July 10, 2009 / Vol. 7, No. 7 / CHINESE OPTICS LETTERS 627
Study of weighted space deconvolution algorithm in
computer controlled optical surfacing formation
Hongyu Li (李李李洪洪洪玉玉玉)
1∗
, Wei Zhang (张张张 伟伟伟)
1
, and Guoyu Yu
2
1
Research Center for Space Optical Engineering, Harbin Institute of Technology,
Harbin 150001, China
2
National Facility of Ultra Precision Surfaces, OpTIC Technium, Ffordd William Morgan,
StAsaph Business Park, North Wales, LL17 0JD, UK
∗
E-mail: hongyuli@hit.edu.cn
Received October 6, 2008
Theoretical and experimental research on the deconvolution algorithm of dwell time in the technology of
computer controlled optical surfacing (CCOS) formation is made to get an ultra-smooth surface of space
optical element. Based on the Preston equation, the convolution model of CCOS is deduced. Considering
the morbidity problem of deconvolution algorithm and the actual situation of CCOS technology, the
weighting spatial deconvolution algorithm is presented based on the non-periodic matrix model, which
avoids solving morbidity resulting from the noise induced by measurement error. The discrete convolution
equation is solved using conjugate gradient iterative method and the workload of iterative calculation in
spatial domain is reduced effectively. Considering the edge effect of convolution algorithm, the method
adopts a marginal factor to control the edge precision and attains a good effect. The simulated processing
test shows that the convergence ratio of processed surface shape error reaches 80%. This algorithm is
further verified through an experiment on a numerical control bonnet polishing machine, and an ultra-
smo oth glass surface with the root-mean-square (RMS) error of 0.0088 µm is achieved. The simulation
and experimental results indicate that this algorithm is steady, convergent, and precise, and it can satisfy
the solving requirement of actual dwell time.
OCIS co des: 240.0240, 220.0220.
doi: 10.3788/COL20090707.0627.
More and more aspheric surfaces are used in optical sys-
tems. Due to the unique geometric shape of aspheric
surfaces, the manufacture difficulty and cost of aspheric
surfaces are greatly higher than that of spherical surface
optical elements
[1,2]
. A breakthrough of modern space
optical element processing technique is computer con-
trolling “small grinding tool” to polish the element. The
processing technology is often referred to as computer
controlled optical surfacing (CCOS) technology, namely
computer-controlled optical surface shaping technology.
CCOS technology is limited to the computer controll of
small grinding tool for polishing, the processing princi-
ple is based on the Preston hypothesis, and the polishing
process can be modeled. So this method is also known as
the deterministic polishing technique
[3]
. Currently, ex-
cept computer-controlled polishing (CCP), the develop-
ing CCOS technology includes many representative tech-
nologies, such as the dynamic stressed lap polishing tech-
nology, magnetorheological finishing technology
[4]
, and
ion beam polishing technology
[5]
. The common charac-
teristics of these technologies are a computer-controlled
“small grinding tool” polishing on the surface of optical
components. In the processing of large-scale optical com-
ponents, particularly in non-spherical optical parts pro-
cessing, these CCOS technologies are increasingly being
applied. Bonnet p olishing with novel polishing tool and
special motion trait is a high precision and efficient opti-
cal component machining method, especially for aspheric
surface machining
[6−8]
. It is a new polishing method pre-
sented by London Optical Science Laboratory and has an
extensive application prospect.
In CCOS technology, the solving algorithm of dwell
function is the key factor affecting the processing ef-
ficiency and error convergence. Based on the Preston
equation, the material removal amount can be derived
from the convolution between dwell time and removal
function. Thus solving the dwell function is a deconvo-
lution essentially. But the deconvolution solution is al-
ways a pathological problem. The popular algorithms
of solving dwell time function are iterative algorithm
and Fourier transform
[9]
. Iterative algorithm adopts ap-
proaching method in numerical calculation, but the dis-
advantage is that the iterative convergence rate is slow.
Furthermore, this approach will oscillate without con-
vergence under certain circumstances. Fourier transform
method changes the convolution to product operation by
Fourier transform, and then the inverse Fourier transform
can be done to obtain the final result. The computation
amount is smaller than that of the iterative method, but
it is difficult for Fourier transform when the removal func-
tion is close to zero, and the results need special treat-
ment, especially it cannot guarantee convergence. The
algorithm to solve dwell time function based on different
optical processing methods has been researched in recent
years. For example, the distribution of dwell time was
obtained according to linear algebra and Tikhonov regu-
larization by Deng et al.
[10]
, and a discrete linear model
was established by Zhou et al. to analyze the relationship
between the dwell time and removal amount by L-curve,
further to calculate the dwell time
[11]
. These two meth-
ods changed the fabrication process from convolution to
matrix product based on linear algebra to solve the ill-
1671-7694/2009/070627-05
c
° 2009 Chinese Optics Letters