Documente Academic
Documente Profesional
Documente Cultură
In the first iteration of the concept evaluation, all proposed concept designs were compared
with the unmodified VJT Inline Robotic X-ray Inspection System that is highly dependent on an
operator to adjust the position of the robotic arm relative to each inspected casting. As the main
goal of all five proposed concepts is to increase throughput by automating the process of measuring
misalignment so that the robot arm can be programmed to readjust as needed, all concepts are
expected to perform faster than the datum. However, the accuracy of the measurements depends
on the resolution of the measuring instrument, which varies between concepts. The concepts which
utilize pressure and laser sensors earn a + rating for accuracy because they are high-resolution
instruments that can more accurately sense alignment than a human can. Unfortunately, the
concepts that use a Kinect system or ultrasonic sensor may not necessarily give readings that
outperform a human operator due to the lower resolutions of the sensors. The image processing
using the MATLAB Image processing toolbox also received a + rating since this method is
extremely accurate and utilizes the precision and speed of the MATLAB program to superiorly
generate feedback regarding the part orientation.
None of the sensors in any design concept will be placed in a way that obstructs the X-ray
path, so the addition of these design modifications will have no effect on the overall X-ray
compatibility of the system. Therefore, all concepts earn an S rating for the “X-ray compatible”
criteria in the Pugh Matrix. Similarly, no sensor will be in direct contact with a hot casting, so all
concepts earn an S rating for the “Heat Resistant” criteria as well. The only design that will
experience any notable loading are the pressure sensors in the third concept, in which pressure
sensors will be attached to the fixture bed that supports each casting. Consequently, the pressure
sensor system is the only design that may have a shorter (i.e. - rating) design life than the
unmodified system, while all others are expected to have the same design life.
Another major priority in designing this device is to decrease dependence on the operator
in making measurements for out-of-plane tilting, so it is reasonable to assume that all concepts
will have better (i.e. +) ratings than the datum for the “User Interface” and “Ease of Use” criterion.
The exception to this is the pressure sensor system, which has a negative “Ease of Use” rating
because its operating principle of locating weight distribution is effective for measuring single
parts only. It would be difficult for the pressure sensor system to measure misalignment in multiple
castings placed on the fixture bed simultaneously, which can normally be accomplished by an
operator-dependent system
Regarding cost, all of these concept designs will have an initial cost to install the
modification onto the existing system. However, by shifting responsibility from the operator to an
automated system, savings in labor cost should offset the initial investment costs within 15 months
or less, provided that each design conforms to requirement 3.4.6 (i.e. product cost should not
exceed $5,000). Therefore, all concepts earn a + rating for the “Cost” criteria.
Summary of Strengths and Weaknesses
The Pugh Matrix has narrowed from five concepts to two concepts. In Section 4.4 below,
these two concepts are iterated. Before design iteration takes place it is important to reflect on
both the previous designs retained as well as those discarded and use this information in a
feedback loop to improve the concept design.
The Xbox Kinect system strengths are that it can image and object and provide depth to
the surfaces and objects in the image. The weakness of this is that it is very computationally
intensive to perform these calculations, which will require special hardware and software to
process the data. The other weakness is that the resolution of the depth sensing is not sufficient to
meet the product design specification.
The strengths of the ultrasonic system are that it is easy to implement with an Arduino
microcontroller, the weaknesses of the ultrasonic sensor are that it does not provide the accuracy
in the depth measurement that we require. The ultrasonic sensor is not a fine measurement device
but rather a coarse instrument.
The strength of the pressure sensor system is that the orientation measurement is based on
the weight distribution, which is more repeatable than a non-contact sensor. The weakness of the
pressure sensor system is that it can only measure the orientation of one object at a time, which
greatly limits throughput.
The two laser based measurement systems can provide both accuracy and precision.
These instruments are reliable and can provide fast measurements. By using the distance data to
place three points on the surface of the part, the amount of out-of-plane rotation can be
determined. This calculation is not difficult to compute and can be carried out on an Arduino
microcontroller. The weakness of the laser system is that the lens will accumulate dust over time
and the beam may reflect off the surface if it is polished. Both of these issues will affect the
measurement.
The strengths of the image processing with MATLAB image processing toolbox is that
he software is incredibly fast and accurate. As shown in the Pugh Matrix in Table 4.1, the image-
processing concept meets requirements for all specifications. Given that the measurement is
digital as opposed to physical, the speed of the system outperforms the sensor designs. Pending
further testing, the ability of the image processing software to identify part orientation is superior
in practice than the remaining concepts. Another strength of this method is that the camera and
included hardware is inexpensive compared to the concepts which require moving parts, as the
camera is fixed during operation. The downside to this concept is that it is software intensive and
requires a good amount of MATLAB experience to write the optimization code for the process.
Detail design highlights the pertinent PDS information relevant for the detail design
subunits. For our selected concept after the iteration process, the various subunits are explained
and components for each are selected to meet specifications from the PDS information without
violating any of the codes and standards listed in section 3.8. For each subunit, options for different
hardware are evaluated to fulfill performance and functional requirements.
The portions of the overall PDS from Chapter 3 that are quantitative in nature and
relevant to the detail design are compiled as follows:
3.3 Performance
3.3.1 The system shall measure out of plane misalignment of any part of a single geometry
to within 1 degree of the XY-plane.
3.3.2 Inspection zone must be able to fit castings of sizes up to 1200mm x 500mm x
400mm.
3.4 Constraints
The following design constraints will limit the design options:
3.4.3 System must have access to 120 V 60 Hz AC power.
3.4.5 Wire and cables should not be placed in the paths of moving parts.
3.4.6 A product cost lower than $5,000.
3.6 Maintenance
The design should comply with the following:
3.6.1 Sensors should be accessible within five minutes for replacement in the event of
defect that cannot be resolved through simple recalibration.
The part will be placed into a fixture to roughly orient the part. This part is expected to be
within 3o of the optimal rotational orientation. At this time the camera will capture and store an
image in the specified MATLAB directory.
Camera Image Manipulation
The first manipulation will be convert the image to grayscale. Grayscale images are faster
to process as they contain less data, and they are able to be filtered into black & white as the
threshold filter can be selected select. The MATLAB function rgb2gray(Image) is used to
transform the image into gray scale. Using the MATLAB function imbinarize(Image)the
picture is converted into a matrix of zeros and ones corresponding to black and white pixels.
These functions have been compiled into a new function IMP(Image, threshold). This
function takes an image converts it to gray scale, filters speckles out of the image, then converts
the image to black and white based on a thresholding value. In Figure 5.1, the results of the
IMP(…) function are shown. The real image is located on the left and the CAD image is located
on the right.
Figure 2.1: Results of Image Processing
CAD Model size and rotation
When comparing the captured image and projected image taken from the CAD file it is
necessary to have the two images the same size, and for the parts in the image to be nearly the
same size this is so the comparison performed later will converge quickly. The size of the model
in cad is changed using the zoom tool in MATLAB, the zoom of the CAD part is changed until
the number of white pixels in the filtered camera image is within the desired range of the number
of white pixels in the CAD image. The zoom process takes an iterative approach to correctly scale
the images.
Once the correct zoom for the cad model is found the part is rotated using the camorbit(…)
function. An image for each rotation is stored, these images will then be compared to the real
image. The pair of images that have the least amount of differences correlate to the position that
the real part is in
Image Properties
MATLAB contains built in functions to get the properties from a selected image matrix,
these properties include the name, centroid of an object, and values at locations of interest. We
will use the getfield(…,’Centroid’) function to locate the centroid of the image so that it can be
cropped to the correct size. Cropping the image to the correct size around the centroid allows the
images to be layered on top of each other. Layering the images shows a differences in pixels.
Figure 2.2: Layered comparison of the Real image and a CAD image
The image in Figure 5.2 shows the comparison between the real image and the CAD
image. Using MATLAB to compare all of the CAD images to the one real the position is
known when the difference between the real and CAD image is minimized.
Programming Flow Chart
Real
Image
No
Converge
Image cropping and Rotate CAD model
Yes
comparison and store images
Error CAD
matrix Images
Yes
No End
In Figure 5.3 the logic for the program is shown in a flowchart. This simplifies the steps
taken to compute the position of the part, the detailed MATLAB code is located in appendix C.
The data in the error matrix is the difference between the real image and each individual CAD
image rotation. MATLAB has convenient built in functions to locate the minimum value in a
matrix. This location directly corresponds to the rotation of the part.
Figure 2.4:
Figure 2.5: Image of the complete prototype with fixed camera and tilted part under inspection
2.4 Bill of Materials with Cost Analysis
A list of required materials for this design, as well as their costs, and methods of
obtainment are summarized in Table 5.1 below.
The prototype allowed the student team to test the measurement functionality of the final
design. Real, casted-aluminum parts were obtained from VJT in order to complete testing and
verification of the prototype. Images of these sand castings were taken at various tilt angles and
the resulting measured azimuth and elevation angles were obtained from MATLAB calculations.
The team was successful in being able to identify the optimal viewing angle using the prototype
setup as well as two uniquely-shaped cast aluminum steering knuckles from VJT. Several plots,
shown in Figures 5.22 through 5.24, were created in order to show how the code converges on the
optimal viewing angle. The minimum point of pixel error, marked in red, for each 3D surface can
be matched to the matrix indices within the 7x7 error array, which correspond to azimuth and
elevation in 1o increments from -3o to +3o. Figure 5.22 illustrates that the system measurement of
a real casting has an inherent tilt of +1o in the azimuth and -3o in elevation when resting on a flat
plane ([5, 0] in the error matrix). Figures 5.23 and 5.24 show how the minimal point then changes
when this same casting is further tilted at +1o and +2o about the elevation axis, respectively, by
adjusting the tilt table gauge blocks pictured in Figure 5.20. As expected, the MATLAB calculated
minimum error correspond to [5,1], or +1o in the azimuth and -2o in elevation for the +1o tilt case;
and [5,2], or +1o in the azimuth and -1o in elevation for the +2o tilt case. Therefore, these results
appropriately reflect the 1o and 2o relative increases in tilt from the control position and verify that
the overall system functions as intended.
Figure 2.6: Plot of Pixel Error vs Tilt - Real Casting Resting on Flat Plane
Figure 2.7: Plot of Pixel Error vs. Tilt – Real Casting Resting on 0o, +1o Tilted Plane
Figure 2.8: Plot of Pixel Error vs. Tilt – Real Casting Resting on 0o, +2o Tilted Plane
2.6 Discussion
From the results, we can verify that by comparing the pixel array of the photograph to the
pixel arrays of the rotated CAD images, an optimal orientation where the pixel error is minimized
can be determined. The design is also able to recognize when the part is tilted as little as 1 degree
and reproduces this in the resulting azimuth and elevation angles. With our prototype, any
automotive or aerospace crucial casting can be X-ray inspected in the proper orientation before
further machining.
Although the design was created for the purpose of X-ray inspection, the image processing
concept can prove to be useful for many other applications. Industrial systems that handle low-
tolerance parts or operate around the intended geometry of the component may malfunction due
to a mistake in the manufacturing process. Even after machining, complex equipment may be
defective and too expensive to simply mass produce. Image processing may be useful in these
situations to determine the difference between the expected and actual dimensions.
Considering more sophisticated and futuristic applications of the design, image processing
may have a medical purpose. By eliminating the automated aspect of the project and adding human
intelligence into the picture, the design can augment existing medical systems such as CT scanners.
Instead of saturating the computer with enough X-ray images to reproduce a 3D structure, our
design can optimize the process and inform the user of the X-ray view angle required to obtain the
desired X-ray image.
2.7 Conclusion
Self-Evaluation of Team Work
The group meets regularly to collaborate on this project in the CAD Lab located on the
first floor of the Old Engineering Building, Professor Longtin’s Thermal Laser Lab, or also
in Jonathan and Tony’s dorm room in the West Apartments. While each member generally
takes charge of his own written section of the report, all members still communicate with
each other during the writing process to ensure that all content in the report is consistent in
voice and direction.
Jonathan is responsible for communicating and setting up meetings with the team’s contact
from VJT Technology, Sam Anjelly. Jonathan also wrote the Product Design Specification
chapter and contributed the ultrasonic sensor actuated by an XY table concept design. He
takes the role as the team’s purchasing liaison, finalizing all sourcing decisions before
submitting purchase orders. Additionally, his duties as the main editor for the report consist
of double checking formatting, grammar, and organization of tables & figures.
Dylan works as an undergraduate research assistant with one of our two faculty advisors,
Professor Jon Longtin. Therefore, Dylan is usually the team member that arranges meetings
with Dr. Longtin and uses his lab access to 3D print scaled-replicas of sample casted model
files given to us by Mr. Anjelly. Dylan has also been investigating methods of accomplishing
the image processing design approach using MATLAB and proposed the sweeping laser
sensor concept.
Tony provided all introductory statements and has completed all research on existing
models as well as patent research from Chapter 2 and contributed the pressure sensor
modification of the fixture bed concept. He assisted Dylan in exploring the feasibility of the
MATLAB image processing approach. Tony was the main contributor of the results and
verifications.
Leo oversaw all CAD modeling of design components, as well as the concept design
drawing of the laser XY table, and is adept in the use of Solid Works. He detailed the majority
of the concept generation and evaluation sections and contributed the Kinect concept.
Regarding the detail design, Leo also compiled the assembly drawings for the finished
system.
References
[2] VJ Technologies, 2016, “Inline Robotic X-ray Inspection System Product Sheet,” from
http://vjt.com/wp-content/uploads/2016/04/VJT-Inline-Robotic-Product-Sheet.pdf
[3] Nelson Air, 2016, “PIglide AT1: Linear Air Bearing Stage Product Sheet,” from
http://www.pi-usa.us/products/Air_Bearing_Stages/PIglide_AT1_Linear_Air_Bearing.pdf
[4] CNC Indexing, 2016, “Compact Tilt Rotary Table,” from http://www.cncindexing.com/wp-
content/uploads/2013/06/Compact-Tilt-Rotary-Table.pdf
[5] Canon, 2011, “EOS Rebel T3i EF-S 18-135mm IS Lens Kit,” from
https://shop.usa.canon.com/shop/en/catalog/eos-rebel-t3i-ef-s-18-135mm-is-lens-kit
[6] B&H Photo Video Incorporated, 2015, “Bolt VM-160 LED Macro Ring Light,” from
https://www.bhphotovideo.com/c/product/750878-REG/bolt_vm_160_led_macro_ring.html
[8] B&H Photo Video Incorporated, 2013, “Oben BD-0 Mini Ball Head,” from
https://www.bhphotovideo.com/c/product/687361-
REG/Oben_BD_0_BD_0_TABLE_TOP_BALL.html
[9] Klingenbeck-Regn, K., 2008, “Method for Displacing a Superimposed Measuring Surface on
a Sensor Surface of an X-ray Detector and X-ray System for Implementing Said Method,” U.S.
Patent 20080031421.
[10] Ross, M., 2008, “Method and Apparatus for Automated, Digital, Radiographic Inspection of
Aerospace Parts,” U.S. Patent 20080226028.
[11] Ho, C.E., Yang, C.H. and Hsu L.H., 2015, “Adjustable Fixture Structure for 3-Dimensional
X-ray Computed Tomography,” U.S. Patent 20150168316.
[12] Gupta, R. and Webb B.J., 1998, “Fixture for Calibrated Positioning of an Object,” U.S.
Patent 005715167.
Appendix A. VJT Engineering Manager Email
We have prepared a list of questions for you relating to the position sensing and fixture design. We
are preparing the product design specification (PDS) and require additional information.
What metals or plastics are currently used in the table and fixture?
What is the average size of a part that will be imaged? What is the maximum size part that can
imaged with your machine?
What is the average weight of a part that will be imaged? What is the maximum weight that should be
supported by your current design?
What is the temperature range of castings that will be placed on the fixture?
How many parts can be imaged by your machine per unit time (parts/min, parts/hour)?
Are these machines used strictly for sand cast imaging or are they sometimes used for inspecting
forged or machined parts?
Would it be possible for you to send a sample CAD part file for us to inspect and see what geometries
we will be encountering?
Finally, a ballpark estimate on the most you would be willing to pay per fixture? and the price you
would like aim for?
We are excited to be working with you and your company. Please answer the questions to the best of
your knowledge, It is important for us to know exactly what you want so that we may meet your
needs.
We would like to schedule a meeting at your location in the near future. Fridays work best for us but if
that is not possible we can plan to meet another time.
Hi Dylan,
You will be working with Sam (Samir Anjelly) on this project from our side for the ME team. He is
not available this Friday, but perhaps the 2 of you can work out other possible timings.
Regards
Vrindesh
To what accuracy should the sensing system read? 1,5,10 degrees? 1 degree or
better
What metals or plastics are currently used in the table and fixture? Aluminum 6061
T6 at Thickness of .1875” - .125”; Sometimes we use G10/FR4 fiber glass up
to .1875” thick.
What is the average size of a part that will be imaged? What is the maximum size
part that can imaged with your machine? Up to 1.2 m Long X 500mm Wide X 400
mm High. Sand casted parts are typically smaller at about 400mm cubed
What is the average weight of a part that will be imaged? What is the maximum
weight that should be supported by your current design? Maximum Weight
Approximately 50 Lb. & height 450 mm
What is the temperature range of castings that will be placed on the fixture? 60-200
Degree F.
How many parts can be imaged by your machine per unit time (parts/min,
parts/hour)? This depends on part size & customer requirements. Typical is 20
– 30 sec /part
What materials are the castings made of (Aluminum, Iron, Steel)? Most
Automotive Casting are aluminum.
Are these machines used strictly for sand cast imaging or are they sometimes used
for inspecting forged or machined parts? No, It is for any part or process per
customer requirement but let’s consider sand casted parts for this exercise.
Would it be possible for you to send a sample CAD part file for us to inspect and
see what geometries we will be encountering? Yes, we can share example
drawings when you visit us.
Finally, a ballpark estimate on the most you would be willing to pay per fixture? and
the price you would like aim for? Let’s evaluate after we have candidate designs.
An upper limit can be arbitrarily set to $5000 – what is important is to
understand the features and functionality of the solution.