14th Int. Symp. on Nondestructive Characterization of Materials (NDCM 2015) - www.ndt.net/app.

NDCM2015

Optical Distortion Evaluation in Large Area Windows using

More Info at Open Access Database www.ndt.net/?id=18112

Interferometry
Robert C. YOUNGQUIST 1, M iles SKOW1, M ark A. NURGE1
1
NASA, Kennedy Space Center, USA
Phone: (321)867-1829, Fax: (321)867-1177; e-mail: [email protected]
[email protected]; [email protected]

Abstract
It is important that imagery seen through large area windows, such as those used on space vehicles, not be
substantially distorted. Many approaches are described in the literature for measuring the distortion of an optical
window, but most suffer from either poor resolution or processing difficulties. In this paper a new definition of
distortion is presented, allowing accurate measurement using an optical interferometer. This new definition is
shown to be equivalent to the definitions provided by the military and the standards organizations. In order to
determine the advantages and disadvantages of this new approach the distortion of an acrylic window is
measured using three different methods; image comparison, Moiré interferometry, and phase-shifting
interferometry.

Keywords: Distortion, optical interferometry, optical metrology, optical windows, Schlieren, Moiré

1. Introduction
The primary function of a window is to allow observation of, yet protection from, a
potentially hazardous environment. Yet, from the window designer’s point of view, ensuring
protection from weather conditions in home windows; from wind, temperature, and airborne
debris in automotive windows; and from extreme pressures and temperatures in aircraft and
spacecraft windows has almost always taken precedence over image quality. This bias was
most strikingly borne out in a 1981 Air Force report [1] discussing optical distortion
requirements that stated “The F-106, F-111, B-1, T-28, F-5, and F-15 have all exceeded these
requirements, and sacrificing pilot visual performance has been justified by the increased
aerodynamic performance of the aircraft.” Such a design preference is defendable—it is more
important to protect an astronaut from the vacuum of space than to provide clear imagery—
yet these are not exclusive requirements. Advances in materials and material processing
allows the designer to attain better optical performance while not sacrificing important
material specifications such as strength. In addition, increased performance demands on
spacecraft windows, which are now used for photography and telescope observation, and
even for laser communications, requires greater consideration be paid to optical clarity.

Along with the need for better optical performance comes a corresponding need for improved
definition and quantification of the distortion of an optical window. Distortion can be
subjectively determined by a viewer looking through the window [1], but this is difficult to
quantify and is not repeatable. Consequently, over the last 50 years a wide range of window
distortion measurement approaches have been proposed [1, 2]. Some are only applicable to
windows with large distortions that can be determined by measuring surface variations [3]
and some require specialized components, such as an array of micro-lenses [4]. By far, the
most common method for measuring distortion is to photograph an image with and without
the window and then compare them [1, 5, 6, 7] but this approach has limited resolution and is
not applicable to higher quality windows. A newer approach, based on M oiré Interferometry
[8, 9], has higher resolution, but yields imagery that can be difficult to quantify. In this paper
we propose a new method for quantifying distortion based on phase-shifting interferometry.
2. Distortion
The window attribute that causes image distortion is curvature, either introduced through a
variation in the window thickness or by a localized variation in the index of refraction of the
window. It is assumed that these curvatures are very gradual relative to the wavelength of
light and that the window surfaces are very smooth, i.e. polished. This is important. Window
variations on the order of wavelengths will scatter light and cause image aberrations such as
light streaks and diffracted images. These are not attributes of distortion and are ignored in
the present discussion. So it is assumed that all window imperfections are very large
compared to an optical wavelength and smoothly varying.

Figure 1 shows an idealization of a section of a window. One side of the window is assumed
to be perfectly flat while the other side is slightly curved with a radius of curvature, R , that is
very large compared to size of this window segment, i.e. R >> 2a . Also, assume that the
window has a uniform index of refraction, n . Using a set of coordinate axes on the segment
as shown the curved surface can be expressed by the equation

z[ x] = p − x 2 / (2R ) . (1)

(a) (b)

Figure 1. This sketch shows a small section of a window with one planar surface and one constant curvature
surface. Figure (a) provides the descriptive parameters for this window section and (b) shows how rays of light
are refracted by the curved surface.

2.1 Distortion Definition 1

The most common definition of distortion is based on tracking rays of light that pass through
the mirror. Referring to Figure 1(b), a beam of light that passes through the window at
location x1 is deflected by an angle α1 , while a parallel beam, a short distance away at
location x 2 is deflected by an angle α2 . This change in the deflection of light versus position
leads to the first definition of distortion, D , namely

α2 − α1
D≡ . (2)
x2 − x1
If the surface curvature is gradual and the sampling distances small this expression becomes
the derivative of the angular deviation with respect to location, i.e.

dα [x ]
D≡ . (3)
dx

This definition is used by the military in specifying distortion, for example an optical flat
should have less that 1 minute of arc per inch distortion [1] and is the basis of the
International Organization for Standardization (ISO) standard [6].

Using Snell’s Law this definition can be related to the surface curvature and the index of
refraction of the window. Looking at Figure 1(b), an incoming ray of light hits the curved
surface at an angle, θ n , to the normal, shown by the dashed line that leads back to center of
the radius of curvature, R . This ray of light is refracted by an angle, θ , towards the y axis.
The goal is to find the deviation angle, α as a function of x , so its derivative can be
calculated. Start with Snell’s Law n sin[θ n ] = sin[θ ] , which, since the curvatures are all very
gradual, can be converted into a small angle form, i.e. nθ n = θ . Now, note that the slope of
the window’s curved surface is equal to θ n and that this slope is also equal to the derivative
of Equation 1. Finally, looking at the figure it is seen that α = θ −θ n . Combining these yields

d ( n −1)
α[ x ] = θ − θ n = ( n −1)θ n = ( n − 1)
dx
( p − x 2 / (2 R) ) = −
R
x. (4)

So from Equation 3 the distortion is found to be

dα [x ] (n − 1)
D= =− . (5)
dx R

This agrees with our initial requirement that a planar surface, i.e. one with infinite radius of
curvature, has zero distortion. It also indicates that as the curvature becomes smaller the
distortion becomes larger, a reasonable result.

2.2 Distortion Definition 2

A second definition of distortion is given by the American Society for Testing and Materials
(ASTM) [5] and states that distortion is equal to one over the focal length, F , of the lens
formed by the curvature of the window, i.e.

D ≡ (1/ F ) . (6)

This definition describes distortion in terms of local curvature as described by the resulting
focusing, or defocusing, of the window. In order to find the focal length of the lens formed by
the curved window surface in Figure 1(b) we need to ask, where on the z axis do the
refracted rays converge? Using a small angle approximation, we see that α[ x ] ≈ x / F . Using
this result and Equation 4 the second definition of distortion yields

D = (1/ F ) = α[ x ] / x = −( n −1) / R . (7)

As expected these two definitions yield the same result, though they start with very different
physical intuition.

2.3 Distortion Definition 3

The third definition of distortion—presented for the first time in this paper—is based on a
fundamental window attribute, namely the window’s optical path length function, σ [x ] . This
function describes the distance, as seen by the light, as it travels through the window and the
air, from one plane to a second plane. So calculate the optical path length in Figure 1 as light
moves from the lower, planar, window surface at z = 0 , to the z = p plane. For each value of
x the light passes through an amount of glass given by y[ x ] in Equation 1 and then passes
through an amount of air given by p − z[x ] . So the optical path length function is given by

σ [ x ] = n (z[ x]) + ( p − z[ x]) = (n −1)( p − x 2 / (2R )) + p . (8)

The derivative of this function describes the direction that the light takes after passing
through the window and air, i.e.

dσ [x ] / dx = −( n −1) x / R = α[ x ] . (9)

The second derivative of this function provides the third definition of window distortion

d 2σ [x ] (n − 1) dα[ x ]
D≡ 2
=− = (10)
dx R dx

showing that the distortion of a window can be expressed as the second derivative of the
optical path length function. This is important because equipment now exists, namely phase-
shifting interferometers, that can easily and quickly provide the optical path length function
of a window, both accurately and with high resolution.

2.4 Component and Total Distortion

Real world windows extend in two dimensions and the definitions given above for distortion
are only one dimensional. The ASTM [5] resolves this by defining three different distortions,
Dx , Dy , and D . Dx and Dy are defined in terms of the angular deviation of light when
scanning the window in the x and y directions, but no relation between these component
distortions and the distortion, D , is provided. The ISO standard [6] states that optical
distortion on a window is equal to the maximum distortion found by measuring in all
directions. This is better than ASTM, but is still unclear because distortion can change sign.

Our new definition of distortion removes this confusion by the following straightforward
extensions of the one dimensional definition given above in Equation 10:

d 2σ [x , y] d 2σ [x , y]
Dx ≡ , Dy ≡ , D = Dx + Dy = ∇2σ [ x , y ] (11)
dx 2 dy 2
This states that the total distortion, D , is the sum of the two component distortions and is
represented by the Laplacian of the optical path length. This definition is mathematically
consistent with the intuition that the distortion should be related to the window curvature, but
it should be stressed that window pass-fail criteria must be carefully written. For example, if
the window is saddle-shaped, σ [ x , y ] ∝ xy , then Dx , Dy , and D in Equation 11 are all zero,
yet the window is not flat.

3. Measuring Distortion
Having the various equivalent definitions of distortion, we can now compare three different
methods for measuring distortion. We will start with a straightforward technique where a test
pattern is photographed at some distance with and without the window and the two images
compared. Then a newer technique using Moiré interferometry will be tried and the third
approach will be to use a phase-shifting interferometer. In all three cases we will examine
the same window, a roughly 6 inch diameter section of acrylic sheet that has appreciable
distortion. This is of interest because future spacecraft are being designed with plastic
windows instead of fused silica in order to save weight even though plastics typically have
greater window distortion than the fused silica windows they are replacing.

3.1 Measuring Distortion Using Image Comparison

A distortion measurement system was constructed as described in the ASTM standard [5] and
is shown in Figure 2. An image was created of a set of parallel dark lines spaced by 1 cm.
This image was located a distance 3.6 m (distance 2L in Figure 2) from a focusing lens with
focal length, f . A camera was used to take a picture of this reference image and then the
acrylic window was placed between the image and the lens and a second photo taken. The
acrylic window was then rotated by 90 degrees and a third photo taken. The presence of the
acrylic window causes the photographed lines to be shifted and deformed. By measuring the
amount of shift in the line segments the distortion in the corresponding section of the window
can be calculated.

Figure 2. This is a sketch of the distortion measurement system using image comparison.

Using geometrical arguments it can be shown that the change in location of a line segment
image on the focal plane array, ∆z , is related to the window deflection angle, α[ x ] , by
α[ x ] = 2∆ z / f . In our configuration the lens focal length was 74 mm. We wrote code
(Mathematica) to find the center of each 1 cm long line segment in the reference image and in
the two window images. Subtracting these locations then yields two arrays of ∆z values from
which the angular light ray deviations in the x and y directions can be found, see Figure 3.
 (a) (b)

Figure 3. Measured angular deviations seen in the acrylic window as a function of location on the window, for
the (a) y -direction and (b) x -direction using the ASTM image comparison approach.

Due to limited resolution it’s difficult to determine the distortion of the window. A one pixel
shift on the camera corresponds to about 0.4 minutes of arc, so the above images correspond
to a less than 3 pixel shift. By averaging data we were able to obtain sub-pixel resolution, but
some of the structure seen in these images may be imaging artifacts and not true line segment
shifts. Taking the derivative of these two plots amplifies the oscillatory nature and swamps
what might be true distortion. For example Figure 3 (a) shows a definite slope across the
window corresponding to a large scale, though small, y -directed, distortion across the
window, which is many times smaller than the “distortion” seen in smaller regions.

3.2 Measuring Distortion Using Moiré Interferometry

Moiré Interferometry [8, 9], like Schlieren Imagery [10], is an optical technique that
amplifies the intensity variations caused by small angular deviations of light rays. The system
sketch is shown in Figure 4. Light passes through a transparent lined pattern (i.e. a Ronchi
ruling) and travels to a spherical mirror with radius R (48 inches in our system). The light
reflects back and passes through the transparent line pattern a second time, and is imaged
onto a camera focal plane.

Figure 4. A sketch of the Moiré Interferometry system for measuring window deformation.
The Ronchi ruling is composed of 1/100th inch wide dark lines separated by transparent
spaces 1/100th of an inch wide. So in order for the camera to see light, it must look through
the gaps between the dark lines and then, looking into the mirror, see light from a gap in front
of the light source.

Moving the Ronchi ruling towards the window a distance d causes alternating bands of dark
and light to appear on the surface of the mirror, i.e. a Moiré pattern. The spacing of these
lines is given by the Ronchi ruling line width times (R / d ) , which equals 0.2 inches when
d = 2.5 inches. We took a photograph of this line pattern as a reference and then inserted the
acrylic window, as shown in Figure 4. The window causes angular deviations of the light
passing through it. Now, if the light returning to the Ronchi ruling is deflected up or down by
the window by an angle equal to (1/50th inch)/( R − d )=1.5 minutes of arc, then the image of
the dark and light bands on the mirror will appear to move by one full spacing, i.e. 0.2 inches.
So the presence of the window causes the dark and light bands to be shifted and deformed, by
amounts much larger than for the ASTM line test described above. Taking into account that
the light passes through the window twice, and assuming pixel resolution in tracking the light
and dark bands, our system has approximately 0.02 minutes of resolution. This is 20 times
better than for the ASTM approach.

We took pictures of the shifted line pattern with the acrylic window oriented along both
Cartesian axes and generated the angular deviation functions shown in Figures 5(a) and 5(b).
Comparing these to Figure 3 shows the improvement achieved by using Moiré Interferometry
over a direct image comparison approach. However, this approach, even though it is very
sensitive, has accuracy problems. For example, inserting a perfect window into the system
will yield shifts that are not due to distortion, but due to the refraction of light by the window.
These refraction effects have to be calculated and numerically removed from the data, which
was done in our case, but neglecting this will lead to an offset in the distortion calculation.

When comparing Figures 3 and 5 (as well as Figure 7 below), note each approach generates
an angular deviation offset. Also, we did not strive to match the x and y axes, so they are
offset in the Figures and in one case, Figure 5b, we inverted the x axis.

(a) (b)

Figure 5. Measured angular deviations seen in the acrylic window as a function of location on the window, for
the (a) y -direction and (b) x -direction using the Moiré interferometry approach.
3.3 Measuring Distortion Using Phase-Shifting Interferometry

Phase-shifting interferometry is a standard technique for measuring the optical path length of
optical components. Several companies sell complete systems, typically used for optical
component evaluation. For our work we used a Zygo Corporation VerifireTM ATZ system
[11] as shown in Figure 6(a), with a 6 inch diameter measurement capability. Placing the
acrylic window into this system yields the optical path length plot shown in Figure 6(b). This
plot shows that the acrylic window has a very slight cylindrical shape.

(a) (b)

Figure 6. This is a photo of a Zygo phase-shifting interferometer, the VerifireTM ATZ, alongside the measured
optical path length of the acrylic test window.

The Zygo Interferometer supplies high resolution imagery, approximately 60 pixels per cm,
allowing the derivatives of this function to be calculated and averaged, yielding the angular
deviation functions from Equation 9. These two functions are plotted in Figures 7 (a) and 7
(b). Comparison of these plots with those shown in Figures 3 and 5, provides an immediate
visual indication of the quality of data provided by the three distortion measurement methods
being compared in this paper.

(a) (b)

Figure 7. Measured angular deviations seen in the acrylic window as a function of location on the window, for
the (a) y -direction and (b) x -direction using the phase-shifting interferometry approach.
The quality of the data from the phase-shifting interferometer is high enough that the second
derivatives can be obtained numerically, i.e. the distortion plots, (Equation 11). These are
shown in Figures 8(a) and 8(b). It is interesting to note that the large curvature seen in the
optical path length function, which shows up as the slope in Figure 7(a), has little effect on
the distortion. It corresponds to roughly a -0.1 min/cm offset in Figure 8(a) which is difficult
to see in the presence of the larger, more localized, distortion effects.

(a) (b)

Figure 8. The measured distortion in the acrylic window as a function of location on the window for the (a) y -
direction and (b) x -direction using the phase-shifting interferometry approach.

As opposed to the results shown in the image comparison method, the fine structure shown in
Figures 7 and 8 are real and are not the result of noise or errors introduced in the derivative
process. Phase-shifting interferometers can achieve nanometer optical path length resolution,
so distortions much smaller than an arc-second/cm can be measured.

4. Conclusions
We have introduced a new definition of distortion and shown that it allows a phase-shifting
interferometer to be used to determine the distortion of an optical window. At first glance this
appears to be a preferable technique for measuring distortion over image comparison and
Moiré interferometry, however, each has its strengths and weaknesses. Phase-shifting
interferometry provides the highest resolution measurements, but the system cost is expensive
and the dynamic range is limited to only higher quality windows. Also, scanning large area
windows requires making multiple measurements and then stitching the imagery to obtain a
complete window map. Moiré Interferometry is inexpensive and sensitive, but quantifying
the data can be difficult and requires careful measurement and analysis. Even so, once the
algorithm is developed this approach can be scaled to large windows by using a larger
spherical mirror and the system sensitivity can be adjusted by changing the spherical mirror’s
radius of curvature. So it can be used over a wide range of window qualities. Finally, the
image comparison approach is the least expensive and most straightforward and has been the
standard method for many years. It is adequate when examining relatively low performance
windows, but as shown above, it is limited in performance.
Acknowledgements

We would like to acknowledge Deborah A. Guelzow and Beverly A. Bush of the Kennedy
Space Center Library for literature search and retrieval support.

References

1. J S Harris and K G Harding, ‘Study and Evaluation of Existing Techniques for

Measuring Aircraft Windscreen Optical Quality: Development of New Techniques for
Measuring Aircraft Windscreen Optical Distortion’, Air Force Aerospace Medical
Research Laboratories, Report No. AFAMRL-TR-81-25, February 1981.

2. J S Harris, K G Harding, S H Mersch, ‘Techniques for Evaluation of Aircraft

Windscreen Optical Distortion’, Optical Engineering, Vol 20, No. 1, pp 115-122,
January/February 1981.

3. ‘Standard Test Method for Measurement of Roll Wave Optical Distortion in Heat
Treated Flat Glass’, American Society for Testing and Materials, ASTM C1651-11,
October 2011.

4. M Savolainen, K-E Peiponen, P Savander, R Silvennoinen, and H Vehvilainen.

‘Novel Optical Techniques for Window Glass Inspection’, Measurement Science and
Technology, Vol 6, No 7, pp. 1016-1021, 1995.

5. ‘Standard Test Method for Measuring Optical Distortion in Flat Glass Products Using
Digital Photography of Grids’, American Society for Testing and Materials, ASTM
C1652M-14, May 2014.

6. ‘Road Vehicles—Safety Glazing Materials—Test Methods for Optical Properties’,

International Organization for Standardization, ISO 3538:1997(E).

7. M Dixon, R Glaubius, P Freeman, R Pless, M P Gleason, M M Thomas, W D Smart,

‘Measuring Optical Distortion in Aircraft Transparencies: A Fully Automated System
for Quantitative Evaluation’, Machine Vision and Applications, Vol. 22, pp 791-804,
2011.

8. U Pingel, ‘New Moiré-Fringe-Method to Inspect Transmitted Distortion and Point-

Defects in Sheet-Glass’, Proceedings Glass Processing Days, pp. 120-124, September
1997.

9. A S Redner and G K Bhat, ‘Moiré Distortiometry for the Quantitative Evaluation of

Optical Quality Glass’, Proceedings Glass Processing Days, pp. 166-168, June 1999.

10. GE Elsinga, B W van Oudheusden, F Scarano, and D W Watt, ‘Assessment and

Application of Quantitative Schlieren Methods: Calibrated Color Schlieren and
Background Oriented Schlieren’, Experiments in Fluids, Vol. 36, pp. 309-325, 2004.

11. Zygo web site description of the VerifireTM Series of phase-shifting interferometers.
http://www.zygo.com/?/met/interferometers/verifire/