You are on page 1of 10

Precision Engineering 56 (2019) 293–302

Contents lists available at ScienceDirect

Precision Engineering
journal homepage: www.elsevier.com/locate/precision

Configuration design and accuracy analysis of special grinding machine for T


thin-walled small concave surfaces☆
Tingzhang Wanga, Chunya Wua, Henan Liua, Mingjun Chena,∗, Jian Chenga,∗∗, Zhen Fangb, Bo Yub
a
Center for Precision Engineering, Harbin Institute of Technology, Harbin 150001, China
b
China Electronics Technology Group Corporation No.26 Research Institute, Chongqing 400060, China

ARTICLE INFO ABSTRACT

Keywords: In this work, a special grinding machine tool using a small ball-end grinding wheel is designed to achieve ultra-
Multi-axis machine design precision grinding of complex components with small concave surfaces. A comprehensive error propagation
Error propagation model model is established, considering kinematic errors, setup errors and structure parameters. Sensitivity analysis is
Sensitivity analysis performed by the Sobol' method to quantify the relationship between the errors, structure parameters and
Structure parameter design
geometric accuracy of the machine tool. On this basis, error allocation is carried out, and the inclination angle of
Error allocation
the grinding wheel spindle and the working range of the C-axis are optimized to reduce the influence of the setup
errors on machining accuracy. Then, the predicted geometric accuracy of the machine is 0.3 μm, contributing
only to half of the profile accuracy requirement for peak-to-valley PV < 0.6 μm. The developed machine is
tested by grinding ψ-shaped components to achieve an average profile accuracy of 0.589 μm and a surface
roughness of 40.2 nm, which means that the machine tool developed by the error model and sensitivity analysis
meets the design requirements. By compensating the setup errors and linear errors based on the error propa-
gation model, the profile accuracy of the ground component reaches 0.34 μm, indicating that the error propa-
gation model can accurately represent the geometric accuracy of the machine tool.

1. Introduction sensitivity analysis are adopted to assist the design. FE analysis is a


widely used aided design method, which can shorten development time
With the rapid development of advanced finishing methods, the and help optimize the configuration of machine tools to a certain extent
structure of optical components required by the aerospace, optical and [2,3]. Liang et al. [4] established an FE model for a fly-cutting machine
electronics industries becomes increasingly complex, which places tool considering joint parameters, on which the static, modal, harmonic
higher demands on processing technology. A ψ-shaped thin-walled and rotor dynamic analyses were carried out. Li et al. [5] proposed a
complex structural component made of hard and brittle materials is one method to optimize an ultra-precision micro machine tool to improve
of the typical components with small concave surfaces. Although the ψ- the dynamic performance based on FE analysis. Yang et al. [6] used FE
shaped component composed of a central pole and a thin-walled analysis to help with the design and dynamic performance analysis of
spherical shell is difficult to machine, ultra-precision grinding has an ultra-precision ion beam figuring machine.
proven to be a promising option for processing it [1]. In conventional Geometric accuracy of machine tools is also a core problem, which
grinding, small concave surface grinding is a challenge and interference is affected by kinematic errors and setup errors. The setup errors have
between the grinding wheel and the component is unavoidable. been studied as independent factors by some researchers, without
Therefore, there is a need for a special grinding machine tool with a considering other factors. Xiao et al. [7] investigated the effects of some
small ball-end grinding wheel to meet the processing requirements of setup errors on the geometric accuracy of a 5-axis machine, with a lack
the component. of any setup error model. The error model of multi-axis machine tools is
In the design phase of machine tools, it is necessary to compre- regarded as the basis for two main methods to improve the accuracy of
hensively consider function, configuration, kinematic errors and setup machine tools, namely, precision design and error compensation [8]. A
errors. Therefore, finite element (FE) analysis, error models and series of distinct and valuable modeling methods have been proposed


This paper was recommended by Associate Editor Libo Zhou.

Corresponding author. P.O. Box 413, Harbin 150001, China.
∗∗
Corresponding author.
E-mail addresses: chenmj@hit.edu.cn (M. Chen), cheng.826@hit.edu.cn (J. Cheng).

https://doi.org/10.1016/j.precisioneng.2018.12.009
Received 19 September 2018; Received in revised form 26 December 2018; Accepted 28 December 2018
Available online 31 December 2018
0141-6359/ © 2018 Elsevier Inc. All rights reserved.
T. Wang et al. Precision Engineering 56 (2019) 293–302

and developed, such as homogeneous transformation matrices (HTM)


[9,10], differential transform theory [11], rigid body kinematic method
[12], matrix summation method [13], Denavit-Hartenberg (D-H) [14],
stream of variation [15], multi-body system (MBS) [16] and screw
theory [17,18]. MBS was proposed by Houston [19] and has been
widely utilized to model, identify and compensate the geometric errors
of multi-axis machine tools [20]. However, in this method, it is ne-
cessary to assign a local coordinate system to each axis, and kinematics
inverse solutions are difficult to be obtained. In contrast, screw theory is
a zero-reference position method that is applied widely in the field of
robotics rather than machine tools. Until 2001, Moon et al. [21]
adopted the screw theory to establish a kinematics model of a machine
tool, taking each error as a twist relative to the global reference frame.
Since it is easy to obtain the kinematic inverse solutions based on the Fig. 1. ψ-shaped thin-walled complex structural component.
screw theory, which is of great significance for error compensation,
Xiang et al. [22] proposed a method to model and compensate both 2.1. Processing characteristics analysis and realization
geometrically dependent and independent volumetric errors of machine
tools by the screw theory. However, an integrated error model, con- There are two main problems needed to be resolved in the machine
sidering kinematic errors, setup errors and the effects of structure tool design, i.e., the grinding of the small concave surfaces and the
parameters simultaneously, has never been reported before. interference between the grinding wheel and the workpiece. The small
Sensitivity analysis is a promising approach to identify the vital errors in concave surface grinding can be realized by small ball-end diamond
precision design and error compensation, which can be classified into global grinding wheels. The grinding wheel spindle should be inclined at a
sensitivity analysis and local sensitivity analysis by definition [23,24]. Since certain angle to the X-Y plane in the workpiece coordinate frame and
local sensitivity analysis can only inspect one point at a time, global sen- rotate freely around the Z-axis to avoid the interference between the
sitivity analysis, such as differential vector method [25], Fourier amplitude grinding wheel and the workpiece. Therefore, the grinding wheel
sensitivity test [26] and Monte Carlo digital simulation [27], seems to be spindle needs to be mounted on a turntable (hereafter referred to as C
more suitable for identifying the critical geometric errors. Cheng et al. [28] turntable). The angular displacement of the C turntable is designed to
proposed a method for analyzing the global sensitivity of a machine tool vary with the grinding position, so that the direction of the grinding
based on the Morris method. Miro et al. [29] used the elementary effect wheel can be adaptively adjusted to avoid the interference. Since the
sensitivity analysis method modified from the Morris method to investigate diameter of the ball-end diamond grinding wheel is only 3.8 mm, a
key subsoil parameters. Lagerwall et al. [30] introduced the Sobol' method high-speed motorized spindle should be the best choice for increasing
to realize the global sensitivity analysis of a spatially distributed ecological the peripheral velocity of the grinding wheel to improve the grinding
model. Sobol' method is a variance-based method, which combines para- quality. However, as observed in previous experiments, the wear of the
meter interaction and relatively straightforward interpretation, making it small ball-end grinding wheel is quite severe, which will significantly
very popular in many fields. Sobol' method can consider different types of affect the processing quality. Therefore, an on-machine truing device
errors simultaneously, but it has less application in error sensitivity analysis for the ball-end diamond grinding wheel is designed to keep the
of machine tools. grinding ability and profile accuracy of the grinding wheel the same. By
Henceforth, this paper is organized to present the design, error taking the movement in tool setting and grinding process in con-
modeling, structure parameter analysis and error allocation methods to sideration, a four-axis motion scheme is finally adopted.
guarantee the machining accuracy of the special grinding machine tool.
The configuration and structure of the machine tool are designed in
Section 2. In Section 3, based on the screw theory, an error model is 2.2. Configuration and structure design
established, considering kinematic errors, setup errors and the effects of
structure parameters. Section 4 details the sensitivity analysis of errors Configuration design is an important part of machine tool design.
and structure parameters based on the Sobol' method, error allocation Two configurations, i.e., gantry structure and T-shaped structure, are
and the optimization of the working range of the C-axis and structure considered in the initial design phase. FE analysis is performed by
parameters. Grinding experiments are performed in Section 5 to verify ANSYS software to assist the configuration design, with simulating the
the performance of the designed machine and accuracy of the proposed linear guide joints and ball screw joints by spring damping elements.
error model. At last, the paper is concluded in Section 6. The harmonic responses of the two configurations under the same static
stiffness are analyzed, and the results are compared in Fig. 2. It can be
2. Design of function and configuration for the grinding machine found that when the frequency is within 1000 Hz, there are two for-
mants for the gantry structure and three for the T-shaped structure,
Fig. 1 displays the ψ-shaped complex structural component con- respectively. The amplitude of the first formant of the T-shaped struc-
sisting of a central rod and a thin-walled spherical shell, the material of ture is larger than that of the gantry structure, and the first natural
which is fused silica. The minimum curvature radius is less than 2 mm, frequency of the gantry structure is higher than the one of the T-shaped
the sphere diameter is about 30 mm, and the wall thickness is less than structure. The structure loop of the gantry structure is closed, whereas
0.9 mm. Ultra-precision grinding and polishing are employed, con- the T-shaped structure remains open, which indicates that the thermal
sidering the specifications of the component that the surface roughness deformation of the gantry structure is somewhat less than that of the T-
(Ra) is less than 5 nm and the profile accuracy (PV) is less than 0.6 μm. shaped structure. Comprehensively considering the above analysis and
To reduce the polishing time, in ultra-precision grinding, the achieved available operating space, the gantry structure is selected as the con-
surface roughness and profile accuracy should be less than 50 nm and figuration of the designed machine.
0.6 μm, respectively. If conventional grinding is adopted to process such As shown in Fig. 2 a), the workpiece spindle device and the on-
a typical thin-walled complex structural component, it is difficult to machine electric discharge truing device are fixed alongside on the
avoid interference between the grinding wheel and the component, so work table laid on the X-Y axis, and the Z-axis is mounted on the gantry
that the small concave surfaces cannot be successfully ground. For this structure with the C turntable hung on it. The grinding wheel spindle is
reason, it is necessary to design a special grinding machine tool. installed on the C turntable at a certain angle of inclination. If the

294
T. Wang et al. Precision Engineering 56 (2019) 293–302

Fig. 4. Relative positions of different axes in the reference coordinate.

(defined as A-axis) is used to represent the inclination angle of the


grinding wheel spindle, and W represents the workpiece spindle.
The distance between each axis and the reference coordinate is
defined to study the effects of the structure parameters on the ma-
chining accuracy, as shown in Fig. 4. The center of the component is set
to be the reference coordinate origin. dx, dy, dz, dc, and da represent the
distances of the X, Y, Z, C and A axes from the reference coordinate in
the Z direction. Lw and La denote the distances of the W and A axes from
the reference coordinate in the Y direction. Lz represents the distance
between the Z-axis and the reference coordinate in the X direction. l is
the distance between the center of the ball-end grinding wheel and the
Fig. 2. Comparing two kinds of configurations. A-axis.
For the designed ultra-precision grinding machine, there are six
kinematic errors for each axis, including three angular errors and three
linear errors. Moreover, setup errors are inevitable in the assembly
process. Therefore, a total of 24 kinematic errors and 9 setup errors are
taken into account, as listed in Table 1. The errors can be classified into
linear error, angular error, squareness error and parallelism error ac-
cording to their respective characteristics. δmn and εmn stand for the
linear error and the angular error, respectively, with the first subscript
denoting the motion direction and the second one representing the
error direction. The setup errors mainly include the parallelism error
(Pmn) and the squareness error (Smn), and the parallelism error can be
regarded as two squareness errors. As for Smn and Pmn, the first sub-
script is the motion direction and the second subscript is the reference
direction.

3.1. Kinematics model

Since each type of movement of a rigid body can be expressed as a


kinematic screw, a kinematics model describing the motion and errors
of machine tools can be established by the screw theory. In screw
theory, reference coordination is necessary to describe all motions of
the rigid body [31]. The relative position of the nth joint to the re-
ference coordinate can be expressed by a 4 × 4 homogeneous trans-
formation matrix
Fig. 3. Diagrammatic sketch of grinding wheel spindle installation.
ˆ 1 ˆ 2 ˆ
grn ( 1, 2 n) =e 1 e 2 e n n [grn (0)] (1)
center of the ball-end grinding wheel does not coincide with the axis of
the C turntable, when it rotates, the spatial position of the ball-end where grn (0) represents the rigid body transformation between the nth
grinding wheel will change, as shown in Fig. 3. Under this circum- joint and the reference coordinate; ˆ i (i = 1,2, n) is a twist, and θi is
stance, it will undoubtedly bring challenges to the grinding trajectory the motion command, kinematic or setup error of the ith joint. For a
planning. Two fine-tuning positioners are designed to adjust the center revolute joint, the twist ξi can be defined as
of the grinding wheel towards the axis of the C turntable before vt
grinding for eliminating this effect. = , vi = × qi
i t
i
(2)

where i = [ 1i , 1i , 1i ]T R3 is a unit vector in the positive direc-


3. Comprehensive error modeling based on screw theory tion of the rotary axis; qi = [q1i , q2i , q3i] R3 is an arbitrary point on the
axis, which can describe the relative position between the ith twist and
Based on the screw theory, a comprehensive error model of the the reference coordinate; ωi (i = 1,2…n) and qi (i = 1,2…n) are spe-
proposed machine tool is established, by taking kinematic errors and cified relative to the reference coordinate and determine the full kine-
setup errors as movements in the error directions. Due to the installa- matic characteristics. qi helps to add the distances between the joints
tion characteristics of the grinding wheel spindle and the workpiece and the reference coordinate into the error model, which has never
spindle, the rotation of the grinding wheel spindle in the X direction been considered by other researchers in the error modeling process. The

295
T. Wang et al. Precision Engineering 56 (2019) 293–302

Table 1
Error components of the designed machine.
Errors Number of errors Symbols

Linear errors 12 δxx, δxy, δxz, δyx, δyy, δyz, δzx, δzy, δzz, δcx, δcy, δcz
Angular errors 12 εxx, εxy, εxz, εyx, εyy, εyz, εzx, εzy, εzz, εcx, εcy, εcz
Setup errors 9 Syx, Szx, Szy, PAY, PAZ, PCX, PCY, PWY, PWZ

transformation matrix of joint i is The homogeneous transformation matrix of the grinding wheel re-
T
lative to the reference coordinate is as follow:
ˆ e ˆ i i (I3 × 3 e ˆ i i) ( i × vi ) + i i vi i
ei i= ˆ Z ˆ ˆ
01 × 3 1 (3) grt (z , C, A) =eZ e C C e A A [g
rt (0)] (8)

where where θA is the installation angle of the grinding wheel spindle, which
is the measurement of the rotary motion around the X-axis; z and θC
e ˆi i = I3 × 3 + ˆ i sin i + ˆ i2 (1 cos i ) (4) represent the motion commands for the Z and C axes, respectively.
where ˆ i so(3) is defined as a skew-symmetric matrix of ωi and can be Since the homogeneous transformation matrix of the workpiece relative
given by to the reference coordinate is expressed by
ˆ x ˆ y ˆ
0 3i 2i grw (x , y, w) =e x e y e w w [g
rw (0)] (9)
ˆi = 3i 0 1i
where W represents the workpiece spindle; θw is the installation angle
2i 1i 0 (5)
of the workpiece spindle; x and y denote the motion commands for the
For a prismatic joint X and Y axes, respectively; the homogeneous transformation matrix of
the reference coordinate with respect to the workpiece can be obtained
vi I3 × 3 i vi
= , ei i=
i 0 0 1 (6) gwr (x , y, w) = [grw (0)] 1e
ˆ
w w e
ˆ y
y e
ˆ x
x (10)
Combining Eqs. (8) and (10), the homogeneous transformation
3.2. Comprehensive error model matrix of the grinding wheel relative to the workpiece can be de-
termined as
According to the structure of the designed machine and the screw ˆ ˆ y ˆ x ˆ z ˆ ˆ
theory, the kinematic chains are established, as shown in Fig. 5. grw (0) gwt (x , y, z , w, C, A ) = [gbw (0)] 1e w w e y e x e z e C C e A A [gbt (0)]
and grt (0) refer to the workpiece-to-reference coordinate and the (11)
grinding wheel-to-reference coordinate rigid body transformation ma- According to Eq. (11), if the kinematic errors and setup errors are
trix, respectively. As shown in Fig. 4, the vector between the workpiece not considered, the ideal kinematics model can be expressed as
and the reference coordinate is denoted by w0 = [0,0,0]T, and the po-
ˆ
sition of the grinding wheel is represented by t0 = [0,-La,da-l]T.
ˆ ˆ ˆ ˆ ˆ
gwt ideal = [gbw (0)] 1 e wi wi e yi yi e xi xi e zi zi e Ci Ci e Ai Ai [gbt (0)]
Therefore, the rigid body transformation matrices are expressed as (12)
I w0 I t However, the kinematic errors and setup errors are actually in-
gwt (0) = 3 × 3 , gbt (0) = 3 × 3 0
01 × 3 1 01 × 3 1 (7) evitable in the manufacturing process, so the actual kinematics model
of the designed machine is established
ˆ ˆ ˆ ˆ ˆ ˆ
gwt actual = [g bw (0)] 1 e wa wa e ya ya e x a xa e za za e Ca Ca e Aa Aa [g
bt (0)]

(13)
ˆ
The actual motion e na na of the nth joint can be expressed as follow

e
ˆ
na na =e
ˆ
ne ne e
ˆ
nie nie (14)
ˆ ˆ
where e ne ne and e nie nie describe the actual effect of the kinematic
errors and setup errors on the motion of the nth joint, respectively. To
establish the error model of the machine tool, it is essential to first
ˆ ˆ
calculate e ne ne and e nie nie of each axis. Here, the strategy for kine-
matic error modeling of the Z -axis and setup error modeling of the C-
axis is taken as an example.
Six kinematic errors present on the Z-axis can be taken as six
ˆ
movements of the Z-axis, so e ne ne can be expressed as

e
ˆ
ze ze =e
ˆ
zx zx e
ˆ
zx
ˆ
zx e zy zy e
ˆ ˆ ˆ
zy zy e zz zz e zz zz (15)
T T
As for εzx, since qεzx = [0,0,dz] , ωεzx = [1,0,0] (shown in Fig. 4),
ˆ
e zx zx is obtained:
1 0 0 0
ˆ 0 cos zx sin zx d z sin zx
e zx zx =
0 sin zx cos zx d z (1 cos zx )
0 0 0 1 (16)
Note that the structure parameter dz has been included in the error
Fig. 5. Kinematic chains of the designed ultra-precision grinding machine tool. model. Since the errors are extremely small, high-order minuteness can

296
T. Wang et al. Precision Engineering 56 (2019) 293–302

Ey = l sin c [PAY sin A PAZ (1 cos A)] dc PCX sin c + d c PCY (1 cos c ) + cx sin c
dc cy sin c + cy cos c + d c cx cos c + Lz zz + zy + dz zx z Szx d z Szy + d z Szx
+ z( xx + yx + PWX ) + x ( xz + yz + PWZ ) yy + dy yx + dx xx xy

(22)

Ez = cz + zz Lz zy Lz Szx x( xy + yy ) + y( yx + PWX ) yz + Lw wx xz

(23)
The error model established in this section has great significance for
error allocation, error compensation and the effects of structure para-
meters on machining accuracy analysis.

4. Accuracy analysis for the specialized grinding machine

Since the machine tool is designed for grinding thin-walled com-


ponents with small concave surfaces, it is necessary to analyze the ef-
fects of the kinematic errors and setup errors on the machining accu-
racy to guarantee the profile accuracy of the components. The shape of
the component is hemispherical with a radius (r) of 16 mm. In grinding,
Z direction is not the sensitive direction of error, so the effects of Ez and
Fig. 6. Schematic of the setup errors of C-axis. the linear errors in the Z direction are negligible. The normal direction
of the component surface at any grinding point is
ˆ
be ignored using the small-angle approximation, then e ze ze can be x y
r = [ , , 0]
simplified to r r (24)

1 zz zy zx d z zy Since Z-axis always keeps in the same position during the grinding
ˆ zz 1 zx zy + Lz zz+ dz zx process, the kinematic errors of the Z-axis can be ignored, thus the
e ze ze =
zy zx 1 zz Lz zy
composite error in the normal direction of the component surface can
0 0 0 1 (17) be expressed as
En = [Ex , Ex , 0]T r
In the actual assembly process, the axis of the C turntable cannot be
= [ x xx y xy + ydx xx xdx xy + xy xz x yx y yy + ydy yx xdy yy xL w PWZ
perfectly parallel to the Z-axis, so the setup errors of the C turntable is + cx (x cos c + y sin c ) + cy ( x sin c + y cos c ) + cx d c ( x sin c + y cos c ) xd z Szx
illustrated in Fig. 6. The parallelism errors represented by squareness cy dc (x cos c + y sin c ) lPAY sin A (x cos c + y sin c ) PCX d c (x x cos c y sin c )
errors (e.g., PCX and PCY in X and Y directions) can change the actual + lPAZ (x cos c + y sin c )(1 cos A) + PCY d c (y + x sin c y cos c ) + ydz Szy + xySxy]/r
motion direction of the C-axis, which can be analyzed by the screw
(25)
theory. As a result, the parallelism errors change the ideal direction of
the C-axis (ωc = [0,0,1]) into ωca as follows
sin PCX 4.1. Effect analysis of structure parameter
ˆ ˆ
=e PCY PCY e PCX PCX = sin PCY cos PCX
ca c
Equation (25) indicates that the structure parameters contribute to
cos PCX cos PCY (18)
the geometric error of the machine tool, so it is necessary to analyze the
As for qca = [0,0,dc]T, the actual kinematics model of the C-axis actual effects in detail. Here the Sobol' method [32,33] and Latin Hy-
affected by the parallelism errors can be modeled as: percube Sampling are adopted to calculate the sensitivity of the struc-
Since the above modeling method can be used to calculate the ki- ture parameters. The error model can be described as

cos c sin c PCY sin c + PCX (1 cos c ) dc PCX (1 cos c ) + dc PCY sin c
ˆ sinc cos c PCY sin c PCX (1 cos c ) dc PCX sin c + dc PCY (1 cos c )
e Cie Cie =
PCY sin c + PCX (1 cos c ) PCX sin c PCX (1 cos c ) 1 0
0 0 0 1 (19)

nematic errors and setup errors of the other axes to establish the actual
f (X) = f (x1, , xk ) (26)
kinematics model (represented by gwt.acutal), the deviation of the
grinding wheel can be expressed by the following formula: where X = (x1, ,xk ) is the input parameter set. The function f (X) can
be decomposed into summands of increasing dimensionality, and f0 is a
[ Ex E y Ez 1]T = (gwt actual gwt ideal ) [0 0 0 1]T (20) constant.
k k k
Ex, Ey and Ez can be obtained by combining Eqs. (11), (12) and (20)
f (x1, xk ) = f0 + fi (x i ) + fij (x i , xj ) + + f1, , k (x1, xk )
together. With adjusting the center of the grinding wheel to the axis of i=1 i=1 j=i+1
the C turntable (dB = l·cosA and LB = l·sinA), Ex, Ey and Ez can be fur-
(27)
ther simplified as follows:

Ex = l cos c [PAY sin A PAZ (1 cos A)] + d c PCX (1 cos c ) + d c PCY sin c + cx cos c f0 = k f (X ) d X (28)
d c cy cos c cy sin c d c cx sin c + zx dz zy + z Szx d z Szx z ( xy + yy )
+ y Sxy y( yz + PWZ ) yx dy yy Lw wz dx xy xx where stands for a k-dimensional unit hyperspace. The total and
k

partial variance of function f (X) can be obtained using Eqs. (29) and
(21)

297
T. Wang et al. Precision Engineering 56 (2019) 293–302

(30), respectively. The total unconditional variance can be obtained by


Eqs. (31) and (32), using the Monte Carlo integrals.

D= k f 2 (X) dX f02 (29)


1 1
Di1, is =
0 0
fi12, i2, is (xi , xi
1 2 x is ) dx i1 dx is (30)
n
1
fˆ0 = f (X m)
n m =1 (31)
n
1 2
Dˆ = f 2 (Xm) fˆ0
n m=1 (32)
where Xm is a sampled set of input parameters and n is the number of
samples. Hence, the partial variance can be obtained
n
1 2
Dˆ i = f (X (p i) m, x im
p
) f (X (q i ) m,
p
x im ) fˆ0
n m=1 (33)
where the superscript p and q represent two sampled sets of input
parameters. In the first function f (X) , p set is brought in. q set except x i Fig. 8. Average sensitive coefficient of structure parameters.
is considered in the second function, and x i is still the one in p set.
Therefore, the effects of all parameters except x i can be given by which can consider the linear and angular errors simultaneously. When
n the proposed structure parameters are incorporated into the error
1 2
Dˆ i = f (X q( i ) m,
q
xim ) f (X q( i ) m,
p
x im ) fˆ0 model represented by Eq. (25), the calculation is carried out under the
n m =1 (34) assumption that the linear errors remain within 1 μm and the angular
Finally, the total sensitivity of parameter x i can be obtained by errors are within 1 μrad. The average sensitivity coefficient for different
errors is presented in Fig. 9, where the sensitivity coefficient of the
Dˆ i same error varies with different displacements in the X, Y and C di-
TSi = 1
Dˆ (35) rections. Szy, Szx, PCY, PCX and εxx vary dramatically with the dis-
In the calculation, it can be found that for a single structure para- placements in the X, Y and C directions. When a movement in the X or Y
meter, the sensitivity will be different when the displacement of the X, direction occurs, εxy varies evidently, while εcx and εcy show a marked
Y or C axis changes, as shown in Fig. 7. The sensitivity of dz varies variation with the rotation of the C-axis. It is obvious that the sensitivity
dramatically with X/Y displacement, followed by dc and Lw. Never- coefficients of Szy, PCY, Szx, εxx, εcx, εcy, εxy and PCX are larger than the
theless, the variation of dc versus the displacement of the C-axis comes sensitivity coefficients of the other errors, which deserve more atten-
the first, dz is the second, and Lw is the third. The sensitivity of dy and dx tion.
shows a similar trend of variation, demonstrating their equal con- When the displacement of the corresponding axis is different, the
tribution to the effects on machining accuracy. Hence, the average value of kinematic error will change, but setup errors will never vary
value of the sensitivity in different positions is used to represent the with the movement of any axis. Therefore, errors can be divided into
sensitivity coefficient of the structure parameter, as illustrated in Fig. 8. two groups, i.e., position-dependent errors and position-independent
It can be concluded that the effect of a series of structure parameters on errors, which are respectively represented by En-dependent and En-in-
machining accuracy is in the order of dz > dc > Lw > dy ≈ dz > l. dependent in the error model. When implementing error allocation, due to

Therefore, in the design stage, it is necessary to pay close attention to their distinct characteristics, various types of errors should be studied in
parameters dz, dc and Lw to reduce the effects on machining accuracy. different methods.
The values of the structure parameters with larger sensitivity coefficient En dependent =[ x xx y xy + yd x xx xdy xy + xy xz x yx y yy + ydy yx xdy yy
are reduced to the utmost by considering the sensitivity and function of ( cx cy d c )(x cos c + y sin c )+ ( cy + cx d c )( x sin c + y cos c )]/r
the machine tool. Table 2 summarizes the optimization results. (36)

4.2. Analysis and allocation of errors for the designed machine En independent = [ xd z Szx + ydz Szy lPAY sin A (x cos c + y sin c )
+ lPAZ (x cos c + y sin c )(1 cos A)
Sensitivity analysis of kinematic errors and setup errors is of great
+ PCX dc (x x cos c y sin c ) + PCY dc (y + x sin c y cos c )
significance for error allocation in the design stage and error compen-
sation in the machining process. The Sobol' method and Latin xL w PWZ + xySxy]/r
Hypercube Sampling are appropriate to analyze the sensitivity of errors, (37)

4.2.1. Allocation of position-dependent errors


A position-dependent error model is established as described by Eq.
(36), which can also be expressed by a function g(X), where X is a
vector of position-dependent error items. The maximum value of the
error can occur at arbitrary position of the corresponding axis, which

Table 2
Optimization results of structure parameters.
Parameter dz dc Lw dy dx l
Fig. 7. Sensitive coefficient of structure parameters varies with X/Y and C
Values (mm) 372 232 90 149 249 61
displacement.

298
T. Wang et al. Precision Engineering 56 (2019) 293–302

Table 3
Position-dependent error values.
Error item Error values

Linear error δxx δxy δyx δyy δcx δcy


(μm/16 mm) 0.05 0.1 0.1 0.05 0.1μ/90° 0.1μ/90°
Angular error εxx εxy εxz εyx εyy εcx εcy
(μ rad/16 mm) 0.2 0.2 0.5 0.2 0.2 0.2μ rad/90° 0.2μ rad/90°

Table 4
Allocation of position-independent errors.
Error items PWZ PAZ PAY PCX PCY Sxy Szx Szy
Values (μ rand) −1.5 1.5 −1.5 0.5 0.3 1.5 0.5 −0.3
Fig. 9. Average sensitive coefficient of different error terms.

means that the probability of different errors reaching the maximum at axis and the machining accuracy. When the installation angle of the
the same position is extremely low. Therefore, the root sum square grinding wheel spindle is set to 40° and the angular displacement of the
(RRS) error analysis equation is adopted [34,35], and the function is C-axis changes from 0° to 360°, the machining error can be described by
defined as follows: the red curve in Fig. 12. When the displacement of the X-axis varies
from 0 to 16 mm, the machining error varies regularly, which can be
n g 2 2 compared by the different color curves in Fig. 12. In the actual grinding
grss = ( ) xi
i=1 xi (38) process, C-axis needs to rotate from 0° to 90° or from −90° to 0°, and
the machining error in the range of 0°–90° is significantly larger than
In grinding, the displacement ranges for the X, Y, and C axes are the one in the range of −90°–0° (shown in Fig. 12). Therefore, the
16 mm, 16 mm, and 90° in sequence, so only the variations of the linear grinding process should be performed, under the condition that the
errors or the angular errors within 16 mm or 90°are considered. The rotation angle of the C-axis ranges from −90° to 0° to reduce the effects
linear errors and angular errors are handled separately in the error of the setup errors.
allocation, as a result of their distinct characteristics. From the per- Even if the C-axis is held at a certain position, the displacement in
spective of comprehensive consideration, the sensitivity coefficients of the X or Y direction can also cause changes in the influence of the setup
the linear errors can be divided into two ranges, i.e., 0–1 × 10−6 and errors, as described in Fig. 13. When the angular displacement of the C-
1 × 10−6∼2 × 10−6. The linear errors in the first and second ranges axis is 0°, the machining error varies dramatically with the movement
are sequentially allocated as 0.1 μm and 0.05 μm, wherein the same in X/Y direction, while the variation of machining error reaches the
result can be obtained by multiplying the error values with the max- minimum when the angular displacement is about −60°. However,
imum sensitivity coefficients in the two ranges, respectively. Similarly, despite the angular displacement of the C-axis, the machining error
the sensitivity coefficients of the angular errors can also be divided into varies between the upper limit and lower limit.
two parts, one ranging from 0 to 0.01 and the other ranging from 0.01 In the actual assembly process, the setup error cannot be as ideal as
to 0.1. The results of the position-dependent error allocation are listed the specific allocated value. The assembly process should be designed
in Table 3. and optimized to substantially reduce the deviation between the actual
The combination of Eqs. (36) and (38) and the error values can setup errors and the allocated ones. As shown in Fig. 14, when the setup
predict that the profile error of the component affected by the position- errors vary within the range of −20% to 20%, the variation of the
dependent errors is less than 0.16 μm. machining error is not completely consistent with the variation of all
the setup errors. When the values of Szx, PAY and PAZ become smaller,
4.2.2. Allocation of the position-independent errors the machining error turns larger instead. Therefore, the average value
The sensitivity coefficients of the position-independent errors can be of 0.14 μm is determined as the machining error caused by the setup
divided into three ranges (i.e., 0–0.05, 0.05–0.15, 0.15–0.25), so 1.5 errors.
μrand, 0.5 μrand and 0.3 μrand are allocated to the setup errors in the From the above analysis and calculation, it can be concluded that
above three ranges in sequence. The results of the setup error allocation the machining errors caused by the position-dependent errors and po-
are given out in Table 4. sition-independent errors are 0.16 μm and 0.14 μm, respectively. The
The inclination angle of the grinding wheel spindle can be regarded
as an angular displacement of the A-axis to include the setup errors of
the grinding wheel spindle in the error model. When the inclination
angle of the grinding wheel spindle and the rotation angle of the C-axis
change, the effects of the setup errors on the machining error will be
different, as shown in Fig. 10. Therefore, it is important to find the
appropriate installation angle of the grinding wheel spindle to minimize
the effects of the setup errors. When the installation angle of the
grinding wheel spindle changes from 0° to 90° and the setup errors are
just equal to the values listed in Table 4, the machining error will vary
as the red line shown in Fig. 11. If the setup errors vary within 20% of
their allocated values, the results are depicted by the other color lines,
as shown in Fig. 11. The minimum points of the curves fluctuate around
the polar angle of 40°, indicating that 40° is the best choice for the
installation angle of the grinding wheel spindle.
When the angular displacement of the C-axis changes, the effects of
the setup errors on the machining accuracy are different, so it is ne-
cessary to quantify the relationship between the rotation angle of the C- Fig. 10. Effects of motion in A and C directions on machining error.

299
T. Wang et al. Precision Engineering 56 (2019) 293–302

Fig. 13. Effects of displacement of X/Y on setup errors.

Fig. 11. Optimization for installation angle of grinding wheel spindle.

Fig. 14. Machining error varies with the variation of setup errors.

Fig. 12. Effects of C axis displacement on setup errors.

total machining error affected by the geometric errors of the designed


machine is about 0.3 μm, so the geometric error contributes only half of
the accuracy requirement (0.6 μm) of the component. When the dis-
tances between different axes, i.e., the structure parameters, are ne-
glected, the error can be expressed by Eq. (39). Under this condition,
the machining error will be reduced to 0.15 μm.
En’ = [ x xx y xy + xy xz x yx y yy + cx (x cos c + y sin c )
Fig. 15. The designed special grinding machine tool.
+ cy ( x sin c + y cos c ) + xySxy ]/r (39)
machine tool was made of granite, equipped with a vibration isolation
5. Machining trials device to isolate the external vibration and an electric discharge device to
realize the on-machine truing of ball-end diamond grinding wheels. The
The developed grinding machine tool using a small ball-end diamond CCD monitoring device was adopted for tool setting and monitoring the
grinding wheel is special for grinding the ψ-shaped component with small wear condition of the grinding wheel. Moreover, the control software of the
concave surfaces, as shown in Fig. 15. The machine tool could realize four- machine tool was self-developed.
axis linkage (including X, Y, Z and C axes), with two fine-tuning positioners In grinding experiments, the ψ-shaped complex structural compo-
adjusting the center of the grinding wheel to the axis of the C turntable. The nents with small concave surfaces were ground to investigate the
grinding wheel spindle was tilted at an angle of 40°. The body of the grinding performance of the developed machine tool. Metal bond,

300
T. Wang et al. Precision Engineering 56 (2019) 293–302

Table 5
Detailed grinding parameters.
Parameters Grinding wheel spindle speed Workpiece spindle speed Grinding depth Feed rate
Values 60000 rpm 180 rpm 1 μm 30 μm/s

(4) The developed machine tool satisfies the design requirements,


which is verified by the grinding experiment results that the
average profile accuracy (PV) and surface roughness (Ra) of the ψ-
shaped component are less than 0.589 μm and 40 nm, respectively.
With error compensation, the achieved profile accuracy of 0.34 μm
indicates that the proposed error modeling method is accurate.

Acknowledgments

Fig. 16. Comparing processing quality of the component before and after ultra- This work is supported by the National High Technology Research
precision grinding a) and b) before ultra-precision grinding, c) and d) after and Development Program (“863” program) of China [2015AA043301]
ultra-precision grinding.
and the National Key Research and Development Program of China
[2018YFB1107600].
diamond grits of 3 μm and 120% concentration were adopted, which is
beneficial for getting high surface quality and profile accuracy. Before References
grinding, the small ball-end diamond grinding wheel was trued on-
machine to reduce the effect of eccentricity. All the detailed grinding [1] Brinksmeier E, Mutlugünes Y, Klocke F, Aurich JC, Shore P, Ohmori H. Ultra-pre-
parameters used in the experiments are listed in Table 5. The Talysurf cision grinding. CIRP Ann Manuf Technol 2010;59(2):652–71.
[2] Huo DH, Cheng K, Wardle F. A holistic integrated dynamic design and modeling
PGI 1240 contact probe profilometer was adopted to evaluate the approach applied to the development of ultraprecision micro-milling machines. Int
grinding accuracy and surface roughness. J Mach Tools Manuf 2010;50(4):335–43.
The grinding experiments were performed on several components to [3] Zhao LP, Chen HR, Yao YY, Diao GZ. A new approach to improving the machining
precision based on dynamic sensitivity analysis. Int J Mach Tools Manuf
compare the average processing quality before and after ultra-precision
2016;102:9–21.
grinding, as shown in Fig. 16. Before ultra-precision grinding, the [4] Liang YC, Chen WQ, Bai QS, Sun YZ, Chen GD, Zhang Q, et al. Design and dynamic
profile accuracy (PV) of the component was 3.056 μm, and the surface optimization of an ultraprecision diamond flycutting machine tool for large KDP
roughness Ra was 0.564 μm, which did not satisfy the accuracy re- crystal machining. Int J Adv Manuf Technol 2013;69:237–44.
[5] Li W, Li BZ, Yang JG. Design and dynamic optimization of an ultra-precision micro
quirement and took too long to perform polishing. After ultra-precision grinding machine tool for flexible joint blade machining. Int J Adv Manuf Technol
grinding, the surface roughness Ra of the component was improved to 2017;93:3135–47.
about 40 nm and the profile accuracy PV reached 0.589 μm. Then the [6] Yang B, Xie XH, Zhou L, Hu H. Design of a large five-axis ultra-precision ion beam
figuring machine: structure optimization and dynamic performance analysis. Int J
setup errors and linear errors were compensated by the proposed error Adv Manuf Technol 2017;92:3413–24.
propagation model. According to Eqs. (21) and (22), the corresponding [7] Xiao Y, Chen MJ, Chu X, Tian WL. Research on accuracy analysis and performance
errors in the X and Y directions were calculated, and the machining verification test of micro-precise five-axis machine tool. Int J Adv Manuf Technol
2013;67:387–95.
code at each grinding point was modified. Finally, the profile accuracy [8] Kvrgic V, Dimic Z, Cvijanovic V, Vidakovic J, Kablar N. A control algorithm for
after grinding was reduced to 0.34 μm, confirming that the machine improving the accuracy of five-axis machine tools. Int J Prod Res
tool developed by using the design and optimization method can well 2014;52(10):2983–98.
[9] Cai LG, Zhang ZL, Cheng Q, Liu ZF, Gu PH, Qi Y. An approach to optimize the
meet the design requirements and the error model is accurate. machining accuracy retainability of multi-axis NC machine tool based on robust
design. Precis Eng 2016;43:370–86.
6. Conclusions [10] Bohez ELJ, Ariyajunya B, Sinlapeecheewa C, Shein TMM, Lap DT, Belforte G.
Systematic geometric rigid body error identification of 5-axis milling machines. Int
J Comp Aided Des 2007;39(4):229–44.
A special grinding machine with a small ball-end diamond grinding [11] Chen JX, Lin SW, Zhou XL. A comprehensive error analysis method for the geo-
wheel is designed to process thin-walled small concave surfaces. The metric error of multi-axis machine tool. Int J Mach Tools Manuf 2016;106:56–66.
function and configuration of the machine are designed, by analyzing [12] Hocken RJ, Simpson JA, Borchardt B. Three dimensional metrology. CIRP Ann
Manuf Technol 1997;26(2):403–8.
the structural characteristics and processing requirements of the com- [13] Lamikiz A, de Lacalle LLN, Ocerin O, Diez D, Maidagan E. The Denavit and
ponents. The structure parameter design and error allocation are Hartenberg approach applied to evaluate the consequences in the tool tip position
completed with error modeling and sensitivity analyzing. The following of geometrical errors in five-axis milling centres. Int J Adv Manuf Technol
2008;37:122–39.
conclusions are drawn: [14] Jha BK, Kumar A. Analysis of geometric errors associated with five-axis machining
centre in improving the quality of cam profile. Int J Mach Tools Manuf
(1) An error propagation model is established by comprehensively 2003;43(6):629–36.
[15] Tang H, Duan JA, Lan SH, Shui HY. A new geometric error modeling approach for
considering kinematic errors, setup errors and structure parameters. multi-axis system based on stream of variation theory. Int J Mach Tools Manuf
The sensitivity analysis of the errors and structure parameters in- 2015;92:41–51.
dicates that the influence degree of each factor is proportional to [16] Fan JW, Guan JL, Wang WC, Luo Q, Zhang XL, Wang LY. A universal modeling
method for enhancement the volumetric accuracy of CNC machine tools. J Mater
the fluctuation rate of the corresponding sensitivity affected by the Process Technol 2002;129:624–8.
movements in the X, Y and C directions. [17] Fu GQ, Fu JZ, Shen HY, Xu YT, Jin Y. Product of exponential formulas for precision
(2) The influence of the structure parameters on the machining accu- enhancement of five-axis machine tools via geometric error modeling and com-
pensation. Int J Adv Manuf Technol 2015;81:289–305.
racy determines that the optimum inclination angle of the grinding
[18] Fu GQ, Fu JZ, Xu YT, Chen ZC. Product of exponential model for geometric error
wheel spindle is 40° and the optimum working range of the C-axis is integration of multi-axis machine tools. Int J Adv Manuf Technol 2014;71:1653–67.
−90°–0°. [19] Huston RL, Passerello CE. Multibody structural dynamics including translation
(3) After error allocation, the ultimate geometric accuracy of the de- between the bodies. Comput Struct 1980;12(5):713–20.
[20] Zhong GY, Wang CQ, Yang SF, Zheng EL, Ge YY. Position geometric error modeling,
signed machine is expected to reach 0.3 μm, which only accounts identification and compensation for large 5-axis machining center prototype. Int J
for half of the profile accuracy requirement of the component Mach Tools Manuf 2015;89:142–50.
(0.6 μm). [21] Moon SK, Moon YM, Kota S, Lancers RG. Screw theory based metrology for design

301
T. Wang et al. Precision Engineering 56 (2019) 293–302

and error compensation of machine tools. Proc DETC 2001;1:697–707. [29] Miro S, Hartmann D, Schanz T. Global sensitivity analysis for subsoil parameter
[22] Xiang ST, Altintas Y. Modeling and compensation of volumetric errors for five-axis estimation in mechanized tunneling. Comput Geotech 2014;56:80–8.
machine tools. Int J Mach Tools Manuf 2016;101:65–78. [30] Lagerwall G, Kiler G, Munoz-Carpena R, Wang N. Global uncertainty and sensitivity
[23] Cossarini G, Solidoro C. Global sensitivity analysis of a trophodynamic model of the analysis of a spatially distributed ecological model. Ecol Model 2014;275:22–30.
Gulf of trieste. Ecol Model 2008;212:16–27. [31] Murray RM, Li Z, Sastry SS. A mathematical introduction to robotic manipulation.
[24] Shahsavani D, Grimvall A. Variance-based sensitivity analysis of model outputs Florida: CRC Press; 1994.
using surrogate models. Environ Model Softw 2011;26(6):723–30. [32] Quillet A, Garneau M, Frolking S. Sobol' sensitivity analysis of the Holocene Peat
[25] Caro S, Wenger P, Bennis F, Chablat D. Sensitivity analysis of the orthoglide: a Model: what drives carbon accumulation in peatlands. J Geophys Res Biogeo
three-DOF translational parallel kinematic machine. J Mech Des 2013;118:203–14.
2006;128(2):392–402. [33] Nossent J, Elsen P, Bauwens W. Sobol' sensitivity analysis of a complex environ-
[26] Cheng Q, Sun BW, Liu ZF, Li JY, Dong XM, Gu PH. Key geometric error extraction of mental model. Environ Model Softw 2011;26(12):1515–25.
machine tool based on extended Fourier amplitude sensitivity test method. Int J [34] Zhang YM, Ji ST, Zhao J, Xiang LJ. Tolerance analysis and allocation of special
Adv Manuf Technol 2017;90:3369–85. machine tool for manufacturing globoidal cams. Int J Adv Manuf Technol
[27] Melchers RE, Ahammed M. A fast approximate method for parameter sensitivity 2016;87:1597–607.
estimation in Monte Carlo structural reliability. Comput Struct 2004;82(1):55–61. [35] Wang TZ, Chen J, Liu HN, Chen MJ, Wu CY, Fang Z, et al. Effects of kinematic
[28] Cheng Q, Feng QN, Liu ZF, Gu PH, Zhang GJ. Sensitivity analysis of machining parameters on electric discharge truing of small ball-end diamond wheels for small
accuracy of multi-axis machine tool based on POE screw theory and Morris method. concave surfaces grinding. Precis Eng 2018;51:117–27.
Int J Adv Manuf Technol 2016;84:2301–18.

302

You might also like