3-D Measurement and Evaluation of Surface Texture Produced
3-D Measurement and Evaluation of Surface Texture Produced
3-D Measurement and Evaluation of Surface Texture Produced
Measurement
journal homepage: www.elsevier.com/locate/measurement
a r t i c l e i n f o a b s t r a c t
Article history: The scraping process involves traditional manual work and is an important technique for
Received 25 November 2010 producing at bearing surfaces with lubricating grooves on a sliding surface. In order to
Received in revised form 13 October 2011 meet the requirements of precision engineering, the scraped surface should have an
Accepted 13 November 2011
equally distributed pattern with a required number of high points per unit area while
Available online 23 November 2011
retaining good atness. In the machine tool manufacturers workshop, however, the quality
inspection of scraped surfaces still depends on human eyes. In this study, a 2-D evaluation
Keywords:
system is rst developed using image processing so that the peak points per area of square
Scraped surface
Surface quality parameter
inch (PPI) and the percentage of points (POP) can be quantied as parameters. A vision-
Image measurement system assisted laser focus probe system is then developed to measure the 3-D form of the scraped
Laser focus probe system proles. The laser probe is made of a DVD pickup head based on the astigmatic principle.
3-D topography Driven by an XY stage, the entire scraped prole can rapidly be scanned. The quality of the
scraped surface can thus be interpreted in a more scientic manner. Based on the mea-
sured 3-D data, new evaluation methods are proposed for ve parameters, namely the
PPI, POP, height of points (HOP) or depth of surroundings (DOS), atness, and oil retention
volume. Experiments show that the 3-D system is consistent with the 2-D system. It not
only reveals more surface quality parameters but also uncovers more characteristic surface
phenomena than the 2-D image system.
2011 Elsevier Ltd. All rights reserved.
0263-2241/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.measurement.2011.11.006
K.-C. Fan et al. / Measurement 45 (2012) 384392 385
on the scraped surface [1]. The two most important param- conguration and the software and the computations are
eters are the peak points per area of square inch (PPI) and described in the following sections.
the percentage of points (POP). This means that the surface
must contain enough high points and a sufcient area to
2. Measurement of 2-D scraped surfaces by computer
bear the load on the surface top. The indented portion is
vision
used to contain oil for lubrication. The current 2-D inspec-
tion method for these two parameters in the industry in-
2.1. Digital image processing
volves judging the tinted pattern distribution by eye. The
other three parameters are the height of points (HOP) or
Digital image processing techniques have been widely
depth of surroundings (DOS) to estimate oil containment,
used in pattern recognition for some time. Characteristic
the shape of the cavity edge to avoid oil push-out and dirt
features can be identied through a series of operations,
push-in, and the atness of the scraped surface to assure a
such as noise reduction [16], auto threshold selection
smooth motion. These three parameters can only be
[17], edge detection [1820], and mathematical morphol-
judged by a 3-D view. However, none of the existing meth-
ogy [21]. A variety of applications can be implemented
ods can provide this information.
by the use of image processing techniques, such as defect
Using computer vision to examine the scraped work-
inspection of PC boards [23,24] and powder metallurgy
piece requires some pre-processing techniques to remove
products [25]. In this study, the parameters PPI and POP
unwanted noises and enhance some important image fea-
of the scraped surface are examined by a computer vision
tures for further post-processing. Higher points on the
technique. The image system consists of a common CMOS
scraped surface can then be identied. Tsutsumi et al.
sensor with 640 480 pixels and 8-bit gray levels, a prop-
developed an image system with a CCD camera and a color
erly selected lens, and a PC. The sequence of image pro-
lter that could recognize the higher points [2]. Yoshimi
cessing techniques for this purpose is shown in the ow
et al. recognized the higher points by a CCD line sensor
diagram of Fig. 1.
and removed them by a grinder [3]. Such image processing,
This owchart shows that in automatic optical inspec-
which offers a quick way to assess the quality of scraped
tion, the camera needs to be calibrated with a commer-
surfaces, is useful for developing an automatic scraping
cially available standard template at the rst step. After
machine [4]. In order to improve the quality of the scraping
calibration, we know the dimensions of each pixel and
process, Tsutsumi et al. further developed an automatic
the correction of the lens distortion. The second step is to
scraping machine [5]. Tadanori et al. investigated the
threshold the gradient magnitude of edge points in order
sequence of scraping operations to establish a standard
to extract the high-point areas. The potential edges of
manufacturing ow [6]. All of these works focus on 2-D im-
these high-point areas need to be identied rst. These
age inspection followed by re-scraping of the surface to
are done by applying Cannys edge detection technique
improve the quality.
[18]. From our long-term experiments over almost 1 year,
Modern DVD players are well developed and their pick-
however, we found that the detected edge points could
up heads have good focusing characteristics. Many applica-
not be extracted very repeatedly. It could be due to the var-
tions use these low-cost DVD pickup heads for measuring
ied room lighting conditions. To solve this problem, after
features such as prole [79,22], velocity [10], straightness
[11], angle [12], and acceleration [1315]. Although there
are some commercially available sensors or measurement
Start
systems that can measure the 3-D proles of microstruc-
tured surfaces (e.g., the laser triangulation sensor, laser
Camera calibration
confocal sensor, laser confocal microscope, etc.), their costs
are high. Furthermore, there are no reports on their appli-
Taking an image
cation to scraped surface measurements thus far.
In this report, a 2-D image vision system that can assess
only two quality parameters of the scraped surface, namely Thresholding the gradient
magnitude for Edge detection
PPI and POP, is rst studied. Subsequently, a 3-D prole
measurement system to measure scraped surfaces
Opening operation
consisting of a DVD pickup head, an image processing sys-
tem, and an XY stageis presented. This system is able to
analyze ve parametersnamely the PPI, POP, HOP (or Grouping and marking
DOS), surface atness, and oil content. In order to verify
the 3-D measurement results, a separate image processing Is the group area Y
system is also developed to measure the PPI and POP, not too large?
only for the entire surface but also for each local area. This N
system can ensure the evenness of scraped spots for a uni- POP and PPI
form bearing load and a uniform velocity when the slider Calculation
moves across the guideway. Experiments show that the
two measurement systems yield consistent results, in End
which the 3-D system reveals more information about
the quality of the scraped surface. Details of the hardware Fig. 1. Flow diagram of image processing.
386 K.-C. Fan et al. / Measurement 45 (2012) 384392
the Cannys edge detection, we then adopted Sobel opera- to case, a better scraped surface should have evenly dis-
tors to mask all potential edge pixels to nd the gradients tributed patterns for the sake of a constant friction on
(Gx and Gy) of each pixels gray level in the X and Y direc- the entire surface, although denitions of the PPI and
tions, respectively. The magnitude of the gradient at each POP refer only to the averaged values for the entire surface.
pixel is the vector sum of these gradients, i.e. Therefore, a moving window corresponding to a surface
rf G2x G2y 1=2 . For all potential edge pixels in the image area of one square inch is generated. In the program, this
plane, a large gradient magnitude indicates a potential window is automatically swept across the full image area
edge point as the gray level changes rapidly at this pixel in zigzag motion with a feed size of one pixel. At each po-
point. A suitable rf value (t) is then selected as the thresh- sition, the parameters of PPI and POP are calculated. This is
old value to identify the edge points of high-point areas. to assure that in addition to the full area, each local area is
For any pixel i, when its rf is higher than t its gray level also evaluated. Although expert rules indicate that the
is set to zero (black, which indicates a high point) and vice best-scraped condition should have a PPI of 40 points
versa. The binarization function can be expressed by Eq. and a POP of 50%, in practice, due to the manufacturing
(1), where gi is the original gray level, t is the threshold cost concern in industry acceptable ranges for the PPI
of rf, and H(gi) is the binarized gray level of each pixel. and POP are between 2025 points and 4060%, respec-
tively [1].
0 rfi P t
Hg i 1
255 rfi < t
2.2. Experimental tests
After thresholding the gradient magnitude for edge
detection, the high points can be identied. The next step A piece of a scraped surface made by a skilled craftsman
is to apply a morphological opening operation to remove is shown in Fig. 3. The picture on the left is the original im-
the noise of the binary image. Then, an 8-connected oper- age and the one on the right is the image after processing.
ation is used to group and mark connected high points, nd The centroid marking of each black zone represents a peak
the area of the group, and mark its centroid. The distribu- area. In practice, the conventional tinted paste coating pro-
tion of contact areas (high-point groups) in the image cess is used to enhance the image contrast of the high
plane can be realized. It is noted that these contact areas points with respect to the low points. The results of ve
are recognized by the tinted paste that generates low gray repeatability tests on the full area are shown in Table 1.
levels in the captured image. The paste, however, may cov- The variation of PPI is 2.3 points and that of POP is 2%,
er a portion of the surrounding area of each high-point which demonstrates that the 2D image processing method
group and connect vicinal groups. This yields an exces- is highly replicable. The pictures also reveal that the spot
sively large contact area. An iterative process of opening distribution is uneven. The use of a moving window to
and grouping is carried out to separate connected high evaluate local parameters is deemed necessary for the sake
points into individual sub-groups. Based on the expert of a constant friction force along the full stroke sliding dis-
rules of the scraping technique, the threshold of the tance. Fig. 4 shows another example evaluated by a moving
high-point contact area is set to 500 pixels in our computer window. It shows that ve local areas do not satisfy the ex-
program. Finally, the last step is to calculate the following pert rules, as marked by the square windows in the gure.
parameters: number of high points per area of square inch Their actual positions are indicated by the (X, Y) coordi-
(PPI) and the percentage of points (POP). nates of their respective upper left corners, and the types
This algorithm is validated by a number of computer of no-go (NG) parameters are remarked in the last column
generated scraped patterns. Fig. 2 is an example of a gener- of Table 2. It is seen that all ve local regions fail to meet
ated scraping pattern. Each high-point area is shown in the required POP range (4060%), being either lower (win-
black and its central point is recorded by a white dot. dows 1, 2, 3, and 5) or slightly higher (window 4). In addi-
The areas of peak and valley zones can clearly be sepa- tion, windows 1 and 5 require slightly more high points to
rated. In addition, as the size of the surface varies from case meet the requirement (2025 points). Note that although
the averaged PPI and POP of the entire surface could be
acceptable according to [1], these local unacceptable areas
may cause velocity variation due to uneven friction forces
when the carriage moves across these areas. This system
provides a stricter rule than human justication. In prac-
tice, however, a slightly out-of-rules area such as window
4 could be regarded acceptable.
Fig. 3. Images of a scraped surface: (left) original and (right) after image processing.
Table 1 Table 2
Repeatability tests for PPI and POP. Unacceptable local areas (NG means no go).
through the polarization beam splitter, the quarter wave- photodiode becomes elliptically shaped in different orien-
length plate, the collimator lens, and the objective lens, tations (Fig. 6, planes 1 and 3). If the measured surface is in
and nally focus onto the disc surface. The reected beam focus, the image becomes circular (Fig. 6, plane 2). Accord-
passes through the same original path, and after two polar- ing to the beam spot distribution among the four quad-
izations by the quarter wavelength plate, projects onto the rants, the FESs are used to measure the prole of the disc
quadrant detector through the cylindrical lens. The photo- after calibration. A voice coil motor is used to trace the
diode outputs focus error signals (FESs) based on the astig- ight height change of the rotating disc. This focus or
matic principle. If the measurement surface is near to or far auto-focus probe technique has been developed over the
away from the focal point, the image of the beam on the course of more than 10 years by the authors group [8,9].
Fig. 9. Experimental setup of 3-D scraped surface measurement. Fig. 10. Linear range of the normalized FES.
390 K.-C. Fan et al. / Measurement 45 (2012) 384392
(a) (b)
Fig. 12. Measured 3-D microstructure prole: (a) raw data plot and (b) smoothed data plot.
K.-C. Fan et al. / Measurement 45 (2012) 384392 391
(a) (b)
Fig. 13. Evaluation of PPI and POP: (a) measured by 2-D CCD camera and (b) measured by 3-D focus probe.
(a)
(b)
(c)
Fig. 15. Three scraped plates: (a) photographs, (b) corresponding 2-D image processing patterns, and (c) corresponding 3-D topography plots.
392 K.-C. Fan et al. / Measurement 45 (2012) 384392