You are on page 1of 14

Measurement 188 (2022) 110486

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

A novel cylindrical profile measurement model and errors separation


method applied to stepped shafts precision model engineering
Yongmeng Liu , Yingjie Mei , Chuanzhi Sun , Ruirui Li , Xiaoming Wang , Hongye Wang ,
Jiubin Tan , Qing Lu
Center of Ultra-precision Optoelectronic Instrument Engineering, Harbin Institute of Technology, Harbin 150080, China
Key Lab of Ultra-precision Intelligent Instrumentation Engineering (Harbin Institute of Technology), Ministry of Industry and Information Technology, Harbin 150080,
China

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

Keywords: A large-scale stepped shaft is an important component in precision machinery manufacturing. Aiming at the
Stepped shaft model engineering of large-scale stepped shafts and improving the surface profile measurement accuracy of
Model engineering large-scale stepped shafts, a cylindrical profile measurement model containing seven systematic errors is
Coaxiality
designed, in which the eccentricity error, tilt error, sensor probe radius error, probe offset error, probe support
Cylindrical profile measurement model
rod tilt error, and horizontal and vertical rail tilt error are considered. Further, a multi-systematic errors sepa­
Error separation
ration method is proposed based on the seven-systematic errors in the cylindrical profile measurement model.
First, the autocollimator and image processing are used to accurately extract the verification parameters, and
then the stepwise estimation method and the equalisation optimizer (EO) are used to obtain the sectional and
spatial parameters. The measurement experiment is based on the large-scale stepped shaft profile measurement
device, and uses the coaxiality indicator to evaluate the profile measurement accuracy of the large-scale stepped
shaft, and compares our model with the dual-systematic errors model and the five-systematic errors model. The
experimental results show that compared with the other two methods, the coaxiality measurement accuracy of
standard stepped shaft No. 1 can be increased by 26.03% and 15.99%, respectively; the cylindricity measurement
accuracy of standard stepped shaft No. 1 can be increased by 16.03% and 9.17%, respectively; the coaxiality
measurement accuracy of the No. 2 standard stepped shaft can be increased by 29.86% and 16.29%, respectively;
the cylindricity measurement accuracy of standard stepped shaft No. 2 can be increased by 17.32% and 10.46%,
respectively –– this verifies the effectiveness of our measurement method. The measurement method used in this
study provides an accurate theoretical basis for the implementation of high-precision model engineering of cy­
lindrical parts for national metrology institutions and key laboratories.

1. Introduction can reach sub-micron level or even nanometre level, and it is an effective
way to ensure the high-precision realization of precision model engi­
In the field of high-end intelligent manufacturing, object surface neering [7]. Large-scale, high-precision stepped shafts are typical parts
profile data is the basis for precision machining and model engineering in the field of precision machinery manufacturing. Key mechanical
[1,2]. At present, advanced intelligent manufacturing systems such as components, such as high-precision machine tool spindles and rocket
digital twins and cloud manufacturing have gradually been deeply turbo pumps, all have a stepped structure [8]. Taking the stepped shaft
applied to ultra-precision fields such as high-end intelligent machine model project as an example, when measuring the surface profile of the
tools, lithography machine manufacturing, and rocket system engi­ stepped shaft, the measurement device such as the ultra-precision
neering [3–5]. This means that the accuracy of object modelling needs to cylindricity measuring instrument has multi-systematic errors. The
reach the micron level or even the nanometre level to ensure the accu­ coupling of these systematic errors will significantly affect the accuracy
racy of the cutting-edge manufacturing system [6]. Ultra-precision of surface profile measurement. Therefore, improving the surface profile
measurement technology is a method to achieve high-precision recon­ measurement model and effectively separating the multi-systematic
struction of object profile. At present, the contact measurement accuracy errors of the measurement model is of great significance for improving

E-mail address: czsun@hit.edu.cn (C. Sun).

https://doi.org/10.1016/j.measurement.2021.110486
Received 20 August 2021; Received in revised form 4 November 2021; Accepted 15 November 2021
Available online 25 November 2021
0263-2241/© 2021 Published by Elsevier Ltd.
Y. Liu et al. Measurement 188 (2022) 110486

the accuracy of ultra-precision measurement and realizing precision


model engineering.
The appearance of the stepped shaft is similar to a cylinder, and the
influence of the systematic errors of the measuring device is mainly
reflected in the cylindrical profile measurement model of the stepped
shaft [9]. At present, researchers have performed significant work on
coaxiality measurement and cylindrical profile models. The eccentricity
error is the primary source of error in the cylindrical profile measure­
ment. Chetwynd [10] first established the Limacon model for eccen­
tricity error, which is widely used in cylindrical instruments. Owing to
the obvious influence of tilt in cylindrical profile measurements, Murthy
et al. [11] proposed a dual-systematic error cylindrical profile mea­
surement model that includes eccentricity and tilt errors for the defects
of the Limacon model. The dual-systematic error measurement model
only considers two common errors, but different measurement schemes
also introduce other non-negligible systematic errors in the measure­
ment of precision devices [12]. For contact probe measurement, Morel Fig. 1. Profile measurement system for large-scale stepped shaft.
[14] pointed out that the probe radius causes not only the measurement
data to have a nonlinear effect on the tracking trajectory of the surface
indicators of the standard stepped shaft is used to analyze the mea­
profile, but also the measurement angle to change in a small range from
surement accuracy. At the same time, the coaxiality and cylindricity
the ideal situation. Based on the guide rail contact measurement device,
indicators also reflect the needs of stepped shafts in actual applications
Sun et al. [15] analysed the systematic errors of cylindrical profile
[24,25]. For example, rocket engine turbo pumps need to measure
measurements and proposed a five-systematic error measurement
coaxiality to ensure the centering characteristics of the rotor during
model, but the model did not consider the errors of the vertical rail and
installation and reduce friction and instability during high-speed rota­
the horizontal rail and the coupling relationship of the tilt error of the
tion. The model engineering of the step shaft can use the coaxiality and
probe support rod is not clear. Regarding the common equipment for
cylindricity indicators to evaluate the accuracy of the contour data, and
coaxiality measurement, the state of the art models still fail to consider
at the same time better respond to the actual application requirements.
more systematic errors coupling, especially the significant influence of
The comparison experiment was carried out with the typical dual-
guide rail error in the measurement.
systematic errors model [11] and the advanced five-systematic errors
The multi-systematic errors cylindrical profile measurement model
model [15]; good results were obtained in both simulation verification
requires a multi-systematic errors separation method to decouple the
and actual measurement verification. The accuracy of the seven-
model owing to the coupling of errors. Haitjema [16] designed a multi-
systematic errors model and the multi-systematic errors separation
step error separation method for roundness calibration for the turntable
method proposed in this study is proven.
error; Tan [17] proposed a geometric axis tilt error separation method to
The structure of this paper is as follows: The section 1 of this paper
separate the position error between the rotary reference and the linear
comprehensively expounds the current research status of stepped shaft
reference; Whitehouse [18] studied the influence mechanism of the tilt
profile model and error separation. In the section 2, a seven-systematic
error of the vertical guide rail during the measurement and compensated
errors cylindrical profile measurement model for stepped shaft is pro­
for its linear influence. Liu et al. [19] proposed the five-point cylindricity
posed. The section 3 proposes the error separation method of the seven-
error separation technology to solve the double integral effect in the
systematic errors cylindrical profile measurement model, introduces the
traditional multi-point EST. Chen et al. [20] proposed a method to
characteristics of the equalization optimizer (EO algorithm), and in­
separate random and systematic errors for the 3D profile rotation
troduces the basis of the coaxiality evaluation in this paper. In the sec­
measurement system. Anandan [21] proposed a multi-directional error
tion 4, the model and method of this paper are simulated and analyzed.
separation technology for ultra-high-speed micro-spindle measure­
The section 5 is the actual experiment of this paper. The effectiveness of
ments. Liu et al. [22] proposed a parallel error separation method for
the model and method in this paper is verified on a super-large
online measurement and reconstruction of cylindrical profile while
coaxiality measuring device using standard stepped shafts. Section 6
separating the radial and tilt errors of the rotary spindle. Shi et al. [23]
gives the conclusion of this paper.
considered the ellipse eccentricity and measurement error during
docking, established a corresponding control model, and realised error
2. Seven-systematic errors cylindrical profile measurement
compensation by using a control strategy. Different measurement
model
models use different error separation methods. For the coaxiality mea­
surement model, the error decoupling needs to consider the spatial
The large-scale stepped shaft coaxiality measurement system consists
distribution characteristics of each geometric parameter, and perform
of a stepped shaft, a contact-type inductive sensor probe, a precision air-
classification processing according to the data source.
floating rotary spindle, a high-precision turntable, a horizontal rail, and
Based on the large-scale stepped shaft profile measurement device in
a vertical rail, all of which is shown in Fig. 1. In contrast to the tradi­
Fig. 1, this study integrates the eccentricity error, tilt error, sensor probe
tional dual-systematic errors model that only considers eccentricity and
radius error, probe offset error, probe support rod tilt error, and hori­
tilt. Aaccording to the actual measurement system analysis in this study,
zontal and vertical rail tilt error in cylindrical profile measurement ––
the systematic errors include the eccentricity error of the stepped shaft,
and establishes a cylindrical profile measurement model with seven
tilt error, sensor probe radius error, probe offset error, tilt of the probe
systematic errors. This is because of the mutual coupling problem of
support rod error, and horizontal and vertical rail tilt error. Compared
various systematic errors in the cylindrical profile measurement model.
with the latest five-systematic errors model, this study considers the
Based on the seven-systematic errors cylindrical profile measurement
probe support rod tilt error and the horizontal tilt error. The tilt errors of
model, this study realises multi-systematic errors separation by using an
the stepped shaft can cause secondary eccentricity of the measurement
autocollimator and image processing, as well as a combination of step­
section and distortion of the measurement profile. The tilt of the probe
wise estimation and optimisation methods, which ensures the accuracy
support rod causes coupling between the sensor probe radius error and
of the cylindrical profile measurement model. In order to evaluate the
the machining error of the stepped shaft. The horizontal and vertical rail
validity of the measurement model, the coaxiality and cylindricity

2
Y. Liu et al. Measurement 188 (2022) 110486

Fig. 2. Cylindrical profile measurement model with seven systematic errors (a) Influence of eccentricity error (b) Influence of tilt error (c) Influence of probe offset
error (d) Influence of probe radius error (e) Influence of the tilt error of the probe support rod (f)Schematic of measured part (g) Measurement principal diagram of
the cylindrical section j.

3
Y. Liu et al. Measurement 188 (2022) 110486

tilt errors cause the sensor probe to shift twice and thrice, respectively. ( )
This paper is mainly based on the seven system errors of eccentricity, roij +
Δrij + rc ( ) ( )
( ) sin φij − θij = dj + Lj tan(w)sin τij
tilt, probe offset, probe radius, probe support rod tilt, horizontal and cos ∂j (3)
vertical rail tilt, to construct a seven-systematic errors cylindrical profile ( ) ( )
+zj tan(ϕ)sin εij + ej sin θij − αj
measurement model. As shown in Fig. 2 (a), the eccentricity error is
caused by the misalignment between the center of the bottom surface of Thus, φij can be expressed as:
the tested part and the rotation center of the turntable, and the mea­ ( ) ( ) ( )
dj + Lj tan(w)sin τij + zj tan(ϕ)sin εij + ej sin θij − αj
surement error is e⋅cos(θij − α). As shown in Fig. 2 (b), the tilt error is φij = θij + arcsin √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
mainly due to the misalignment of the geometric axis of the measured rsj
cos2 (φij − βj ) (
+ sin2 φij − βj + cosij ∂ c
) Δr +r
part and the measurement rotation axis, which leads to changes in pa­ cos2 γ ( j)
rameters such as eccentricity and sampling angle, as shown in Eq.(2). As (4)
shown in Fig. 2 (c), the probe offset error makes the sampling direction Based on the above derivation and power series expansion of Eq. (1),
misaligned, and the measurement error caused is (1/cos(sin-1[dj/ the measurement equation can be obtained as:
(r0 + Δrij)]) − 1) × Δrij. As shown in Fig. 2 (d), the error of the probe ⎧ ( ( ))
radius is mainly due to the fact that the measurement center of the probe ⎪ dj + dj1 + ej sin θij − αj 2
⎪ ρij = roij + Δrij −
⎪ ( )
should pass the center of the probe during measurement, rather than the ⎪


⎪ cos∂j Δrij + rc
⎪ 2 roij +
contact point between the probe and the cylindrical profile. The mea­ ⎪

⎪ cos∂j

surement error caused by the measurement error and the tilt angle of the ⎪



( )
⎪ +ej cos θij − αj + zj tanϕcosεij
probe support rod are Coupling; as shown in Fig. 2 (e), the tilt angle of ⎪


⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ̅
the probe rod makes the influence of the probe radius magnified to rc ⎪

⎨ cos2 ηij ( ) cos2 ηij ( )
/cos(∂j), and the machining error magnifies to Δrij /cos(∂j). As shown in (5)
2 2
roij = rsj + sin ηij = ro + sin ηij
⎪ cos2 γ cos2 γ
Fig. 2 (f), the horizontal rail mainly drives the probe to cause the second ⎪


⎪ ⎛ ⎞

deviation, the size is Lj⋅tanw⋅sinτij; the vertical rail causes the third de­ ⎪

⎪ ( )
⎪ ⎜ ⎟
viation of the probe, the size is zj⋅tanϕ⋅sinεij; At the same time, there will ⎪ η = φ − β = θij + arcsin⎜dj + dj1 + ej sin θij − αj ⎟ − β


⎪ ij
⎪ ij j ⎝ Δrij + rc ⎠ j
be a deviation in the measuring direction, the magnitude is zj⋅tanϕ⋅cosεij. ⎪

⎪ roij + ( )
⎪ ∂
With a comprehensive analysis of Fig. 2, based on the five-systematic ⎪



cos j

errors model [15] and according to the geometric relationships between dj1 = Lj tanwsinτij + zj tanϕsinεij
the errors, the seven-systematic errors cylindrical profile measurement
models can be obtained: From Eq. (5), there are obvious coupling relationships between
multi-systematic errors. According to the characteristics of the mea­
surement system parameters, the contact type inductive sensor probe

⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )

⎪ Δrij + rc 2 ( ( )) ( ) rc
⎨ρ = roij + − dj + dj1 + ej sin θij − αj 2 + ej cos θij − αj − + zj tanϕcosεij
ij
cos∂j cos∂j (1)



dj1 = Lj tanwsinτij + zj tanϕsinεij

radius rc, the tilt angle of the probe rod ∂j, the tilt angles w of the hori­
The parameter nomenclature is presented in Table 1. Tilt error can zontal rail and ϕ of the vertical rail, and the tilt angles τij of the hori­
cause the ellipse of the measuring section; thus, the measuring point of zontal rail and εij of the vertical rail––can be classified as verification
the section has geometric relationships with the axis of the ellipse and parameters; the initial eccentricity e0 of the stepped shaft, the initial
the sampling angle. At the same time, the tilt also causes a change in the eccentric angle α0, the least square cylinder radius ro, and the tilt angle γ
section eccentricity, which is related to the initial eccentricity of the between the geometric axis of the stepped shaft and the rotary axis of the
stepped shaft and the direction of the geometric axis. The expression is turntable can be classified as spatial parameters; the eccentricity ej of the
as follows: measuring section j and the eccentric angle αj, initial offset dj of the
⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( ) sensor probe, the included angle βj between the major axis direction of



⎪ roij = r2lj cos2 φij − βj + r2sj sin2 φij − βj the fitted ellipse on the measuring section j and the initial measurement



⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ̅ direction, and the semi-major axis rlj and semi-minor axis rsj of the fitted





cos2 φij − βj 2
( ) ellipse on measuring section j can be classified as sectional parameters.
⎪ = r + sin φ − β
The parameter matrices are listed in Table 2.
⎪ sj ij j

⎪ cos2 γ


⎨ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( )̅
ej = lzj + e0 cosα0 2 + mzj + e0 sinα0 2 (2) 3. Multiple systematic errors separation



⎪ [ ]

⎪ mzj + e0 sinα0

⎪ αj = tan− 1

⎪ lzj + e0 cosα0 Table 2 divides the coupling parameters in the seven-systematic er­




⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅ rors measurement model into three types. In order to obtain pure

measurement data and separate the errors from the measurement model,
′2 ′ ′ ′

⎪ l +m2 m l

⎩ tanγ = ,m = ′ ,l = ′
the three types of parameters need to be separated in batches. Among
′2
n n n
them, the verification parameters can be obtained by referring to the
The vector (l, m, n) is the direction vector of the geometric axis,
specifications and direct measurement, and spatial parameters and
which is used to characterise the tilt angle γ of the geometric axis. Based
sectional parameters need to be separated step by step and parameter
on the analysis in Fig. 2 (b), the relationship between the sampling angle
estimation.
φij of the geometric centre and the sampling angle θij based on the
rotation centre can be expressed as:

4
Y. Liu et al. Measurement 188 (2022) 110486

Table 1 Table 2
Nomenclature. Parameter category and parameter matrix
O Rotation centre of the bottom Oj1 Rotation centre of the Parameter category Parameter matrix
surface of stepped shaft measuring section j of
Verification parameters rc, ∂j, w, ϕ, τij, εij
stepped shaft
Spatial parameters e0, α0, ro, γ
O’0 Geometric centre of the bottom Oj Geometric centre of the
Sectional parameters ej, αj, dj, βj, rlj, rsj
surface of stepped shaft measuring section j of
stepped shaft
O’j Center of the fitted ellipse of the zj Sampling height of the
measuring section j measuring section j The image processing method can be used to quickly process and
Δrij Machining error of stepped shaft θij Actual sampling angle acquire the tilt angle of the probe support rod. The processing flow in­
φij Sampling angle based on geometric γ Tilt angle of geometric axis volves the use of a CCD camera to obtain the optical image of the rod,
centre and then uses image interception, image grayscale, image segmentation
βj Included angle between the major ∂j Tilt angle of probe support
axis direction of the fitted ellipse rod
(the general method uses threshold segmentation or clustering seg­
on the measuring section j and the mentation), image binarization, image edge extraction, etc. to obtain the
initial measurement direction. edge features of the measuring rod. Then, the fitting straight line is used
e0 Initial eccentricity ej Eccentricity of the measuring to obtain the vertical reference line and the measuring rod reference
section j
line. Finally, the tilt angle of the probe support rod can be obtained from
α0 Initial eccentricity angle αj Eccentricity angle of the
measuring section j the included angle by using the image correction and fitting straight line
ρij Distance from the probe to the rc Contact type inductive for calculation.
rotation centre sensor probe radius
rsj Semi-minor axis of the fitted ellipse rlj Semi-major axis of the fitted
on measuring section j ellipse on measuring section j
3.2. Separation of sectional and spatial parameters based on parameter
roij Distance from the measuring point ro Least squares cylindrical
i to the geometric centre Oj on the radius estimation
elliptical measurement section j
caused by tilt The spatial parameter matrix is {e0, α0, ro, γ}, and the sectional
dj Initial offset of the sensor probe dj1 The offsets of the probe parameter matrix is {ej, αj, dj, βj, rlj, rsj}. The analysis of Eq. (2) shows that
caused by the horizontal and
vertical rails
γ is related to l and m, and thus the spatial parameter matrix can be
w Tilt angles of the horizontal rail ϕ Tilt angles of the vertical rail expressed as {e0, α0, ro, l, m}. Owing to the large number of parameters,
τij Angles of the horizontal rail in the εij Angles of the vertical rail in stepwise estimation was used to gradually separate the systematic
measurement direction the measurement direction errors.
C Concentration in the control G Mass generation rate in the
First, for the estimation of the sectional parameters, the inherent tilt
volume control volume
F Exponential term t0 Initial time component of the section may be confused with the elliptization of the
C0 Initial concentration λ Flow rate section caused by the tilt error of the cylinder; thus, it is necessary to
Cmax The maximum values of particles Cmin The minimum values of identify the tilt feature of the section. The section tilt feature recognition
dimensions d particles dimensions method follows two basic features: (1) The cylindrical profile measures
multiple sections. When the sampling height zj of section j has an
3.1. Verification parameter separation equidistant distribution, the eccentricity {ej} caused by the tilt of each
section is an arithmetic sequence, and the eccentricity angle {αj’} caused
The verification parameter matrix is {rc, ∂j, w, ϕ, τij, εij}. The contact- by the tilt of each section is a constant number. (2) The eccentricity {e0j}
type inductive sensor probe is a ruby ball with a fixed specification, in caused by the inherent elliptization of the section is a series of fluctu­
which the radius parameter can be directly checked according to the ating constant values, and the corresponding eccentricity {α0j} is an
specification. The tilt angles of the horizontal and vertical rails were unknown quantity. The inherent elliptic eccentricity {e0j} has two types
measured using an autocollimator. The tilt angle of the probe rod was of numerical distribution. The first is the constant distribution, and the
second is the general distribution with a mean value of e = P-1 e / P

measured using image processing. 0 j=0 0j
According to the measurement principle of the autocollimator, it is (P is the number of measuring sections). The two eigenvector graphs are
known that the offset angle α is proportional to the focal length f and shown in Fig. 3.
inversely proportional to the optical image offset t. Based on the coor­ The vector relationship between the actual eccentricity of the section
dinate system shown in Fig. 2, the tilt angle of the horizontal and vertical and the eccentricity caused by the tilt of the section and the eccentricity
rails needs to be measured from both the x-axis and y-axis. The available caused by the inherent elliptization of the section is ej = ej + e0j. By
parameters of the horizontal and vertical rails are: decomposing the actual eccentricity of the section according to the
⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ vector relationship, the section parameter βj = tan-1 ((ej sinαj − h) / ej
⎨ w = (tanα )2 + ( tanα )2 , τ = θ − tan− 1 ( tanα /tanα ) cosαj) can be obtained. The parameter h is determined by the initial
(6)
wx wy ij jj wy wx
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎩ ( )2 ( )2 − 1
( / ) eccentricity of the cylindrical profile and the βj angle. According to Eq.
ϕ= tanαϕx + tanαϕy , εij = θjj − tan tanαϕy tanαϕx
(5), the semi-major axis rlj corresponding to the elliptization of the
measurement section can be derived from the semi-minor axis rsj; thus,
where αwx and αwy represent the angles measured by the autocollimator
the sectional parameter matrix can be expressed as {ej, αj, dj, rsj}.
along the x-axis and y-axis of the horizontal rail, respectively; αϕx and
Based on the obtained verification parameters and substituting them
αϕy represent the angles measured by the autocollimator along the x- and
into Eq. (1) to eliminate the linear influence of the measurement
y-axes of the vertical rail.
equation, the measurement equation ρij1 can be obtained as

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
[ ]
Δrij + rc 2 [ ( )] ( ) rc
ρ1ij = roij + − dj + Lj Tw sinτij + zj Tϕ sinεij + ej sin θij − αj 2 + ej cos θij − αj + (7)
cos∂j cos∂j

5
Y. Liu et al. Measurement 188 (2022) 110486

Fig. 3. Eigenvector graph of {e0j} (a) The eigenvector graph when {e0j} is a constant distribution (b) The eigenvector graph when {e0j} is a general distribution.

where Tw = tanw, and Tϕ = tanϕ. According to Eq. (6), the verification


parameters {rc, ∂j, w, ϕ, τij, εij} are all known. roij needs to use the feature The sectional parameters { ̂r lj , ̂r sj } can be estimated using an opti­
recognition method to determine whether the elliptization of the section misation algorithm. Thus, the sectional parameter matrix {ej, αj, dj, βj, rlj,
is caused by the axis tilt error. In the preliminary estimation, roij can be rsj} are all estimated. According to Eq. (1), the sectional parameters {ej,
regarded as an overall parameter to participate in parameter estimation; αj} are related to the spatial parameter matrix; therefore, the next step is
that is, the parameters {ej , αj, dj, roij} are estimated by the least squares to estimate based on the sectional parameters. Based on the estimated
method. Thus, the least squares objective function ξj1 is as Eq. (8): sectional parameter matrix and the measurement equation ρij2,

⎧⎛√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ⎞ ⎫

n ∑n ⎨ [ ] ⎬
r [ ]
(8)
⎝ c 2 + d + L T sinτ + z T sinε + e 2 − r ⎠cos∂ − r
ξ1j = 2
Δrij1 = 1
ρij − ejm − j j w ij j ϕ ij jv oij j c
2
= min
i=1 i=1
⎩ cos∂j ⎭

excluding the parameters that have no relevance with the spatial pa­
where ejv = ej sin(θij − αj), ejm = ej cos(θij − αj). Using the optimisation rameters, the measurement equation ρij3 can be obtained as
̂ ̂ ̂ √[(
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅
algorithm proposed in Section 3.3, the parameters { ̂ e j , Î ± j , d j , r oij }
⎧ ) ] [ ( )] ( )

⎨ ρ3ij = ρ2ij − roij cos∂j + ̂r sj 2 − ̂e j sin θij − ̂ α j 2 + ̂e j cos θij − ̂
αj
̂ ̂
are estimated, and the sectional parameters { e j , Î ± j , d j } are
̂ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎪ [ ( )] [ ( )]̅
substituted into Eq. (7) for error separation. The sectional parameters ⎩
roij = ̂r lj cos θij − βj 2 + ̂r sj sin θij − βj 2
̂r oij which contain the section elliptization caused by the axis tilt error
(11)
cannot be eliminated, and the semi-major and semi-minor axes of the
ellipse need to be estimated. Then, the measurement equation ρij2 is as To estimate the spatial parameters {e0, α0, ro, l, m}, the measurement
Eq. (9): equation ρij4, including the machining error and axis tilt error combined

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
[ ] ̅
Δrij + rc ( ) rc 2 [ ( )]
ρ2ij = + roij = ρ1ij − ̂e j cos θij − ̂ αj − + ̂ d j + Lj Tw sinτij + zj Tϕ sinεij + ̂ e j sin θij − ̂ αj 2 (9)
cos∂j cos∂j

To separate the major and minor axes of the ellipse, according to Eqs.
(1) and (9), the least square objective function ξj2 is as Eq. (10):


n n {(
∑ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ) }
[ ( )] [ ( )]
ξ2j = 2
Δrij2 = ρ2ij − rlj cos θij − βj 2 + rsj sin θij − βj 2 cos∂j − rc 2 = min (10)
i=1 i=1

6
Y. Liu et al. Measurement 188 (2022) 110486

machining error of the cylinder, is


⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
[
⎪ 5
( )] [ ( )]
⎪ ρ
⎪ ij

= ρ4ij − ̂e j cos θij − ̂ α j 2 + ̂e j sin θij − ̂ α j 2 − ̂r o



⎨ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( )
̂e j = ̂lzj + ̂ e 0 coŝ α0 2 + m ̂ zj + ̂ e 0 sin̂ α0 2 (14)



⎪ [ ]

⎪ m̂ zj + ̂e 0 sin̂ α0

⎩α ̂ j = tan− 1
̂lzj + ̂e 0 coŝ α0

3.3. Equalization optimizer (EO) optimization algorithm

Aiming at the measurement equations (7) (9) (11) (12) (14) in


stepwise estimation method, the paper uses an optimisation algorithm to
approximate the solution. We use the measurement equation as the
objective functions of the optimisation algorithm, and the estimated
parameters are used as variable values. As shown in Fig. 4, the equal­
isation optimizer (EO) is an optimisation algorithm inspired by the
physical dynamic source and sink model [26]. Its concept is to achieve
Fig. 4. Schematic of equalization optimizer (EO).
optimisation based on the process of controlling the volume dynamic
mass balance. By solving the first-order differential mass balance
equation, the balance relationship of the control volume concentration
with Eq. (11) can be obtained as:
can be obtained.
⎧ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎪ ( ) ( ( )) ( )
⎪ 4
Δrij + ro 2 − ej sin θij − αj 2 + ej cos θij − αj = ρ3ij ( ) G
⎪ ρij =

⎪ C = Ceq + C0 − Ceq F + (1 − F), F = e− λ(t− t0 )
(15)

⎪ √(̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ λV
⎨ ) ( )̅
ej = lzj + e0 cosα0 2 + mzj + e0 sinα0 2 (12) The parameter nomenclature is presented in Table 1. The algorithm

⎪ [ ]



⎪ − 1 mzj + e0 sinα0
flow is as follows:

⎩ αj = tan
lzj + e0 cosα0
(1) Initialisation: The initial concentration is constructed based on
To estimate the spatial parameters, the least-squares objective the number and dimensions of particles, and is uniformly and
function ξj3 combined with Eq. (12) can be obtained as: randomly initialised in the search space according to the upper
⎧ 3 ∑p ∑n
2 and lower limits. The expression is as follows:

⎪ ξj = j=1 Δrij3

⎪ i=1


⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Ciinitial = Cmin + randi (Cmax − Cmin ), i = 1, 2, …, n (16)


⎪ ∑p ∑n { ([ ( )] [ ( )] )
}2

⎪ ρ4ij − ej cos θij − αj 2 + ej sin θij − αj 2 − ro =min
⎪ = j=1
⎨ i=1 where randi is the selection of i random quantities, and n is the total
⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( )̅ number of particles.



⎪ ej = lzj +e0 cosα0 2 + mzj +e0 sinα0 2




⎪ [ ] (2) Calculate the fitness value: Determine the objective function and
⎪ mzj +e0 sinα0


⎩αj =tan− 1 calculate the fitness. The study is calculated based on the objec­
lzj +e0 cosα0
tive function given in Section 3.2.
(13)
Using the optimisation algorithm, the spatial parameters {
̂
e 0 , α 0 ,̂r o ,̂l, m
̂ ̂ } can be estimated; thus, all parameters are separated.
The available measurement equation ρij5, which only contains the

⎧⎛√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ⎞ ⎫

n ∑n ⎨ [ ] ⎬
r c [ ]
Function 1 : ξ1j = 2
Δrij1 = ⎝ ρ1ij − ejm − 2 + d + L T sinτ + z T sinε + e 2 − r ⎠cos∂ − r
j j w ij j ϕ ij jv oij j c
2
= min
i=1 i=1
⎩ cos∂j ⎭


n n {(
∑ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ) }
[ ( )] [ ( )]
Function 2 : ξ2j = 2
Δrij2 = ρ2ij − rlj cos θij − βj 2 + rsj sin θij − βj 2 cos∂j − rc 2 = min
i=1 i=1


⎪ ∑ p ∑ n




⎪ ξ3j = Δrij32

⎪ j=1 i=1





⎪ n {
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ }2

⎪ p ∑
∑ ([ ( )] [ ( )] )

⎪ = ρ4ij − ej cos θij − αj 2 + ej sin θij − αj 2 − ro = min

Function 3 : j=1 i=1



⎪ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( )̅



⎪ ej = lzj + e0 cosα0 2 + mzj + e0 sinα0 2





⎪ [ ]

⎪ mzj + e0 sinα0

⎪ αj = tan− 1

⎩ lzj + e0 cosα0

7
Y. Liu et al. Measurement 188 (2022) 110486

( )
⎧→ → → →→

⎪ G = G cp C eq − λC

{ → (19)
⎪ 0.5 r 1 if r2 ⩾0.5
⎩→

G cp =
0 otherwise


where G cp is the control parameter vector of the generated rate, →
r 1 is a
random number vector whose dimension is consistent with the opti­
mised space dimension, and each element value is a random number
from 0 to 1, and r2 is a random number in the range of 0 to 1.

(6) Solution update: Update the solution based on Eq. (15).

The stopping condition of the equalisation optimizer is 300 iterations


Fig. 5. Evaluation method for the stepped shaft coaxiality. or the difference between the results of the two iterations is less than
0.001. The calculation process based on the equalisation optimizer (EO)
⎡ ⎤ can quickly estimate the sectional and the spatial parameters, which

e 4̃
e
s.t. ej ∈ ⎣
j
,
j ⎦, effectively separate the seven-systematic errors, to obtain the ‘pure’ data
3 3 of the cylindrical profile measurement.
⎡ ⎤
α 4̃
2̃ α
αj ∈ ⎣ j , j ⎦,
3 3 3.4. Evaluation method for the stepped shaft coaxiality
[ ]
dj ∈ 0, 0.01̃ro , rsj ,
[ ] Coaxiality is used as an indicator to evaluate measurement accuracy
ro ∈
2̃ro 4̃ro
, , l, in this paper. To evaluate the coaxiality of the stepped shaft, it is
3 3 necessary to determine the reference elements and establish a reference
m ∈ [0, tañγ] axis. In this study, the reference element of the stepped axis is the upper
and lower end region, and the centreline of the two regions is deter­
̃ej ̃
αj , ̃ro , ̃γ represent the data measured before errors are separated. mined as the reference axis using the least square method. The mea­
surement element is the rest of the stepped axis. As shown in Fig. 5, L is
(3) Determine the equilibrium state pool: To search for the optimal the reference axis, and ϕT represents the smallest zone of the stepped
value, the equilibrium candidate provides a search mode for the axis.
particles. That is, the four optimal particles and their arithmetic
average form an equilibrium state pool of five particles, and the 4. Simulations
probability of the five candidate solutions being selected is 0.2.
The expression is as follows: 4.1. The influence of integrated systematic errors on cylindrical profile

{
→ → → → →
} measurement
C eq,pool = C eq(1) , C eq(2) , C eq(3) , C eq(4) , C eq(ave) (17)
To analyse the influence of integrated systematic errors on the cy­
lindrical profile measurement, it is necessary to compare the proposed
seven-systematic errors model with the dual-systematic errors model
(4) Exponential term F: To better balance the local search and global and the five-systematic errors model. The dual-systematic errors model
search of the algorithm, F in Eq. (15) is improved as follows: and the five-systematic errors model are shown in Eqs. (20) and (21),
[ → ] respectively:

F = a1 sign(→
r − 0.5) e− λ t − 1 (18) ( )
ρ̇ = ej cos θij − αi + Δrij + rsj (20)
(
ρ̈ = ej cos θij
( ( ))
(5) Mass generation rate G: To strengthen the local optimisation ) dj + zj tanϕsinεij + ej sin θij − αi
− αi + zj tanϕεij + roij + Δrij − ( )
capability of the algorithm, the mass generation rate is designed 2 roij + Δrij + rc
as follows: (21)
Assuming that the number of sampling points n is 1024, the average
contour machining error Δrij is 0.001 mm, which conforms to the stan­
dard normal distribution (μ = 0.001, σ = μ/3), the cylinder radius rsj is
50 mm, the initial eccentricity angle α0 is π/3, the included angle βj
between the major axis direction of the fitted ellipse on the measuring
section j and the initial measurement direction is π/6; the horizontal rail
Table 3 movement length Lj is 500 mm, The sampling height zj is taken three
Error level from level 1 to 5. times (respectively 0 mm, 100 mm, and 300 mm) to reflect the change of
Level e0 (μm) dj (μm) γ rc (mm) ∂j w ϕ calculation errors at different heights. We refer to [15], the error level
Level 1 1 50 1′ 2.5 1◦ 1′′ 1′′ was set, which gradually increased from level 1 to level 5, as shown in
Level 2 5 100 3′ 2 3◦ 3′′ 3′′ Table 3. The seven-systematic errors model subtracts with the dual-
Level 3 10 200 5′ 1.5 7◦ 8′′ 8′′ systematic errors model and the five-systematic errors model, respec­
Level 4 20 300 8′ 1 10◦ 12′′ 12′′ tively; and the calculation errors between the models are analysed to
Level 5 30 500 12′ 0.5 15◦ 20′′ 20′′
reflect the influence of the systematic errors on the measurement and the

8
Y. Liu et al. Measurement 188 (2022) 110486

Table 4
Parameter values corresponding to each error level.
Error Eccentricity e0 (μm) Probe offset dj (μm) Tilt angle γ (′′ ) Probe rod tilt angle ∂j (◦ )

Standard stepped shaft No. 1 No. 2 No. 1 No. 2 No. 1 No. 2 No. 1 No. 2

Level 1 2.31 1.13 8.43 9.34 4.38 8.30 16.30 19.54


Level 2 4.20 4.85 28.59 33.94 63.33 60.20 21.05 23.51
Level 3 9.13 9.63 57.82 58.53 104.07 96.22 25.27 28.06
Level 4 14.56 16.24 93.56 101.19 197.16 180.54 32.91 34.76
Level 5 23.77 24.90 156.29 184.81 242.93 234.02 39.43 38.68

Fig. 6. The law of calculation errors between the seven-systematic errors model, and the dual-systematic errors model (the five-systematic errors model) at
different heights.

accuracy of the multi-systematic errors model. significant decrease in measurement accuracy. Furthermore, under the
Fig. 6 (a-c) shows the law of calculation errors between the seven- same error level, the calculation error increases with an increase in the
systematic errors model and the dual-systematic errors model at sampling height. For example, for error level 5, from Fig. 6 (a-c), when
different heights, and Fig. 6 (d-f) shows the law of calculation errors the sampling height increases from 0 mm to 300 mm, the maximum
between the seven-systematic errors model and the five-systematic er­ calculation error increases from 3.23 μm to 57.08 μm, and from Fig. 6 (d-
rors model at different heights. As shown in Fig. 6, the calculation errors f), when the sampling height increases from 0 to 300 mm, the maximum
of each contour are all close to zero at the same sampling height when calculation error increases from 0.48 μm to 2.27 μm.
the error level is level 1. For example, the maximum calculation error in
Fig. 6 (a) is 0.03 μm, and the maximum calculation error in Fig. 6 (d) is
4.2. The combined effect of integrated systematic errors, radius, and
0.009 μm, which shows that the measurement accuracy of the model in
height on measurement
this study is similar to the dual-systematic errors model and the five-
systematic errors model at level 1 –– this verifies the validity of the
The combined effect of the integrated systematic errors, radius, and
measurement model in this study.
height on the measurement, and the conditions are the same as those in
In addition, when the error level gradually increased from level 1 to
Section 4.1. The radius of the cylinder was in the range of 1–60 mm. The
level 5, the calculation error also gradually increased, and the calcula­
maximum calculation error represents the largest calculation error
tion error difference between adjacent error levels gradually increased.
among the models under the same error level. Fig. 7 (a-c) show the
As shown in Fig. 6 (a), when the error level is level 5, the calculated error
relationship between the maximum calculation error and the cylinder
is 3.23 μm, as shown in Fig. 6 (d); when the error level is 5, the calcu­
radius of the seven-systematic errors model and the dual-systematic
lated error is 0.48 μm. As shown in Fig. 6 (c), when the error levels are 3
errors model at different heights; while Fig. 7 (d-f) show the relation­
and 4, the maximum calculation error difference between them is
ship between the maximum calculation error and the cylinder radius of
15.35 μm; when the error levels are 4 and 5, the difference is 35.61 μm,
the seven-systematic errors model and the five-systematic errors model
as shown in Fig. 6 (f). When the error levels are 3 and 4, the maximum
at different heights.
calculation error difference between them is 0.51 μm. When the error
It can be seen from Fig. 7 that under the same radius and height, the
levels are 4 and 5, the difference is 1.40 μm, indicating that as the error
maximum calculation error increases with an increase in the error level.
level increases, both the dual-systematic errors model and the five-
As shown in Fig. 7 (b), when the radius is 1 mm and the error levels are 1
systematic errors model can produce larger errors, resulting in a
and 5, the maximum calculation errors are 0.24 μm and 154.40 μm

9
Y. Liu et al. Measurement 188 (2022) 110486

Fig. 7. The relationship between the maximum calculation error and the cylinder radius between the seven-systematic error model and the dual-systematic errors
model (the five-systematic errors model) at different heights.

Fig. 8. Coaxiality error evaluation simulation of the multi-systematic errors model.

respectively; as shown in Fig. 7 (e), when the radius is 1 mm and the from 0 to 300 mm, the maximum calculation error increases from
error levels are 1 and 5, the maximum calculation errors are 0.03 μm and 111.98 μm to 271.80 μm, and when the radius is 60 mm, the maximum
24.96 μm respectively. At the same height, as the radius increases, the calculation error stable value increases from 2.72 μm to 14.06 μm; it can
influence of the comprehensive systematic errors and the maximum be seen from Fig. 7 (d) (e) (f) that when the radius is 1 mm and the
calculation error decreases, and the difference in the maximum calcu­ sampling height increases from 0 to 300 mm, the maximum calculation
lation error between different error levels also gradually decreases, and error increases from 18.32 μm to 48.25 μm; and when the radius is
the maximum calculation error eventually tends to move toward a fixed 60 mm, the maximum calculation error value increases from 0.42 μm to
value. As shown in Fig. 7 (c), the maximum calculation errors of error 0.62 μm.
levels 4 and 5 at the radius of 1 mm are 55.61 μm and 271.80 μm In summary, the influence of the integrated systematic errors on the
respectively, and they are reduced to 8.48 μm and 14.06 μm at the radius contour measurement decreases with an increase in the contour radius
of 60 mm respectively; As shown in Fig. 7 (f), the maximum calculation and gradually tends move to a stable value, but the influence increases
errors of error levels 4 and 5 at the radius of 1 mm are 6.75 μm and with the increase in the sampling height simultaneously.
48.25 μm respectively, and they are reduced to 0.21 μm and 0.62 μm at
the radius of 60 mm respectively. Under the same radius and error level, 4.3. Coaxiality error evaluation simulation of multi-systematic errors
the maximum calculation error increases with an increase in the sam­ model
pling height. When the error level is level 5, it can be seen from Fig. 7 (a)
(b) (c) that when the radius is 1 mm and the sampling height increases To verify the feasibility and effectiveness of the seven-systematic

10
Y. Liu et al. Measurement 188 (2022) 110486

Fig. 9. Drawing of super-large coaxiality measuring device.

Fig. 10. Feature recognition method to solve βj.

error cylindrical profile measurement model in coaxiality measurement, increases from level 1 to 5, the accuracy of the coaxiality evaluation
a comprehensive simulation analysis was carried out based on the using the model in this study is significantly higher and more stable, and
simulation data. First, a set of stepped shaft profile data containing seven close to the theoretical value. For example, at level 4, compared with the
types of offset errors was generated according to Eq. (5). Then, the dual-systematic errors model and the five-systematic errors model, the
traditional dual-systematic errors model, five-systematic errors model, coaxiality evaluation accuracy of the large-sized stepped shaft is
and our model are used to separate the error parameters to complete the increased by 46.15% (1.32 μm) and 6.67% (0.11 μm), respectively, and
simulation of stepped shaft coaxiality. It is assumed that the upper and the coaxiality evaluation accuracy of the small-sized stepped shaft is
lower cylinders of the small-sized stepped shaft have radii of 20 mm and increased by 47.18% (1.09 μm) and 6.87% (0.09 μm), respectively. It
30 mm with a height of 450 mm and an ideal coaxiality of 1.21 μm, the can be seen that the multi-systematic errors separation method in this
upper and lower cylinders of the large-sized stepped shaft have radii of study can obtain relatively accurate coaxiality results when large-size
40 mm and 60 mm, respectively, with a height of 900 mm and an ideal stepped shafts are used.
coaxiality of 1.51 μm. Ten sections were uniformly sampled, and the
number of sampling points in each section was 1024. The average 5. Experiments
machining error Δrij is 0.001 mm which obeys the standard normal
distribution (μ = 0.001, σ = μ/3). According to the collected data, the The experiment was based on the large-scale stepped shaft profile
error level gradually increased from level 1 to 5. measurement device shown in Fig. 9. The precision air-bearing turntable
The coaxiality error represents the difference between the measured includes an air-bearing turntable and a centre tilting table. The radial
coaxiality value and the nominal coaxiality value of the standard step­ and axial accuracy of the air-bearing turntable is 38 nm. The centre
ped shafts. For the error separation of the dual-systematic errors model tilting table has an angle and displacement adjustment range of ± 0.5◦
and the five-systematic errors model, the method is similar to that of the and ± 3 mm, respectively. The displacement ranges of the horizontal
study. The dual-systematic errors can be eliminated by linearisation in and vertical guide rails are 800 mm and 2500 mm, respectively. The
the dual-systematic errors model, and the five-systematic errors can be resolution of the inductive sensor is 5 nm, the range ± 110 μm, and the
separated in a stepwise manner in the five-systematic errors model. maximum allowable error ± 0.11 μm. The sensor probe uses a ruby ball
From Fig. 8 (a) and (b), when the error level is low, the coaxiality under with a radius of 2.5 mm. A brand of industrial camera with a resolution
the three methods is close to the theoretical value, indicating that the of 2448 × 2048 was used for image acquisition. Two types of standard
error separation method in this study is effective. As the error level stepped shafts were selected for the measurement. The upper- and

11
Y. Liu et al. Measurement 188 (2022) 110486

Fig. 11. A set of measurement data of standard stepped shaft No. 1 and No. 2.

Table 5 Table 6
Coaxiality and cylindricity errors of No. 1 standard stepped shaft at error level Coaxiality and cylindricity errors of No. 2 standard stepped shaft at error level
from level 1 to level 5. from level 1 to level 5
Level ∙ ∙ ∙ ∙ ∙ ∙
Level ∙ ∙ ∙ ∙ ∙ ∙
Cyl/Cyl Cyl/Cyl Cyl/Cyl Coa/Coa Coa/Coa Coa/Coa Cyl/Cyl Cyl/Cyl Cyl/Cyl Coa/Coa Coa/Coa Coa/Coa
of 2SEM of 5SEM of 7SEM of 2SEM of 5SEM of 7SEM of 2SEM of 5SEM of 7SEM of 2SEM of 5SEM of 7SEM
(μm) (μm) (μm) (μm) (μm) (μm) (μm) (μm) (μm) (μm) (μm) (μm)

1 8.62/ 8.59/ 8.55/ 10.79/ 10.75/ 10.73/ 1 9.49/ 9.48/ 9.42/ 12.82/ 12.69/ 12.64/
0.12 0.09 0.05 0.09 0.05 0.03 0.09 0.08 0.02 0.22 0.09 0.04
2 8.70/ 8.63/ 8.58/ 11.09/ 10.80/ 10.75/ 2 9.69/ 9.55/ 9.45/ 13.14/ 12.81/ 12.66/
0.20 0.13 0.08 0.39 0.10 0.05 0.29 0.15 0.05 0.54 0.21 0.06
3 8.85/ 8.79/ 8.62/ 11.46/ 11.04/ 10.86/ 3 9.87/ 9.64/ 9.57/ 13.87/ 13.58/ 12.79/
0.35 0.29 0.12 0.76 0.34 0.16 0.47 0.24 0.17 1.27 0.98 0.19
4 9.81/ 8.99/ 8.77/ 12.83/ 11.71/ 11.21/ 4 10.68/ 10.31/ 9.76/ 15.64/ 14.10/ 13.05/
1.31 0.49 0.27 2.13 1.01 0.51 1.28 0.91 0.36 3.04 1.50 0.45
5 10.73/ 9.92/ 9.01/ 15.48/ 13.63/ 11.45/ 5 12.01/ 11.09/ 9.93/ 19.12/ 16.02/ 13.41/
2.23 1.42 0.51 4.78 2.93 0.75 2.61 1.69 0.53 6.52 3.42 0.81

lower-cylinder radii of standard stepped shaft No. 1 were 50 mm and Measurement data is taken in a metrology laboratory controlled envi­
60 mm, respectively, with a height of 1000 mm and coaxiality of ronment: 20 ◦ C ± 1 ◦ C (68◦ F ± 1.8◦ F), 65% ambient relative humidity,
10.70 μm. The upper-cylinder of standard stepped shaft No. 1 is used to draft free, and isolated from low frequency floor borne vibration. The
evaluate the cylindricity, and its cylindricity is 4.0 μm. The upper- and concentricity evaluation is in accordance with section 3.4, and the
lower-cylinder radii of the No. 2 standard stepped shaft were 80 mm and cylindricity evaluation is in accordance with the reference [9]. A set of
100 mm, respectively, with a height of 1500 mm and coaxiality of measurement data of standard stepped shaft No. 1 and No. 2 is shown as
12.60 μm. The upper-cylinder of standard stepped shaft No. 2 is used to Fig. 11.
evaluate the cylindricity, and its cylindricity is 9.5 μm. The material of In the experimental records, we abbreviated the seven-systematic
the standard stepped shafts was 38CrMoAl, and the quenching and errors model as 7SEM, the five-systematic errors model as 5SEM, the
tempering hardness was HBS200-220. The aging treatment was carried traditional dual-systematic errors model as 2SEM. The coaxiality value is
out after quenching and tempering, and the surface nitriding hardness represented by Coa, the coaxiality error value is represented by Coaʹ; the
was HRC66-65, which must be placed vertically after processing. cylindricity value is represented by Cyl, the cylindricity error value is

12
Y. Liu et al. Measurement 188 (2022) 110486

Table 7 errors model (5SEM), and uses the same multi-systematic errors sepa­
Analysis of the coaxiality and cylindricity measurement uncertainty. ration method. The coaxiality error represents the difference between
No. Sources of uncertainty Standard Sensitivity Uncertainty the measured coaxiality value and the nominal coaxiality value of the
uncertainty u coefficient component standard stepped shafts. The parameter values corresponding to each
(xi) (μm) ci ui(y) (μm) error level are shown in the Table 4. The measurement results are listed
1 The standard deviation 0.069 1 0.069 in Table 5 and Table 6, respectively. The measurement uncertainty of
of coaxiality coaxiality and cylindricity is shown in Table 7. From the first set of data
measurement in Table 5 and Table 6, it can be seen that when the eccentricity, tilt
repeatability is 0.17 μm
2 The standard deviation 0.057 1 0.057
angle, probe offset, and rod tilt angle are relatively small, the coaxiality
of cylindricity accuracy of the dual-systematic errors model, the five-systematic errors
measurement model, and the seven-systematic errors model are close to the theoretical
repeatability is 0.14 μm value, and the measurement error is low. This indicates that our model
3 The resolution of 0.001 1 0.001
and error separation method of this study are effective. When the ec­
inductive sensor is 5 nm
4 Maximum permissible 0.064 1 0.064 centricity, tilt angle, probe offset, and rod tilt angle gradually increase,
error of probe the measurement accuracy of the other two models decreases. The
is ± 0.11 μm maximum coaxiality and cylindricity deviation caused by the models, as
5 Radial error of turntable 0.098 1 0.098 seen in Table 5, is 4.78 μm and 2.23 μm respectively, and the maximum
is (0.038 + 0.0003H) μm
6 Axial error of turntable 0.011 1 0.011
coaxiality and cylindricity deviation, as seen in Table 6, is 6.52 μm and
is 0.038 μm 2.61 μm respectively. However, compared with the dual-systematic er­
7 The vertical rail error is 0.154 1 0.154 rors model and the five-systematic errors model, the coaxiality and
0.533 μm cylindricity measurement accuracy of the seven-systematic errors model
8 The horizontal rail error 0.087 1 0.087
was significantly improved. For example, in the fourth set of data in
is 0.301 μm
9 The maximum allowable 0.040 1 0.040 Table 5, the coaxiality measurement accuracy of the seven-systematic
error of the two standard errors model is improved by 1.62 μm relative to the dual-systematic
stepped shafts errors model, and 0.50 μm relative to the five-systematic errors model;
is ± 0.07 μm the cylindricity measurement accuracy of the seven-systematic errors
Coaxiality combined standard uncertainty uc(μm) 0.227
model is improved by 1.04 μm relative to the dual-systematic errors
Cylindricity combined standard uncertainty 0.224
uc(μm) model, and 0.22 μm relative to the five-systematic errors model. We use
Coaxiality expanded uncertainty U(μm) 0.45(k = 2) percentages to reflect the increase in model accuracy. The percentage
Cylindricity expanded uncertainty U(μm) 0.45(k = 2) value of the increase in accuracy corresponds to the value obtained by
dividing the error value between the seven-systematic errors models and
the compared errors models by the compared errors model. In the fifth
represented by Cylʹ. Among them, the coaxiality error and cylindricity
set of data in Table 5, the coaxiality measurement accuracy of the seven-
error are calculated as:
systematic errors model is improved by 26.03% (4.03 μm) relative to the
{
Coa’ = Coa − Ċoa dual-system error model, and 15.99% (2.18 μm) relative to the five-
(22)
Cyl’ − Ċyl system model; the cylindricity measurement accuracy of the seven-
systematic errors model is improved by 16.03% (1.72 μm) relative to
where Ċoa, Ċyl represent the nominal values of coaxiality and cylin­ the dual-system error model, and 9.17% (0.91 μm) relative to the five-
dricity. We use percentages to reflect the increase in model accuracy. system model. In the fifth set of data in Table 6, the accuracy of the
The percentage value of the increase in accuracy corresponds to the seven-systematic errors model is improved by 29.86 % (5.71 μm) rela­
value obtained by dividing the error value between the seven-systematic tive to the dual-systematic errors model, and 16.29 % (2.61 μm) relative
errors models and the compared errors models by the compared errors to the five-systematic errors model; the cylindricity measurement ac­
model. The equation is expressed as follows: curacy of the seven-systematic errors model is improved by 17.32%
{ (2.08 μm) relative to the dual-system error model, and 10.46%
PCoa = |Coa1 − Coa2/(Coa2)*100%|
(23) (1.16 μm) relative to the five-system model. Therefore, the accuracy of
PCyl = |Cyl1 − Cyl2/(Cyl2)*100%|
the coaxiality measurement is effectively improved, by using the seven-
systematic errors model and the multi-systematic errors separation
where pCoa and pCyl represent the relative value of concentricity and
method proposed in this study.
cylindricity. Coa 1 and Coa 2 represent the coaxiality value to be eval­
uated and the coaxiality value to be compared. The expressions of Cyl 1
and Cyl 2 are similar. 6. Conclusions
First, the verification parameters were measured. The tilt angle of the
horizontal rail was 0.1◦ , and the tilt angle of the vertical rail was 0.2′′ For the model engineering of the stepped shaft, this paper designs a
measured by the autocollimator. The tilt angle of the probe rod may seven-systematic errors cylindrical profile model with eccentricity error,
change in each measurement, and image processing is used to solve the tilt error, sensor probe radius error, probe offset error, probe support rod
problem in real time. The results are shown in Table 5 and Table 6. In tilt error, and horizontal and vertical rail tilt error, and improves the
this study, five sections were sampled at equal intervals in the upper and accuracy of the model engineering of the stepped shaft by improving the
lower sections of standard stepped shafts. The number of sampling accuracy of the profile data. At the same time, based on the seven-
points in each section was 1024, and the sampling height was between systematic cylindrical profile measurement model, a multi-systematic
0 and 900 mm. The angle βj between the major axis direction of the fitted errors separation method is proposed. First, the autocollimator and
ellipse and the initial measurement direction is obtained by the feature image processing are used to accurately extract the verification pa­
recognition method described in Section 3.2. As shown in Fig. 10, the rameters, and then the stepwise estimation method and the equalisation
initial measurement direction determines the relative relationship of βj, optimizer (EO) are used to achieve the separation of the sectional and
and the height also significantly affects the size of βj. spatial parameters.
The experiment compares our proposed model (7SEM) with the Based on the simulation parameters, we conducted a simulation
traditional dual-systematic errors model (2SEM) and the five-systematic analysis on the measurement models and the error separation methods.
The simulation experiment clearly reflects the influence of

13
Y. Liu et al. Measurement 188 (2022) 110486

comprehensive error on different measurement models, and confirms [2] C. Lu, J. Wang, P. Yin, L. Wang, Error identification of measurement software based
on digital twin of gear measuring center, Measurement 173 (2021) 108666,
the necessity of systematic errors separation. Based on the large-scale
https://doi.org/10.1016/j.measurement.2020.108666.
stepped shaft profile measurement device and two types of standard [3] S. Dai, G. Zhao, Y. Yu, P. Zheng, Q. Bao, W. Wang, Ontology-based information
stepped shafts to verify the validity of the measurement model, the modeling method for digital twin creation of as-fabricated machining parts, Robot
coaxiality indicator is used to evaluate the profile measurement accu­ Comput.-Integr. Manuf. 72 (2021) 102173, https://doi.org/10.1016/j.
rcim.2021.102173.
racy of the large-scale stepped shaft, and our model is compared with the [4] J.Y. Chen, Accelerate lithography improvement for high performance computing,
dual-systematic errors model and the five-systematic errors model. The Int. Conf. Extreme Ultrav. Lithogr. 2018; 10809.
experimental results show that compared with the other two methods, [5] F. Xiao, F. Qian, B. Su, Y. Fan, Digital Twin of Solid Rocket Motor, Problem and
Challenge, 2018 11th ISCID 2018; 7-11.
the coaxiality measurement accuracy of standard stepped shaft No. 1 can [6] H. Cui, D. Lei, X. Zhang, H.e. Lan, Z. Jiang, L. Kong, Measurement and analysis of
be increased by 26.03% (4.03 μm) and 15.99% (2.18 μm), respectively; the radial motion error of aerostatic ultra-precision spindle, Measurement 137
the cylindrici-ty measurement accuracy of standard stepped shaft No. 1 (2019) 624–635.
[7] L. Chen, B. Li, Z. Jiang, Inspection of assembly error with effect on throat and
can be increased by 16.03% (1.72 μm) and 9.17% (0.91 μm), respec­ incidence for turbine blades, J. Manuf. Syst. 43 (3) (2017) 366–374.
tively; the coaxiality measurement accuracy of the No. 2 standard [8] R. Paul, S. Anand, Optimal part orientation in Rapid Manufacturing process for
stepped shaft can be increased by 29.86% (5.71 μm) and 16.29% achieving geometric tolerances, J. Manuf. Syst. 30 (4) (2011) 214–222.
[9] C. Sun, C. Li, Y. Liu, H. Wang, B. Wang, X. Wang, J. Tan, A cylindricity evaluation
(2.61 μm), respectively; the cylindricity measurement accuracy of approach with multi-systematic error for large rotating components, Metrologia 57
standard stepped shaft No. 2 can be increased by 17.32% (2.08 μm) and (2) (2020) 025020, https://doi.org/10.1088/1681-7575/ab55c4.
10.46% (1.16 μm), respectively. It can be shown that the reliability of [10] D.G. Chetwynd, Roundness measurement using limacons, Precis. Eng. 1 (3) (1979)
137–141.
the seven-systematic errors cylindrical profile model and multi-
[11] T.S.R. Murthy, S.Y. Rao, J. Peters, A simple approach for evaluation of cylindrical
systematic errors separation method proposed in this paper, which surfaces, CIRP Ann.-Manuf. Technol. 30 (1) (1981) 441–444.
effectively improves the accuracy of the cylindrical profile measurement [12] H. Haitjema, H. Bosse, M. Frennberg, A. Sacconi, R. Thalmann, International
of the stepped shaft, and provides an accurate data source for the model comparison of roundness profiles with nanometric accuracy, Metrologia 33 (1)
(1996) 67–73.
engineering of the stepped shaft. [14] M.A.A. Morel, Uncertainty Estimation of Shape and Roughness Measurement,
It should be noted that for the large-scale stepped shaft model en­ University of Technology Eindhoven, 2006.
gineering, we used a contact rotary measuring device to sample and [15] C. Sun, H. Wang, Y. Liu, X. Wang, B. Wang, C. Li, J. Tan, A cylindrical profile
measurement method for cylindricity and coaxiality of stepped shaft, Int. J. Adv.
evaluate the surface profile of the stepped shaft. Although its accuracy is Manuf. Technol. 111 (9-10) (2020) 2845–2856.
higher than that of line lasers and surface lasers, and the measurement [16] H. Haitjema, 2020 Straightness, flatness and cylindricity characterization using
uncertainty is smaller, its measurement efficiency is relatively low for discrete Legendre polynomials, CIRP Ann.-Manuf. Technol. 69 (1) (2020) 457–460.
[17] J. Tan, Several hot issues in the research of ultra-precision measurement
large-size surface profile because of a point contact. At the same time, in technology and instrument engineering, China Mech. Eng. 3 (2000).
terms of the measurement model, it is currently explained from the [18] D.J. Whitehouse, Handbook of Surface and Nanometrology, Institute of Physics
overall errors of the sub-system of the device and does not consider some Publishing, 2002, 882–934.
[19] W. Liu, J. Fu, B. Wang, S. Liu, Five-point cylindricity error separation technique,
uncertain factors in the rotation process, and there is a lack of more Measurement 145 (2019) 311–322.
detailed analysis of the error sources. Therefore, the measurement [20] B.o. Chen, X. Zhang, H. Zhang, X. He, M. Xu, Investigation of error separation for
model still needs to be refined in the future, and the corresponding error three dimensional profile rotary measuring system, Measurement 47 (2014)
627–632.
separation methods also need to be further adjusted.
[21] K.P. Anandan, O.B. Ozdoganlar, A multi-orientation error separation technique for
spindle metrology of miniature ultra-high-speed spindles, Precis. Eng. 43 (2016)
CRediT authorship contribution statement 119–131.
[22] W. Liu, K. Fan, P. Hu, Y.i. Hu, A parallel error separation method for the on-line
measurement and reconstruction of cylindrical profiles, Precis. Eng. 51 (2018) 1–9.
Yongmeng Liu: Conceptualization. Yingjie Mei: Writing – original [23] K. Shi, C. Liu, J.D. Biggs, Z. Sun, X. Yue, Observer-based control for spacecraft
draft. Chuanzhi Sun: Methodology. Ruirui Li: Software. Xiaoming electromagnetic docking, Aerosp. Sci. Technol. 99 (2020) 105759, https://doi.org/
Wang: Investigation. Hongye Wang: Supervision. Jiubin Tan: Formal 10.1016/j.ast.2020.105759.
[24] Z. Chai, Y. Lu, X. Li, G. Cai, J. Tan, Z. Ye, Non-contact measurement method of
analysis. Qing Lu: Writing – review & editing. coaxiality for the compound gear shaft composed of bevel gear and spline,
Measurement 168 (2021) 108453, https://doi.org/10.1016/j.
Declaration of Competing Interest measurement.2020.108453.
[25] Y. Liu, G. Li, H. Zhou, Z. Xie, F. Feng, W. Ge, On-machine measurement method for
the geometric error of shafts with a large ratio of length to diameter, Measurement
The authors declare the following financial interests/personal re­ 176 (2021) 109194, https://doi.org/10.1016/j.measurement.2021.109194.
lationships which may be considered as potential competing interests: [26] F. Afshin, M. Heidarinejad, B. Stephens, S. Mirjalili, Equilibrium optimizer: A novel
optimization algorithm, Knowl.-Based Syst. 191 (5) (2020), 105190.
Liu Yongmeng reports financial support was provided by National
Natural Science Foundation major research pro-jects of China (grant
Further reading
number 91960109).
[13] J. Gu, J.S. Agapiou, S. Kurgin, Error compensation and accuracy improvements in
Reference 5-axis machine tools using the global offset method, J. Manuf. Syst. 44 (2) (2017)
324–331.
[1] L. Zhang, L. Zhou, B.K.P. Horn, Building a right digital twin with model
engineering, J. Manuf. Syst. 59 (2021) 151–164.

14

You might also like