conditions such as running cutting conditions. Figure 7 shows flank wear predicted with
cost, the evaluation of machine tool characteristics gives cutting speed and feed rate at ten minutes machining. The
us good guidelines for usage of machine tools. width of flank wear on lathe B is less than that on lathe A
when cutting speed and feed rate are low. Flank wear process on
3.1. Effects of mnchine tool chrncteristics on cutting lathe B, however, is sensitive to cutting speed in high speed
processes area.
Figure 8 and Figure 9 show grooving wear and surface
Two NC lathes, as shown in Table 3, are examined to roughness with cutting time respectively. The network of
identify their machine tool characteristics with the same grooving wear prediction for each machine tool has been
workpiece and tool. Adaptive prediction can adapt wear adapted to 2508 machining data. The network of surface
characteristic constants, initial wear offsets, and weight roughness prediction for lathe A has been adapted to 712 data;
parameters of the networks to six practical operations on each and that for lathe B has been adapted to 658 data. The results in
machine tool independently. Machine tool characteristics, these figures verify that adaptive prediction can adapt the
therefore, can be acquired in the form of the parameters. weight parameters of networks to practical processes very
T. Matsumura et al/Journal of Materials Processing Technology 62 (1996) 440--447 445
mm
0.20 0.20
w Lathe A Measured
0 Lathe B Measured
-.-__ Lathe A Predicted
- Lathe B Predicted
0.10 0.10
100 200 300
100 200 300
10 15 Cutting speed m/min Cutting speed m/min
CSting time min
(a) Lathe A (b) Lathe B
Fig. 8 Prediction of grooving wear
Cutting conditions are same as in Table 2. Fig. 10 Prediction of grooving wear at 10 minutes machining
s Cutting conditions are same as in Table 2, where cutting
speed and feed rate are variable.
w
0.20
:
E
a
Lathe A
Measured
n Lathe B
Measured
-.-._ Lathe A
Predicted
- Lathe B
Predicted
/ 0.10
10 100 200 300 100 200 300
C%ting time min Cutting speed m/min Cutting speed m/min
Fig. 9 Prediction of surface roughness (a) Lathe A (b) Lathe B
Cutting conditions are same as in Table 2.
Fig. 11 Prediction of surface roughness at 10 minutes machining
Cutting conditions are same as in Table 2, where cutting speed
and feed rate are variable.
well. Figure 10 and Figure 11 show the prediction of grooving cannot be changed during the operation.
wear and surface roughness with cutting speed and feed rate, at (c) Three positions on the turret can be assigned to turning
ten minutes machining. The effects of machine tool tools used in the operation (j&,=3) according to the
characteristics on grooving wear and surface roughness are less number of tool used in other operations.
than those on initial flank wear, The results also enable us to Tolerable maximum values of flank wear V,,, and surface
understand that both of the processes on lathe Aare worse than
roughness U&J,, are given as shown in Table 4. For cutting
those on lathe B. Adaptive prediction, then, proves to be
speed and feed rate, tool life Tlir, can be given when either
effective in identifying the machine tool characteristics
accurately, even if the difference of machining process is flank wear or surface roughness predicted exceeds its tolerable
little. value. For Nproducts machined with a cutting edge, machining
cost E,(N)can be expressed as the sum of actual cutting cost,
3.2. Effect of machine tool characteristics on operation exchange costs of tool and workpiece, and tool price C,:
planning
E,(N) = C, N. T, +C, T, +C, N * T, + C, (11)
The effect of machine tool characteristics on the optimum
where 7’, is cutting time for one product; C, is running cost; Ti
cutting conditions that minimize machining cost is discussed
and T, are time for tool index and that for work exchange
in the operation planning. Each lathe machines N,,, bar type
respectively. When Np products are machined during tool life
products sequentially in the turning operation shown in Table
4. The constraints of the operation are as follows: T)+, the number of tools Tnum used for machining all the
(a) Depth of cut is constant and products are finished with one products (N,,,) can be calculated as follows:
feed cutting along the length of workpiece. Tool material
and its shape are fixed. N,, 1 N, (N,=k. N,,)
(b) Tool life for tool exchange cycle and cutting conditions T ,mm = (12)
(int)(N,,,l NJ + 1 (N, f k. N,,)
446 T. Matsumura et al./Journal of Materials Processing Technology 62 (1996) 440-447
Table 4
Planning conditions and cutting conditions
Planning conditions
Running cost CT $/min 1
Price of tool C, $/edge 5
Time for tool index Ti min 1
Time for tool setting Td min 5
Time for work setting T, min 1
Number of position assigned T set 3
to tool on turret
0.10 0.10
Number of work NW 300 100 200 300 100 200 300
Length of workpiece 44 - 50.0 Cutting speed m/min Cutting speed mlmin
Diameter of workpiece % - 50.0 (a) Lathe A (b) Lathe 3
Tolerable maximum flank wear vBmnv - 0.14
Fig, 12 Effect of machine tool characteristics on machining cost
Tolerable maximum surface roughness (R,,,,Y),,m,x pm 14 Planning conditions and cutting conditions are shown in Table 4.
Cutting conditions
Material cut 0.45% carbon steel with increase ductility of workpiece material; and the decrease
Tool material Carbide tool P20 of feed rate makes theoretical roughness and cutting force
Tool geometry (-5,-5,5,5,15,15,0.8) down. The decrease of feed rate also compensates for excessive
Cutting speed V m/min to be optimized flank wear in the machining on lathe A when cutting speed is
Feed rate f mm/rev to be optimized
high.
Depth of cut dmm 1.0
Lubrication dry
3.3. Eflect of machine tool chnmcteristics on machining
scheduling
Table 5
Job information
Job 1 Job 2 Job 3 Job 4 Job 5 Job 6
Workpiece length mm 30.0 30.0 50.0 50.0 70.0 70.0
Workpiece diameter mm 50.0 100.0 50.0 100.0 50.0 100.0
Number of product 600 600 400 850 700 200
Tolerable maximum flank wear mm 0.16 0.12 0.14 0.16 0.14 0.18
Tolerable maximum surface roughness pm 12.0 14.0 14.0 12.0 12.0 16.0