Mode-Coupled Regenerative Machine Tool Vibrations

Tamás Kalmár-Nagy
1
, Francis C. Moon
2
1
United Technologies Research Center, 411 Silver Lane, East Hartford, CT 06108
2
Sibley School of Mechanical and Aerospace Engineering
Cornell University, Ithaca, NY 14853, USA
Abstract
In this paper a new 3 degree-of-freedom lumped-parameter model for machine tool vibrations is developed and
analyzed. One mode is shown to be stable and decoupled from the other two, and thus the stability of the system
can be determined by analyzing these two modes. It is shown that this mode-coupled nonconservative cutting
tool model including the regenerative effect (time delay) can produce an instability criteria that admits low-level
or zero chip thickness chatter.
1 Introduction
One of the unsolved problems of metal cutting is the existence of low-level, random-looking (maybe chaotic)
vibrations (or pre-chatter dynamics, see Johnson and Moon [17]). Some possible sources of this vibration are
the elasto-plastic separation of the chip from the workpiece and the stick-slip friction of the chip over the tool.
Recent papers of Davies and Burns [9], Wiercigroch and Krivtsov [43], Wiercigroch and Budak [41] and Moon
and Kalmár-Nagy [27] have addressed some of these issues. Numerous researchers investigated single degree-of-
freedom regenerative tool models (Tobias [39], Hanna and Tobias [13], Shi and Tobias [34], Fofana [11], Johnson
[18], Nayfeh et al. [28], Kalmár-Nagy et al. [20], Stépán [36], Kalmár-Nagy [21], Stone and Campbell [38], Stépán
et al. [37]). Even though the classical model (Tobias [39]) with nonlinear cutting force is quite successful in
predicting the onset of chatter (Kalmár-Nagy et al., [19]), it cannot possibly account for all phenomena displayed
in real cutting experiments. Single degree-of-freedom deterministic time-delay models have been insufficient so
far to explain low-amplitude dynamics below the stability boundary. Also, real tools have multiple degrees of
freedom. In addition to horizontal and vertical displacements, tools can twist and bend. Higher degree-of-freedom
models have also been studied in turning, as well as in boring, milling and drilling (Pratt [32], Batzer et al.
[2], Balachandran [1], van de Wouw et al. [44]). In this paper we will examine the coupling between multiple
degree-of-freedom tool dynamics and the regenerative effect in order to see if this chatter instability criteria will
permit low-level instabilities.
Coupled-mode models in aeroelasticity or vehicle dynamics may exhibit so-called ’flutter’ or dynamics insta-
bilities (see e.g. Chu and Moon [8]) when there exists a non-conservative force in the problem. One example is the
follower force torsion-beam problem as in Hsu [15]. In the present work we assume that the chip removal forces
rotate with the tool thereby introducing an unsymmetric stiffness matrix which can lead to flutter and chatter.
Tobias called this mode-coupled chatter. Often this model of chatter is analyzed without the regenerative effect.
In this paper we will show that the combination of mode-coupling nonconservative model and a time delay can
produce an instability criteria that admits low-level or zero chip thickness chatter. There is no claim in this paper
to having solved the random- or chaotic low level dynamics since only linear stability analysis is presented in this
paper. But the results shown below provide an incentive to extend this model into the nonlinear regime. A dy-
namic model with the combination of 2-degree-of-freedom flutter model with time delay may also be applicable to
aeroelastic problems in rotating machinery where the fluid forces in the current cycle depend on eddies generated
in the previous cycle. However the focus of this paper is on the physics of cutting dynamics.
The structure of the paper is as follows. In Section 2 an overview of the turning operation is given, together
with the description of chatter and the regenerative effect. The equations of motion are developed in Section 3.
The model parameters are estimated in Section 4. Analysis of the model is performed in Section 5 and conclusions
are drawn in Section 6.
2 Metal Cutting
The most common feature of machining operations (such as turning, milling, and drilling) is the removal of a
thin layer of material (the chip) from the workpiece using a wedge-shaped tool. They also involve relative motion
1
pp. 129-149, in Guenter Radons and Raimund Neugebauer eds.: Nonlinear Dynamics of Production
Systems. Wiley-VCH, Berlin, 2004
Machined surface
Workpiece
vto
o
l

Figure 1: Turning
between the workpiece and the tool. In turning the material is removed from a rotating workpiece, as shown in
Figure 1.
The cylindrical workpiece rotates with constant angular velocity Ω [rad/s] and the tool is moving along the
axis of the workpiece at a constant rate. The feed f is the longitudinal displacement of the tool per revolution of
the workpiece, and thus it is also the nominal chip thickness. The translational speed of the tool is then given by
v
tool
=


f (1)
The interaction between the workpiece and the tool gives rise to vibrations. One of the most important source
of vibrations in a cutting process is the regenerative effect. The present cut and the one made one revolution
earlier might overlap, causing chip thickness (and thus cutting force) variations. The associated time delay is the
time-period τ of one revolution of the workpiece
τ =


(2)
The phenomenon of the large amplitude vibration of the tool is known as chatter. A good description of chatter is
given by S. A. Tobias [39], one of the pioneers of modern machine tool vibrations research: ’The machining of metal
is often accompanied by a violent relative vibration between work and tool which is called the chatter. Chatter
is undesirable because of its adverse affects on surface finish, machining accuracy, and tool life. Furthermore,
chatter is also responsible for reducing output because, if no remedy can be found, metal removal rates have to
be lowered until vibration-free performance is obtained.’
Johnson [18] summarizes several qualitative features of tool vibration
• The tool always appears to vibrate while cutting. The amplitude of the vibration distinguishes chatter from
small-amplitude vibrations.
• The tool vibration typically has a strong periodic component which approximately coincides with a natural
frequency of the tool.
• The amplitude of the oscillation is typically modulated and often in a random way. The amplitude modu-
lation is present in both the chattering and non-chattering cases.
Tool vibrations can be categorized as self-excited vibrations (Litak et al. [24], Milisavljevich et al. [25]) or
vibrations due to external sources of excitation (such as resonances of the machine structure) and can be periodic,
quasiperiodic, chaotic or stochastic (or combinations thereof). A great deal of experimental work has been carried
out in machining to characterize and quantify the dynamics of metal cutting. Recently a number of researchers
have provided experimental evidence that tool vibrations in turning may be chaotic (Moon and Abarbanel [26],
Bukkapatnam et al. [5], Johnson [18], Berger et al. [3]). Other groups however now disavow the chaos theory for
cutting and claim that the vibrations are random noise (Wiercigroch and Cheng [42], Gradišek et al. [12]).
2
W
f
f
2
i

C
F
C
F
T
F
R
workpiece
cutting
tool insert
machined
surface
v
C
Figure 2: Oblique chip formation model
2.1 Oblique Cutting
Although many practical machining processes can adequately be modeled as single degree-of-freedom and orthog-
onal, more accurate models demand a chip formation model in which the cutting velocity is not normal to the
cutting edge.Figure 2 shows the usual oblique chip formation model, where the inclination angle i (measured
between the cutting edge and the normal to the cutting velocity in the plane of the machined surface) is not zero,
as in orthogonal cutting. The cutting velocity is denoted by v
C
, the chip flow angle is η
c
, the thickness of the
undeformed chip is f, the deformed chip thickness is f
2
and the chip width is w. The three dimensional cutting
force acting on the tool insert is decomposed into three mutually orthogonal forces: FC, FT , FR. The cutting
force F
C
is the force in the cutting direction, the thrust force F
T
is the force normal to the cutting direction
and machined surface, while the radial force F
R
is normal to both F
C
and F
T
. While orthogonal cuting can be
modeled as a 2-dimensional process, oblique cutting is a true three-dimensional plastic flow problem (Oxley [30]).
3 3 DOF Model of Metal Cutting
Figure 3 shows a tool with a cutting chip (insert) both in undeformed and deformed state of the tool.The three
z

y
x

cutting edge
tool insert
r
x
r
z
v
C
Figure 3: 3 DOF metal cutting model
3
degrees of freedom are horizontal position (x), vertical position (z), and twist (φ). In the lumped parameter model
(Figure 4) all the mass m of the beam is placed at its end (this effective mass is equivalent to modal mass for a
distributed beam).
x
z

chip
m k ,c

k
x
c
x
k
z
c
z
F
T
F
C
Figure 4: 3 DOF lumped-parameter model
The equations of motion are the following
m¨ z + c
z
˙ z + k
z
z = F
z
(3)
m¨ x + cx ˙ x + kxx = Fx (4)
I
¨
φ + c
φ
˙
φ + k
φ
φ = M
y
(5)
Figure 5 shows the forces acting on the tooltip.
As the tool bends about the x axis, the direction of the cutting velocity (and main cutting force) changes, as
shown in Figure 6.In order to derive the equations of motion, two coordinate systems are defined. An inertial
frame (I, J, K) fixed to the tool and a moving frame (i, j, k) fixed to the cutting velocity.The force acting on the
insert can then be written as
F = −FT I + FRJ −FCK (6)
or
F = Fxi + Fyj + Fzk (7)
where i, j, k are unit vectors in the x, y, z directions, respectively.
Fx = −FT (8)
F
y
= F
C
sinβ + F
R
cos β (9)
F
z
= F
R
sinβ −F
C
cos β (10)
The bending also results in a pitch ψ (shown in Figure 6). This is not a separate degree of freedom, but nonetheless
it will influence the inclination angle.
The following assumptions are used in deriving the equations of motion
• The forces that act on the insert are steady-state forces
• The width of cut w (y-position) is constant
• All displacements are small
• Yaw is negligible
Steady-state forces refer to time averaged quantities. The effect of rate-dependent cutting forces were studied
by Saravanja-Fabris and D’Souza [33], Chiriacescu [7], Moon and Kalmár-Nagy [27]. Next we find the position of
the tooltip in the fixed system of the platform. To do so we have to find the rotation matrix R that describes the
relationship between the moving frame (i, j, k) and the fixed frame (I, J, K).

i j k

= R

I J K

(11)
Using the Tait-Bryant angles {ψ, φ} we express R as a product of two consecutive planar rotations (Pitch-Roll
system)
R = R
2
R
1
(12)
4
x
y
z
F
T
F
R
F
C
I
J
K
i
j
k
v
C
Figure 5: Forces on the tooltip
F
C
F
R
z
y

Figure 6: Direction of cutting velocity
5
The cross section is first rotated about I by the pitch angle ψ. The corresponding rotation matrix is
R1 =

¸
1 0 0
0 cψ sψ
0 −sψ cψ

(13)
where the abbreviations c = cos, s = sin were used. The second rotation is about the J2 (the rotated J) axis
through the roll angle φ (with respect to the toolholder)
R2 =

¸
cφ 0 sφ
0 1 0
−sφ 0 cφ

(14)
R can then be calculated by (12)
R =

¸
cφ −sφsψ cψsφ
0 cψ sψ
−sφ −cφsψ cψcφ

(15)
The position of the tooltip can be expressed in the fixed frame as
r

= R

¸
r
x
0
r
z

=

¸
r
x
cφ + r
z
cψsφ
rzsψ
r
z
cψcφ −r
x

(16)
The roll producing moment can then be calculated as
M
y
= (r

×F) · j = F
T
(r
x
sφ −r
z
cφcψ) + F
C
cβ (r
x
cφ + r
z
cψsφ) −F
R
sβ (r
x
cφ + r
z
cψsφ) (17)
In the following we assume small displacements and small angles and neglect nonlinear terms. The angle β is
taken to be proportional to the vertical displacement, i.e. β = −nz (n > 0) and so is the pitch, i.e. ψ = kz
(k > 0).
m¨ x + c
x
˙ x + k
x
x = −F
T
(18)
m¨ z + c
z
˙ z + k
z
z = −

F
C
+ nz
¯
F
R

(19)
I
¨
φ + c
φ
˙
φ + k
φ
φ = M
y
= φ

r
x
¯
F
T
+ r
z
¯
F
C

−r
z
F
T
+ r
x
F
C
+ nzr
x
¯
F
R
(20)
where
¯
F
C
,
¯
F
R
,
¯
F
T
denotes the constant term in F
C
, F
R
and F
T
, respectively.
3.1 Cutting Forces
Generally we assume that the cutting forces FC, FT , FR depend only on the inclination angle i and chip thickness
f (see Figure 2), and the rake angle α (see Figure 3). We again emphasize that the chip width w is considered
constant in the present analysis. Our hypothesis here is that F
C
and F
T
depend linearly on both the rake angle
and chip thickness (see Section 4.2) in the following manner
F
C
= −l
C
α + m
C
t
1
+ F
C0
(21)
F
T
= −l
T
α + m
T
t
1
+ F
T0
(22)
where m
C
and m
T
are cutting force coefficients, while l
C
and l
T
are angular cutting force coefficients (they show
how strong the force dependence is on rake angle). The variable t
1
is the chip thickness variation (the deviation
from the nominal chip thickness). The constant forces FC0 and FT0 arise from cutting at a nominal chip thickness.
The radial cutting force can be expressed as (Oxley [30])
F
R
= sini
F
C
cos i (i −sinα) −F
T
sin
2
i sinα + cos
2
i
(23)
where Stabler’s Flow Rule (Stabler [35]) η
C
= i was used. The effective rake angle depends on the initial rake
angle and the roll
α = α0 −φ (24)
while the inclination angle will depend on the initial inclination angle (i0) as well as the pitch
i = i
0
−ψ (25)
The chip thickness depends on the nominal feed and the position of the tooltip (both the present and the delayed
ones). The displacement of the tooltip is due to translational and rotational motion as shown in Figure 7.Here
the dashed line corresponds to the position vector of the tooltip in the undeformed configuration, while the solid
6
r
x
r
z

i
k
z
x
Figure 7: Motion of the tooltip
line depicts how this vector rotates (φ) and translates (due to the displacements x and z). The chip thickness is
then given by
t1 = t10 + x −xτ + rz sin(φ −φτ ) ≈ t10 + x −xτ + rz (φ −φτ ) (26)
where x
τ
and φ
τ
denote the delayed values x(t −τ) and φ(t −τ), respectively. Then the cutting forces can be
written as
FC = mC (x −xτ ) + (lC + rzmC) φ −rzmCφτ +
¯
F
C
. .. .
mCt10 + FC0 −α0lC (27)
F
T
= m
T
(x −x
τ
) + (l
T
+ r
z
m
T
) φ −r
z
m
T
φ
τ
+ m
T
t
10
+ F
T0
−α
0
l
T
(28)
If the initial inclination angle is assumed to be zero, the expression for F
R
will simplify
F
R
= k

¯
F
T
+ (sinα
0
−1)
¯
F
C
+ t
10
(m
C
(1 −sinα
0
) −m
T
)

z (29)
3.2 The Equations of Motion
Substituting (27-28) into equations (18-20) and eliminating the constant (by translation of the variables) results
m¨ z + cz ˙ z + kzz = −n
¯
FRz −mC (x −xτ ) −(lC + rzmC) φ + rzmCφτ (30)
m¨ x + c
x
˙ x + k
x
x = −m
T
(x −x
τ
) −(l
T
+ r
z
m
T
) φ + r
z
m
T
φ
τ
(31)
I
¨
φ + c
φ
˙
φ + k
φ
φ = −rzmT (x −xτ ) −rz (lT + rzmT −mCt10 −FC0 + α0lC) φ + r
2
z
mT φτ (32)
where now (x, z, φ) represent deviations from the steady values of the original displacements. As we can see, the
x and φ equations are uncoupled from the z equation, so the stability of the system is determined by (31, 32).
Equations (31, 32) can also be written as
¨ x + 2ζ
x
ω
x
˙ x +

ω
2
x
+
m
T
m

x +
1
m
(l
T
+ r
z
m
T
) φ =
m
T
m
x
τ
+ r
z
m
T
m
φ
τ
(33)
¨
φ + 2ζ
φ
ω
φ
˙
φ +
rzmT
I
x +

ω
2
φ
+
rz
I
(l
T
+ r
z
m
T
−m
C
t
10
−F
C0
+ α
0
l
C
)

φ =
rzmT
I
x
τ
+ r
2
z
mT
I
φ
τ
(34)
where
ω
x
=

kx
m
, ω
φ
=

k
φ
I
(35)
By introducing the nondimensional time and displacement
ˆ
t = t/T ˆ x = x/X (36)
ˆ x
00
+ 2ζ
x
ω
x
T ˆ x
0
+

ω
2
x
+
m
T
m

T
2
ˆ x +
1
m
(l
T
+ r
z
m
T
)
T
2
X
φ =
m
T
m
T
2
x
τ
+ r
z
m
T
m
T
2
X
φ
τ
(37)
φ
00
+ 2ζ
φ
ω
φ

0
+
rmT
I
T
2
Xˆ x +

ω
2
φ
+
rz
I
(l
T
+ r
z
m
T
−F
C0
−m
C
t
10
+ α
0
l
C
)
¸
T
2
φ =
r
z
m
T
I
T
2
Xx
τ
+ r
2
z
m
T
I
T
2
φ
τ
(38)
7
With the choice of the following scales
T =
1
ω
x
X =

I
m
(39)
the equations assume the form (ˆ τ = ω
x
τ)
ˆ x
00
+ 2ζ
x
ˆ x
0
+ k
11
ˆ x + k
12
φ = r
11
ˆ x
ˆ τ
+ r
12
φ
ˆ τ
(40)
φ
00
+ 2
ˆ
ζ
φ
φ
0
+ k
21
ˆ x + k
22
φ = r
21
ˆ x
ˆ τ
+ r
22
φ
ˆ τ
(41)
where
k
11
= 1 +
mT
ω
2
x
m
k
12
=
lT + rzmT
ω
2
x

Im
(42)
k21 =
r
z
m
T
ω
2
x

Im
ˆ
ζ
φ
= ζ
φ
ω
φ
ω
x
(43)
k
22
=

ω
φ
ωx

2
+
rz
ω
2
x
I
(l
T
+ r
z
m
T
−m
C
t
10
−F
C0
+ α
0
l
C
) (44)
r11 =
m
T
ω
2
x
m
r12 = r21 =
r
z
m
T
ω
2
x

Im
(45)
r22 =
r
2
z
m
T
ω
2
x

Im
(46)
Note that the stiffnesses k
12
and k
21
are different. This is characteristic of nonconservative systems (Bolotin [4],
Panovko and Gubanova [31]). In many mechanical systems this nonconservativeness is due to the presence of
following forces.
4 Estimation of Model Parameters
In the following we estimate different terms in (42-46) to establish their relative strengths in order to simplify the
model.
4.1 Structural Parameters
The toolholder is assumed to be a rectangular steel beam. The length of the toolholder is relatively short for
normal cutting, while it can be longer for boring operations (see Kato et al. [22]). So we assume l to be between
0.05 m and 0.3 m. The width and height are usually of order of a centimeter. The stiffnesses for such a cantilevered
beam can be in the following ranges
k
x
' 10
4
÷10
7
N
m
(47)
kz ' 10
5
÷10
7
N
m
(48)
k
φ
' 1000 ÷10000
N
rad
(49)
Since a lumped parameter approximation is used, the mass at the end of the massless beam is assumed to be the
modal mass. The vibration frequencies are then
ωx ' 100 ÷5000
rad
s
(50)
ω
z
' 100 ÷10000
rad
s
(51)
ω
φ
' 1000 ÷10000
rad
s
(52)
The ratio
ω
φ
ω
x
varies between 2 and 10 (the shorter the tool is the higher the ratio).
4.2 Cutting Force Parameters
Experimental cutting force data during machining of 0.2% carbon steel is shown in Figure 8 (Oxley [30]).The graph
shows the forces FC and FT for different rake angles (α = −5

and 5

for top and bottom Figures, respectively).
The width of cut and chip thickness were 4 mm and 0.25 mm, respectively. Since our model assumes constant
8
Figure 8: Forces in oblique cutting of 0.2% carbon steel. α = −5

(top) and α = 5

(bottom). f = 0.125 mm. After
Oxley (1989)
[rad]
0.5
1
1.5
2
2.5
3
3.5
4
0.25
0.5
0.75
1
1.25
1.5
1.75
2
F [kN]
C
F [kN]
T
[rad]
Figure 9: Forces vs. rake angle (derived from Oxley [30]) a, cutting force b, thrust force.
9
cutting speed, forces were taken from these graphs at the value 200 m/s of the cutting speed and plotted against
rake angle (Figure 9).The constants l
C
and l
T
were found as the slope of the lines corresponding to t
1
= 0.25 mm
l
C
= 1580
N
rad
, l
T
= 3150
N
rad
(53)
A linear relationship is assumed between forces at zero rake angle and chip thickness, i.e.
FC = FC0 + mct1 (54)
FT = FT0 + mT t1 (55)
where these coefficients were determined to be
m
C
= 6 ∗ 10
6
N
m
F
C0
= 458 N (56)
mT = 1.65 ∗ 10
6
N
m
FT0 = 784 N (57)
4.3 Model Parameters
Since
r
z
ω
2
x
I
(l
T
+ r
z
m
T
−m
C
t
10
−F
C0
+ α
0
l
C
) ¿

ω
φ
ω
x

2
(58)
this term will be neglected, i.e.
k
22
=

ω
φ
ωx

2
(59)
Also, the term r
22
is very small, so it is neglected
r
22
= 0 (60)
5 Analysis of the Model
With the approximations (59, 60) the model (40, 41) can be written as the matrix equation
¨ x +C˙ x +Kx = Rx
τ
(61)
where
x =

ˆ x
φ

and the matrices are given by
C =
¸
2ζx 0
0 2ζ
φ

, K =
¸
1 + p a + pq
pq k
22

, (62)
R =
¸
p q
q 0

(63)
Here we introduced the parameters
p =
m
T
ω
2
x
m
, q =
r
z
X
= r
z

m
I
(64)
where constants a and k
22
are
a =
l
T
ω
2
x

Im
, k
22
=

ω
φ
ω
x

2
(65)
It is characteristic of systems with nonsymmetric stiffness matrix, that they can lose stability either by divergence
(buckling) or by flutter. Chu and Moon [8] examined divergence and flutter instabilities in magnetically levitated
models. Kiusalaas and Davis [23] studied stability of elastic systems under retarded follower forces. Recently
several numerical methods were proposed to investigate stability of linear time-delay systems (see Chen et al. [6],
Engelborghs and Roose [10], Insperger and Stépán [16], Olgac and Sipahi [29]).
5.1 Classical Limit
If q = 0 the equations reduce to
x
00
+ 2ζxx
0
+ (1 + p) x + aφ = pxτ (66)
φ
00
+ 2ζ
φ
φ
0
+ cφ = 0 (67)
The φ-equation is uncoupled from the x-equation and reduces to that of a damped oscillator. Its equilibrium
φ = 0 is asymptotically stable and thus it does not affect the stability of the x-equation. In this case we recover
the 1 DOF classical model (Tobias and Fishwick [40]).
10
5.2 Stability Analysis of the Undamped System without Delay
First we perform linear stability analysis of the system
¨ x +Kx = 0 (68)
where the matrix K is non-symmetric and of the form (k
22
> 0)
K =
¸
k
11
k
12
k
21
k
22

(69)
Assuming the solutions in the form
x = de
iωt
(70)
we obtain the characteristic polynomials

K−ω
2
I

d = 0 (71)
which have nontrivial solution if the determinant of K−ω
2
I is zero

k
11
−ω
2
k
12
k
21
k
22
−ω
2

=

k
11
−ω
2

k
22
−ω
2

−k
21
k
12
= 0 (72)
The characteristic equation for the coupled system becomes
ω
4
−(k
11
+ k
22
) ω
2
+ k
11
k
22
−k
21
k
12
= 0 (73)
Divergence (static deflection, buckling) occurs when ω = 0 (or det K = 0), that is when
k11k22 −k21k12 = 0 (74)
If ω 6= 0, then the characteristic equation (73) can be solved for ω
2
as
ω
2
=
1
2

k11 + k22 ±

(k11 + k22)
2
−4 (k11k22 −k21k12)

(75)
For stable solutions, both solutions should be positive. Since k
22
> 0, this is the case if
0 ≤ k11k22 −k21k12 ≤

k
11
+ k
22
2

2
(76)
The two bounds correspond to divergence and flutter boundaries, respectively. With the stiffness matrix in (62)
k
11
= 1 + p, k
12
= a + pq (77)
k21 = pq (78)
In the plane of the bifurcation parameters q, p the divergence boundaries are given by
p =
1
2q
2

k
22
−aq ±

4k
22
q
2
+ (k
22
−aq)
2

(79)
and the flutter boundary is characterized by
p =
1
1 + 4q
2
(80)

k22 −2aq −1 ±2

q

a (1 −k22) + a
2
q −q (k22 −1)
2

(81)
Figure 10 shows these boundaries on the (q, p) parameter plane for a = 1, k22 = 2.The different stability regions
are indicated by the root location plots.
5.3 Stability Analysis of the 2 DOF Model with Delay
In this section we include the delay terms in the analysis. In order to be able to study how these terms influence
the stability of the system, we introduce a new parameter, similar to the overlap factor (Tobias [39]).
First we analyze the system with no damping:
¨ x +Kx = µRx
τ
(82)
When µ = 0 we recover the previously studied (68), while µ = 1 corresponds to equation (61) without damping.
The characteristic equation is
det

−λ
2
I +K−µe
−λτ
R

= 0 (83)
11
q
p
S
U
U
U
Figure 10: Stability boundaries of the undamped 2 DOF model without delay
λ
4
+

k
11
+ k
22
−µpe
−λτ

λ
2
+ k
11
k
22
−k
12
k
21
+
µe
−λτ
(q (k12 + k21) −pk22) −µ
2
q
2
e
−2λτ
= 0 (84)
Substituting λ = iω, ω ≥ 0 yields a complex equation that can be separated into the two real ones (the second
equation was divided by µsin(τω) 6= 0)
ω
4
−(k11 + k12) ω
2
+ k11k22 −k12k21+ (85)
µcos (τω)


2
+ q (k
12
+ k
21
) −pk
22

−µ
2
q
2
cos (2τω) = 0

2
+ q (k
12
+ k
21
) −pk
22
+ 2µq
2
cos (τω) = 0 (86)
We solve the second equation for cos (τω)
cos (τω) =

2
+ q (k
12
+ k
21
) −pk
22
−2µq
2
(87)
Using this relation and the identity cos (2τω) = 2 cos (τω)
2
−1 in the real part (85) results
ω
4
−(k
11
+ k
22
) ω
2
+ k
11
k
22
−k
12
k
21
+ µ
2
q
2
= 0 (88)
Divergence occurs where ω = 0, that is where
k11k22 −k12k21 + µ
2
q
2
= 0
Substituting the elements of the stiffness matrix as given in (62) yields
−(q (a + q (p −µ)) (p −µ)) + k
22
(1 + p −p µ) = 0 (89)
which can be solved for p as
1
2 q
2

k
22
(1 −µ) + q (2 q µ −a) ±

(k
22
(µ −1) + q (a −2µq))
2
+ 4 q
2
(k
22
+ q µ (a −q µ))

(90)
The change of the divergence boundary is shown in Figure 11 (top, middle, bottom) for µ = 0.1, 0.5 and 1 while
the delay was set to 1. Flutter occurs for ω > 0, and the boundary can be found by numerically solving equations
(86, 88) for p and q for a given µ. Figure 12 shows the flutter boundary for a small µ (0.01) together with the
flutter boundary (80). Figure 13 shows how this boundary changes with increasing µ (µ = 0.1, 0.5, 1). Figure 14
shows the full stability chart, complete with both the divergence and flutter boundaries, for µ = 1. To validate this
stability chart the parameter space (p, q) was gridded and the delay-differential equation (61, 62) was integrated
with constant initial function (note that the amplitude does not matter for linear stability) at the gridpoints.
The integration was carried out for 15τ intervals of which the first 5τ intervals were discarded. Stability was
determined by whether the amplitude of the solution grew or decayed. Dark dots correspond to stable numerical
solutions. This figure can also explain a practical trick used in machine shops: sometimes, to avoid chatter, the
tool is placed slightly ABOVE the centerline. We note that increasing q moves the system into the stable region
of the chart.
12
0.5
1
1.5
2
2.5
0.5
1
1.5
2
2.5
0.5
1
1.5
2
2.5
3
3
3
p
q
q
q
p
p
U U
S
U U
U
S
U U
U
U
S
Figure 11: The change of the divergence boundary for system (80), τ = 1. µ = 0.1, 0.5, 1 for top, middle and bottom
Figures, respectively
13
- 2 0 2
q
0
2 p
f

f
S
S
S
U
U
U
Figure 12: Flutter boundary of (80) with µ = 0.01. S and U denote Stable and Unstable regions
- 1 0 1
q
0
2 p

Figure 13: Flutter boundary as a function of µ (µ = 0.1, 0.5, 1)
14
Figure 14: Stability chart for the undamped system (82), µ = 1, τ = 1
Now we examine the effect of damping on the size of stability regions. It is an important step, as it is known
(Herrmann and Jong [14]) that damping can have a destabilizing effect in nonconservative systems. The damping
coefficients ζ
x
and ζ
φ
are taken to be 0.01, while the ratio of frequencies ω
φ

x
was changed in Figure 15 (this
is the same as keeping this ratio fixed and increasing ζ
φ
). As the figure shows, the size of the stability regions
increases with added damping. And finally, we show how the lobes of the conventional stability chart deform with
the added parameter q (0 ≤ q ≤ 1). Figure 16 shows that increasing q results in the ’birth’ of unstable regions.
These upside-down lobes are actually lobes of the classical model for p < 0 (p is the nondimensional cutting force
coefficient which is positive). In our model these lobes become a new source of instability, where the classical
model would predict stable behavior.
6 Conclusions
A new 3 DOF model derived may help explain at least two phenomena in metal cutting. The first is that off-
centering the tool might help avoiding chatter. The second phenomenon is the observation that small amplitude
tool vibrations can arise below the classical stability boundary. As shown, the added degrees of freedom result
in unstable regions below the one predicted by the one DOF classical model. To summarize the important
observations:
• The 3-DOF model results in coupling between twist and lateral bending
• The model can exhibit both divergence and flutter instabilities
• Damping seems to increase the size of stability regions
• The tool offset produces new regions of instability (the upside-down lobes)
This model is based on the assumption of rate-independent cutting forces, i.e. forces that do not exhibit
hysteresis (Moon and Kalmár-Nagy [27]). It does not include temperature effects either (Davies and Burns [9]).
Finally, only the analysis of a full nonlinear model could characterize the nature of vibrations and provide
estimates of vibration amplitudes for the low chip thickness unstable regions.
References
[1] B. Balachandran, Nonlinear Dynamics of Milling Processes, Philosophical Transactions of the Royal Society,
359, 793—819 (2001).
15
Figure 15: Stability chart for the 3 DOF model. a, ω
φ

x
= 2 b, ω
φ

x
= 10
16
Figure 16: Stability charts for the 3 DOF model with increasing q
[2] S. A. Batzer, A. M. Gouskov, and S. A. Voronov, Modeling the Vibratory Drilling Process. Proceedings of
the 17th ASME Biennial Conference on Vibration and Noise 1-8 (1999).
[3] B. S. Berger, I. Minis, Y. H. Chen, A. Chavali, and M. Rokni, Attractor Embedding in Metal Cutting.
Journal of Sound and Vibration, 184(5), 936—942 (1995).
[4] V. V. Bolotin, Nonconservative Problems of the Theory of Elastic Stability. New York: The MacMillan
Company (1963).
[5] S. T. S. Bukkapatnam, A. Lakhtakia, and S. R. T. Kumara, Analysis of Sensor Signals Shows Turning on a
Lathe Exhibits Low-Dimensional Chaos. Physical Review E, 52(3), 2375—2387 (1995).
[6] S.-G. Chen, A. G. Ulsoy, and Y. Koren, Computational Stability Analysis of Chatter in Turning. Journal of
Manufacturing Science and Engineering, 119, 457—460 (1997).
[7] S. T. Chiriacescu, Stability in the Dynamics of Metal Cutting. Elsevier (1990).
[8] D. Chu, and F. C. Moon, Dynamic Instabilities in Magnetically Levitated Models. Journal of Applied Physics,
54(3), 1619—1625 (1983).
[9] M. A. Davies, and T. J. Burns, Thermomechanical Oscillations in Material Flow During High-Speed Ma-
chining. Philosophical Transactions of the Royal Society, 359, 821—846 (2001).
[10] K. Engelborghs, and D. Roose, Numerical Computation of Stability and Detection of Hopf Bifurcations of
Steady State Solutions of Delay Differential Equations. Advances in Computational Mathematics, 10, 271—289
(1999).
[11] M. S. Fofana, Delay Dynamical Systems with Applications to Machine-Tool Chatter. Ph.D. thesis, University
of Waterloo, Department of Civil Engineering (1993).
[12] J. Gradišek, E. Govekar, and I. Grabec, I, Time Series Analysis in Metal Cutting: Chatter versus Chatter-Free
Cutting. Mechanical Systems and Signal Processing, 12(6), 839—854 (1998).
[13] N. H. Hanna, and S. A. Tobias, A Theory of Nonlinear Regenerative Chatter. Transactions of the American
Society of Mechanical Engineers - Journal of Engineering for Industry, 96(1), 247—255 (1974).
[14] G. Herrmann, and I.-C. Jong, On the Destabilizing Effect of Damping in Nonconservative Elastic Systems.
Transactions of the ASME, 592—597 (1965).
[15] C. S. Hsu, Application of the Tau-Decomposition Method to Dynamical Systems Subjected to Retarded Fol-
lower Forces. Transactions of the American Society of Mechanical Engineers - Journal of Applied Mechanics,
37(2), 259-266 (1970).
[16] T. Insperger, and G. Stépán, Semi-discretization Method for Delayed Systems. International Journal for
Numerical Methods in Engineering, 503-518 (2002).
17
[17] M. Johnson, and F. C. Moon, Nonlinear Techniques to Characterize Prechatter and Chatter Vibrations in
the Machining of Metals. International Journal of Bifurcation and Chaos, 11(2), 449—467 (2001).
[18] M. A. Johnson, Nonlinear Differential Equations with Delay as Models for Vibrations in the Machining of
Metals. Ph.D. thesis, Cornell University (1996).
[19] T. Kalmár-Nagy, J. R. Pratt, M. A. Davies, and M. D. Kennedy, Experimental and Analytical Investigation
of the Subcritical Instability in Turning. In: Proceedings of the 1999 ASME Design Engineering Technical
Conferences. DETC99/VIB-8060 (1999).
[20] T. Kalmár-Nagy, G. Stépán, and F. C. Moon, Subcritical Hopf Bifurcation in the Delay Equation Model for
Machine Tool Vibrations. Nonlinear Dynamics 26, 121-142 (2001).
[21] T. Kalmár-Nagy, Delay-Differential Models of Cutting Tool Dynamics with Nonlinear and Mode-Coupling
Effects. Ph.D. thesis, Cornell University (2002).
[22] S. Kato, E. Marui, and H. Kurita, Some Considerations on Prevention of Chatter Vibration in Boring
Operations. Journal of Engineering for Industry, 717-730 (1969).
[23] J. Kiusalaas, and H. E. Davis, On the Stability of Elastic Systems Under Retarded Follower Forces. Inter-
national Journal of Solids and Structures, 6(4), 399—409 (1970).
[24] G. Litak, J. Warmi´nski, and J. Lipski, Self-Excited Vibrations in Cutting Process. 4th Conference on
Dynamical Systems - Theory and Applications, 193-198 (1997).
[25] B. Milisavljevich, M. Zeljkovic, and R. Gatalo, The Nonlinear Model of a Chatter and Bifurcation. Pages
391—398 of: Proceedings of the 5th Engineering Systems Design and Analysis Conference. ASME (2000).
[26] F. C. Moon, and H. D. I. Abarbanel, Evidence for Chaotic Dynamics in Metal Cutting, and Classification of
Chatter in Lathe Operations. Pages 11—12,28—29 of: Moon, F. C. (ed), Summary Report of a Workshop on
Nonlinear Dynamics and Material Processing and Manufacturing, organized by the Institute for Mechanics
and Materials, held at the University of California, San Diego ( 1995).
[27] F. C. Moon, and T. Kalmár-Nagy, Nonlinear Models for Complex Dynamics in Cutting Materials. Philo-
sophical Transactions of the Royal Society, 359, 695—711 (2001).
[28] A. Nayfeh, C. Chin, and J. Pratt, Applications of Perturbation Methods to Tool Chatter Dynamics. In:
Moon, F. C. (ed), Dynamics and Chaos in Manufacturing Processes. John Wiley and Sons (1998).
[29] N. Olgac, and R. Sipahi, An Exact Method for the Stability Analysis of Time-Delayed LTI Systems. IEEE
Transactions on Automatic Control, 47(5), 793-797 (2002).
[30] P. L. B. Oxley, Mechanics of Machining: An Analytical Approach to Assessing Machinability. New York:
John Wiley and Sons (1989).
[31] Y. G. Panovko, and I. I. Gubanova, Stability And Oscillations of Elastic Systems. New York: Consultants
Bureau (1965).
[32] J. Pratt, Vibration Control for Chatter Suppression with Application to Boring Bars. Ph.D. thesis, Virginia
Polytechnic Institute and State University (1996).
[33] N. Saravanja-Fabris, and A. F. D’Souza, Nonlinear Stability Analysis of Chatter in Metal Cutting. Transac-
tions of the American Society of Mechanical Engineers - Journal of Engineering for Industry, 96(2), 670—675
(1974).
[34] H. M. Shi, and S. A. Tobias, Theory of Finite Amplitude Machine Tool Instability. International Journal of
Machine Tool Design and Research, 24(1), 45—69 (1984).
[35] G. V. Stabler, Advances in Machine Tool Design and Research. Pergamon Press Ltd., Oxford (1964). Chap.
The Chip Flow Law and its Consequences, page 243.
[36] G. Stépán, Modelling Nonlinear Regenerative Effects in Metal Cutting. Philosophical Transactions of the
Royal Society, 359, 739—757 (2001).
[37] G. Stépán, R. Szalai, and T. Insperger, Nonlinear Dynamics of High-Speed Milling Subjected to Regenerative
Effect. This volume (2003).
[38] E. Stone, and S. A. Campbell, Stability and Bifurcation Analysis of a Nonlinear DDE model for Drilling.
Submitted to Journal of Nonlinear Science (2002).
[39] S. A. Tobias, Machine Tool Vibration. London: Blackie and Son Limited (1965).
[40] S. A. Tobias, and W. Fishwick, The Chatter of Lathe Tools Under Orthogonal Cutting Conditions. Trans-
actions of the American Society of Mechanical Engineers, 80(5), 1079—1088 (1958).
[41] M. Wiercigroch, and E. Budak, Sources of Nonlinearities, Chatter Generation and Suppression in Metal
Cutting. Philosophical Transactions of the Royal Society, 359, 663—693 (2001).
18
[42] M. Wiercigroch, and A. H-D. Cheng, Chaotic and Stochastic Dynamics of Orthogonal Metal Cutting. Chaos,
Solitons & Fractals, 8(4), 715—726 (1997).
[43] M. Wiercigroch, and A. M. Krivtsov, Frictional Chatter in Orthogonal Metal Cutting. Philosophical Trans-
actions of the Royal Society, 359, 713—738 (2001).
[44] N. van de Wouw, R. P. H. Faassen, J. A. J. Oosterling, and H. Nijmeijer, Modelling of High-Speed Milling
for Prediction of Regenerative Chatter. This volume (2003).
19