Sie sind auf Seite 1von 9

# Ye Ding

LiMin Zhu
State Key Laboratory of
Mechanical System and Vibration,
School of Mechanical Engineering,
Shanghai Jiao Tong University,
Shanghai 200240, China

XiaoJian Zhang
State Key Laboratory of Digital Manufacturing
Equipment and Technology,
School of Mechanical Science and Engineering,
Huazhong University of
Science and Technology,
Wuhan 430074, China

Han Ding1
State Key Laboratory of
Mechanical System and Vibration
School of Mechanical Engineering,
Shanghai Jiao Tong University,
Shanghai 200240, China
e-mail: hding@sjtu.edu.cn

## Numerical Integration Method

for Prediction of Milling Stability
This paper presents a numerical scheme to predict the milling stability based on the
integral equation and numerical integration formulas. First, the milling dynamics taking
the regenerative effect into account is represented in the form of integral equation. Then,
the tooth passing period is precisely divided into the free vibration phase during which the
analytical solution is available and the forced vibration phase during which an approximate solution is needed. To obtain the numerical solution of the integral equation during
the forced vibration phase, the time interval of interest is equally discretized. Over each
small time interval, Newton-Cotes integration formulas or Gauss integration formulas are
employed to approximate the integral term in the integral equation. After establishing the
state transition matrix of the system in one period, the milling stability is predicted by using
Floquet theory. The benchmark examples are utilized to verify the proposed approach. The
results demonstrate that it is highly efficient and accurate. [DOI: 10.1115/1.4004136]
Keywords: stability prediction, milling, integral equation, numerical integration,
Floquet theory

Introduction

## It is well recognized that the chatter stability for metal cutting

is a vital issue in current research, which has gained considerable
attention in recent years. From the viewpoint of machining dynamics, chatter is one of the most severe limitations for surface
quality and productivity in milling operations due to choosing
improper machining parameters. To fulfill the aim of high performance milling, it is the fundamental base for optimizing the
machining process to predict the stability lobe diagrams of the
process parameters accurately and efficiently.
The model-based chatter stability analysis methodology usually
consists of two steps. First, the mathematical model of the dynamics of milling is established on the basis of the modal analysis
technology and cutting force coefficient calibration experiment
[1]. The insightful review on the history and state of the art of
mathematical modeling of chatter vibrations in metal cutting was
provided by Altintas and Weck [2]. As far as the milling process
is concerned, the governing equations of the milling dynamics
taking the regenerative effect into account are generally formulated as delay-differential equations (DDEs) with time-periodic
coefficients [24]. To obtain the frequency response function of a
cutter in a machine tool, the practical substructure analysis based
methods for prediction of the tool point dynamics were developed
by Schmitz [5] and Schmitz et al. [6]. The cutting force coefficients in milling can be identified by using the mechanistic
approaches proposed by Altintas [1] or Budak et al. [7]. Second,
the stability lobe diagrams for the machining parameters are constructed based on the mathematical model. Over the past two decades, much effort has been devoted to obtain the stability lobe
diagrams analytically or semi-analytically to reduce the computational complexity of numerical schemes. Altintas et al. [8]
1
Corresponding author.
Contributed by the Manufacturing Engineering Division of ASME for publication
in the JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received
March 23, 2010; final manuscript received April 13, 2011; published online June 8,
2011. Assoc. Editor: Patrick Kwon.

## provided a comprehensive review for prediction of the milling

chatter stability from the viewpoint of frequency and discrete time
domain. Minis and Yanushevsky [9] discussed the problem of
milling chatter stability under the framework of periodic differential equations. Altintas and co-workers [1013] proposed the
widely accepted frequency domain methods. Davies et al. [14]
proposed a time domain method by using the analytical solution
for the noncutting vibrations coupled with an approximate solution for the vibrations during cutting. On the basis of the technique
that the original DDE is approximated by a series of ordinary differential equations (ODEs), Insperger and Stepan [15,16] presented the semidiscretization method (SDM) and the first-order
SDM [17], which have been applied to various problems such as
varying spindle speed milling [18,19], milling with variable pitch
or variable helix cutters [20], and milling with serrated tools [21]
etc. Bayly et al. [22,23] suggested the temporal finite element
analysis (TFEA) method, which can be used for simultaneous prediction of the surface location error [24,25]. Butcher et al. [26]
presented the Chebyshev polynomials based method. Ulsoy and
co-workers [27,28] proposed the Lambert function based method,
and then Olvera et al. [29] improved the result for milling stability
analysis. Olgac and Sipahi [30,31] explored the cluster treatment
of characteristic roots for predicting milling chatter stability.
More recently, Butcher et al. [32] presented the Chebyshev collocation method. Ding et al. [33] proposed the full-discretization
method (FDM) based on the direct integration scheme, which can
also be generalized to simultaneously predict the surface location
error [34]. Using the similar technique as the SDM which approximates the original DDE by a series of ODEs, Insperger [35] gave
another description of the FDM, and then compared the FDM
with the SDMs in terms of the rate of convergence and computational time.
From the fundamental formulations of the governing equations for dynamic systems in engineering, there generally
exist three kinds of equations to describe the same problem,
i.e., differential equations, variational equations, and integral
equations. To characterize the same dynamic problem, they

C 2011 by ASME

## may be mathematically equivalent, but the computational

complexity and accuracy of their corresponding numerical
methods may be quite different [36,37]. Inspired by the numerical integration methods for the classical integral equations (without time delays) [38,39], this paper focuses on the
problem of prediction of milling stability via the integral
equation and numerical integration formulas. The primary
contribution of the paper is to provide an alternative semi-analytical method employing numerical integration schemes to
calculate the stability lobe diagrams, which applies to the
problems of milling with low radial immersion, near-full
immersion, and variable pitch cutters. The main feature of the
proposed method is that the original DDE is approximated by
a series of algebraic equations constructed directly according
to the theory of integral equations. After transforming the
original DDE into the integral equation and dividing equally
the forced vibration time interval into small intervals, the
algebraic equations are obtained via the classical NewtonCotes or Gauss integration formulas approximating the integral term in the integral equation, respectively. After the state
transition matrix of the system over one period being constructed by using the algebraic equations, the milling stability
is then predicted via Floquet theory. The benchmark examples
in the literature are utilized to verify the proposed approach.
The results show that the proposed numerical algorithms are
highly efficient and highly accurate.
The rest of this paper is organized as follows. Section 2 sets up
the mathematical framework necessary for describing the dynamics of milling taking the regenerative effect into account. Section
3 presents the basic algorithms for predicting of the milling stability. In Sec. 4, the benchmark simulation examples are given, and
the method is extended to the dynamics of milling with variable
pitch cutters. Section 5 states the conclusions.

Mathematical Model

## Without loss of generality, the motion of the two degrees of

freedom (DOF) milling system is governed by [1,24,40]
_ Kqt ap Kc tqt  qt  T
M
qt Cqt

N
X

 





g /j t cos /j t Kt sin /j t Kn cos /j t
j1

(6)
In Eqs. (3)(6), Kt and Kn are the tangential and the normal linearized cutting force coefficients, respectively, and /j t is the angular position of the jth tooth defined by
/j t 2pX=60t j  1  2p=N


The switching function g /j t is defined as



1 if /st < /j t < /ex
g /j t
0 otherwise

(7)

(8)

where /st and /ex are the start and exit angles of the jth cutter
tooth, respectively. For down-milling, /st arccos2a=D  1 and
/ex p; for up-milling, /st 0 and /ex arccos1  2a=D,
where a=D is the radial immersion ratio, a is the radial depth of cut


and D is the diameter of the cutter.
qt
_ Cqt=2 and xt
Letting pt Mqt
, Eq. (1)
pt
can be re-written as
_ Axt ap Btxt  xt  T
xt

(9)

where

A




0
0
M1 C=2
M1
and
Bt

Kc t 0
CM1 C=4  K CM1 2
(10)

By regarding the term ap Btxt  xt  T as the nonhomoge_ Axt, the soluneous term of the homogeneous equation xt
tion of Eq. (9) can be expressed as
t n
o
eAtn Bnxn  xn  T dn
xt eAtt0 xt0 ap
t0

(1)

(11)

where M, C, K, qt, and ap denote the system mass, damping, stiffness matrices, displacement vector, and axial depth
of cut, respectively. In Eq. (1), T denotes the tooth passing
period, i.e., T 60=NX, where N is the number of the cutter teeth and X is the spindle speed in revolutions per minute
(rpm). Kc t in Eq. (1) denotes the directional cutting force
coefficient matrix (for more details of the formulations, refer
to Ref. [1]), which varies periodically in time, i.e.
Kc t Kc t T, and is defined as


hxx t hxy t
(2)
Kc t
hyx t hyy t
where
hxx t

hyy t

N
X

 





g /j t sin /j t Kt cos /j t Kn sin /j t

## where xt0 denotes the state value at t t0 .

Since the function xn under the integral symbol is unknown,
Eq. (11) is a kind of integral equations. If the time delay term
xn  T in Eq. (11) vanishes, it would become the standard Volterra integral equation of the second kind [38,39]. It is well known
that the key step for the semidiscretization method [16,17] and
TFEA method [23,24] is to construct the transition matrix on one
tooth passing period conducting from a discrete dynamical map
which approximates the original system Eq. (1). They are based
on the techniques from the theory of differential equations and the
method of weighted residuals, respectively. Starting with the integral equation Eq. (11), a novel numerical scheme will be developed to construct a discrete dynamical map based on the classical
numerical integration techniques of integral equations for prediction of the milling stability in Sec. 3.

j1

(3)
hxy t

N
X

 





g /j t cos /j t Kt cos /j t Kn sin /j t
j1

(4)
hyx t

N
X


 





g /j t sin /j t Kt sin /j t Kn cos /j t

Numerical Algorithms

## 3.1 Newton-Cotes Formula Based Algorithm. Without loss

of generality, denoted by t0 and tf , the time the cutter leaves the
workpiece and the duration of the free vibration. When the cutter
is not in contact with the part, the term Bn in Eq. (11) vanishes,
and Eq. (11) becomes xt eAtt0 xt0 . At the end of the free
vibration duration, i.e., at t t0 tf , the state becomes

j1

(5)
031005-2 / Vol. 133, JUNE 2011

(12)

## Transactions of the ASME

To
 deal with the forced vibration phase, the time interval
t0 tf ; t0 T is equally divided
 into n small intervals, and the
time step is denoted by s T  tf n. The corresponding
sampled time points are ti t0 tf i  1s, where
i 1; ; n 1 and xt1 xt0 tf .
Following the classical numerical algorithm for the Volterra integral equations of the second kind [38,39], we adopted NewtonCotes formulas to approximate the integral term in Eq. (11).
At the time point t2 , xt2 can be obtained from Eq. (11) as
xt2 eAt2 t1 xt1 ap

t2 n
o
eAt2 n Bnxn  xn  T dn
t1

(13)
Similar to the numerical integration method for the classical integral equations [38,39] and by using the trapezoidal rule [41], Eq.
(13) is simplified as
ns h

## eAt2 t1 Bt1 xt1

xt2 eAt2 t1 xt1 ap
2

i o
xt1  T Bt2 xt2  xt2  T
:
(14)

## Then, the transition matrix U1 over one tooth passing period

can be expressed as

ap s 1
ap s
D1
D1 E

U 1 I  C1 
2
2

(18)

## Thereafter, the chatter stability can be determined via Floquet

theory [42]. If the modulus of any eigenvalue of the transition matrix is greater than one for a given depth of cut ap and spindle
speed X, the milling process is unstable.
Remark 1. To determine t0 and tf , one simple method is to sample the directional cutting force coefficient matrix Eq. (2) during
one tooth passing period with a high sampling frequency to find
out them approximately. As for the case that there always exists
one tooth in the cut during one tooth passing period, the algorithm
is still applicable by choosing t0 as an arbitrary time point when
the cutter is in contact with the part and setting tf 0.
Alternatively, xti (i 3; ; n 1) can be computed by using
the Simpsons rule [41]. It follows from Eq. (11), i.e.,
xti eAti ti2 xti2
ti n
o
eAti n Bnxn  xn  T dn
ap

(19)

ti2

## Similarly, at the other time points ti (i 3; ; n 1), xti can be

calculated by
xti eAti ti1 xti1
ti n
o
eAti n Bnxn  xn  T dn
ap

(15)

ti1

## Equation (15) for ti (i 3; ; n 1) can also be approximated by

xti eAti ti xt
 i1

s eAti ti1 Bti1 xti1  xti1  T
ap
2 Bti xti  xti  T
(16)

## According to the Simpsons rule, xti can be approximated by

xti e2As xt8
i2
9
2As
<
=
i2 xti2  xti2  T
s e Bt
ap
4eAs Bti1 xti1  xti1  T
;
3:
Bti xti  xti  T

By combining Eqs. (12), (14), and (20), an alternative discrete dynamical map is obtained as
3
3
2
2
xt1  T
xt1
7
a s
6 xt2  T 7

ap s 6
7
6 xt2 7
6
p
D2 6 . 7 
D2 E 6
I  C2 
7
..
5
4 .. 5
4
3
3
.
xtn1

From Eqs. (12), (14), and (16), the following discrete dynamical
map can be obtained as
2

xt1
xt2
..
.

ap s 6
6
D1 6
I  C1 
4
2

7
a s
6
7
6
p
D1 E 6
7 
5
4
2

xtn1

where

7
7
7
5

0
eAs

0
..
.

..

As

0
3

B2
eAs B2

7
7
7
7
7
5

B3
.. ..
.
.
eAs Bn

where
3

0
6 eAs
6 2As
e
C2 6
6
4

0
0
..
.

0
..
.

..

e2As

0
6 3eAs B1 =2 3B2 =2
6 2As
6
4eAs B2
and D2 6 e B1
6
..
4
.

7
7
7,
7
5

xtn1  T
(21)

2
3
0    0 eAtf
7
7
60  0 0 7
7, E 6 . .
.. 7
7
4 ..
. . ...
. 5
5
0  0 0
3

e
0
6 eAs B1
6
6
D1 6
6
4

xtn1  T
(17)

0
6 eAs
6
C1 6
6
4

xt1  T
xt2  T
..
.

(20)

Bn1

3
0    0 eAtf
60  0 0 7
E6
, and Bi denotes Bti for i 1; ; n 1:
.. 7
4 ... . . . ...
. 5
0  0 0
Journal of Manufacturing Science and Engineering

B3
..
.
e2As Bn1

7
7
7
7
7
5

..

.
4eAs Bn

Bn1

## Then, an alternative transition matrix U2 on one tooth passing

period can be expressed as

ap s 1
ap s
D2
D2 E

U 2 I  C2 
3
3

(22)

## Remark 2. According to Ref. [38], the state xt2 can also be

calculated by using the Simpsons rule with adding the state
xt1 2s in the formulation, which can be obtained via the trapezoidal rule.
JUNE 2011, Vol. 133 / 031005-3

## 3.2 Gaussian Formula Based Algorithm. To numerically

calculate Eq. (13) and Eq. (15), the Gauss quadrature methods can
also be applied. In this section, the two-point Gauss-Legendre
integration formula [41] is employed. At the time points ti
(i 2; ; n 1), Eq. (13) and Eq. (15) are approximated by
xti eAti ti1 xti1
 
 At t   
s
e i i1 Bti1 xti1  xti1  T


ap

2
eAti ti Bti xti  xti  T
(23)
where
ti1



 
ti1 ti s
1
ti1 ti s
1


  p and ti
 p :
2
2
2
2
3
3
xti1

xti

## The state terms

and
can be linearly approximated
via the boundary values xti1 and xti , i.e.,


t  ti1
t  ti1
xti1 i1
xti
xti1 1  i1
s
s


1
1
1
1
1 p xti1
1  p xti
(24)

2
2
3
3
and


 ti1
t  ti1
xti 1 
xti1 i
xti
s

 s 
1
1
1
1
1  p xti1
1 p xti

2
2
3
3

and
xti  T

## Substituting Eqs. (24)(27) into Eq. (23) yields the result as

xi eAs xi1

8
9
Exp1  C1  Bti1 Exp2  C2  Bti xi1
>
>
>
>


>
>
=
s < Exp1  C2  Bti1 Exp2  C1  Bti xi


ap


2>
 Exp1  C1  Bti1 Exp2  C2  Bti xti1  T >
>
>
>
>

: 
;
 Exp1  C2  Bti1 Exp2  C1  Bti xti  T
(28)

A2s1p1
As1p1
Ati ti
3 ,
3 ,
e 2
where
Exp1 eAti ti1 e

Exp2 e
C1 12 1 p13 and C2 12 1  p13 . According to Eqs. (12)
and (28), the following discrete dynamical map is constructed

ap s 6
6
D3 6
I  C3 
4
2

xt1
xt2
..
.

xtn1

 T and xti
Similarly, the time delayed terms
be linearly approximated via the boundary values xti1

(25)

where

 T can
 T and

xti  T, i.e.,




1
1
1
1
1 p xti1  T
1  p xti  T
2
2
3
3
(26)

3
0
eAs

0
.. ..
.
.
eAs

2
0 
7
7
60 
7; E 6 . .
7
4 ..
..
5
0 

and

3
Exp1  C2  Bt1
Exp2  C1  Bt2
Exp1  C1  Bt2
Exp2  C2  Bt3

Exp1  C2  Bt2
Exp2  C1  Bt3
..
.

Then the transition matrix U3 over one tooth passing period can
be expressed as

ap s 1
ap s
D3
D3 E

(30)
U 3 I  C3 
2
2
Remark 3. If the inverse matrices in Eqs. (18), (22), or (30) do
not exist, the Moore-Penrose pseudo-inverses [43] of the corresponding matrices can be used instead.
Remark 4. As for the structures of the transition matrices, U1 in
Eq. (18), U2 in Eq. (22), and U3 in Eq. (30), all have very similar
matrix forms as the TFEA method [23,24]. However, the basic
procedures and the elements of the transition matrices for the proposed method and the TFEA method are totally different. To solve
the DDE when the tool is in the cut, the present approach obtains
031005-4 / Vol. 133, JUNE 2011

7
7
7
5

3
0 eAtf
0 0 7
..
.. 7
.
. 5
0 0

0

6 Exp1  C1  Bt1
6 Exp  C  Bt
6
2
2
2
6
6
D3 6
6
6
6
6
4

xtn1  T
(29)

0
6 eAs
6
C3 6
6
4

xt1  T
xt2  T
..
.

7
a s
6
7
6
p
D3 E 6
7 
5
4
2

ti

xti1

xti1  T





1
1
1
1
1  p xti1  T
1 p xti  T (27)
2
2
3
3

..

.
Exp1  C1  Btn
Exp2  C2  Btn1

Exp1  C2  Btn
Exp2  C1  Btn1

7
7
7
7
7
7
7
7
7
7
5

the approximate solution by using the numerical integration techniques for the integral equation, while the TFEA method [23,24]
by the application of finite elements in time.

Numerical Results

## All computer programs of the proposed algorithms are written in

MATLAB 7.x and run on a personal computer [Intel Core (TM) 2 Duo
Processor, 2.1 GHz, 1GB] for the benchmark data available in the literature [16,23,44] to test the accuracy and the efficiency, respectively.
4.1 Example 1: Single DOF Milling Model. If one DOF of
the cutter is much stiffer than the other DOF, or the cutter is
assumed to be flexible in the feed direction only [45], the two
DOF system described by Eq. (1) can be reduced as a single DOF
Transactions of the ASME

The system parameters are: a two fluted cutter, the natural frequency is fn xn =2p 922Hz, the relative damping is
f 0:011, the modal mass is mt 0:03993kg, and the cutting
force coefficients are Kt 6  108 N=m2 and Kn 2  108 N=m2 .
4.1.1 Rate of Convergence Analysis. As the reliable and inexpensive time domain methods, the local discretization errors of
the zeroth-order SDM [16] and the first-order SDM [17] are
proved as Os2 and Os3 , respectively, where s denotes the discretization step. By using the error estimation method for integral
equations [38] and following the similar way as the SDMs
[16,17], the local discretization error of the trapezoidal rule based
method, i.e., Eq. (14) and Eq. (16), can be proved as Os3 , similarly to that of the first-order SDM [17]. As for the Gauss-Legendre integration formula based method, i.e., Eq. (28), the local
discretization error is also Os3 . The local discretization error
comparison of the numerical integral methods with the two SDMs
for single DOF system is illustrated in Fig. 1, where the operation
parameters are chosen as the same as Ref. [35]. The down-milling
operations are used with a=D 1, X 5000rpm, and three different depths of cut ap 0:2; 0:5; 1mm, respectively. In Fig. 1, the
convergence of the critical eigenvalues jlj are calculated with
respect to different approximation parameters n (i.e., the number
of the discretization intervals over one tooth passing period due to
using the two fluted cutter and a=D 1) for Eq. (18), Eq. (30),
the zeroth-order SDM [16], and the first-order SDM [17], where
jl0 j is determined by using the first-order SDM with n 500 for
the reference.
4.1.2 Stability Diagrams. The stability lobe diagrams are calculated by using the original MATLAB code of the FDM [33] and
the present method using Eq. (18) over a 400  200 sized grid of
parameters of the spindle speed X and depth of cut ap . The computational results are compared in Table. 1 for down-milling with
radial depth of cut ratios a=D 1 and 0.05, respectively. It is
shown that the present method has high computational efficiency
without loss of any computational precision. Compared with the
FDM in Ref. [33], it is shown that the proposed method is more
efficient. Note that as mentioned by Insperger [35], the computational efficiency of the FDM can be improved by employing some
efficient matrix multiplication algorithms such as the method proposed in Ref. [46].

Fig. 1 Convergence of the eigenvalues for different approximation parameters n for the proposed numerical integration
method (NIM) via Eq. (18) and Eq. (30), the zeroth-order (0th)
SDM [16], and the first-order (1st) SDM [17]

## system. The dynamic equation of a single DOF milling model in

the state-space form is [16,23,45]
_ Axt ap Btxt  xt  T
xt

(31)

where
2
A4

fxn
mt fxn 2  mt x2n

3

1
0
mt 5; Bt
hxx t
fxn

0
0

(32)

## 4.2 Example 2: Two DOF Milling Model. In this example,

the computational precision of the proposed method is compared
in detail with the TFEA method [24] and the first-order SDM
[17]. The system parameters are selected as the same as those
employed in the single DOF milling model case, and they are
assumed to be equal at x and y directions. Under the same program framework as for the zeroth-order SDM [16], the first-order
SDM is implemented, and the technique for separation of the free
vibration and forced vibration is employed. The stability lobe diagrams obtained by using the first-order SDM with the number of
the discretization intervals over one period m 30 and the proposed numerical integration method (NIM) via Eq. (30) with
n 10 over a 400  200 sized grid of parameters are drawn together in Fig. 2 for up-milling with the radial depth of cut ratio
a=D 0:05. It is seen that the results of the two methods are
highly consistent. The elapsed time is about 408 seconds for the
first-order SDM, while about 47.0 seconds for the NIM via Eq.
(30). It should be noted that the computational time for the firstorder SDM can be reduced if some efficient matrix multiplication
algorithms and other numerical skills [46] are utilized.
As mentioned in Sec. 3, the structures of the transition matrices of the proposed approach are similar to the TFEA method
[23,24]. To further compare TFEA method [24] with the present
one [the NIM via Eq. (30) with n 10], four elements in the cut
are used for the TFEA method and the program is implemented
by using the symbolic calculations in MATLAB. The other operation and computational parameters are the same as that used in
JUNE 2011, Vol. 133 / 031005-5

Table 1

## FDM [33] with the number of the discretization

intervals over one tooth passing period m 40

## Eq. (18) with the number of the discretization

intervals over one forced vibration duration n 40

358.0

129.6

## FDM [33] with the number of the discretization

intervals over one tooth passing period m 40

## Eq. (18) with the number of the discretization

intervals over one forced vibration duration n 10

291.8

13.9

a=D 1

Computational
time (s)

a=D 0.05

Computational
time (s)

## Fig. 2 Comparison of the proposed NIM via Eq. (30) (with

n 5 10) and the first-order (1st) SDM (with m 5 30) [17] for two
DOF milling model

## Fig. 3 Comparison of the proposed NIM via Eq. (30) (with

n 5 10) and the TFEA method (with four elements in the cut) for
two DOF milling model

Table. 2

Direction
X
Y

## Modal parameters of cutter from Ref. [44]

Mode

Natural
frequencies (Hz)

Modal effective
masses (kg)

Damping
ratios

1
2
3
1

441.64
563.6
778.56
516.21

11.125
1.4986
13.063
1.199

0.028722
0.055801
0.058996
0.025004

Fig. 2. The comparative result is shown in Fig. 3, which illustrates a good agreement between the two methods. The computational time is about 1510 s for the TFEA method. It should be
noted that the computational time can be reduced, if the numerical calculations are totally employed to implement the TFEA
method.
4.3 Example 3: Stability Prediction for Variable Pitch
Cutters. Milling with variable pitch cutters is a practical technique for eliminating chatter vibrations. Altintas et al. [44] proposed the analytical method in frequency domain to predict the
stability lobe diagram which was well validated by the experimental results. Budak [47,48] proposed an analytical design method
for variable pitch cutters to increase stability on the basis of the
frequency domain method. Sims et al. [20] and Olgac and Sipahi
[30] used the semidiscretization method and the cluster treatment
of characteristic roots to reconsider the problem, respectively.
Due to the multiple regenerative time delay effect, Eq. (9) should
be modified as
N
X


_ Axt ap
Bj t xt  xt  Tj
(33)
xt

## Fig. 4 Stability prediction for the variable pitch cutter


M1 C=2
M1
, Tj is the time delay
CM1 C=4  K CM1 2
between teeth j and j 1 due to spindle rotation, and the sum
PN
j1 Tj equals to the spindle period T 60=X. In Eq. (33), Bj t
is periodic at the spindle period T and defined as

where A


Bj t

0
Kj t

0
0


(34)

where Kj t is expressed as

j1



Kj t g /j t












 
cos /j t Kt cos /j t Kn sin /j t 
sin /j t Kt cos /j t Kn sin /j t 












sin /j t Kt sin /j t Kn cos /j t  cos /j t Kt sin /j t Kn cos /j t 

## The algorithms presented in Sec. 3 can be generalized to predict

the chatter stability of the system governed by Eq. (33). The main
difference is that the discrete dynamical map should be constructed over one spindle period. To demonstrate this, Eq. (18)
using the trapezoidal rule is generalized here. To simplify the procedure, the whole spindle period are divided equally. Denote by ti
(i 1; ; n 1) and s the sampled time points and the sampling
period, respectively. All Tj for (j 1,...,N) are integral multiples
of s, i.e., Tj nj s for (j 1,...,N). At the sampled time points, the
state in Eq. (33) has the expression as
xti eAti ti1 xti1
N ti n
X

o
ap
eAti n Bj n xn  xn  Tj dn (36)
j1

ti1

## xti eAti ti1 xti1


 
N  Ati ti1
sX
e
Bj ti1 xti1  xti1  Tj

ap
Bj ti xti  xti  Tj
2 j1
(37)

## xti Fi Gi1 xti1

N 
X
Hj;i xti  nj s
j1


eAs Hj;i1 xti1  nj s
where

#1
N
sX
Bj ti
Fi I  ap
2 j1

(38)

"

Gi1 e

As

!
N
sX
I ap
Bj ti1
2 j1

s
Hj;i ap Bj ti
2

## From Eq. (37), xti can be explicitly computed, i.e.,

(35)

(39)

(40)

(41)

s
Hj;i1 ap Bj ti1
(42)
2

P
Denoting that yi col xti ; xti  s; ; xti  s Nj1 nj , the
discrete dynamical map Eq. (38) can be re-written as
yi Wi yi1

(43)

## JUNE 2011, Vol. 133 / 031005-7

where

Fi Gi1
6I
6
6
6
6
6
6
6
6
6
Wi 6
6
6
6
6
6
6
6
6
4

0 
0
I 0
I

Fi H1;i

Fi eAs H1;i1

..
..

(44)

Remark 5. In the case that one or more time delays Tj are not integral multiple of the sampling period s, there is no difficulty to
interpolate the time delayed terms linearly using the corresponding boundary values with the way similar to those employed in
Eqs. (26)(27).
The benchmark example from Ref. [44] is utilized to verify the
algorithm. The tool diameter is 19.05 mm. The cutter has four
flutes with the helix angle 30 deg. The pitch angles are 70110
70110 deg. The radial immersion is 9.525 mm. The milling
mode is down milling. The modal parameters of the cutter are
listed in Table 2. The cutting force coefficients are Kt 697MPa
and Kn Kr  Kt 0:367  697MPa. The first two modes in the x
direction and the first mode in the y direction are taken into
account. By using the computational parameter n 72 with the
120  80 equally sized grid of parameters for the spindle speed
and depth of cut, the stability lobe diagram is illustrated in Fig. 4,
which agrees very well with the result by using the frequency
domain solution experimentally validated in Ref. [44].

Conclusions

In this work, based on the integral equation and numerical integration formulas, a concise semi-analytical method for constructing a discrete dynamical map approximating the delay-differential
equations governing the milling dynamics is developed. By using
Floquet theory, the chatter stability can be determined according to
the moduli of all the eigenvalues of the transition matrix of the discrete map. Three benchmark examples which have been validated
in the literature are utilized to verify the proposed method. Several
distinct features of the algorithm are summarized as follows. The
algorithm is general in the sense that it applies to end milling under
different conditions including high radial immersion ratio, small
radial immersion ratio, single time delay, and multiple time delays.
The algorithm has high computational efficiency, which has been
demonstrated via example 1. The algorithm also has high computational accuracy which has been demonstrated via example 2.
Moreover, the method is easy to understand and implement
because the formulas used are very simple in form.

Acknowledgment
This work was partially supported by the National Key Basic
Research Program under Grant No. 2011CB706804, and the Science and Technology Commission of Shanghai Municipality
031005-8 / Vol. 133, JUNE 2011

Fi eAs H4;i1



0 0 0

.
.

0
I 0
I 0

It should be noted that if two of the time delays are equal, e.g.,
Tj1 Tj2 , their corresponding coefficients Fi Hj;i and Fi eAs Hj;i1
for j j1 ; j2 should be summed respectively in the matrix Wi .
At last, one obtains the following transition matrix U4 over one
spindle period
U4 Wn Wn1    W2 W1

   Fi H4;i

3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5

## under Grant Nos. 09QH1401500 and 10JC1408000. The authors

would like to thank Profesor E. Budak for his assistance in implementing the MATLAB program with variable pitch cutters by using
the frequency domain method and appreciate the anonymous
reviewers for their insightful comments and constructive suggestions to improve the manuscript.

References
[1] Altintas, Y., 2000, Manufacturing Automation: Metal Cutting Mechanics, Machine
Tool Vibrations, and CNC Design, Cambridge University Press, Cambridge.
[2] Altintas, Y., and Weck, M., 2004, Chatter Stability of Metal Cutting and
Grinding, CIRP Ann.Manuf. Technol., 53(2), pp. 619642.
[3] Balachandran, B., 2001, Nonlinear Dynamics of Milling Processes, Philos.
Trans. R. Soc. London, Ser. A, 359, pp. 793819.
[4] Wiercigroch, M., and Budak, E., 2001, Sources of Nonlinearities, Chatter
Generation and Suppression in Metal Cutting, Philos. Trans. R. Soc. London,
Ser. A 359(1781), pp. 663693.
[5] Schmitz, T. L., 2000, Predicting High-Speed Machining Dynamics by Substructure Analysis, CIRP Ann. Manuf. Technol., 49(1), pp. 303308.
[6] Schmitz, T. L., Davies, M. A., and Kennedy, M. D., 2001, Tool Point Frequency Response Prediction for High-Speed Machining by RCSA, ASME J.
Manuf. Sci. Eng., 123(4), pp. 700707.
[7] Budak, E., Altintas, Y., and Armarego, E. J. A., 1996, Prediction of Milling
Force Coefficients from Orthogonal Cutting Data, ASME J. Manuf. Sci.
Eng., 118(2), pp. 216224.
[8] Altintas, Y., Stepan, G., Merdol, D., and Dombovari, Z., 2008, Chatter Stability of Milling in Frequency and Discrete Time Domain, CIRP J. Manuf. Sci.
Technol., 1(1), pp. 3544.
[9] Minis, I., and Yanushevsky, R., 1993, A New Theoretical Approach for the
Prediction of Machine Tool Chatter in Milling, J. Eng. Ind., 115(1), pp. 18.
[10] Altintas, Y., and Budak, E., 1995, Analytical Prediction of Stability Lobes in
Milling, CIRP Ann.Manuf. Technol., 44(1), pp. 357362.
[11] Budak, E., and Altintas, Y., 1998, Analytical Prediction of Chatter Stability
in MillingPart I: General Formulation, ASME J. Dyn. Syst,. Meas., Control, 120(1), pp. 2230.
[12] Budak, E., and Altintas, Y., 1998, Analytical Prediction of Chatter Stability in MillingPart II: Application of the General Formulation to
Common Milling Systems, ASME J. Dyn. Syst., Meas., Control,
120(1), pp. 3136.
[13] Merdol, S. D., and Altintas, Y., 2004, Multi Frequency Solution of Chatter
Stability for Low Immersion Milling, ASME J. Manuf. Sci. Eng., 126(3), pp.
459466.
[14] Davies, M. A., Pratt, J. R., Dutterer, B., and Burns, T. J., 2002, Stability Prediction for Low Radial Immersion Milling, ASME J. Manuf. Sci. Eng.,
124(2), pp. 217225.
[15] Insperger, T., and Stepan, G., 2002, Semi-Discretization Method for Delayed
Systems, Int. J. Numer. Methods Eng., 55(5), pp. 503518.
[16] Insperger, T., and Stepan, G., 2004, Updated Semi-Discretization Method for
Periodic Delay-Differential Equations with Discrete Delay, Int. J. Numer.
Methods Eng., 61(1), pp. 117141.
[17] Insperger, T., Stepan, G., and Turi, J., 2008, On the Higher-Order Semi-Discretizations for Periodic Delayed Systems, J. Sound Vib., 313(12), pp. 334341.
[18] Seguy, S., Insperger, T., Arnaud, L., Dessein, G., and Peigne, G., 2010, On
the Stability of High-Speed Milling with Spindle Speed Variation, Int. J.
Adv. Manuf. Technol., 48(912), pp. 883895.
[19] Seguy, S., Dessein, G., Arnaud, L., and Insperger, T., 2010, Control of Chatter by Spindle Speed Variation in High-Speed Milling, Adv. Mater. Res.,
112(1), pp. 179186.
[20] Sims, N. D., Mann, B., and Huyanan, S., 2008, Analytical Prediction of Chatter Stability for Variable Pitch and Variable Helix Milling Tools, J. Sound
Vib., 317(35), pp. 664686.

## Transactions of the ASME

[21] Dombovari, Z., Altintas, Y., and Stepan, G., 2010, The Effect of Serration on
Mechanics and Stability of Milling Cutters, Int. J. Mach. Tools Manuf.,
50(6), pp. 511520.
[22] Bayly, P. V., Halley, J. E., Mann, B. P., and Davies, M. A., 2001,
Stability of Interrupted Cutting by Temporal Finite Element Analysis,
Proceedings of the ASME Design Engineering Technical Conference,
Vol. 6C, pp. 23612370.
[23] Bayly, P. V., Mann, B. P., Schmitz, T. L., Peters, D. A., Stepan, G., and
Insperger, T., 2002, Effects of Radial Immersion and Cutting Direction on
Chatter Instability in End-Milling, American Society of Mechanical Engineers, Manufacturing Engineering Division, MED, Vol. 13, pp. 351363.
[24] Mann, B. P., Young, K. A., Schmitz, T. L., and Dilley, D. N., 2005,
Simultaneous Stability and Surface Location Error Predictions in Milling,
ASME J. Manuf. Sci. Eng., 127(3), pp. 446453.
[25] Mann, B. P., Edes, B. T., Easley, S. J., Young, K. A., and Ma, K., 2008,
Chatter Vibration and Surface Location Error Prediction for Helical End
Mills, Int. J. Mach. Tools Manuf., 48(34), pp. 350361.
[26] Butcher, E. A., Ma, H., Bueler, E., Averina, V., and Szabo, Z., 2004, Stability
of Linear Time-Periodic Delay-Differential Equations Via Chebyshev Polynomials, Int. J. Numer. Methods Eng., 59(7), pp. 895922.
[27] Maghami Asl, F., and Ulsoy, A. G., 2003, Analysis of a System of Linear
Delay Differential Equations, ASME J. Dyn. Syst., Meas., Control, 125(2),
pp. 215223.
[28] Yi, S., Nelson, P. W., and Ulsoy, A. G., 2007, Delay Differential Equations
Via the Matrix Lambert W Function and Bifurcation Analysis: Application to
Machine Tool Chatter, Math. Biosci. Eng., 4(2), pp. 355368.
[29] Olvera, D., Calva, V., Gonzalez, J. L., Pacheco, J., Elias-Zuniga, A., and De
Lacalle, L. N. L., 2008, Milling Stability Lobes Computation through the
Lambert W Function, Transactions of the North American Manufacturing
Research Institution of SME, Vol. 36, pp. 233239.
[30] Olgac, N., and Sipahi, R., 2005, A Unique Methodology for Chatter Stability
Mapping in Simultaneous Machining, ASME J. Manuf. Sci. Eng., 127(4), pp.
791800.
[31] Olgac, N., and Sipahi, R., 2007, Dynamics and Stability of Variable-Pitch
Milling, J. Vib. Control, 13(7), pp. 10311043.
[32] Butcher, E. A., Bobrenkov, O. A., Bueler, E., and Nindujarla, P., 2009,
Analysis of Milling Stability by the Chebyshev Collocation Method: Algorithm and Optimal Stable Immersion Levels, J. Comput. Nonlinear Dyn.,
4(3), pp. 031003.

## Journal of Manufacturing Science and Engineering

[33] Ding, Y., Zhu, L., Zhang, X., and Ding, H., 2010, A Full-Discretization
Method for Prediction of Milling Stability, Int. J. Mach. Tools Manuf., 50(5),
pp. 502509.
[34] Ding, Y., Zhu, L., Zhang, X., and Ding, H., 2011, On a Numerical Method
for Simultaneous Prediction of Stability and Surface Location Error in Low
Radial Immersion Milling, ASME J. Dyn. Syst., Meas. Control, 133(2),
p. 024503.
[35] Insperger, T., 2010, Full-Discretization and Semi-Discretization for Milling
Stability Prediction: Some Comments, Int. J. Mach. Tools Manuf., 50(7),
658662.
[36] Feng, K., and Shi, Z.-C., 2001, Mathematical Theory of Elastic Structures,
Springer, Berlin; New York.
[37] Yang, D., and Zhao, Z., 2006, The Boundary Element Theory and Application,
Beijing Institute of Technology Press, Beijing, in Chinese.
[38] Delves, L. M., and Mohamed, J. L., 1985, Computational Methods for Integral
Equations, Cambridge University Press, Cambridge.
[39] Li, X., 2008, Integral Equations, Science Press, Beijing, in Chinese.
[40] Gradisek, J., Kalveram, M., Insperger, T., Weinert, K., Stepan, G., Govekar,
E., and Grabec, I., 2005, On Stability Prediction for Milling, Int. J. Mach.
Tools Manuf. 45(78), pp. 769781.
[41] Yang, W. Y., Cao, W., Chung, T.-S., and Morris, J., 2005, Applied Numerical
Methods Using Matlab, Wiley-Interscience, Hoboken, NJ.
[42] Farkas, M., 1994, Periodic Motions, Springer-Verlag, New York.
[43] Serre, D., 2002, Matrices: Theory and Applications, Springer-Verlag, New York.
[44] Altintas, Y., Engin, S., and Budak, E., 1999, Analytical Stability Prediction
and Design of Variable Pitch Cutters, ASME J. Manuf. Sci. Eng., 121(2), pp.
173178.
[45] Insperger, T., Mann, B. P., Stepan, G., and Bayly, P. V., 2003, Stability of
up-Milling and Down-Milling, Part 1: Alternative Analytical Methods, Int. J.
Mach. Tools Manuf., 43(1), pp. 2534.
[46] Henninger, C., and Eberhard, P., 2008, Improving the Computational Efficiency and Accuracy of the Semi-Discretization Method for Periodic DelayDifferential Equations, Eur. J. Mech. A=Solids, 27(6), pp. 975985.
[47] Budak, E., 2003, An Analytical Design Method for Milling Cutters with Nonconstant Pitch to Increase Stability, Part I: Theory, ASME J. Manuf. Sci.
Eng., 125(1), pp. 2934.
[48] Budak, E., 2003, An Analytical Design Method for Milling Cutters with Nonconstant Pitch to Increase Stability, Part 2: Application, ASME J. Manuf. Sci.
Eng., 125(1), pp. 3538.