Mechanical Systems and Signal Processing 169 (2022) 108750

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Generalized modeling of milling dynamics for the 4DOF

machining system with asymmetric flexibility
Xing Zhang , Zhao Zhao , Kunhong Chen , Wanhua Zhao *
State Key Laboratory for Manufacturing Systems Engineering, Xi’an Jiaotong University, Xi’an, Shaanxi Province 710054, PR China

A R T I C L E I N F O A B S T R A C T

Communicated by Kihong Shin Asymmetric flexibility of the machining system exists widely in the practical milling process. In
this paper, a generalized milling dynamics modeling method for an asymmetric flexible 4DOF
Keywords: cutter-workpiece system is proposed, which is further used to reveal how the asymmetric flexi­
Asymmetric flexibility bility affects the milling dynamics characteristics. The regenerative milling force model is firstly
Milling dynamics
established through the coupling coefficient matrix, in which the milling condition with non-
Feed direction-dependent
parallel direction between the axis directions of coordinate system, the vibration directions of
Stability
SLE system DOF and the acting directions of milling force are well handled. Then, a generalized 4DOF
milling dynamics model considering a new dimension, namely the cutter and workpiece feed
direction angle combination ψ (ψ c, ψ w), is presented. The studies point out that the milling dy­
namics characteristics of the asymmetric flexible machining system, including the milling sta­
bility and surface location error (SLE), have a significant feed direction-dependent feature.
Moreover, the generation mechanism of this kind of asymmetric milling dynamics has been
successfully clarified. The proposed method has been experimentally validated by a series of
milling tests in an asymmetric flexible machining system. Lastly, the influences of the asymmetry
degree of the system flexibility, milling type and radial depth of cut on the milling dynamics are
both analyzed by numerical simulation.

1. Introduction

Machining dynamics [1] is an important research topic in the field of CNC machining. The self-excited vibration [2] and forced
vibration [3] are the main disadvantageous factors that limit the improvement of machining accuracy and efficiency. It is very
important to study the machining dynamics mechanism for milling parameter selection [4], machine tool dynamic characteristic
design [5] and tool parameter optimization [6], which may provide the possibility to ensure the machining quality and improve the
machining efficiency. In the past few decades, a large number of research have been carried out around the two themes of self-excited
vibration and forced vibration in milling process.
Self-excited vibration, also known as chatter, is the unstable cutting caused by the regenerative effect resulting from the dynamic
coupling between the transient response and the milling force in a flexible machining system. In order to distinguish the stable and
unstable milling status, the stability lobe diagram (SLD) is often used to characterize the critical state. Thanks to the continuous
exploration by scholars, the frequency domain methods, such as the zero-order analysis method [7], multi-frequency method [8], and
the time domain methods, such as the transient time finite element method [9], semi-discretization method [10] and full-discretization

* Corresponding author at: Room A404, The South No. 1 Building, Qujiang Campus, Xi’an Jiaotong University, Xi’an, Shaanxi Province, PR China.
E-mail addresses: [email protected] (X. Zhang), [email protected] (W. Zhao).

https://doi.org/10.1016/j.ymssp.2021.108750
Received 11 July 2021; Received in revised form 9 November 2021; Accepted 11 December 2021
Available online 4 January 2022
0888-3270/© 2021 Elsevier Ltd. All rights reserved.
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

method [11] have been proposed up to today to realize the accurate prediction for SLD. Recently, the state dependent differential
equation methods [12–14] have also been studied to consider the coupling effect between the forced and self-excited vibration.
Forced vibration is an important issue in stable milling, because it can directly determine the machining quality. Montgomery and
Altintas et al. [15] put forward a method of surface topography simulation and machining error analysis based on the milling dy­
namics. Ismail et al. [16] proposed a surface topography generation model and analyzed the influence of tool eccentricity, vibration
and flank wear. Schmitz et al. [17,18] presented a closed analytical model for tool vibration response, and analyzed the influence of
tool eccentricity on roughness and surface location error (SLE). Denkena et al. [19] established the frequency response coupling
dynamic model for the tool system, reconstructed the actual movement of tool using the measured milling force data, and finally
extracted the formed surface morphology. Honeycutt and Schmitz [20] provided the time domain simulation results for SLE when
machining under unstable condition with period-2 bifurcation. It is shown that the zero or low error conditions may be selected even
for the period-2 bifurcation. Zhang et al. [21] systematic studied the influences of system dynamics, cutter runout and milling pa­
rameters on the form errors based on an improved time simulation technique. Niu et al. [22] proposed a unified method to study the
surface generation mechanism and to predict the machining quality for variable pitch tools considering runout.
Although many studies have been carried out on the milling dynamics, there are still some problems worthy of attention. The
spindle rotor-holder-cutter can be regarded as an axisymmetric structure, but the spindle box and its driving structure are difficult to be
symmetrical along the spindle axis due to the limitations in design and assembly. Additionally, the workpiece system is always
asymmetrical along the spindle axis due to the variable shape of parts and the clamping complexity. This kind of asymmetric structure
will cause the asymmetry of dynamic characteristics for the tool tip and the cutting point of workpiece, that is, the frequency response
functions (FRFs) would be different along an arbitrary direction in the cross-section of the cutter. It is found that the asymmetric
flexibility of the machining system has a significant effect on the milling dynamics.
In the research of milling dynamics modeling with asymmetric flexibility, Zhang et al. [23] found that the inconsistent modal
parameters of a 2DOF milling system will affect the milling stability through numerical simulation. Comak et al. [24] presented a
stability model of dynamic milling operation with consideration of the stationary and rotating dynamics, the results shown that the
asymmetric dynamics of a two fluted end mill with uneven geometry has an important influence on the stability pockets. Honeycutt
and Schmitz [25] used the time domain simulation to predict milling behavior for strongly asymmetric, symmetric, and weakly
asymmetric dynamics system, in which the asymmetric dynamics would add significant complexity to the milling behavior and
identify the period-n bifurcations based on the Poincaré and stability maps. Law et al. [5,26] derived the oriented FRFs by using the
feed-plane transfer function matrix to predict the milling stability with consideration of the feed direction, and lastly minimized the
machining time by a pocket milling strategy. Cao et al. [27] developed an asymmetric stiffness spindle-milling system based on the
piezoelectric actuator, and found that the asymmetric stiffness would affect the stability in up and down milling. On the basis of this, an
asymmetric stiffness control based chatter control strategy was further expounded. Baumann et al. [28] presented a new approach for
disturbing the regenerative effect by using the asymmetric dynamic properties. Here, a four-fluted milling cutter was modified with
parallel slots in the tool shank in order to establish the asymmetric dynamic characteristics. At last, the milling experiment cases
indicated a remarkable increase in the stability limit for 69% through this kind of preparations.
In summary, literature reviews show that there are still two disadvantages in the up-to-date research.

• Firstly, as far as the author knows, there is an implicit premise assumption in the existing milling dynamics modeling. That is, the
axis directions of the cutter and workpiece coordinate system, the acting directions of milling force, and the vibration directions of
system DOF are always parallel to each other. But in fact, the cutter will move along any feed direction in the feed plane, and the
axis direction of the cutter coordinate system may not be parallel to that of workpiece. Nevertheless, the traditional milling dy­
namics model is difficult to be applied to analyze this kind of milling condition with non-parallel direction.
• Secondly, how the asymmetric flexibility of the machining system affects the milling dynamics characteristics is not fully revealed
in the above works. Especially when both the cutter and workpiece systems have flexibility, it will aggravate the complexity of the
problem, which is often ignored in the existing research.

Therefore, this paper proposes a generalized modeling method for milling dynamics in an asymmetric flexible machining system.
The novelty of this paper are as follows:

(1) The milling condition with non-parallel direction is considered in the milling dynamics model. To the best of the author’s
knowledge, the new dimension, namely the cutter and workpiece feed direction angle combination ψ (ψ c, ψ w), is studied in the
proposed milling dynamics for the first time. With the aid of the generalized regenerative milling force model, the workload for
modal parameter fitting and identification can be significantly reduced.
(2) The study points out that the milling dynamics characteristics of the asymmetric flexible machining system have a prominent
feed direction-dependent feature, and its generation and evolution mechanism has been successfully clarified.

The rest content of the article is organized as follows: In section 2, the generalized milling dynamics model is derived. In section 3,
the experimental verification is carried out. In section 4, the generation and evolution mechanism are both expounded. In section 5, the
milling dynamics characteristics are further discussed. Finally, the conclusion of this paper is given in section 6.

2
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

2. Generalized milling dynamics modeling

2.1. Regenerative milling force model

Fig. 1 is the schematic diagram of milling process for an asymmetric flexible 4DOF machining system along an arbitrary feed
direction in the feed plane. XmOYm is defined as the global coordinate system of machine tool. XcOYc and XwOYw are designated as the
cutter and workpiece coordinate system, respectively. Here, the vibration DOF of the cutter and workpiece system are set to the di­
rection along their respective coordinate axes. XfOYf is the feed coordinate system, or called the local coordinate system describing the
movement of the cutter relative to the workpiece, where the Xf axis points to the feed direction. The cutter-workpiece engagement
process and milling force are usually characterized in this coordinate system.
At this time, what should be pointed out is that the axes of the cutter coordinate system are no longer parallel to that of the
workpiece coordinate system, and the cutter feed direction is also no longer parallel to any coordinate system axis, just along an
arbitrary feed direction in the feed plane. That is to say, this kind of milling process is conducted under the non-parallel condition. As a
result, the axis directions of the cutter coordinate system, the axis directions of the workpiece coordinate system, and the acting di­
rections of milling force are not parallel to each other. Hence, in order to describe the feed direction quantitatively, we define the feed
direction angle combination ψ (ψ c, ψ w) in Fig. 1, in which the ψ c is the cutter feed direction angle equaling to the intersection angle
between Xf axis and Xc axis, and the ψ w is the workpiece feed direction angle equaling to the intersection angle between Xf axis and Xw
axis.
Based on the above definition, this paper further establishes the regenerative milling force model under the non-parallel condition.
To calculate the milling force acting on the cutter and workpiece conveniently, the cutter within the effective axial depth of cut range is
discretized into a series of micro cutting units along its axis with a thickness of db. According to the basic cutting force mechanism
model [29], the milling force acting on the jth cutting unit of the ith tooth consists of the shearing and ploughing force components. It is
shown in the following formula.
{
dFt,i,j (t) = (kts h∼
i,j (t)db + ktp db)Wi,j (t)
(1)
dFt,i,j (t) = (krs h∼
i,j (t)db + krp db)Wi,j (t)

where the subscript t and r denote the tangential and radial direction, respectively.kts and krs are the tangential and radial shearing
cutting force coefficients, respectively.ktp and krp are the tangential and radial ploughing cutting force coefficients, respectively.Wi,j (t)
is the window function to judge whether the tooth is participating in cutting or not. Its formula is as follows.
{
1 θs,i,j ⩽ϕfi,j (t)⩽θe,i,j
Wi,j (t) = (2)
0 others

where θs,i,j and θe,i,j are the start and end cutting angles of tooth, respectively.
Based on the regenerative chatter mechanism, the h∼ i,j (t) in above equation means the regenerative instantaneous uncut chip
thickness. It is composed of the nominal thickness and the regenerative one, in which the nominal term can be obtained from the
trochoid motion of tooth, and the regenerative term would have relationship with the dynamical displacement modulation of the

Fig. 1. Milling process of an asymmetric flexible 4DOF machining system along an arbitrary feed direction.

3
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

machining system. In the feed coordinate system, the nominal thickness for the jth cutting unit on the ith tooth can be expressed as the
sine function model [30] as below.

hi,j (t) = ft sin(ϕfi,j (t))Wi,j (t) (3)

f
where ϕi,j (t) is the rotation angle for the jth cutting unit on the ith tooth at time t in the feed coordinate system. Its representation can be
derived as
πnt
ϕfi,j (t) = − (j − 1)β0 − (i − 1)χ 0 (4)
30

where n is the spindle speed. β0 = dbtanα/R0 is the lag angle of one cutting unit, in which α is the helix angle, R0 is the nominal radius
of cutter. χ 0 = 2π/Nt is the spacing angle between the adjacent cutter teeth, in which Nt is the tooth number.
Meanwhile, the regenerative thickness can be calculated by the Eq. (5) below with consideratoin of the regeneration effect caused
by the transient vibration of cutter and workpiece [31].
[ ]
Δhi,j (t) = [sin(ϕfi,j (t)) cos(ϕfi,j (t))] Δqcf (t) − Δqwf (t) Wi,j (t) (5)

where Δqcf (t) = qcf (t) − qcf (t − τ), in which qcf (t) = [xcf (t), ycf (t)]T is the vibration displacement vector of cutter system in the feed
coordinate system. Δqwf (t) = qwf (t) − qwf (t − τ), in which qwf (t) = [xwf (t), ywf (t)]T is the vibration displacement vector of workpiece
system in the feed coordinate system. τ denotes the average delay time taking cutter runout into account [32].
Combining Eq. (3) and Eq. (5), on can yield the regenerative instantaneous uncut chip thickness of the jth cutting unit on the ith
tooth by
h∼
i,j (t) = hi,j (t) + Δhi,j (t)
( [ ]) (6)
= ft sin(ϕfi,j (t)) + [sin(ϕfi,j (t)) cos(ϕfi,j (t))] Δqcf (t) − Δqwf (t) Wi,j (t)

Moreover, the transformation relationship of the vibration DOF among the feed coordinate system, the cutter coordinate system,
and the workpiece coordinate system can be given as follows.
{
qcf (t) = Rotψ c qc (t)
(7)
qwf (t) = Rotψ w qw (t)
[ ] [ ]
cos(ψ c ) sin(ψ c ) cos(ψ w ) sin(ψ w )
where Rotψ c = and Rotψ w = are the rotation matrices. qc (t) = [xc (t), yc (t)]T is the vi­
− sin(ψ c ) cos(ψ c ) − sin(ψ w ) cos(ψ w )
bration displacement vector of cutter in its coordinate system. qw (t) = [xw (t), yw (t)]T is the vibration displacement vector of workpiece
in its coordinate system.
Substituting the above equation into Eq. (6) yields
( )
h∼
i,j (t) = ft sin(ϕfi,j (t)) + [sin(ϕfi,j (t)) cos(ϕfi,j (t))](Rotψ c Δqc (t) − Rotψ w Δqw (t)) Wi,j (t) (8)

Then, by substituting Eq. (8) into Eq. (1) and summing up all milling force on cutting units, it can be concluded that the regen­
erative milling force acting on the cutter in the feed coordinate system can be obtained by
[ ] [ ] ([ ] [ ])
Fcfx (ψ , t) ∑∑ f
dFt,i,j (t) ∑∑ f
kts f
ktp ∑∑
= Rotϕ = db Rotϕ ft sin(ϕi,j (t)) + Wi,j (t) + db Rotfϕ
Fcfy (ψ , t) i j dF r,i,j (t) i j k rs krp i j
([ ] )
kts
[sin(ϕfi,j (t)) cos(ϕfi,j (t))](Rotψ c Δqc (t) − Rotψ w Δqw (t)) Wi,j (t) (9)
krs
⎡ ⎤
f f
f
− cos(ϕi,j (t)) − sin(ϕi,j (t))
where Rotϕ = ⎣ f f
⎦ is the cutter rotation matrix.
sin(ϕi,j (t)) − cos(ϕi,j (t))
Going a step further, one can achieve the milling force acting on the cutter in the cutter coordinate system as follows.

4
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

[ ] [ ]
Fcx (ψ , t) Fcfx (ψ , t)
= Rotψ-c 1
Fcy (ψ , t) Fcfy (ψ , t)
([ ] [ ])
∑∑ kts ktp ∑∑
= db Rotψ-c 1 Rotfϕ ft sin(ϕfi,j (t)) + Wi,j (t) + db Rotψ-c 1 Rotfϕ
([ ]i j k rs k rp ) i j
kts
[sin(ϕfi,j (t)) cos(ϕfi,j (t))](Rotψ c Δqc (t) − Rotψ w Δqw (t)) Wi,j (t) (10)
krs

Similarly, the milling force acting on the workpiece in the workpiece coordinate system can be derived as
[ ] [ ]
Fwx (ψ , t) - 1
Fwfx (ψ , t)
= Rotψ w
Fwy (ψ , t) Fwfy (ψ , t)
([ ] [ ])
∑∑ kts ktp ∑∑
= − db -
Rotψ w Rotϕ 1 f f
ft sin(ϕi,j (t)) + Wi,j (t) − db Rotψ-w 1 Rotfϕ
i j krs k rp i j
([ ] )
kts
[sin(ϕfi,j (t)) cos(ϕfi,j (t))](Rotψ c Δqc (t) − Rotψ w Δqw (t)) Wi,j (t) (11)
krs

Lastly, the regenerative milling force acting on the cutter and workpiece system can be expressed by

F(ψ , t) = F0 (ψ , t) + ΔF∼ (ψ , t) (12)

The first term on the right of the above formula is the nominal milling force component, which can be dropped by rearranging Eq.
(10) and Eq. (11) as follows.
⎡ ⎤ ⎛
0
⎢ Fcx (ψ , t) ⎥ ⎜ ([ ] [ ])
⎢ 0 ⎥ ⎜ ∑∑ ∑∑
⎢ Fcy (ψ , t) ⎥ kts ktp
F 0 (ψ , t) = ⎢ ⎥ = ⎜
⎜ db Rot - 1 Rotf f sin(ϕf
(t)) + Wi,j (t) − db Rotψ-w 1 Rotfϕ
⎢ F0 (ψ , t) ⎥ ⎜ ψc ϕ
k
t i,j
k
⎢ wx ⎥ ⎝ i j rs rp i j
⎣ 0 ⎦
Fwy (ψ , t)
⎞

([ ] [ ]) ⎟
⎟
kts k ⎟
ft sin(ϕfi,j (t)) + tp Wi,j (t) ⎟ (13)
krs krp ⎟
⎠

Besides, the second term on the right of Eq. (12) represents the regenerative milling force component with its detail in the following
equation.
⎡ ⎤
∼
⎢ ΔFcx (ψ , t) ⎥
⎢ ⎥
⎢ ΔF ∼ (ψ , t) ⎥
⎢ cy ⎥
ΔF∼ (ψ , t) = ⎢ ∼ ⎥
⎢ ΔFwx (ψ , t) ⎥
⎢ ⎥
⎣ ΔF ∼ (ψ , t) ⎦
wy (14)

= Hψ (Qψ (t) − Qψ (t − τ))

[ ]
Hψ ,11 Hψ ,12
= (Qψ (t) − Qψ (t − τ))
Hψ ,21 Hψ ,22
[ T T ]T [ ]
where Qψ = qc , qw = xc , yc , xw , yw T .Hψ is the coupling coefficient matrix, the elements of the matrix are given by

5
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ [ ]
⎨ ∑∑ kts ∑∑
Hψ ,11 = db - 1 f
Rotψ c Rotϕ [sin(ϕfi,j (t)) cos(ϕfi,j (t))]Rotψ c Wi,j (t)H ψ ,12 = − db Rotψ-c 1 Rotfϕ
⎪
⎪ krs
⎪
⎪ i j i j
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
[ ] ∑∑
kts
[sin(ϕfi,j (t)) cos(ϕfi,j (t))]Rotψ w Wi,j (t)Hψ ,21 = − db Rotψ-w 1 Rotfϕ
krs i j

[ ] ∑∑ [ ]
kts k
[sin(ϕfi,j (t)) cos(ϕfi,j (t))]Rotψ c Wi,j (t)Hψ ,22 = db Rotψ-w 1 Rotfϕ ts [sin(ϕfi,j (t)) cos(ϕfi,j (t))]Rotψ w Wi,j (t)
krs i j
krs

2.2. Generalized milling dynamics model

In view of the above formula derivations, the milling dynamics equation of the asymmetric machining system can be defined by the
following equation. The generalization of the proposed model lies in that it is established in the global coordinate system and the feed
direction angles are considered. This is the biggest difference from the traditional models.

Fig. 2. Comparison between the proposed and traditional modeling method.

6
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Mψ Q̈ψ (t) + Cψ Q̇ψ (t) + K ψ Qψ (t) = F0 (ψ , t) + ΔF∼ (ψ , t)

(15)
= F0 (ψ , t) + Hψ (Qψ (t) − Qψ (t − τ))

where Mψ = diag(mcx , mcy , mwx , mwy ), Cψ = diag(ccx , ccy , cwx , cwy ), K ψ = diag(kcx , kcy , kwx , kwy ), in which diag means the diagonal­
ization operation function.
As shown in Fig. 2, it is well known that the milling dynamics modeling based on the time domain methods mainly includes three
steps: the first step is to measure and identify the modal parameters of the tool and workpiece system, the second one is to establish the
milling dynamics equation in the feed coordinate system, and the third one is to solve the equation for stability and SLE prediction.
It should be pointed out that the traditional models does not consider the feed direction between cutter and workpiece. Therein, the
analysis of cutter-workpiece engagement, the calculation of milling force, and the definition of the vibration DOF in the machining
system are always to be carried out in the feed coordinate system (or local coordinate system). Hence, it is convenient to establish the
milling dynamics equation in the second step.
However, if the traditional modeling method is used to consider the feed direction, the modal hammering test and parameter
identification need to be carried out under different feed direction angles in the first step. That is to say, under each feed direction angle
of cutter and workpiece, two times of modal hammering and parameter identification along the X and Y directions are needed. The
final total number of times is about 2(Nψ c + Nψ w ), where Nψ c and Nψ w are the number of cutter and workpiece direction angles,
respectively.
Noteworthy, the proposed method establishes the relationship between the vibration displacement and the regenerative milling
force by using the coupling coefficient matrix with consideration of the feed direction, and the milling dynamics equation in the second
step will be more complex than the traditional models. But, as the modal hammering and parameter identification on the cutter and
workpiece system are handled in the global coordinate system, the final total number of times can reduce sharply to only four, which is
much less than that of the traditional methods, and can greatly diminish the experimental workload.

2.3. Calculation of milling stability and SLE

The time-delay dynamics system represented by the Eq. (15) has infinite dimensional state space, and its closed analytical solution
cannot be obtained by the existing mathematical methods. Most scholars are always to use the approximate numerical method to judge
the stability of the system. Hence, the paper solves the equation based on the idea of the semi-discretization method [10] to obtain the
milling stability and SLE of the machining system.
Firstly, the above dynamic equation is further written as a state space model.

Fig. 3. Experimental device.

7
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

U̇(t) = A(t)U(t) + B(t)U(t − τ) + F(t) (16)

[ ] [ ] [ ]
04×4 I4×4 04×4 04×4 [ ]
where A(t) = − 1 − 1 , B(t) = − 1 , F(t) = 04×1 M−ψ 1 F0 T , U = Qψ Q̇ψ T
Mψ (Hψ − K ψ ) − M ψ Cψ − Mψ Hψ 04×4
In the process of solving, the main steps for the semi-discretization derivation are given as Appendix A. Then, the state transition
relationship of the system in one cutter rotation period can be expressed as the following equation. According to the Floquet theory, the
stability of the system can be determined as follows: if the modulus of all eigenvalues of the state transition matrix Φ are less than or
equal to one, the system is stable; Otherwise, the system is unstable.
V Nm + 1 = ΦV 1 + G (17)
∑ m
where Φ = ZNm ZNm − 1 ...Z1 ;G = ENm + Ni=1 Zi+1 Zi+2 ⋯ZNm Ei .
When the system is stable, its response can be achieved by calculating the fixed point of the system. Here, let V* = V Nm + 1 = V1,
one can get

V* = (I − Φ)− 1 G (18)
Furthermore, the vibration displacement response of the cutter and workpiece can be extracted from the above formula. On this
basis, the envelope trajectory surface of the cutter tooth relative to the workpiece can be calculated in the feed coordinate system by
( ) { ( ) ( ) }
Sx,i,j (tk ) sin(ϕi,j (tk ) − ψ c ) cos(ψ c )
St = = Rotψ c Ri,j + fv tk + qc (tk ) − Rot−ψ c1− ψ w qw (tk ) (19)
Sy,i,j (tk ) cos(ϕi,j (tk ) − ψ c ) sin(ψ c )

where Ri,j is the actual cutting radius of tooth considering cutter runout. fv is the feed speed.
Lastly, the machined surface can be obtained by intersection of the trajectory surface and the workpiece entity, and then the SLE
can be calculated by the following formula.
⎧
⎪
⎪
⎪
⎪
⎨ 1 ∑ 1 ∑
SLE = (max(Sy,i,j (tk )) − R0 ) (Up milling) (min(Sy,i,j (tk )) + R0 ) (Down milling) (20)
⎪ N N
⎪ t a i,j N t Na i,j
⎪
⎪
⎩

where Na is the discrete number of cutting units.

Fig. 4. Measured and fitting FRFs of cutter and workpiece system.

8
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Table 1
Identified modal parameters of cutter and workpiece system.
Direction Modal mass(kg) Modal damping(N/(m/s)) Modal stiffness(N/m) Natural frequency(Hz) Damping ratio

Cutter system Xc 0.0998 21.580 2.698 × 106 827.5 0.0208

Yc 0.0791 18.722 2.248 × 106 848.4 0.0222
Workpiece system Xw 6.232 446.676 3.053 × 106 111.4 0.0512
Yw 6.977 462.796 4.478 × 106 127.5 0.0414

3. Experimental verification

3.1. Experimental device

In order to verify the validity of the milling dynamics modeling method proposed in this paper, a series of experiment tests are
carried out on the MDG DMU50 high speed five-axis machine tool. Fig. 3 shows the experimental setup, milling process and data
acquisition instrument. As shown in Fig. 3(a), the tool used in the experiment is a two-tooth cemented carbide cutter with a helix angle
of 35◦ and a diameter of 16 mm, which is adopted to cut the workpiece made of 7050 aluminum alloy. In order to obtain a high flexible
workpiece system, a set of thin-walled structure with a thickness of 2 mm is employed as a fixture to clamp the part, and finally a 2DOF
flexible workpiece system is built up. As indicated in Fig. 3(b), the workpiece feed direction is realized through controlling the C axis
turntable, and the cutter feed direction is realized through controlling the synchronous motion of X and Y axes by its feed drive system.
Finally, the required feed direction angle combination ψ (ψ c , ψ w ) can be easily obtained.
Before machining, the LMS dynamic characteristics measurement system is utilized to perform the hammering tests of the dynamic
stiffness on the tool tip and workpiece as seen in Fig. 3(c), and both the measured FRFs and its fitting results are shown in Fig. 4(a) and
(b), respectively. The identified modal parameters are listed in Table 1 below. By reference [33], the specific cutting force coefficients
are calibrated as kts = 849.5 N/mm2, krs = 388.3 N/mm2, ktp = 21.1 N/mm, krp = 8.5 N/mm, and the cutter runout parameters are
evaluated as ρ = 6.1 μm, λ = 26.7◦ , γ = 0.001◦ , η = 115.9◦
During machining, as seen in Fig. 3(a) and (d), a three-dimensional acceleration sensor (type: PCB356A66, sensitivity: 1.01mv/m/
s2) is employed to measure the vibration acceleration response of the workpiece system. Two eddy current sensors (type:
eddyNCDT3060-ES4, measure range: 0.4 ~ 4 mm, resolution: 0.08 μm, sensitivity: 0.5952 mm/V in X direction and 0.5914 mm/V in Y
direction) are applied to record the vibration displacement of the workpiece system. A set of dynamometer (type: Kistler 9265B) is
adopted to test the transient milling force. The above three kinds of data are collected and stored synchronously by the NI acquisition
card, and the sampling frequency is about 5.0 kHz. After machining, the three-coordinate measuring machine (type: global classic
SR575, resolution: 1.8 μm) was used to test the contour accuracy of the part.

3.2. Verification results of the milling stability

Fig. 5 shows the theoretical and experimental results of milling stability for the asymmetric flexible machining system. It can be
seen from Fig. 5(a) that the SLD will change significantly under different feed direction angles. Among them, the variation of the lobe
pattern along the left–right direction in the spindle speed dimension is relatively small, but along the up-down direction in the axial
depth of cut dimension is very sharp. When the cutter feed direction angle increases, the lobe pattern in the whole parameter domain
displays a first decreasing and then increasing trend. While, it seems that the workpiece feed direction angle only in the peak region of
the lobe pattern, that is, the resonance region of the forced vibration, has a great influence on the stability. The above results indicate
that the milling stability of the asymmetric flexible machining system has a remarkable feed direction-dependent feature.
Based on the above analytical results in Fig. 5(a), several groups of milling parameters are selected near the critical limit of the SLD
for milling experiments, and the chatter is judged by analyzing the vibration response, so as to test the correctness of the theoretical
model. Fig. 5(b)~(e) respectively show the vibration acceleration responses and its analysis results of the workpiece along the Xw
direction for the four cases of A, B, C and D in Fig. 5(a). Here, the amplitude-frequency spectrum of the time domain signal of vibration
acceleration is calculated by the Fast Fourier Transform. One can conclude that in the spectrum of case A and C, there are only the
spindle rotation frequency and its higher-order harmonics, while in case B and D, there are frequencies obviously different from the
spindle rotation frequencies, namely the chatter frequencies, which indicates that the unstable cutting occurs in this milling process.
These results of the four group experiments are well consistent with the theoretical prediction in Fig. 5(a). The above conclusion can
also be fully supported by the judgment of the machined surface morphology image in Fig. 5(b)~(e). In addition, the discrimination
results of several other groups of experiments are also shown in Fig. 5(a), which can also verify the milling stability under different feed
direction angles. For the algorithm performance comparison, one can also find in Fig. 5(a) that a more accurate milling stability with
feed direction angle can be achieved by the proposed method when compared with the traditional ZOA method by using the orientated
FRFs.

3.3. Verification results of the SLE

Fig. 6 shows the theoretical analysis and experimental verification results of the SLE in the asymmetric machining system under
stable milling conditions. Fig. 6(a) and (b) provide the contour lines of the SLE distribution in the stable milling region under three

9
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Fig. 5. Verification results of milling stability (ae = 2 mm, ft = 0.1 mm/tooth, down milling).

10
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Fig. 6. Verification results of SLE under stable milling (ae = 2 mm, ft = 0.1 mm/tooth, down milling).

groups of different feed direction angles. As we can see, there are existing multi-local minimum near the spindle speed and axial depth
of cut at (5.1krpm, 2.8 mm), (6.2krpm, 3.5 mm), and (8.2krpm, 4.0 mm), respectively, and the SLE has a sharp change near the spindle
speed of 8.5krpm. Moreover, due to the influence of the asymmetric flexibility, when the feed direction angle increases, the SLE
contour lines would shift to the direction of low spindle speed and large axial depth of cut. Similar to the milling stability, we can still
realize that the SLE also has a prominent feed direction-dependent feature.
Several groups of machining parameters are selected from Fig. 6(a) for milling experiment. Fig. 6(c) plots the measured vibration

11
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Fig. 7. Analysis results of the asymmetric milling dynamics characteristics (ae = 2 mm, ft = 0.1 mm/tooth, down milling).

12
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

displacement of the workpiece along its Xw and Yw directions, which is acquired by the eddy current sensor. It can be seen that the
simulated vibration displacements of workpiece are in good agreement with the measured values in amplitude and variation trend
shown in Fig. 6(d). In Fig. 6(e), the milling experiments of several groups of spindle speeds are carried out, and the simulated and
measured values of SLE under two groups of feed direction angle combinations are given. From the comparison, we know that although
the measured value has a certain degree of experiment error, it can accurately reflect the change of SLE with the spindle speed.
Similarly, another group of milling parameter are selected from Fig. 6(b) for further experiments, and the analysis results are
gathered in Fig. 6(f-g). Different from Fig. 6(d), the simulated values of two groups of different feed direction angles in Fig. 6(g) are
different. This is because in the workpiece coordinate system, the cutter feed direction angle will not affect the milling force and its
response for the workpiece, but only the workpiece feed direction angle will. Further selecting six groups of axial depth of cuts from
Fig. 6(b) for milling, the comparison results in Fig. 6(h) indicate that the theoretical model can also accurately predict the influence
trend of the axial depth of cut on the SLE.
The above theoretical and experimental results prove that the milling dynamics characteristics of the asymmetric flexible
machining system, including milling stability and SLE, are asymmetric. That is to say, the characteristics are variable under different
feed direction angles, and have a conspicuous feed direction-dependent feature.

4. Generation mechanism of the asymmetric milling dynamics characteristics

4.1. Further analysis on the asymmetric characteristics

In order to clarify the generation mechanism of the asymmetric milling dynamics characteristics for an asymmetric flexible
machining system, this section does some further study through numerical simulation. As shown in Fig. 7(a) and (b), the influence of
cutter and workpiece feed direction angles on the SLD have been separately analyzed, respectively. It can be seen that the cutter feed
direction angle will affect the milling stability in the full spindle speed range, while the workpiece feed direction angle has a greater
impact on the milling stability only in some local spindle speed ranges, such as at 6.1krpm and 8.2krpm, but has a smaller impact in
other regions.
Fig. 7(c) and (d) illustrate the variation of the minimum apclim and the apclim at a spindle speed of 8.0krpm in SLD under different feed
direction angle combinations, respectively. The apclim in the two figures presents a complex change. For more detail, the cutter feed
direction angle has a significant effect on the apclim , while the workpiece feed direction angle only affects the apclim in some local regions.

Fig. 8. Dynamic stiffness of the cutter-workpiece system in the local coordinate system.

13
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

On this basis, Fig. 7(e) plots the variation of apclim with feed direction angle in the polar coordinate. From this figure, the following
conclusions can be observed:

(1) The overall distribution of apclim displays as a typical saddle-shape diagram for the asymmetric flexible machining system.
(2) The saddle-shape diagram is centrosymmetric. That is to say, the change period of apclim is 180◦ . Therefore, when the feed di­
rection is completely reversed, the apclim is the same.
(3) Take the minimum apclim for example, there are two local maximum and two local minimum for the apclim within 180◦ . The
maximums of the apclim appear at about 60◦ and 150◦ , and the minimum ones appear at about 15◦ and 115◦ . The ratio of the
maximum to minimum is more than two times.
(4) The apclim at ψ c = 0◦ is slightly greater than that at ψ c = 90◦ . Moreover, the apclim at ψ c = 0◦ and 90◦ are both less than the
maximum values. This means that the most stable milling feed direction is not along the Xm or Ym axis.
(5) The overall distribution of apclim for the symmetric flexible machining system is a circle-shape. This is equivalent to saying the
feed direction cannot affect the milling dynamics for a symmetric flexible machining system.
(6) Compared with the two symmetric systems, the apclim of the asymmetric system would be greater at some feed direction angles.

Fig. 7 (f) and (g) further analyze the |SLE|max in the spindle speed range of 7.6 ~ 8.7krpm and the change of |SLE| at the speed of
7.6krpm under different feed direction angles during stable milling. For the |SLE|max , it is mainly affected by the cutter feed direction
angle, while for the |SLE| at the speed of 7.6krpm, it is mainly affected by the workpiece feed direction angle. Fig. 7(h) similarly shows
the saddle-shape diagram of |SLE|max . Its overall distribution is almost similar to that of the apclim , but it also has some differences. The
biggest discrepancy is that the angle positions of the maximum and minimum of the |SLE|max are almost opposite to that of the apclim .
That is, the larger the apclim , the smaller the |SLE|max , and vice versa. This is consistent with our conventional cognition. Because when
the apclim is large, the dynamic stiffness of the system will be high, thus the |SLE|max becomes small.
The above in-depth analysis shows us the relationship between milling stability, SLE and feed direction. Meanwhile, it gives us an
important revelation that one can select the reasonable feed direction to achieve high milling performance with higher apclim and lower
|SLE|max .

4.2. Explanation of the generation mechanism

The above analysis results indicate that the milling stability and SLE would be inevitably affected by the feed direction. This section
will explain its generation mechanism from the perspective of the dynamic stiffness in the local coordinate system.
As shown in Fig. 8(a), different from the model by the proposed method, we also establish the milling dynamics equation for the
milling system in the traditional feed coordinate system as follows.

Mf Ẍ f (t) + Cf Ẋ f (t) + Kf Xf (t) = F 0 (t) + ΔF ∼ (t) (21)

[ ] ( ) ( )
T
where Xf (t) = xcf , ycf , xwf , ywf .Mf = diag(mcfx , mcfy , mwfx , mwfy ).Cf = diag ccfx , ccfy , cwfx , cwfy .K f = diag kcfx , kcfy , kwfx , kwfy .
Here, the two milling force terms at the right of above formula are obtained by making Rotψ c = Rotψ w = I (I is the unit matrix) in Eq.
(13) and Eq. (14), respectively.
With the aid of the Laplace transform, the dynamic displacement response of the cutter and workpiece in the feed coordinate system
can be obtained by
⎡ ⎤
ϕcfx 0
⎢ ϕcfy ⎥{ 0 }
Xf (s) = ⎢
⎣
⎥ F (s) + ΔF ∼ (s)
⎦ (22)
ϕwfx
0 ϕwfy

where ϕpfq = 1/(mpfq s2 + cpfq s + kpfq )(p = c, w; q = x, y) are the FRFs of cutter and workpiece system in the feed coordinate system.
One can calculate these oriented FRFs by using the original FRFs in the machine coordinate system based on the decomposition and
synthesis principle of the force and displacement. The relationship can be expressed as follows.
⎧
⎪ ϕcfx = ϕcx cos2 ψ c + ϕcy sin2 ψ c
⎪
⎨ϕ 2 2
cfy = ϕcx sin ψ c + ϕcy cos ψ c
(23)
⎪
⎪ ϕ = ϕwx cos ψ w + ϕwy sin2 ψ w
2
⎩ wfx 2 2
ϕwfy = ϕwx sin ψ w + ϕwy cos ψ w

It can be seen from the above formula that the oriented FRFs in the feed coordinate system can be calculated by the nonlinear
combination of the original FRFs in the global coordinate system. In particular, when the flexibility in the global coordinate system is
asymmetric, the flexibility in the feed coordinate system will change with the feed direction angle. Only when the flexibility in the
global coordinate system is totally symmetric, this kind of change would disappear.
To verify this kind of change, we have carried out a number of tests on the cutter and workpiece system, and the theoretical and
measured results with feed direction angles of 25◦ , 45◦ and 65◦ are shown in Fig. 8(b). It can be found that the asymmetric flexibility of
the machining system, including the natural frequencies and amplitudes, will change with the feed direction angle. The synchronous

14
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

change of the simulated and measured value indeed verify the above theoretical derivation of the FRFs in the feed coordinate system.
Furthermore, as one of the frequency domain analysis methods for milling stability, the average cutter tooth angle approach [1,34]
simplifies a multi DOF machining system into a single DOF system along the normal direction of the cutter-workpiece engagement

Fig. 9. Influence of asymmetric stiffness on the asymmetric milling dynamics characteristics (ae = 2 mm, ft = 0.1 mm/tooth, down milling).

15
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

area, so as to use the frequency domain analytical method to calculate the SLD. Here, we apply this method to calculate the critical
limit of axial depth of cut, as shown in the following formula.
− 1
apclim = (24)
2kts Re (ϕn )Nt*

where ϕn is the oriented FRF along the normal direction of the cutter-workpiece engagement area. Re (ϕn ) is the real part of ϕn . Nt* is the
average cutter tooth number, which is equal to N*t = Nt (θe − θs )/360◦ .
It can be seen from the above equation that the real part of the FRF needs to be negative in order to make the apclim has a practical
significance. Moreover, we can also know that the apclim is proportional to the coefficient − 1/(Re (ϕn )Nt* ).
Here, the normal direction of the engagement area shown in Fig. 8(a) is defined by the average engagement angle, and its
calculation formula is as follows.
{
αn = (θs + θe )/2 (Up milling)
(25)
αn = π − (θs + θe )/2 (Down milling)
Similar to Eq. (23), we can further obtain the FRFs of the cutter and workpiece system along the normal direction by

Fig. 10. Influence of milling type and radial depth of cut on the asymmetric milling dynamics characteristics (ft = 0.1 mm/tooth, ψ w = 0◦ ).

16
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

{
ϕcn = ϕcx cos2 (ψ c − αn ) + ϕcy sin2 (ψ c − αn )
(Up milling)
ϕwn = ϕwx cos2 (ψ w − αn ) + ϕwy sin2 (ψ w − αn )
{ (26)
ϕcn = ϕcx cos2 (ψ c + αn ) + ϕcy sin2 (ψ c + αn )
(Down milling)
ϕwn = ϕwx cos2 (ψ w + αn ) + ϕwy sin2 (ψ w + αn )

It will be seen from this that the oriented FRFs along normal direction are not only related to the feed direction angle, but also to the
average engagement angle. Since the dynamic stiffness of the cutter and workpiece system at the cutting point is in series, the
comprehensive oriented FRFs of the cutter-workpiece system along the normal direction can be derived as follows. Finally, the apclim can
be calculated by substituting Eq. (27) into Eq. (24).
ϕn = ϕcn + ϕwn (27)

Based on the above derivation, Fig. 8(c) describes the variation of the coefficient − 1/(Re (ϕn )Nt* ) of the cutter and workpiece system
with the feed direction angles. Among them, the top figure is calculated by the dominant mode of the cutter system, and the bottom
figure is calculated by the dominant mode of the workpiece system. Compared with Fig. 7(c) and (d), we will be able to find out that
there is a significant positive correlation relationship between the apclim and the coefficient − 1/(Re (ϕn )N*t ). Moreover, due to the higher
flexibility of the cutter system, the minimum apclim in Fig. 7(c) will be mainly affected by the dominant mode of the cutter system. While
for the peak region of Fig. 7 (b), the chatter is difficult to occur here, thus it is mainly affected by the workpiece system with low
flexibility.
In addition, the SLE of milling process is fundamentally determined by the dynamic displacement response of cutter and workpiece
system in the normal feed direction, and the dynamic responses of the cutter and the workpiece system are mainly generated by the
milling force exciting on the FRF of ϕcfy and ϕwfy . Fig. 8(d) shows the maximum flexibility of the FRFs of cutter system (top) and the
workpiece system (bottom). We can still realize that there is also a significant positive correlation relationship between Fig. 8(d) and
Fig. 7(f ~ g). This is mainly because that the larger the flexibility along the normal direction of the machined surface is, the greater the
vibration response of the system would be, thus the greater the SLE has.
Through the above analysis of the system flexibility under the local coordinate system, we can successfully explain the generation
and evolution mechanism of milling stability and SLE in Fig. 7.

5. Discussions

In this section, the influences of the asymmetric stiffness, milling type and radial depth of cut on the asymmetric milling dynamics
characteristics have been studied through numerical simulation. In order to analyze the influence trend easily, we only take the
asymmetric stiffness of the cutter system as an example, where the masses and dampings are symmetric, and the flexibility of the
workpiece system is considered to be completely symmetric with ϕwfy = ϕwfx.

5.1. The influence of the asymmetric stiffness

Fig. 9(a) and (b) bring insight into the influence of different asymmetric stiffnesses on the milling stability and SLE, respectively. It
can be seen that the shape of lobe pattern in Fig. 9(a) and the distribution of SLE in
Fig. 9(b) are both changing saliently.
In Fig. 9(c) and (d), the saddle-shape diagrams of the minimum apclim and |SLE|max are both drawn, respectively. The following
several points can be summarized from the figures:

(1) The mirror property of the asymmetric stiffness will cause the apclim and |SLE|max to vary in 90◦ period. For example, when the
stiffness ratio between X and Y direction isrkcx/kcy = 0.9, the angle positions of the sharp saddle of the minimum apclim in Fig. 9(c)
are at 105◦ and 285◦ , and that of the smooth saddle of the minimum apclim are at 15◦ and 195◦ . While whenrkcy/kcx = 0.9, the angle
positions of the above sharp and smooth saddle would exchange.
(2) The asymmetric stiffness has a prominent local regulation effect on the minimum apclim and |SLE|max . In more detail, during
rkcx/kcy increases from 0.5 to 0.95, the saddle-shape diagram only within 330◦ ~60◦ and 150◦ ~240◦ has a dramatic change, and
maintains almost invariant in the range of 60◦ ~150◦ and 240◦ ~330◦ . This indicates that the stiffness kcx mainly plays a reg­
ulatory role in this local regions at 330◦ ~60◦ and 150◦ ~240◦ . Meanwhile, the main regulation region of the kcy is opposite to
that of the kcx. It needs to be supplemented that the regulation effect of asymmetric stiffness on the |SLE|max is just on the
contrary to theapclim .
(3) In the significant local regulation region, with the decrease of stiffness asymmetry degree, the minimum apclim and |SLE|max are
both closer to the symmetric stiffness system.
(4) In the non-significant local regulation region, the change of asymmetry stiffness hardly causes the variation of the local min­
imum of the apclim and |SLE|max .

Fig. 9(e) and (f) show the sensitivities of the minimum apclim and |SLE|max to the asymmetric stiffness, respectively. It can be
concluded that:

17
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

(1) Under different asymmetric stiffnesses and ψ c, the sensitivity will be different.
(2) With the decrease of stiffness asymmetry degree, the position of high sensitivity firstly appears in the significant local regulation
region, then appears in the non-significant region, and the value of maximum sensitivity keeps growing. This means that when
there is a slight asymmetry for a symmetric system, it will cause a dramatic change on the minimum apclim and |SLE|max in the
non-significant region, and a relatively gentle change in the significant region.

Fig. 9(g) and (h) summarize the variation of the global maximum value of minimum apclim and the global minimum value of |SLE|max
under different asymmetric stiffnesses. It can be seen that whenrkcx/kcy = 0.75, the global maximum value of the minimum apclim is
largest. Whenrkcx/kcy = 0.85 and 0.70, the minimum value of the |SLE|max is smallest. The results have an important guiding significance
for the active design for the dynamic characteristics of the machining system, especially for the static stiffness design of the spindle
box-spindle-cutter system and fixture-workpiece system.

5.2. The influence of the milling type and radial depth of cut

Fig. 10 analyzes the influence of milling type and radial depth of cut on the asymmetric milling dynamics characteristics of the
machining system. It can be seen from Fig. 10(a) and (b) that the distribution of milling SLD and SLE contour are completely different
under different milling types and radial depth of cuts. This can be consistent with our conventional cognitive for machining dynamics.
In Fig. 10(c), with the increase of the radial depth of cut, the saddle-shape diagram of minimum apclim gradually shrinks and deflects
clockwise. This deflection is mainly due to the change of αn and ϕn with the increase of the cutting width. Here, the changes of αn caused
by the four given radial depth of cuts are 8.1◦ ,12.7◦ ,11.5◦ and 12.7◦ , calculated by Eq. (25), which have a good agreement with the
values of 7.2◦ , 14.4◦ , 10.5◦ and 14.1◦ marked in the figure. Similarly, in Fig. 10(d), there is such a same deflection phenomenon on the
|SLE|max , and the deflection angles are almost the same as that in Fig. 10(c).
From Fig. 10(e), the milling types produce a mirror symmetry phenomenon on the minimum apclim , and the angle position of the
symmetry axis is about 32◦ 。The reason for this symmetry phenomenon is that the change trend of Φn along the normal direction in
both up and down milling are the same, and there only has a certain phase lag between them. According to the Eq. (26), we know that
the value of the phase lag angle is about 2αn. By Eq. (25), the values of the phase lags will be 36.8◦ , 78.5◦ and 126.9◦ when the radial
depth of cut are 1.6 mm, 6.4 mm and 12.8 mm, respectively. It’s very close to the results labeled 35.2◦ , 75.7◦ and 125.3◦ in the figure.
In Fig. 10(f), although the milling types cause an obvious change on the |SLE|max , the saddle-shape diagram also has the characteristics
of mirror symmetry in terms of the topological pattern.

6. Conclusions

In this paper, a generalized milling dynamics modeling method for an asymmetric flexible 4DOF cutter-workpiece system is
proposed, and the influence mechanism of the asymmetric flexibility on milling dynamics has been well revealed. The Experimental
results verify the validity of the proposed method. The main conclusions are as follows:

(1) The generalized regenerative milling dynamics model with a new dimension, namely feed direction angle combination ψ (ψ c,
ψ w), is established. In the model, the coupling coefficient matrix is defined to well handle the milling condition with non-parallel
direction between the axis directions of coordinate system, the vibration directions of system DOFs and the acting directions of
milling force. The proposed model not only has a better adaptability, but also can reduce the modeling workload when
compared with the traditional ones.
(2) It is found that the milling dynamics characteristics of an asymmetric machining system, including the milling stability and SLE,
have a significant feed direction-dependent feature, which are described by the saddle-shape diagrams of milling stability and
SLE versus the feed direction angles.
(3) The generation mechanism of the asymmetric milling dynamics characteristics has been explained successfully. On this basis,
the centrosymmetry property, extremum characteristics and local regulation effect are both analyzed.
(4) The influence of milling type and radial depth of cut on the asymmetric milling dynamics characteristics has been studied by
numerical simulation, in which the deflection and the mirror symmetry properties of the saddle-shape diagrams are pointed out.

CRediT authorship contribution statement

Xing Zhang: Methodology, Software, Validation, Writing original draft. Zhao Zhao: Experiment, Supervision. Kunhong Chen:
Investigation, Writing – review & editing. Wanhua Zhao: Conceptualization, Methodology, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

18
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

Acknowledgements

This work was financially supported the National Natural Science Foundation of China (No. 51905410, 52075426), and the Major
Science and Technology Project of Shaanxi Province (No. 2019zdzx01-01-02), the China Postdoctoral Science Foundation (No.
BX20180253, 219945), and the Fundamental Research Funds for the Central Universities (No. xzy012019009, xxj022019025).

Appendix A

The main steps for the derivation of semi-discretization method are given as follows.

(1) Discretizing time into series with an interval of Δt = T/Nm , where T is the cutter rotation period, Nm is the discrete number, then
the above governing equation in one time interval can be approximated to

U̇(t) = Ak U(t) + Bk U(t − τ) + F(t) (A.1)

1
∫ tk+1 1
∫ tk+1
where Ak = Δt tk A(t)dt, Bk = Δt tk B(t)dt.

(2) For the initial condition U(tk ) = Uk , the general solution of the above equation can be expressed as

U(t) = eAk (t− tk )

Uk + (eAk (t− tk )
− I)A−k 1 (Bk U(t − τ) + Fk ) (A.2)

(3) Interpolating linearly the time delay term by U(t − τ) ≈ 0.5Uk− τk +1 + 0.5Uk− τk , where τk = int(τ+0.5Δt
Δt ) with int( ) as the rounding
function, on can further obtain the Eq. (A.3) at time t = tk+1 .
Uk+1 = Pk Uk + 0.5Rk Uk− τk +1 + 0.5Rk Uk− τk + D k Fk (A.3)

where Pk = eAk Δt , Rk = (eAk Δt − I)A−k 1 Bk , Dk = (eAk Δt − I)A−k 1 .

(4) The above equation can be transformed into a discrete map as shown below.
V k+1 = Zk V k + Ek (A.4)
[ ]
T T
where V k = QTψ ,k Q̇ψ ,k , ⋯, Q̇ψ ,k− Nm
T
.

⎡ ⎤
Pk ⋯ 0.5Rk 0.5Rk ⋯ 0 0
⎢ I ⋯ 0 0 ⋯ 0 0⎥
⎢ ⎥
Zk = ⎢
⎢ 0 ⋯ I 0 ⋯ 0 0⎥⎥
⎣⋮ ⋱ ⋮ ⋮ ⋱ ⋮ ⋮⎦
0 0 0 0 0 I 0

Ek = [Dk Fk , 01×4Nm ]T

References

[1] T.L. Schmitz, K.S. Smith, Machining Dynamics, Springer US, 2009.
[2] Y. Altintas, G. Stepan, E. Budak, T. Schmitz, Z.M. Kilic, Chatter stability of machining operations, Transactions of the ASME, J. Manuf. Sci. Eng. 142 (11) (2020)
1–46.
[3] D. Montgomery, Y. Altintas, Mechanism of cutting force and surface generation in dynamic milling, Transactions of the ASME, J. Eng. Industry 113 (2) (1991)
160.
[4] K. Ringgaard, Y. Mohammadi, C. Merrild, O. Balling, K. Ahmadi, Optimization of material removal rate in milling of thin-walled structures using penalty cost
function, Int. J. Mach. Tools Manuf 145 (2019), 103430.
[5] M. Law, Y. Altintas, A.S. Phani, Rapid evaluation and optimization of machine tools with position-dependent stability, Int. J. Mach. Tools Manuf 68 (2013)
81–90.
[6] A. Comak, E. Budak, Modeling dynamics and stability of variable pitch and helix milling tools for development of a design method to maximize chatter stability,
Precis. Eng. 47 (2017) 459–468.
[7] Y. Altintaş, E. Budak, Analytical prediction of stability lobes in milling, CIRP 44 (1) (1995) 357–362.
[8] S.D. Merdol, Y. Altintas, Multi frequency solution of chatter stability for low immersion milling, Transactions of the ASME, J. Manuf. Sci. Eng. 126 (3) (2004)
459–466.

19
X. Zhang et al. Mechanical Systems and Signal Processing 169 (2022) 108750

[9] P.V. Bayly, M.A. Davies, J.E. Halley, J.R. Pratt, Stability analysis of interrupted cutting with finite time in the cut, ASME Internat. Mech. Eng. Congress
Exposition (2001).
[10] T. Insperger, G. Stépán, Semi-discretization method for delayed systems, Int. J. Numer. Meth. Eng. 55 (5) (2002) 503–518.
[11] Y.e. Ding, LiMin Zhu, XiaoJian Zhang, H. Ding, A full-discretization method for prediction of milling stability, Int. J. Mach. Tools Manuf 50 (5) (2010) 502–509.
[12] Y. Sun, S. Jiang, Predictive modeling of chatter stability considering force-induced deformation effect in milling thin-walled parts, Int. J. Mach. Tools Manuf.
135 (2018) 38–52.
[13] G. Totis, T. Insperger, M. Sortino, G. Stépán, Symmetry breaking in milling dynamics, Int. J. Mach. Tools Manuf 139 (2019) 37–59.
[14] J. Niu, J. Jia, R. Wang, J. Xu, Y. Sun, D. Guo, State dependent regenerative stability and surface location error in peripheral milling of thin-walled parts, Int. J.
Mech. Sci. 196 (2021), 106294.
[15] D. Montgomery, Y. Altintas, Mechanism of cutting force and surface generation in dynamic milling, Trans. ASME, J. Eng. Industry 113 (1991) 160–168.
[16] F. Ismail, M.A. Elbestawi, R. Du, K. Urbasik, Generation of milled surfaces including tool dynamics and wear, Trans. ASME, J. Eng. Industry 115 (1993)
245–252.
[17] T.L. Schmitz, B.P. Mann, Closed-form solutions for surface location error in milling, Int. J. Mach. Tools Manuf 46 (12-13) (2006) 1369–1377.
[18] T.L. Schmitz, J. Couey, E. Marsh, N. Mauntler, D. Hughes, Runout effects in milling: surface finish, surface location error, and stability, Int. J. Mach. Tools Manuf
47 (2007) 841–851.
[19] B. Denkena, M. Kruger, D. Bachrathy, G. Stepan, Model based reconstruction of milled surface topography from measured cutting forces, Int. J. Mach. Tools
Manuf 54–55 (2012) 25–33.
[20] A. Honeycutt, T.L. Schmitz, Surface location error and surface roughness for period-N milling bifurcations, Transactions of the ASME, J. Manuf. Sci. Eng. 139 (6)
(2017), 061010.
[21] X. Zhang, W. Zhang, J. Zhang, B.o. Pang, W. Zhao, Systematic study of the prediction methods for machined surface topography and form error during milling
process with flat-end cutter, Proc. Inst. Mech. Eng., Part B: J. Eng. Manuf. 233 (1) (2019) 226–242.
[22] J. Niu, J. Jia, Y. Sun, D. Guo, Generation mechanism and quality of milling surface profile for variable pitch tools considering runout, Trans. ASME J. Manuf. Sci.
Eng. 142 (12) (2020) 1–35.
[23] X. Zhang, J. Zhang, B.o. Pang, DiaoDiao Wu, XiaoWei Zheng, WanHua Zhao, An efficient approach for milling dynamics modeling and analysis with varying
time delay and cutter runout effect, Internat. J. Adv. Manuf. Technol. 87 (9-12) (2016) 3373–3388.
[24] A. Comak, O. Ozsahin, Y. Altintas, Stability of milling operations with asymmetric cutter dynamics in rotating coordinates, J. Manuf. Sci. Eng. 138 (8) (2016),
https://doi.org/10.1115/1.4032585.
[25] A. Honeycutt, T. Schmitz, Milling bifurcations for strongly asymmetric, symmetric, and weakly asymmetric system dynamics, Precis. Eng. 55 (2019) 1–13.
[26] M. Gaur, M. Law, Pocket milling strategies using combined-mode and feed-direction-dependent stability criteria, Proc. CIRP 82 (2019) 261–266.
[27] D. Li, H. Cao, J. Liu, X. Zhang, X. Chen, Milling chatter control based on asymmetric stiffness, Internat. J. Machine Tools Manuf. 147 (2019) 103458.
[28] J. Baumann, A. Wirtz, T. Siebrecht, D. Biermann, Disturbance of the regenerative effect by use of milling tools modified with asymmetric dynamic properties,
J. Manuf. Mater. Process. 4 (2020) 67.
[29] S. Engin, Y. Altintas, Mechanics and dynamics of general milling cutters: part I: helical end mills, Int. J. Mach. Tools Manuf 41 (15) (2001) 2195–2212.
[30] K.A. Desai, P.K. Agarwal, P.V.M. Rao, Process geometry modeling with cutter runout for milling of curved surfaces, Int. J. Mach. Tools Manuf 49 (2009)
1015–1028.
[31] J. Li, Z.M. Kilic, Y. Altintas, General cutting dynamics model for five-axis ball-end milling operations, Trans. ASME J. Manuf. Sci. Eng. 142 (12) (2020) 1–25.
[32] X. Zhang, J. Zhang, W. Zhang, J. Li, W. Zhao, A non-contact calibration method for cutter runout with spindle speed dependent effect and analysis of its
influence on milling process, Precis. Eng. 51 (2018) 280–290.
[33] X. Zhang, W. Zhang, J. Zhang, B. Pang, W. Zhao, General modeling and calibration method for cutting force prediction with flat-end cutter, Transactions of the
ASME, J. Manuf. Sci. Eng. 140 (2) (2018), 021007.
[34] J. Tlusty, W. Zaton, F. Ismail, Stability lobes in milling, Ann. CIRP 32 (1) (1983) 309–313.

20