Research on Radii of Curvature Measurement for Micro-accessory in

Precision and Ultra-precision Machining

Tian Xinli, Wang Jianquan∗, Zhang Baoguo, Tang Xiujian, Li Fuqiang
National Key Laboratory for Remanufacturing, No.21 Dujiakan, Chang Xindian, Beijing, China
100072

ABSTRACT

To measure the radii of curvature (ROC) for micro-accessories in precision and ultra-precision with high accuracy, the
thesis put forward a new technology based on digital images of scanning electron microscope (SEM), in which the
contour line of specimen was attained and the ROC was calculated after data fitting. The Canny edge detection and some
other image processing techniques were successively adopted to extract the edge profile of single diamond particle. Then
a high order polynomial was used in view of the least square law to fit the sampling point coordinates of contour line
nonlinearly. Lastly, the ROC could be computed according to the metric ruler of SEM and the proposed formula. The
measurement result shows that the deviation between the technology and current methods is no more than 10%, as the
magnification rate of SEM, the amount of sampling points on contour line and the order of fitting model are 1200, 80, 20
respectively. Besides, the measurement accuracy can reach a nanometer scale. This research also indicates that it is
suitable for the ROC measurement of non-optical small parts or other micro-accessories with high feasibility and
application value.

Keywords: precision and ultra-precision machining, radius of curvature (ROC), image processing, edge detection, data
fitting, micro-accessory

1. INTRODUCTION

With the development of precision and ultra-precision machining and detecting technologies, the measurement of radii of
curvature (ROC) has become a hot issue for various tiny spherical or aspherical surfaces1, the pinpoint of AFM probes2,
and the detector of contact profilometer3. The knife-edge method, spherometer technology and template method4, 5 are
chief means in measuring of large ROC of optical components, which are not suitable for tiny ROC measurement owing
to a few limitations. A new technique using an interferometer and laser tracker is invented to measure the ROC of
concave mirror with the diameter of 0.5m by Arizona university6. The Linnik interferometer is applied to investigate the
ROC of precision sphere in another research7. However, these methods have not been verified in the ROC measurement
of non-optical micro-accessories, since some complicated and precise optical measurement system is essential. A special
high-accuracy technology for ROC is demanded as being short of correlative equipments. Therefore, this thesis puts
forward a novel method to characterize the ROC of tiny parts and micro-accessories with the field emission scanning
electron microscope (FE-SEM).
The ROC of diamond particle has a remarkable impact on the grinding efficiency and surface quality during the
precision machining of hard brittle materials such as engineering ceramic and glass. Taking the single diamond particle
for instance, this research detects the contour line with specific algorithm after preprocessing of SEM images.
Afterwards, a nonlinear polynomial fitting of coordinate data is carried out, which provides a preparation for the
following ROC calculation. And the measuring results indicate the absolute feasibility of the new technology.

2. EXPERIMENT

2.1 Image acquisition

The current image acquisition systems have some a series of technical deficiencies while gathering clean-cut digital
images. The high resolution FE-SEM of model Nova NanoSEM 450 is set as the image acquisition device in this

∗
[email protected]; phone 1 340 015-5353

6th International Symposium on Advanced Optical Manufacturing and Testing Technologies: Design, Manufacturing, and
Testing of Smart Structures, Micro- and Nano-Optical Devices, and Systems, edited by T. Ye, X. Luo, S. Hu, X. Bao, Y. Li,
Proc. of SPIE Vol. 8418, 841817 · © 2012 SPIE · CCC code: 0277-786/12/$18 · doi: 10.1117/12.966405

Proc. of SPIE Vol. 8418 841817-1

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms

 research. The diamond particle brazed on a cylindrical matrix material is placed horizontally in the specimen chamber of
SEM. And the central axis of observed sample should be strictly parallel with the objective table in chamber. The result
shows that the SEM images have an advantage over the pictures acquired by digital microscope and Olympus
metallurgical microscope. Moreover, the details of diamond contour can be easily obtained through SEM observation
when the experiment conditions are right.
2.2 Edge detection
The edge is one of the most important and simple features for digital image8. The object of edge detection must be
grayscale image, thus the information about the hue and position should be deleted for JPEG or TIF images captured by
SEM. The grayscale value Y of a specific pixel can be calculated using the equation (1).
Y=0.299Rij+0.587Gij+0.114Bij (1)
Where Rij, Gij, Bij are the RGB values of a pixel (i, j) in SEM image.
There will be multiple interferences and noises inevitably during the formation of SEM image. Consequently, the
median filtering technique should be employed to reduce the noise level and smooth images without local details loss.
The median filtering model G can be defined as following9.
G (i, j ) = median[ f (i, j )] (2)
( i , j )∈A

Where A is the used template matrix, f(i, j) is the grayscale value of pixel (i, j).
Additionally, the histogram equalization based on cumulative distribution function can be utilized to resolve the
matter of small dynamic range. It can get a uniform distribution of grayscale value from 0 to 255, which makes the image
seem much more distinct and lively.
According to the former achievements, it is difficult to gain smooth and fine contour profile from edge detection
with first order derivative and second order derivative10. The Canny algorithm expresses its 3 criterions in mathematical
language, and searches the edge line in the light of local maximum value of image gradient, which has some prominent
characteristics such as high ratio of single to noise and single edge response. Fig.2 shows the edge detection results of
several typical differential operators for single diamond particle. It can be seen from the figure that the Canny algorithm
is very effective and exact to get the contour outline of tiny accessories.

(a) Sobel algorithm (b) LOG algorithm (c) Canny algorithm

Figure 2: Image contour lines created by several typical detection algorithms.
2.3 Curve fitting and ROC calculating
It is not hard to find that the profile of diamond particle point presents a nonlinear feature. The power function model,
exponential function model, logarithmic function model, trigonometric function model and their compound form are
often used to fit and approach the intended data sets. Nevertheless, most models can be converted into high order
polynomials in Taylor series expansion. So that the research employs equation (3) to fit the sampling coordinates of
diamond particle contour line, which has been preprocessed before.
y = f ( x) = c0 + c1 x + c2 x 2 + L + cn x n (3)
Where ci is the coefficient of fitting curve.
n
The mean square error ε = ∑ ( y − yˆ ) between fitting model and actual profile curve achieves the minimum
2
i i
i =1

while the least square method is applied. Thus this theory may be perfect for characterize the profile of single diamond
particle. Later, the curvature k of each point on fitting curve along the tangent can be computed using equation (4), which
also is the reciprocal of the minimal ROC Rf.

Proc. of SPIE Vol. 8418 841817-2

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms

 ⎧ y′′
⎪k =
(1 + y′2 ) (4)
3/2
⎨
⎪
⎩ Rf = 1/ k
The Rf is just the curvature of fitting function according to the above expression, but not the real ROC of diamond
particle. There is a definitive relationship between the calculated ROC and actual ROC. The SEM provides accurate
measuring ruler for every image, and then the amount of pixels np at certain magnification ratio and the corresponding
size of one pixel in fitting curve f(x) are investigated. The relevant relationship on actual ROC Ro and Rf can be expressed
by equation (5).
Rl
Ro = f r (5)
Nn p
Where lr is the actual size of metric ruler in SEM image.

3. INFLUENCING FACTORS ANALYSIS

Before the image acquisition and successive image processing, we should use the ultrasonic cleaner to clear away all
superficial foreign matters of diamond particle. The image is going to be separated to two parts that contain diamond and
its background after grayscale converting, image enhancing and smoothing. Then, two threshold of Canny algorithm are
set as T1=80 and T2=130 based on the edge detection effect. To give facilities for observation and sampling on contour
outline, the image should be color inverting that makes the background be white and the diamond profile be black line.
After that, the coordinate sampling is implemented point by point along the profile line before the nonlinear polynomial
fitting, and the ROC of diamond particle can be gained in the light of equation (4) and (5). The experiment indicates that
the measuring precision is able to achieve the nanometer and even sub-nanometer level. Therefore, this method has a
considerable precision in the measurement of single diamond particle. The concrete procedures and results of the method
are shown in Fig. 3.

a I mage acquisition b Image enhancing c Canny edge

or,a t
f Calculation of e Curve fitting of d Color inverting and
micro -ROC sampling points coordinate sampling
Figure 3: Schematic on the measurement of ROC for diamond particles.
It seems that the magnification ratio of SEM image, the distribution and number of sampling points on the contour
outline, and the polynomial order of fitting function have a significant impact on the ROC measurement result of micro-
accessories according to repeated experiments. So the detailed study should be done upon those relationships which may
affect the determination of measuring parameters.
3.1 Magnification ratio of SEM rm
An improper magnification ration will degrade the image acquisition effect directly and markedly while measuring the
ROC of single diamond particle. The research discovers that it can attain profile-complete and clear-cut images when the
SEM magnification ranges from 300 to 1500. First of all, the ROCs of three types of diamond particle are measured with

Proc. of SPIE Vol. 8418 841817-3

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms

 the universal tool microscope by the Photoshop method. And the measuring results are 2μm, 5μm, 10μm respectively.
The relation between different magnifications and diamond particles is revealed in Fig. 4.
200 400 600 800 1000 1200 1400
14 14

1# Diamond Particle
K.:,,\,2#
2# Diamond Particl e 12
U
O 3# Diamond Particle
10 10
®

a8 a

14
2
2 i ya A. .. .. .. _: 5.2 2
200 400 600 800 1000 1200 1400 1600
Magnification of SEM n,

Figure 4: Measured ROC value of diamond particles at different magnification.

It can be seen from Fig. 4 that there is a large deviation between calculated value and actual value of ROC while the
magnification of SEM varies from 300 to 600. As the magnification ratio rises, the measuring value goes constant and
the calculative deviation is less and less. The deviation reaches least at rm=1200, and the margin of SEM image becomes
indistinct and virtual with the continuous enlargement of magnification. Meanwhile, the loss of contour details is on the
increase and the calculative deviation rises again. Thus rm=1200 is defined as the optimal magnification ratio and image
acquisition parameter. Table 1 shows the calculative deviation of ROC for diamond particles at diverse magnification.
Table 1: The measurement deviation for ROC value of diamond particles at different magnification.
Sample rm=300/(%) rm=600/(%) rm=800/(%) rm=1200/(%) rm=1500/(%)
1# 77.05 43.95 10.15 9.35 12.5
2# 58.86 19.42 8.04 1.62 3.12
3# 20.99 17.81 8.54 2.7 1.32

3.2 Distribution and number of sampling point NS on contour line

The ginput command of Matlab allows users to sample the coordinate of points in a figure positioning the cursor with the
mouse, and discretizes the contour lines into points in the Cartesian coordinate system. The distribution and amount of
sampling points on the line will affect the fitting result greatly. Fig. 5 shows a typical profile line of diamond particle that
resembles a parabola. The minimal ROC of this line involves in section Ⅰ, and more sampling points should locate in
this key area. There is just a small variation of ROC in section Ⅱ and Ⅲ, thus we can set less sampling points in these
parts. When the ratio of sampling number equals to 3:1:1, the fitting curve is much closer to the original contour line, and
the measuring deviation reaches the least.

Sampling Sampling Contour

Point Cursor Line

Figure 5: Distribution of sampling points on contour outline of diamond particle.

The number of sampling points NS is related to the consuming time and measuring precision. That is, the lesser
sampling number can not represent the practical profile of researching object, and the calculated ROC of fitting curve
may be deviate from the actual ROC heavily. The larger sampling number would also prolong the time of data fitting and
curvature computing. Fig.6 shows the relation about different sampling number NS and ROC of a diamond particle.

Proc. of SPIE Vol. 8418 841817-4

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms

 20 40 60 80 100 120 140 160 180 200
50 50

45
t -0- 1# Diamond Particle 45

Ea f 2# Diamond Particle 40
-A- 3# Diamond Particle
V 35 35
in
E 30 30
Ó -

y 25 25
=
20 20

= 15_ 15

10
OS
CD
5
Ñ

20 40 60 80 100 120 140 160 180 200

Sampling Number NS

Figure 6: Relationship between sampling number NS and measured ROC of diamond particles.
The Fig. 6 indicates that there is a threshold value of sampling number for high efficiency and measuring accuracy.
As the sampling number is small, the calculative deviation seems large than any time. But when the number rises to NS′,
the deviation decreases and almost remains unchanged. At the same time, the consuming time and measuring accuracy
achieve a good balance. According to the Fig. 5, we can consider that the most suitable number of sampling point may be
NS=80.
3.3 Order of fitting model m
The determination of the order of fitting model is a key problem for this new technology, since it does not only affect the
approximation level of original sampling points, but also it is the foundation of precise ROC calculation. Generally,
higher order of fitting model is in favor of getting exact fitting result and greater workload. Fig. 7 shows the fitting
results of several typical polynomial function for 2# diamond particle as rm=1200 and NS=80.
220
Sampling Points
200 0 2 Orders Fitting Model
-O -3 Orders Fitting Model
180 -A -5 Orders Fitting Model
io< 160 -v-10 Orders Fitting Model
T
45 140
-4 -20 Orders Fitting Model
± -50 Orders Fitting Model
d 120
c
100
ó
80

60

40

150 200
Coordinate of X -axis

Figure 7: Curve fitting results of sampling points for diamond particle by several polynomial functions.
The fitting results accords with the original sampling data satisfactorily as the order of fitting polynomial is below
20. Contrarily, the average error of fitting curves and practical profile cannot be accepted from the Fig. 7. It indicates that
higher order of fitting model may be not appropriate for the data fitting during the ROC measurement. In order to judge
the degree of fitting, the research applies the coefficient of determination R2, the modified coefficient of determination R2 ,
the sum of the squares of errors SSE, and root of mean square errors RMSE to test the goodness of all fitting models. The
test results are shown in Table 2.
Table 2: The goodness of test for several typical polynomial functions.
Order of fitting model R2 test R 2 test SSE RMSE
2 0.9871 0.9865 1196 5.274
3 0.9876 0.9867 1152 5.237
5 0.9988 0.9986 111.7 1.671
10 0.9996 0.9995 35.38 1.005
20 0.9998 0.9997 16.25 0.8062
50 -1.498 -4.621 23140 107.6
With the ascendance of order of fitting polynomial, the R2 and R
2
increase slowly but the SSE and RMSE. When

Proc. of SPIE Vol. 8418 841817-5

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms

 the order of fitting function equals to 20, the R2 and R2 reach their maximal values, which means that the fitting effect is
best of all for 2# diamond particle. The similar experiments done upon 1# and 3# diamond particle also prove the
identical pattern in measurement of ROC. Consequently, the 20 orders nonlinear polynomial is determined to be the
optimal fitting model in this measuring technology.

4. CONCLUSIONS

The precise measurement of ROC for micro-accessories is always an important subject for industrial application. The
thesis uses the image processing technologies to extract the contour outline of micro-accessory ground on the accurate
scaling and measuring functions of FESEM. And the nonlinear polynomial fitting is introduced to fit the coordinate sets
of contour line with the theory of least square law. The actual ROC of research object can be calculated according to the
equations posed in the study. Furthermore, the thesis takes single diamond particle as example, and investigates deeply
the relationship between the observing magnification, the number of sampling points, the order of fitting model and the
measurement results. The new measuring technology shows quite high precision and prominent advantages comparing
with current methods for micro-ROC. The next stage of this research is about to develop more efficient and accuracy
edge detection algorithms, as well as realize the smart and instrumented operation in the ROC measuring of micro-
accessories.

ACKNOWLEDGMENT

This work was financially supported by the National Natural Science Foundation of China (51075399, 51105378), and
the National High Technology Research and Development Program of China (SQ2008AA04XK1478977).

REFERENCES

1. L. M. Yang and H. X. Ye, High-precision metrology for optical components with large-apertures and large radii of
curvature, Optics and Precision Engineering, 2011, 19(6): 1207-1212.
2. Y. X. Xu, L. Chen and R. H. Zhu, et al., Study on the measurement of radii of curvature of mini-spheres, Chinese
Journal of Scientific Insturment, 2006, 27(9): 1159-1162.
3. Z. J. Zhang, F. Du and X. Jin, et al., Complex edge recognition algorithm of micro-accessory, Optics and Precision
Engineering, 2009, 17(2): 356-361.
4. Y. Y. Yang and J. Chen, A novel real-time testing system used for large curvature radius, Opto-Electronic
Engineering, 2008, 3: 29-33.
5. Q. Wang and X. H. Xu, Precision analysis of measurement of radius of curvature by interferometer, Journal Of
Southeast University (Natural Science Edition), 2009, 39(Sup.II): 55-59.
6. C. Y. Zhao, R. Zehnder and J. H. Burge, Measuring the radius of curvature of a spherical mirror with an
interferometer and laser tracker, Optical Engineering, 2005, 44(9): 090506.
7. T. Narumi, T. Nakamura and Y. Ichihara, et al., Measuring system for micro radius of curvature, Proc. SPIE:
Optomechatronic Systems III, 2002, 4902: 262-269.
8. J. L. Zhou, X. Q. Zhan and K. He, et al., Image edge detection based on statistical estimation, Journal of Shenyang
University of Technology, 2010, 32(6): 665-671.
9. Q. Q. Ruan. Foundation of digital image processing, Tsinghua University Press, Beijing, 2009.
10. S. C. Deng, T. G.Liu and Z. X. Xiao, Extraction of enclosing image edge for metallographic structure based on
Canny and grayscale contour line, Optics and Precision Engineering, 2010,18(10): 2314-2323.

Proc. of SPIE Vol. 8418 841817-6

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/17/2013 Terms of Use: http://spiedl.org/terms