section 2 - university of alabama at birmingham · section 2 materials, design ... this...
TRANSCRIPT
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 37
SECTION 2
MATERIALS, DESIGN
& MANUFACTURING
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 39
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
PULTRUDED GLASS-GRAPHITE/EPOXY HYBRID COMPOSITES SUBJECTED TO TRANSVERSE HIGH STRAIN-RATE COMPRESSION LOADING
Mohammad Afrough, Tejas S. Pandya
The University of Mississippi Oxford, Mississippi, USA
ABSTRACT
In a previous investigation, the energy absorption and
dynamic response of different combinations of fiber-reinforced
pultruded hybrid composites made of unidirectional glass and
graphite fiber/epoxy, was investigated. It was found that placing
glass fibers in the inner core of composites results in a higher
ultimate compressive strength and specific energy absorption.
In the present work, the dynamic response of pultruded glass-
graphite/epoxy hybrids has been investigated, with specimens
of rectangular cross-section being subjected to compression
loading in a direction transverse to the fiber orientation. Crack
initiation and propagation was monitored with a high-speed
video camera, and the effects of hybridization were analyzed. It
was found that the location of glass or graphite fibers inside the
pultruded composites has no significant effect on the ultimate
compressive strength under transverse compression loading.
The energy absorption in all the hybrid specimens was almost
identical. The graphite/epoxy composite showed higher specific
energy absorption due to its lower density, and the glass/epoxy
composite had the lowest specific energy absorption.
INTRODUCTION Composites are utilized because they have desired
properties that cannot be attained by other types of constituent
materials. Fibrous composites, including reinforcing fibers
embedded in a matrix material, are commonly used in different
applications. Fiber-reinforced composite materials present
different features in terms of stiffness, specific strength,
deformation etc. Their usage encompasses a wide range of
automotive, aerospace and marine applications.
The two most common reinforcing fibers are glass and
graphite. Glass fibers have high tensile strength and low tensile
modulus. On the other hand, graphite fibers possess very high
tensile modulus, low weight and low impact resistance. A
number of investigations have been carried out on pultruded
composites [1-13]. Their results illustrate that hybridization of
both glass and graphite, with different percentages within the
same epoxy matrix, has a significant effect on the mechanical
properties.
In many applications, composite materials are subjected to
dynamic loading, which is produced by vibration or wave
propagation [14]. Therefore, it is necessary to investigate strain
rate sensitivity, failure modes and other behaviors of composite
material under dynamic loading. Several studies have been
performed on strain rate sensitivity [15-17]. Ochola et al. [18]
tested a glass fiber reinforced polymer at different strain rates.
The results show that the value of mean compressive ultimate
stress for this composite increases by 20.9% as the strain rate is
increased from 10-3
to 450 s-1
. Various set-ups like drop-weight
and split Hopkinson pressure bar (SHPB) have been used [19,
20].
Furthermore, the direction of loading and the fiber
orientation play important roles in the dynamic behavior of
composite materials. Yokoyama et al. [21] investigated the
behavior of cubic specimens of carbon/epoxy laminates in all
three principal material directions under high strain rate
compression testing and discussed the composites’ failure
mechanism. They described that by increasing strain-rate, while
the compression is along the fiber direction, ultimate strength
increases, and energy absorbed up to failure strain decreases. In
previous investigations by the authors [22], the energy
absorption and dynamic response of different combinations of
glass and graphite pultruded composites under longitudinal
compression loading was studied. It was found that locating
glass fibers in the inner core raises the ultimate compressive
strength and results in better dynamic behavior of the
composite. Since there are many applications where loads are
applied in a transverse direction, evolution and optimization of
pultruded composite materials from this point of view is a need.
The purpose of this study is to analyze the dynamic
behavior and energy absorption of pultruded glass-
graphite/epoxy hybrid composites by applying transverse
compression load at high strain-rate. A modified SHPB test
system has been used for producing dynamic load.
Additionally, a high-speed camera was deployed to record the
events and monitor crack initiation, propagation and failure of
samples during the tests.
SPECIMEN DETAILS Samples from a long beam with rectangular cross-section
composites were manufactured by a pultrusion process (Pulstar
804 machine). Glass and graphite were the reinforced fibers.
Volume fraction of epoxy was constant at 40%. The graphite
fibers were AS4W-12K (Hercules), the glass fibers were E-
Glass (PPG 2001, #12), and the epoxy was EPON 862/W/537
(Shell Chemical Company, USA). All the samples were cut
precisely in 6.6 mm x 6.6 mm sections from a long rectangular
beam of 3.3 mm thickness (Figure 1).
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 40
Figure 1. Dimensions of tested samples
Table 1. Fiber combinations with layup sequence for hybrid composites [13]
The six fiber combinations with layup sequence for hybrid
composites of samples are shown in Table 1. Reference [23]
describes more details of the manufacturing process for these
pultruded samples. The measured bulk densities of the
specimens are shown in Table 2.
Table 2. Measured average bulk densities of the pultruded hybrid specimens
CFMIX01 CFMIX02 CFMIX03
Density
(kg/m3)
1569 1924 1807
CFMIX04 CFMIX05 CFMIX06
1809 1749 1784
EXPERIMENTAL TECHNIQUE High strain-rate testing of the samples was carried out
using a modified SHPB located at the Blast and Impact
Dynamics Laboratory, University of Mississippi. A schematic
of the SHPB set-up is shown in Figure 2. It consists of a striker
bar, two strain gages, an incident/input bar and a
transmitter/output bar. The bars are made of maraging steel.
Specimens are sandwiched between the incident and
transmission bars. A copper pulse shaper was applied to attain a
triangular incident pulse.
Figure 2. Schematic of a Split Hopkinson Pressure bar [20]
Petroleum jelly was used to place specimens between the
incident and transmitter bars, and also to avoid friction and
shear effects on the samples during testing.
A high- speed video camera (HyperVision HPV-2,
Shimadzu Corporation, Japan) was employed to record the
events and monitor the crack initiation, propagation and failure
of samples during the SHPB compression loading tests. One
hundred and two frames of each event were recorded at a frame
rate of 500,000 fps. Two 1000W strobes were used to provide
the required lighting.
RESULTS AND DISCUSSION Six combinations of pultruded glass-graphite/epoxy hybrid
samples were tested. Figure 3 shows the generic stress wave
pulses including incident, reflected and transmitted pulses. For
validating the dynamic equilibrium condition, a specimen must
be in force equilibrium [22]. The stresses in both interfaces of
each specimen were calculated to ensure that equilibrium was
attained (Figure 4).
Five samples were tested for each combination at average
strain rate of 800 s-1
. Figure 5 shows typical strain-stress
behavior for all the combinations. All curves are plotted until
the shattering point of each specimen, with high-speed
photography utilized for capturing this moment. They are
essentially coincident with each other until ultimate
compressive strength, with all of them possessing equal initial
stiffness. This synchronization appears to be from the matrix
domination response of these pultruded composites, with load
applied in the transverse direction.
Figure 6 illustrates the average ultimate compressive
strength for six specimens of each combination. CFMIX01 with
60% graphite possess the highest ultimate compressive strength
and CFMIX02 with 60% glass has the lowest one. The
compressive strength of CFMIX05, the hybrid with 40%
graphite in the outer layer and 20% glass in the core, is slightly
nearer to that of graphite/epoxy. The small difference between
the highest and lowest ultimate compressive strength
demonstrates that using different portions of glass and graphite
fibers does not dramatically change the compressive strength
under transverse compression dynamic loading. This small
difference results from the stronger bond between graphite fiber
and epoxy matrix compared to the bond between glass fiber and
epoxy matrix, perhaps due to the sizing of graphite fibers. As
can be seen in Figure 7, graphite fibers (Figure 7, a), have
mostly disintegrated after the compression event while glass
fibers (Figure 7, b), remained almost intact.
Fiber Orientation
Thickness = 3.3 mm
mm
6.6 mm
6.6 mm
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 41
Figure 3 Typical Incident, reflected and transmitted pulses from SHPB test of a pultruded hybrid composite sample
Figure 4. Validation of stress equilibrium for tested pultruded hybrid composite sample
Figure 5. Typical stress-strain curve for all hybrids tested under transverse SHPB transverse compression loading
Figure 6. Compressive ultimate strength (MPa) for all hybrids tested by SHPB
Figure 7. Micrograph view of specimens after compression event (500x), (a) graphite fibers, and (b)
glass fibers
A comparison between this study and the previous study
[22] shows that in transverse compression dynamic loading, the
ultimate compressive strength of graphite-glass/epoxy
pultruded hybrids is about one-third of the longitudinal
compression dynamic loading.
As can be seen in Figure 5, all specimens absorbed almost
equivalent amounts of energy (integration of the area under
stress-strain curve) during the dynamic compression tests, but
the specimens with greater volume fraction of graphite fibers
absorbed marginally more energy. However, the specific energy
absorption (energy absorbed per unit mass) for each specimen
is distinct due to their different densities (Figure 8). As shown
in Table 2, specimens with more graphite fibers have lower
density, which resulted in higher specific energy absorption.
Specimens with 60% graphite (CFMIX01) and 40% graphite
(CFMIX05) show the highest specific energy absorption, while
60% glass (CFMIX02) exhibits the lowest specific energy
absorption.
(a)
(b)
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 42
Figure 8. Specific energy absorption for all specimens tested by SHPB
As mentioned, a high-speed video camera was deployed to
monitor the crack initiation, propagation and failure of samples
during the SHPB compression loading tests. For these samples,
the event (from time zero to crack initiation of specimen) takes
55μs -75μs. Four images of specimen CFMIX05 are shown in
Figure 9. The stress versus time history of this specimen is
depicted in Figure 10. The maximum load (ultimate strength)
occurs at 51 μs and the crack initiates at 66 μs. The image at 79
μs shows that the fiber glass in the specimen core detaches
from epoxy matrix. The results indicate that the entire sample
has not been completely damaged at the peak stress. It should
be noted that the compression load is applied transverse to the
fiber direction, i.e., load is perpendicular to the epoxy and fiber
orientation for each of the hybrid composites.
Figure 9. Four images of specimen CFMIX05 under compression at different times
Figure 10. Stress versus time history of CFMIX05
CONCLUSION In this investigation, six combinations of pultruded glass-
graphite/epoxy hybrids have been experimentally studied under
transverse high strain-rate compression loading, using a
modified SHPB. It was observed that failure of specimens
tested along a transverse direction is dominated by matrix
failure. It was also observed that 60% graphite (CFMIX01) has
the highest ultimate compressive strength. The ultimate
compressive strength was found to be marginally greater with a
higher percentage of graphite. This marginal difference results
from the stronger bond between graphite fiber and epoxy
matrices compared to a bond between glass fiber and epoxy
matrix, perhaps due to the sizing of graphite fibers. The
location of glass or graphite fibers inside the pultruded
composites had no significant effect on the ultimate
compressive strength under transverse compression dynamic
loading. While all specimens absorbed almost an equivalent
amount of energy, the graphite/epoxy demonstrated
significantly higher specific energy absorption compared to the
other combinations. This was due to its lower density, while
glass/epoxy showed the lowest specific energy absorption.
ACKNOWLEDGEMENT The authors wish to acknowledge funding received from
US Army Research Office-DURIP Grant # W911NF-13-1-
0248 for the high-speed digital cameras used in this research.
The authors would also like to thank Dr. Raju Mantena and Mr.
Seyed Soheil Daryadel for their guidance in performing this
research, and Mr. P. Matthew Lowe, for his help with sample
preparation.
REFERENCES [1] Swaminathan, R. and P. Raju Mantena., 2003, “Axial
Loading and Buckling Response Characteristics of Pultruded
Hybrid Glass-Graphite/Epoxy Composites,” Journal of
Reinforced Plastics and Composites, vol. 22, Issue 7.
[2] Nori C.V., P. Raju Mantena, and McCarty T.A., 1996,
“Experimental and Finite Element Analysis of Pultruded Glass-
Graphite/Epoxy Hybrids in the Axial and Flexural Modes of
Vibration,” Journal of Composite Materials, vol.30, No.18, pp.
[3] Mahmood M. Shokrieh, Majid Jamal Omidi, 2011,
“Investigating the transverse behavior of Glass–Epoxy
composites under intermediate strain rates,” Journal of
0 μs
66 μs
51 μs
79 μs
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 43
Composite Structures, Volume 93, Issue 2, Pages 690–696.
[4] M. V. Hosur, M. Adya, S. Jeelani, U. K. Vaidya, Piyush K.
Dutta, 2004, “Experimental Studies on the High Strain Rate
Compression Response of Woven Graphite/Epoxy Composites
at Room and Elevated Temperatures,” Journal of Reinforced
Plastics and Composites, vol. 23 no. 5 491-514.
[5] Vaughan, J.G., Roux, J.A., and Mantena, P.R., 1992,
“Characterization of Mechanical and Thermal Properties of
Advanced Composite Pultrusion,” Proceedings of the 1992
NSF Design and manufacturing Systems Conference, 1141-
1145.
[6] Mantena, P.R., Vaughan, J.G., Donti, R.P., and Kowsika,
M.V., 1992, “Influence of Process Variables on the Dynamic
Characteristics of Pultruded Graphite-Epoxy Composites,”
Vibro-Acoustic Characterization of Materials and Structures,
ASME, 14, 147-154.
[7] Ahmad Almagableh, Swasti Gupta, P. Raju Mantena and
Ahmed Al-Ostaz, 2008, “Dynamic Mechanical Analysis of
Graphite Platelets and Nanoclay Reinforced Vinyl ester, and
MWCNT Reinforced Nylon 6.6 Nanocomposites,” Proceedings
of the 2008 SAMPE Fall Technical Conference, (Paper # 034
on CD) ROM, Memphis, TN, Sep 8-11.
[8] Prakash Jadhav, P. Raju Mantena and Ronald F. Gibson,
2006, “Energy Absorption and Impact Damage Evaluation of
Grid-Stiffened Composite Panels under Transverse Loading,”
Composites Engineering, Part B, 37, 191-199.
[9] Murthy Kowsika, P. Raju Mantena and K. Balasubramniam,
2002, “Energy Absorption and Dissipation Characteristics of
Pultruded Glass-Graphite/Epoxy Hybrid Composite Beams,”
Journal of Thermoplastic Composite Materials, vol.15, no.3.
[10] P. Raju Mantena, Richa Mann and C.V. Nori, 2003, “Low
Velocity Impact Response and Dynamic Characteristics of
Glass-Resin Composites,” Journal of Reinforced Plastics and
Composites, Vol 20/No. 6/2001, pp. 513-534.
[11] Dev Barpanda and P. Raju Mantena, 1998, “Effects of
Hybridization on the Creep and Stress relaxation
Characteristics of Pultruded Composites,” Journal of
Reinforced Plastics and Composites, Vol 17, number 3, 1998,
pp.234-249.
[12] Dev Barpanda and P. Raju Mantena, 1996, “Dynamic
Mechanical Analysis of Pultruded Glass-Graphite/Epoxy
Hybrid Composites at Elevated Temperatures,” Journal of
Reinforced Plastics and Composites, vol. 15, Number 5,
pp.497-532.
[13] Sujit Kumar and P. Raju Mantena, 1996, “Dynamic and
Static Torsional Characterization of Pultruded Hybrid
Cylindrical Composite Rods,” Journal of Composite Materials,
Vol. 30, No.8, pp.918-932.
[14] Gibson R.F., 2007, “Principles of Composite Material
Mechanics,” 2nd Edition, CRC Press, Boca Raton.
[15] Sierakowski, R.L., 1997, “Strain rate effects in
composites,” Appl. Mech. Rev., Vol. 50, No. 1, Part 1, 741-761.
[16] Hamouda, A.M.S. and Hashmi, M.S.J., 1998, “Testing of
composite materials at high rates of strain,” advances and
challenges, J. Mat. Process. Technol., Vol. 77, 327-336.
[17] Al-Hassani, S.T.S. and Kaddour, A.S., 1998, “Strain rate
effects on GRP, KRP and CFRP composite laminates,” Key
Engineering Materials, Vol. 141-143, Part 2, 427-452.
[18] R.O. Ochola, K. Marcus, G.N. Nurick, T. Franz, 2004,
“Mechanical behaviour of glass and carbon fiber reinforced
composites at varying strain rates,” Journal of Composite
Structures, Vol. 63: 455–467.
[19] Yoshihiko Hangai, Naoyuki Kubota, Takao Utsunomiya,
Hisanobu Kawashima, Osamu Kuwazuru, Nobuhiro
Yoshikawa, 2015, “Drop weight impact behavior of
functionally graded alum4inum foam consisting of A1050 and
A6061 aluminum alloys,” Journal of Materials Science &
Engineering A, 639: 597–603.
[20] Bazel. A. Gama, Sergey. L. Lopatnikov, John. W.
Gillespie. Jr., 2004, “Hopkinson bar experimental technique - a
critical review,” Applied Mechanics Reviews, Vol. 57, Issue 4,
223-250.
[21] Yokoyama, T., Nakai, K. and Odamura, T., 2007, “High
Strain-Rate Compressive Characteristics of a Unidirectional
Carbon/Epoxy Composite: Effect of Loading Directions,”
Experimental Analysis of Nano and Engineering Materials and
Structures, Springer Netherlands, 681-682.
[22] Seyed Soheil Daryadel, Cameron Ray, Tejas Pandya and P.
Raju Mantena. 2015, “Energy Absorption of Pultruded Hybrid
Glass/Graphite Epoxy Composites under High Strain-Rate
SHPB Compression Loading,” Journal of Materials Sciences
and Applications, 6, 511-518.
[23] Mantena, P.R., Vangipuram R., and Vaughan, J.G., 1994
“Dynamic Flexural Properties of Pultruded Glass/Graphite
Hybrid Composites,” 39th International SAMPE Symposium,
174-182.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 44
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
KSU AUTONOMOUS GO-KART FRAME
Andrew Geiman, Dr. Kevin McFall
Kennesaw State University, Marietta, GA, USA
ABSTRACT An autonomous go-kart frame was required in order to
expand the testing capabilities of line detection and line
following. The paper below covers the design and construction
of this frame. The project life cycle was one semester and the
costs could not exceed $650. Every aspect of the design needed
to be analyzed on paper or using finite element analysis to
ensure the design would effectively meet the design
requirements. The autonomous go-kart frame would need to
hold a microcontroller, batteries, motor drivers, a camera and
sensors to perform its autonomous movement on roads. The
two forward wheels would be driven by individual motors,
while the rear wheels are on casters. This configuration allowed
the vehicle to more and steer along pavement. The final design
was successful because it was able to meet both the schedule
and the cost requirements.
Figure 1: Rendering of the final model assembly.
INTRODUCTION During the previous semester a peer designed a “Low-
cost Platform for Autonomous Ground Vehicle Research” [3].
This design was the initial testing bed for software that would
allow the kart to move autonomously through the streets of
Kennesaw State University. The project lacked funding for
structure, but a new semester leads to new prospects and better
funding. This large opportunity for design freedom and
educational expansion could not be ignored. A vision of a
higher performance, modular and easy to use go-kart frame
drove the objectives. The main goals of this project are to
improve the functionality of the go-kart in carrying capability,
strength, speed, and steering, while keeping cost low.
GATHERING INFORMATION To begin the design process, a viewing of the formula one
team’s racecar at KSU was conducted. The race team was
nationally recognized and a natural place to start looking when
dealing with a moving vehicle. Information obtained about the
steering system and wheel base dimension was vital to moving
the project forward. After the viewing it became clear that it
would be too expensive to build a custom steering system at
this time. The race team used professionally manufactured parts
and welds their frame together. The professionally cut
annuluses are then welded together by the team’s most
experienced welders. After examining the racecar material
geometry, a search for material costs was performed. Using
simple square models on paper three different quotes were put
together. The first quote was an all steel frame with thin walled
tubes that would have to be welded together to make a frame.
The second quote was simple aluminum bar stock that was bent
and riveted and the third was a quote for extruded aluminum
8020. The first round of quotes included CV joint for a shock
absorbing system, pivot joints for the steering system, and
shocks. All three quotes came in above price point, the cheapest
being the aluminum bar stock at $1010. The CV joints were
$110 each and the shocks were $15 each. These items along
with the pivot joints were cut. After running the quotes again
the aluminum bar stock came was only at $481 per appendix,
which was back inside the set price limit. The next step was a
decision matrix, to ensure the project on schedule.
PROBLEM STATEMENT Field testing on a previous generation go-kart’s wooden
frame design revealed the need for a sturdier frame. The
previous frame was difficult to drive due to its low torque
output and the tended to veer in one direction. The direct drive
system did not work when turning and the wooden panel
deflects significantly due to the weight of the equipment. Tiny
caster wheels in the front of the kart provided little turning
ability and would jam often due to deflection. All around the
design was effective for the first field test but further progress
required a stable rugged platform.
CUSTOMER REQUIREMENTS The customer required a frame which could mount 100 lbs
of equipment and still clear speed bumps at low speed. This
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 45
would allow the kart to travel along the school road and mount
any configuration of equipment required. The new frame
needed to be light, have seven sensor mount points, and be able
to absorb vibration from the road. The kart would need to be
sufficiently light to be carried by two people along with the
equipment needed to operate it. The seven sensors consisted of
one for the forward camera, an encoder on each of the two
drive wheels, and a collision distance sensor on each of the four
corners of the kart. The frame needed to be modifiable in case
unforeseen challenges arise in the future. Deflection at the
caster wheels needed to be less than two degrees to prevent
rotational jamming and sticking. The kart also had to be able to
turn. Vibration on the equipment needed be kept to a minimum
to ensure equipment life.
DESIGN SPECIFICATIONS The kart would be about four feet by four feet to match the
size of a small car and allow for parking simulations. To
accommodate a lightweight design requirement the frame
needed to weigh less than fifty pounds. The frame also needed
be able to navigate speed bumps; the average speed bump is
five inches high and eighteen inches deep. To reduce deflection,
in the equipment mounting surface and the entire kart to less
than an eighth of an inch, rigid beams with supports were used.
Materials needed to be strong, lightweight with good fatigue
life and corrosion resistance. Many of these choices came down
to three main factors; how affordable is it, can it be
manufactured and will it satisfy the customer’s requirements?
DESIGN MATRIX The design matrix using score multipliers was used to
calculate the best design. The multipliers are based on how
important that parameter is to the project. Then a datum
concept is picked at random and all other concepts are
compared to it. Once a winner is found the datum is switched to
the winner and the matrix is run again. After two matrixes the
Riveted Bar frame concept was chosen.
Table 1: Design Matrix
FRONT WHEELS A minimum of a 10” diameter wheel was required to
traverse speed bumps without the frame bottoming out. If a ten
inch affordable wheel could be found then it would satisfy this
object's selection. Many race car /go-kart manufacturers sell
wheels in sets of four and at very high prices. However, the kart
requires two different types of wheels, since there is no steering
system. Harbor freight sells affordable wheels which allows for
low cost replacement. Since it is a brick and mortar store,
replacement of these parts could be purchased same day if a
problem were to arise.
REAR WHEELS Large ten inch caster wheels are needed for this area of the
kart. The rear wheel assembly can take a maximum load of
three hundred pounds, which far exceed the working load of the
kart. Like the previous parts there needed to be cheap easy
replacement with a high safety factor. Rubber tires are prone to
wearing out and would be a hassle replacing them. With a
vendor already chosen for the front wheels it is simple to
search. Harbor freight also offers a ten inch caster system for
only $15 each with the top mounted double bearing assembly
will allow for easy rotation. Making the choice to purchase
Harbor Freight wheels saved a minimum of $50.
DRIVETRAIN The heritage motors are two 2.5 cm electric motors from
the AndyMark Inc. The 12 volt DC motors can run at 5,310
RPM unloaded. They are low torque high RPM motors. Since
the motors could not turn the kart individually the prototype
required two drivetrains for maneuvering. The data from the
prototype made it clear that a drivetrain was needed. This
combined with the steering system quotes lead to the decision
for two drive train systems. Purchasing the CIMple boxes
raised the overall cost by $100, but still saved money compared
to $350 required for steering equipment. The Andy mark
CIMple box was an obvious choice, it was built for the current
motors, the gear ratio is 16:46, and at $50 apiece they are
affordable. Adding to their allure was the fact that two of the
motors used can be mounted to it giving the kart the ability to
double its power in the future. There were also sensor mount
points drilled out to record the rpm of the wheels, which
satisfied a customer requirement. The drivetrain came with a
precision machined shaft with a keyway already cut out
allowing for a keyed coupling to be used. The mounting plate
had holes drilled ready to mount on purchase to make assembly
even easier. Since the CIMple box provided a solution for all
of the customer requirements two were purchased to be added
to the go-kart.
VIBRATION ABSORBERS COUPLINGS There are many types and styles of shock absorber
available but many of them are expensive and do not allow easy
installation. The chosen design needs to be simple, easy to use,
cheap and effective. These criteria lead to a simple rubber
cylinder with threaded bolts attached to each end for mounting.
The size and effectiveness of the absorber was unknown during
procurement, this lead to two types being purchased, one eight
millimeter large set and one six millimeter small set. The two
sizes would go through two field tests to see if they are
sufficient for use.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 46
MOUNTING PLATFORMS The platform must hold a minimum 100 pounds of
equipment; this includes multiple batteries, motor drives, and a CPU. The preferred way to get a cheap, light weight, strong and durable platform was to get eighth inch aluminum plate and mount the vibration absorber to it. Composites were considered next but the price increases when a large surface area and thickness are required. Carbon fiber plate and G10 composites were two of the best composites available. However carbon fiber was problematic to work with and G10 prices climb quickly when a four foot by one foot sheet was needed. There was also a concern that the stress concentrations in the mounting holes could crack the brittle composites due to the vibration.
SENSOR MOUNTS The material selected to hold the 4 sensors mounted on the
corner of the vehicle was Aluminum 6061-t6. It matched many of the other components and provided a very strong and versatile structure. The mount was a simple square tube that is three feet long and half an inch wide, the walls are eighth inch thick. Since the weight of the sensor and beam weight was less than half a pound combined, the main driving factor for this component was the deflection due to acceleration and deceleration. Five pounds of force were chosen to size the width and wall thickness, this was derived from the mass of the bar and the sensor assembly mass at a conservative acceleration of 7.5 feet per second squared. This was assuming the cart could reach 15 feet per second in 2 seconds. The component becomes a simple beam and the max deflection is 5.96E-5 inches. This does not account for manufacturing tolerances but under its own forces the beam was very stable.
WHEEL ROTATION SENSOR The sensors to monitor the wheel velocity were chosen
from AndyMark website. They mate up to the drivetrain easily and required no tooling holes or modifications. The optical encoder simply mounted directly on to the drive shaft, where the drive train housing had pre drilled holes.
MODELING To begin the process of modeling, the purchased pieces
were drawn in their simplest forms possible, how they shipped. This meant the manufacturing issues were brought to the forefront while still in the design phase and many of the issues could be solved early. This proved to be an advantage and made manufacturing the parts very streamlined later on. An example of this is the main beam which holds the drive train in place. If the part needed a mounting hole, one was added in using Hole Wizard and a drill size was recorded. This insured that all pieces could be made with little use of heavy machinery. The most complicated parts to model where the couplings and the rear wheel bracings. The couplings were hard due to the level of precision required and the rear wheel bracing was hard due to the bend required.
COUPLINGS With the precision ground shaft to work with a coupling
system needed to be manufactured to connect the wheels and drivetrain. This involved designing, analyzing and manufacturing them in house to save cost. Raw material would then need to be chosen and purchased ahead of time so these could be used on the prototype. The first phase was sizing the coupling dimension, things such as thickness, height, diameter and features (figure 2). The wheel bolt pattern was found using caliper and the shaft size and tolerance was given in the drivetrain specification. A key slot was also needed and this was a manufacturing issue at first. Later after talking to a few machinists it was discovered that a post machining process was necessary.
Figure 2: Rendering of coupling 1.0.
After the keyway was performed the coupling needed to be
analyzed to ensure that it was in the ballpark on thicknesses. The bolt preload needed to be considered as well and sizing the new bolts that would hold the wheel and coupling together need to be found too. After running a shear stress calculation on the included bolts, it was discovered they had a factor of safety in the teens. Next the moment and preload tension were calculated along with the combined load of all three. The FEA to support the hand calculations is below in Figures 3 and 4.
The next phase was optimizing the design, finalizing a material and tolerancing the hole for the drive shaft for a loose press fit. For material aluminum 6061-T6 was used for its great properties, the diameter and height of the part was chosen and the raw material was purchased. Coupling 2.0 (Figure 5) came out of the optimization lighter and more efficient.
In Figures 6 and 7 a clear difference can be seen in the displacement of the Rear Wheel bracket designs. For this part the driving analysis factor was the displacement of the end, because if the angle becomes too steep the caster wheels will bind. This was clearly demonstrated again in Figures 10 and 11, where the factor of safety never drops below five.
All components were analyzed using a combination of finite element analysis and hand calculations. The entire frame assembly underwent many simulations to determine if the system would work. (Figures 12, 13, and 14) Ensuring the
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 47
system will not fracture when loaded to the specified weight.
The results of the coupling FEA can be seen in Figures 8 and 9.
The coupling is constrained at the bolt holes with a force and
torque applied to the center drive shaft holes.
Figure 3: Factor of Safety fringe plot of Coupling 1.0.
Figure 4: Fatigue Life fringe plot of Coupling 1.0.
Figure 5: Rendering of coupling 2.0.
FEA
Figure 6: Displacement fridge plot of Rear Wheel Bracket
1.0.
Figure 7: Displacement fridge plot of Rear Wheel Bracket
2.0.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 48
Figure 8: Stress fridge plot of Coupling 2.0.
Figure 9: Displacement fridge plot of Coupling 2.0.
Figure 10: Factor of Safety fridge plot of Rear Wheel
Bracket 1.0.
Figure 11: Factor of Safety fridge plot of the Rear Wheel
Bracket 2.0.
Figure 12: Von Mises Stress fridge plot of the Final
Assembly.
Figure 13: Factor of Safety fridge plot of the Final Assembly, with the lowest FOS being 6.495 around the
platform holes.
UAB
FATU
Figu
Figu
School of Eng
Figure 14: D
UIGE
ure 15: Dama
ure 16: Life C
gineering – Me
Displacement Assem
age Percent aplot of Coup
Cycles fridge p2.0
echanical Engin
fridge plot of bly.
after a millionpling 2.0.
plot of Rear W
neering - ECT
the Final
cycles fridge
Wheel Bracke
TC 2015 Procee
e
t
Figu
Csimulacompobecom
BUCK
F
Tpackagbucklithe kabuckliThis yieldethe lefone orwas aplatfor
PRE-T
motorscheduwas as
edings – Vol. 1
ure 17: Dam
Critical compoation packageonents under
me damaged du
KLING ANAL
Figure 18: Buc
The entire strucge, with data loing was a largeart. Figure 18 ding; the right was further
ed very low disft hand platforr two millimeta critical factrm further.
-ASSEMBLYThe only compr, drivetrain, anule to allow fossembled using
14
age Percent fBracket
onents were e (Figures 15-
cyclical loadiuring the life of
LYSIS
ckling Analys
cture was testoaded directly e concern whedoes show areahand platformanalyzed usinsplacement valrm in Figure 1ters. This was tor in decidin
ponents that nend wheel assemor use on the teg the AndyMar
fridge plot of 2.0.
analyzed us-17). This ensing would nof the vehicle.
is of the Entir
ed using a bufrom the stres
en considering as which could
m deflects twelng hand calculues. It can al8 does not deffurther cause fng whether to
eeded pre assembly. This wasemporary cart. rk Inc. publishe
Page 4
Rear Wheel
ing a fatigusured that key
ot wear out o
re Vehicle
uckling analysis FEA. Isolatethe usability o
d suffer isolatelve millimetersulations whichlso be seen thaflect more thanfor concern ano analysis th
embly were ths done ahead oThe drive trained video.
9
e y
of
is d
of d s. h at n d e
e of n
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 50
MANUFACTURING The first pieces to be manufactured were the two
couplings; this was a request by the customer. Once the
material, tooling and FEA were in and completed
manufacturing could begin. The tolerances and dimensions
were double checked per machining good practices. The CNC
machine schedule was developed before this project but is was
modified to ensure minimal tool damage. However two end
mill bits were broken on the first coupling, the first broke due
to a feed speed error. The aluminum became hot and gummy
due to the temperature, and then bonded to the carbide steel end
mill bit. The second bit was smashed by the machine due to
operator error. After the first bit broke the machine was started
half way through the milling cycle, this was a huge mistake!
The end mill bit was driven deep into the material at the
jogging velocity and the bit was shattered. After these two
mistakes were corrected the milling process went smoothly.
With the coupling done, the tolerances and drive train assembly
were tested before the delivery deadline. The next pieces to be
manufactured were the main 90 degree corner pieces which
allowed for the cart to begin taking shape. The bends were
made on a Chinese pipe bender and then holes were drilled in
the desired pattern with a #30 drill bit. Match holes were then
drilled in the main beams to ensure a clean fitting joint. Once
the pattern was copied and the mating joints were numbered,
Clecos were installed. These are temporary mechanical rivets,
and allow for the joint to be assembled and disassembled with
no damage to the material. The Clecos allowed for a rigid
layout to be done and a few mistakes were found and then
eliminated. Cut lengths were checked and then the two middle
support structures with L-brackets were added to the frame. The
forward camera plate was riveted into place along with the
camera post bracket, followed by the mounting holes for the
drive train. The drive train was not attached yet to allow for
easy maneuvering of the assembly. The rear wheel mounting
brackets were cut and bent followed by the bracings. These
were then mounted to the frame using match holes and then
riveted into place. It was important to allow for all of the
manufacturing imperfection and again, match holes were used
to ensure the rear wheels could be mounted properly. The
wheels and drive train were added next to form an almost
complete cart. The main mounting plates were added using
match and tapped holes, then sensor beam and finally the
camera post was put into place.
PERFORMANCE VALIDATION Three tests were performed on the frame before it was
delivered. The first test was done by analyzing the frame for
excessive deflection when loaded with hand pressure. As
rudimentary as it sounds this was a vital step in checking the
handmade kart. The criteria for passing the test were no visual
damage or errors in the construction. When this test was passed
the kart was loaded to 70% the max recommended load and
pushed at speed. This checked the operations of the vehicle
under load without damaging it in case of issues. The criteria to
pass this test were no excessive or unpredicted displacement
and smooth operation of the vehicle. The third and final test
was to load the vehicle to 140% of the recommended load. The
vehicle was not operated under these conditions but inspected
for any mechanical failures. The FEA analysis showed that this
load would not damage the kart and performing the test ensures
the quality of the manufacturing. The criterion for success was
no failure and no audible straining.
FUTURE RECOMMENDATIONS Many future recommendations are to improve the karts
performance life and increase its weight capacity. The top
recommendation is to add a support beam from the forward
middle rail to the forward beam that has the camera mounted to
it. This would help eliminate vibration in the camera post and
provide a clearer picture. Adding two more motors to the frame
would allow for faster operating speed and more realistic
testing of the visual code. The final recommendation for the
vehicle is to add support material to both rear and front wheel
supports. These four areas were under the most stress. Adding a
welded backbone to all four of the rear brackets would allow
for a 40% increase in operation load. Also adding a bottom
mounting plate supported on both side to the forward motor
mount would allow for the same load increase.
CONCLUSION The fabrication of the frame took over forty man hours,
with another forty-eight hours of CNC time. The frame includes
wheels, tire, motors, structure, drive train and sensors mount
points. The combined weight of these items was less than fifty
pounds allowing for a two man team to easily carry an
unloaded structure. Final costs of the project totaled $612,
which was $130 more than the original estimate but still less
than the price limit. This was a 21% error in estimated cost but,
there was a surplus of material when the project ended. All of
this time and money gave way to a go-kart frame that is light
weight, durable, modular, and stable. The kart can hold one-
hundred pounds of equipment that allows for autonomous
driving testing. It makes it cheaper and easier to test the next
wave of technology that will make the world a safer place.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 51
REFERENCES [1] Budynas, R. G., and Nisbett, J. K., 2011, Shigley's
mechanical engineering design, McGraw-Hill, New York.
[2] Oberg, E., 2012, Machinery's handbook a reference book
for the mechanical engineer, designer, manufacturing engineer,
draftsman, toolmaker, and machinist, Industrial Press, New
York.
[3] Ollukaren, N., and McFall, K, 2014, Low-cost Platform for
Autonomous Ground Vehicle Research, Proceedings of the
14th Early Career Technical Conference, Vol. 13.
[4] Wahab, M. A., 2008, Dynamics and vibration: an
introduction, John Wiley, Chichester, England.
[5] “www.AndyMark.com,” AndyMark Robot Parts Kits
Mecanum Omni Wheels.
APPENDIX A
APPENDIX B
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 52
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
AZI-DRIVE: A NOVEL THRUSTER MAPPER FOR TWIN PROPELLER MARITIME VESSELS
Andrew Gray
University of Florida Gainesville, Florida, U.S.A
Kevin French
University of Florida Gainesville, Florida, U.S.A.
Jacob Panikulam University of Florida
Gainesville, Florida, U.S.A
Zach Goins University of Florida
Gainesville, Florida, U.S.A
Dr. Eric Schwartz University of Florida
Gainesville, Florida, U.S.A
ABSTRACT
Over-actuated surface vessels, while highly maneuverable,
can be challenging to control. This paper describes the process
of implementing the controls of a novel propulsion system onto
a small, fully autonomous surface vehicle (ASV). Two
independent azimuth steered rimless propellers provide
propulsion for the vessel. Using a nonlinear control allocation
with singularity avoidance method, the approach was tested
with simulations and then implemented on an ASV. Using the
singularity avoidance variable, users could adjust the
responsiveness of the system. Testing confirmed that the
method allowed the ASV to move in all three dimensions
required of a surface vehicle.
INTRODUCTION For most maritime vessels, a propeller and rudder are
adequate to allow a boat to transit from one point to another.
However, for complex maneuvers, such as station holding
performed by dynamic positioning (DP) vessels, more
complicated propulsion systems are required. These systems
are generally over-actuated [1] and require complex control
software to utilize the thrusters.
Students from the Machine Intelligence Lab (MIL) at the
University of Florida have a history of developing autonomous
maritime vessels [2][3][4][5]. PropaGator 2, MIL’s most recent
ASV, shown in Figure 1, was designed to have a top speed of
10 knots while maintaining the maneuverability of a DP vessel
[6]. To accomplish these characteristics, a planing hull was
developed with a low drag coefficient in the forward direction.
For propulsion, two student-designed azimuth thrusters were
installed on the back half of the vessel.
While the hull design and propulsion system on
PropaGator 2 allowed for high speeds, the complexity of the
system created a challenging control problem. In simulation,
the two thrusters were able to control the boat simultaneously
in all three dimensions [7], but implementation proved difficult.
Additionally, the placement of the thrusters at the rear
combined with large drag resistance across the perpendicular
plane made strafing a challenge. The thrusters produce more
thrust in the forward direction than in the reverse. While the
drag of the propellers is very small, the drag force experienced
by the boat is far greater in the reverse than in the forward
direction. Another challenge was that due to the plowing
characteristics of the boat, speed must be considered in its
control.
The propulsion system on PropaGator 2 does not fit the
model of a traditional DP ship. First, PropaGator 2 is only six
feet long. Most DP vessels are meant for the open ocean and
are hundreds of feet long. PropaGator 2 is only able to operate
in small bodies of water with limited waves. Traditional DP
ships will have two to three times more thrusters than
PropaGator 2. Numerous thrusters allow for redundancy and
can make the vessel’s movements more responsive. With only
two azimuth thrusters, the ASV cannot afford to lose a thruster;
in some cases, the ASV may be less responsive than a vessel
with four or five thrusters.
To control the propulsion system on PropaGator 2, a
student-implemented constrained nonlinear control allocation
Figure 1. The autonomous boat PropaGator 2
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 53
with singularity avoidance thruster mapper was used. The
thruster mapper allowed the ASV to move in any direction
while maintaining the ability to hold its position and heading.
This ability classified PropaGator 2 as a DP vessel and was
proven by testing the platform in several small lakes.
PLATFORM SUMMARY PropaGator 2 was made to be fast, maneuverable, and
lightweight. A catamaran hull was selected as the hull type
because catamaran hulls typically produce less drag when
compared to a similar sized monohull [8]. Hard chines were
added to the bottom of the vessel to allow for a planing hull,
using previously conducted hydrofoil research [6]. A hard
transom was installed at the back of the boat. The hard transom
caused the water to cleanly separate from the boat in order to
reduce drag. Fiberglass was selected for the hull material
because of its strength and low weight. For propulsion, two rim
driven azimuth thrusters, shown in Figure 2, were installed near
the stern. To reduce drag, both propellers are enclosed in Kort
nozzles and are independently steered with servos inside the
hull.
To allow the boat to “sense” its environment, a suite of
sensors was installed. A Spartan AHRS-8 IMU, paired (through
a Kalman filter) with a GPS module, allowed PropaGator 2 to
determine its position, heading, roll, pitch, and yaw. For object
detection and recognition, an IDS imaging camera and a SICK
LIDAR were installed onto the front of the boat. For
communication between the shore and the boat, a wireless
router was used.
Two types of drive systems were considered: a differential
drive and a constrained azimuth drive system. Originally, a
differential drive system was considered. The thrusters would
have been locked in a forward facing direction during this
configuration. Turning was handled by differential thrust.
While control of the vessel would be simple, station holding
and strafing would not have been possible with this type of
drive system. Because of these limitations, the azimuth system
was chosen, resulting in better maneuverability.
The azimuth thruster drive system took advantage of the
thrusters’ ability to create thrust vectors in different directions.
This allowed the boat to maintain its heading z (rotation) while
traversing in the x (surge) and y (sway) axis as shown in figure
3. However, because the thruster system is over-actuated, the
solutions to produce the desired thrust vectors were not unique.
Additionally, the thrusters could not rotate more than several
full revolutions due to cable binding. To prevent binding, the
thrusters were limited to 180 degrees of rotation at a rotation
speed of 340 degrees per second. With the new constraint on
the thruster rotations, producing thrust vectors became much
more challenging. A method was needed that could determine
appropriate vectors while using a constrained and over-actuated
system. To achieve this goal, a constrained nonlinear controller
was implemented on PropaGator 2.
NONLINEAR CONTROL ALLOCATION The constrained nonlinear control allocator used on
PropaGator 2 was created specifically for surface vessels with
multiple thrusters and was meant to be solved in real time [9].
Additionally, the allocator had a term which penalized solutions
containing singularities. In order to use the allocator, the
propulsion system had to be defined first. Each thruster had to
be characterized by the maximum and minimum amount of
thrust , maximum and minimum angle limitations , and the
maximum and minimum rotation speed . The allocator
initially developed for surface vessels is:
Figure 3. Three axes of motion performed by PropaGator
2
Figure 2. CAD model of the azimuth thruster used on PropaGator 2
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 54
(1)
where
, (2)
, (3)
, (4)
, (5)
The first term in equation (1) is the total power
consumption, of the actuators. The next term, ,
adds a penalty based on the force difference . is found by
subtracting the achieved force from the desired force .
The weights in the matrix are chosen so that whenever
possible. In an effort to limit drastic changes, the third term
penalizes changes in thruster azimuth angle. The matrix is
tuned until drastic changes are acceptably limited. The fourth
term penalizes singularities, thereby affecting the
maneuverability of the vessel through the use of the variable.
A high would result in high maneuverability at a cost of more
power. Whereas a low value would result in a less responsive
but more efficient solution.
The above equation is nonconvex, and solving it is
generally intractable for real-time applications. The
optimization is considered nonconvex because of the
potentially nonlinear power cost in (1), the nonlinear constraint
(2), and the singularity avoidance nonconvex term in (1). The
high computation time caused by the nonconvex equation
would make the optimization problem not useful to applications
that required the solutions in rapid succession like PropaGator
2. In order to make the computation tractable, a linear
approximation of the constraints and dynamics yielded a
tractable quadratic program.
The resulting approximation is as follows [9]:
(6)
linear constraints
, (7)
, (8)
, (9)
, (10)
IMPLEMENTATION To make PropaGator 2 controllable on the water, three
levels of software were required: the path planner, the
controller, and Azi-drive. At the highest level, the path planner
was used. The path planner constantly checked the position
and orientation of the boat. Once given a new goal position
from the user, the path planner produced a desired position and
orientation (DPOSE) to guide the boat towards the goal
position. The DPOSE was placed two meters away from the
boat, between the current position and the goal position. As
the boat moved towards the DPOSE, the DPOSE was moved
closer to the goal position. When DPOSE was less than two
meters from the goal position, the distance between the DPOSE
and the boat was decreased at a linear rate determined by the
path planner, until the boat was within a one meter radius of the
goal position.
Moving the boat in the right direction based upon the
position of the DPOSE was performed by the controller. Using
a proportional derivative (PD) controller, the responsiveness of
the boat could be changed. Differences in headings and
position from the DPOSE were translated into a wrench (i.e., a
force and torque vector). The wrench was then passed to the
Azi-drive shown in figure 4.
Azi-drive converted the wrench values into thruster
positions and thrust amounts by solving the associated
quadratic program. In order to produce a thrust vector that best
fits the given wrench, Azi-drive calculates the best value and
translates that value into thruster forces. The best fit solution is
based upon minimizing the error in the cost function. The
singularity avoidance gain, , governs the cost for producing a
solution in a singularity state, i.e., one where enormous
(potentially infinite) thrust is required to produce particular
wrenches.
Avoiding singularities was important for PropaGator 2
because singularities caused the boat to become unstable. A
Figure 4. Flow chart of maneuverability, starting at the
path planner and ending at thruster commands
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 55
singularity with the thrusters was encountered when the
thrusters were turned toward each other. The thrusters were
still generating individual forces, but the boat did not move
because of the orientations. Gradually, the controller increased
the amount of thrust generated by each thruster in an attempt to
move the boat. The increase in thrust did not move the boat,
and the cycle repeated, forcing the user to step in and stop the
program (i.e., causing the autonomy to fail).
The thruster mapper solution rate was too slow when Azi-
drive was first programmed. Python was used to initially solve
the optimization problem. Python is a convenient system to
quickly develop programs. However, because Python is a
runtime language, programs written in the language run
significantly slower than compiled languages. Using Python,
the computer on PropaGator 2 was able to solve the
optimization problem at a rate of 10 Hz. The slow update rate
caused oscillations in orientation of the boat that could not be
avoided. A faster solution for the optimization problem was
necessary.
The code-generation tool, CVXGEN, provided a way to
rapidly solve the nonlinear control allocation equation. To use
CVXGEN, a convex optimization problem is defined in a high
level form. The high-level definition is passed to the CVXGEN
tool, which in turn produces a quadratic program solver in C
[10]. CVXGEN quickly solves convex optimization problems
by unrolling matrix multiplications, directly accounting for
sparsity, and taking advantage of problem structure at
generation time. The CVXGEN-generated solver found
solutions up to 25x faster than the previous Python
implementation, allowing the solver to avoid numerical issues
as a consequence of linearization. Once applied to the boat,
the responsiveness increased tenfold.
Prior to testing PropaGator 2 in the water, changes applied
were tested in a simulator (see Figure 5). While the simulator
had basic physics capabilities built in, it did not give a perfect
representation of how the boat would perform. However, the
simulator allowed for immediate feedback to changes made in
software. The simulator also provided a chance to practice
tuning the PD controller gains prior to in-water testing.
After testing the boat in simulation, PropaGator 2 was
transported to a local pool for testing. The pool provided a
controlled environment where gain changes to the PD
controller were more apparent due to the placidity of the water.
Another advantage of using the pool was that students could be
in the water with the boat to physically disturb the vessel in
order to observe its reactions to changes in heading and
position.
Because of the thruster configuration, tuning one of the
axes affected the other axes. This required all three axes to be
tuned simultaneously. The processing of tuning all three axis
continued until the boat was acceptably able to maintain its
position and heading.
Once the team was satisfied with the performance of
PropaGator 2 at the pool, the vessel was placed in a small lake.
The lake provided a large enough area to allow the boat to
travel long distances across the water. During the longer
movements, the team was able to observe how PropaGator 2
tracked across a straight line. If the boat tracked poorly, the
gains in the PD controller were slightly adjusted accordingly
until the performance of the boat was satisfactory.
RESULTS In order to prove that Azi-drive was working properly, the
actual wrenches were compared to the desired wrenches. To
find the actual wrench values produced by Azi-drive, the
following equations were used:
(11)
, (12)
, (13)
, (14)
(15)
(16)
Equation (11) is the actual torque, , experienced by the
boat. This is found by taking the sum of the torque created by
both thrusters. The torque created by either the port or
starboard thruster was found by multiplying the force in the x-
direction, , with the thruster’s y-distance from the center of
rotation, . Torque from the y-direction force, is found by
multiplying the force by the x-distance, . Summing the two
products produced the thrusters torque as shown in (12). In
equations (13) and (14), the sum of the forces along the x axis
and y axis are shown. Finally, in order to find the force in each
direction, equations (15) and (16) were.
Data from testing was recorded for post analysis of the
Azi-drive. The topics recorded were: desired wrench, thruster
force, and thruster direction. Since the rates of recording for
each movement were different and inconsistent, a script was
created. The script was created to find the actual wrench in real
time by subscribing to the topics that produced the thruster
Figure 5. Simulator used to test software prior to boat testing
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 56
force and thruster angle. In the following paragraphs, the
results of the script are discussed.
To show that Azi-drive was working, the actual forces and
torque were compared to the desired forces and torque. To
make the graphs easier to read, the actual forces and torque
were averaged with the previous five samples, shown in
Figures 6, 7, and 8, where the y-axis is the force or torque and
the x-axis is time:
As the graphs above show, Azi-drive was able to produce
forces and torques that were close to the desired results given
by the controller. The actual X-force varied slightly from the
desired X-force. This variation can be attributed to the
optimized solution, compensating with the X-force in order to
maintain a closer relation with the Y-force and torque in Figures
6 and 7. In Figure 7, the Y-forces created can be clearly seen to
match the desired forces. A small amount of latency can be
seen as the thruster mapper responds to wrench changes from
the controller. Finally, the actual torque is shown to be very
similar to the desired torque. Once again, latency is seen as the
Azi-drive attempts to match the desired torque request from the
controller. The above graphs prove that Azi-drive works well
as a thruster mapper and has potential for use on other
platforms.
CONCLUSION This paper covered the research, implementation, and
testing of a novel thruster mapper used to control the ASV,
PropaGator 2. Two azimuth thrusters were selected as the
propulsion system because of the maneuverability gained when
compared to a traditional differential drive system. A
constrained non-linear control allocation optimizer was
implemented in order to control the thrusters while constraining
them to a limited amount of rotation. Once Azi-drive was
tested and applied to the boat, PropaGator 2 was able to move
in all three dimensions (surge, sway, and yaw) when traversing
from point to point.
Having only two thrusters in the back of the boat made
controlling of PropaGator 2 difficult. Adding one or more
thrusters to the bow would simplify the control problem. Even
if the thrusters were fixed and could only produce a force along
one axis, this force would assist with swaying the boat, a
movement that proved difficult in the current configuration.
Having a dynamic two stage control system would help to
overcome the issue with the boat handling differently at higher
speeds due to planing. At higher speeds, the boat generally
Figure 8. The solid line is the desired torque and the dashed line is the averaged actual torque
Figure 7. The solid line is the desired Y force and the dashed line is the averaged actual Y force
Figure 6. The solid line is the desired X force and the dashed line is the averaged actual X force
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 57
does not need to strafe along the y-axis and could be
maneuvered using a differential drive behavior. A control
method that allowed movement along the y-axis would be
activated when operating at lower speeds. Finally, improving
the path planner by allowing it to factor in the velocity of the
boat would make transitioning between points smoother.
Currently, the path planner does not account for the momentum
of the boat. As the boat approaches the waypoint, it tends to
overshoot the point due to the boat inertia.
ACKNOWLEDGEMENTS The authors would like to thank the students and faculty
members who helped with the operation of the boat, the
University of Florida College of Electrical and Computer
Engineering and the College of Mechanical and Aerospace
Engineering departments. An extended thank you to Dr.
Schwartz and Dr. Arroyo and to the Machine Intelligence Lab.
Thank you Dr. Crane for transportation and the use of the lab
equipment. A special thank you to Jacob Mattingley for
generously providing an academic license for CVXGEN.
REFERENCES [1] Berge, S. P., & Fossen, T. I. (1997). Robust control
allocation of overactuated ships; experiments with a model
ship. In Preprints of IFAC conference on maneuvering and
control of marine craft, Brijuni, Croatia.
[2] Gray, A., Shahrestani, N., Frank, D., and Schwartz, E.,
2013, “PropaGator 2013: UF Autonomous Surface Vehicle,”
AUVSI Foundation’s 6th Annual RoboBoat Competition,
Virginia Beach, VA.
[3] Walters, P., Sauder, N., Nezvadovitz, J., Voight, F., Gray, A.,
and Schwartz, E., 2014, “SubjuGator 2014: Design and
Implementation of a Modular, Fault tolerant AUV,” AUVSI
Foundation’s 17th Annual RoboSub Competition, San Diego,
CA.
[4]: Frank, D., Gray, A., and Schwartz, E., 2014, “PropaGator
2014: UF Autonomous Surface Vehicle,” AUVSI Foundation’s
7th Annual RoboBoat Competition, Virginia Beach, VA.
[5] Nezvadovitz, J., Griessler, M., Voight, F., Walters, P., and
Schwartz, E., 2015, “SubjuGator 2015: Design and
Implementation of a Modular, High-Performance AUV,”
AUVSI Foundation’s 18th Annual RoboSub Competition, San
Diego, CA.
[6] Frank, D., Gray, A., and Schwartz, E., 2014, “PropaGator 2:
A Planing Autonomous Surface Vehicle With Azimuth Rim-
Driven Thrusters,” Proc. 14th Annual Early Career Technical
Conference, Birmingham, Alabama, pp. 185-190.
[7] Patel, D., Frank, D., & Crane, C. (2014, October).
Controlling an over-actuated vehicle with application to an
autonomous surface vehicle utilizing azimuth thrusters. In
Control, Automation and Systems (ICCAS), 2014 14th
International Conference on (pp. 745-750). IEEE.
[8] Lamb, T., 2004, Ship Design and Construction Vol. II,
Society of Naval Architects and Marine Engineers, Chap. 45.
[9] Johansen, T., Fossen, T., & Berge, S. P. (2004). Constrained
nonlinear control allocation with singularity avoidance using
sequential quadratic programming.Control Systems Technology,
IEEE Transactions on, 12(1), pp. 211-216.
[10] Mattingley, J., & Boyd, S. (2012). CVXGEN: a code
generator for embedded convex optimization. Optimization and
Engineering, 13(1), pp. 1-27.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 58
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
ON-MACHINE MEASUREMENT OF CUTTING TOOL DAMAGE BASED ON OPTICAL KNIFE-EDGE DIFFRACTION
ChaBum Lee, Kwun-Lon Ting
Tennessee Technological University Cookeville, TN, USA
Joshua A. Tarbutton
University of South Carolina Columbia, SC, USA
Sun-Kyu Lee Gwangju Institute of Sci. & Tech.
Gwangju, South Korea
Hwan-Sik Yoon University of Alabama Tuscaloosa, AL, USA
Jonathan D. Ellis University of Rochester
Rochester, NY, USA
ABSTRACT
This research has been proposed to National Science
Foundation CMMI (The Civil, Mechanical and Manufacturing
Innovation) division for 2015. The objective of this research is
to create a new methodology for damage detection of cutting
tools used in machine tools based on the knife-edge diffraction
principle. A detection technique for cutting tool failures is
desperately needed for condition-based maintenance in
machining processes. Cutting tool wear, chipping, and/or
breakage can affect the dimensional accuracy and surface
roughness of products. Firstly, we will characterize the
geometry and edge morphology of various cutting tools for
turning by using high resolution microscopes, and we will
relate those properties with knife-edge diffraction
characteristics. Secondly, we will create a free-space knife-edge
interferometer and establish image correlation methods to
quantify the cutting tool wear, roughness, chipping, and even
breakage by relating edge diffraction fringe curves to cutting
tool conditions. Lastly, we will create an optical fiber-based on-
machine measurement system to enable effective measurement
of the cutting tool conditions and to integrate measurement and
instrument techniques with machine tools. For cutting tool
damage applications, this research will answer fundamental
questions regarding the performance boundaries of knife-edge
interferometry and on-machine measurement technologies, and
produce new knowledge on the measurement, instrumentation
and sensors.
INTRODUCTION Damage of cutting tools is influenced by the stress state
and temperature at the tool surface. This in turn depends on the
cutting mode, such as turning, milling, or drilling, the cutting
conditions and the presence or absence of cutting fluid and type
[1-4]. In machining, the tool damage mode and the rate of
damage are very sensitive to changes in the cutting operation
and the cutting conditions. Thus it is necessary not only to find
the most suitable cutting tool and work material combination
for a given machining operation, but also to reliably predict the
cutting tool life.
Figure 1. A damage classification of the cutting tools according to damage size [4].
Figure 2. Cutting tool edge morphology: new (left) and damaged (right) [5].
Mechanical damage is classified as wear or fracture
depending on its scale. Figure 1 illustrates the different modes,
from a scale of less than 0.1 pm to around 100 pm (much
greater than 100 pm becomes failure). Abrasive wear is
typically caused by sliding hard particles against the cutting
tools. The hard particles come from either the work material’s
microstructure, or they are broken away from the cutting edge.
Abrasive wear reduces as the tool hardness relative to particles
increases, and it generally depends on distance cut. Attrition
Rak
e
Surface
Rak
eS
urface
FlankS
urface
Flank
Surface
0.3 μm 0.3μm
Roundness withinRoundness within 50 nm
Cutting edge Micro Chipping
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 59
wear occurs on a scale larger than abrasion. Particles or grains
of the tool material are mechanically weakened by micro-
fracture as a result of sliding interaction with the work, before
being removed by wear. Next comes chipping (sometimes
called micro-chipping at its small-scale limit) as seen in Fig. 2.
This is caused by mechanical shock loading on a scale that
leads to large fluctuations in cutting force, as opposed to the
inherent local stress fluctuations that cause attrition. Finally,
fracture is larger than chipping, and is classified into three
types: early stage, unpredictable and final stage. The early stage
occurs immediately after beginning a cut, if the tool shape or
cutting condition is improper; or if there is some kind of defect
in the cutting tool or in its edge preparation. Unpredictable
fracture can occur at any time if the stress on the cutting edge
changes suddenly, for example by chattering or by an
irregularity in the workpiece hardness. Final stage fracture can
be observed frequently at the end of a tool’s life in milling, and
fatigue due to mechanical or thermal stresses on the cutting
edge is the main cause of this damage [1,2].
With the progress of micro/nanoscale wear, the surface
roughness and dimensional accuracy of the machined part
deteriorates, which consequently leads to changing the worn
cutting tool. In practice, the most important consideration in
selecting cutting tools is the tool life, which defines the time
corresponding to the prevailing tool-life criterion. At the end of
the cutting tool’s life, it must be replaced or reground to
maintain workpiece accuracy, surface roughness or integrity.
However, cutting tool life exhibits significant scatter and is
problematic in nature. The cutting tool conditions are visually
inspected by machine operators in the majority of cases. In
addition, the scanning electron microscope (SEM), atomic force
microscope (AFM), touch-triggered stylus probe (TSP), stereo
microscopes with digital camera attachment, computer vision
systems, and coordinate measuring machine (CMM) can be
used to measure cutting tool wear parameters [6-9]. For
nanoscale analysis [10-13], more advanced measuring
techniques using white light interferometry or confocal
microscopy can be applied. However, there is a lot of
discomfort with characterizing and quantifying cutting tool
conditions with these methods because machine operators have
to separate the cutting tools from the tool fixtures for
inspection.
The integration of metrology into the manufacturing
process is becoming increasingly important for production
technology. Measurement directly done on machine tools is
especially improving manufacturing strategies. So far, optical
or tactile measuring systems are the state-of-the-art for
dimensional measurements in process metrology [12-16].
Many studies related to on-machine measurement (OMM)
techniques have been focused on the machined surface
characterization. However, the machined surface topography
may be influenced by not only machining conditions, but also
cutting tool conditions such as wear. Unfortunately, the OMM
for cutting tools has not been reported, and operators inspect
the cutting tools visually to ensure that they are still within
specific geometry tolerances. However, micro-nanoscale
cutting tool damage cannot be clearly observed by visual
inspection. Alternatively, operators remove the cutting tool
from the fixture and inspect it for damage by using CMM or
high resolution microscopes such as SEM. Unfortunately, these
methods are not sufficient to provide the cutting tool
information such as morphology, wear, and chipping
conditions, in a quantitative way. In addition, operators suffer
from relocation of the cutting tools, and this error will also
badly affect the machining results. Thus the measurement
process must be integrated into the manufacturing process on
machine tools for production technology. This paper describes a
new methodology based on a knife-edge diffraction
interferometry (KEI) technique for OMM that can be used for
nonprobe-compatible CNC turning centers.
KNIFE-EDGE SENSING PRINCIPLE
Figure 3. Knife-edge diffraction by randomly rough edge surface morphology.
The knife-edge diffraction method was recently introduced,
and a deterministic model to characterize edge diffraction
phenomenon according to edge roughness was created [17,18].
Assuming that a Gaussian beam with the beam diameter α is
incident on the knife-edge, the total field to the front of the
knife-edge is simply the incident field, and the beam propagates
along the z axis as illustrated in Fig. 3. The incident field sE
can be defined as,
.),(),,(2
22
0
ss yx
Ajk
sssssss eeyxEzyxE
(1)
where, Es is the amplitude of the incident field, k0 is a wave
vector in air, α is the beam width, and A and B are the distances
between the optics components. The first and second terms are
an expression of the incident wave with Gaussian intensity
distribution, and the last term is for an aperture of the field. The
incident field sE
creates constructive and destructive
interference due to phase function with respect to the knife-
edge displacement. The superimposed wave can be derived by
using Fourier transform (FT) and the total field dE
(sum of
transmitted and diffracted field) at the detector, can be obtained
by applying an inverse FT of the superimposed wave [19-21].
The total field to be measured along the z axis can be defined
from the inverse FT relation of the incident field as,
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 60
.
)2(
1),,(
0
0
2
dydxdkydkxe
eEzyxE
d
s
rkj
spacek
rkj
saperture
dddd
(2)
where jir
is a vector from i to j coordinate system.
Figure 4. 2 axis flexure mechanism with knife edge sensor.
Figure 5. Calibration results of KES
Optical metrology systems can be extremely precise but
are fundamentally limited by the surface quality of optical
components. As an approximated model to a rough knife-edge
surface is presented, the phase difference between the
transmitted and diffracted fields is related to the roughness,
assuming that the roughness is randomly distributed in nature.
The roughness on the knife edge boundary is assumed to
comprise a nonzero mean with Gaussian probability density
function (pfd) [20]. Based on this assumption, knife edge
surface roughness is modeled in a pdf form as,
2
2
00
2
2
2
1)(,)(
.)2(
1),,(
h
spacek
rkjrkj
saperture
ddd
R
d
ehpdfkxdydxdhdkydhpdf
eeEzyxE ds
(3)
where R represents a rough surface and h is the random height
along the y0 axis and σ is roughness, standard deviation of h as
illustrated in Fig. 3. Here, only the roughness along the x0 axis
was considered, independent of the y0 axis.
*
0)( R
d
R
dd EEyI
(4)
where, the total power Id induced by the knife-edge at detector
will be calculated by multiplying the total field and the
conjugated total field at each detector plane. The asterisk (*)
denotes a complex conjugate.
Figure 6. Schematics of fiber-coupled OMM system.
This knife-edge technique has been recently applied to
displacement sensing. The displacement sensor [17,18], or so-
called, knife-edge sensor (KES), was implemented into the XY
nanopositioning system (150×150mm2, 20mm thickness) based
on parallel kinematics. The monolithic configuration was
designed and fabricated to yield a compact layout as seen in
Fig. 4 and to provide more than ±1.0mm range. The two KES
were embedded into the stage and measure XY displacement to
control the positioning of the stage in real time. The laser diode
(LD, λ 650nm, α 5.0mm) light is 50:50 separated at BS and
travels to each KES with surface roughness ~0.1µm. Voice coil
motors were used as actuators. Four photodetectors (PD,
3×3mm2) were installed at the center of the stage to receive the
transmitted and diffracted light behind the knife-edge. Figure 5
showed that the KES sensitivity is 8.99mV/µm and
9.02mV/µm within ±500µm with nonlinearity 0.60% and
0.73% along the X and Y axes, respectively.
The phase-locked loop (PPL) based on metrology concept
has also been introduced using lensed fibers for on-machine
surface topography measurement [12,13]. Methods for
measuring the machined surface topography using fiber-optics
have been widely employed because it is cheap and installation
is easy compared to others [16]. Furthermore, advances in
fiber-optic technology have allowed a single-mode fiber (SMF)
to become useful in evaluating the machined surface in an on-
machine condition.
Many kinds of SMF shapes have been reported, such as the
graded index lensed fiber, wedge shaped fiber, and photonic
crystal lensed fiber [22, 23]. However, they are not appropriate
XY
LD PrismBS
PD
OKE
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 61
for measuring the surface topography directly in a micrometer
scale because the beam spot size was relatively larger than that
of the surface topography.
Figure 7. Measurement results of machined Cu surface: (a) Zygo and (b) our OMM
The experimental setup of the proposed PLL-based
interferometer is shown in Fig. 6. The light emitted from the
pigtailed LD (λ 650 nm) is coupled into a 2 × 2 fiber coupler
and split into two beams; one goes to the lensed fiber and the
other is detected at PD2. The light reflected from the sample
surface is recollected by the lensed fiber. The whole system was
composed of all-SMF optic components and an SMF operating
within a wavelength range of 630–680nm was used. The
fluctuation of the light source could be monitored at PD2, and
the power fluctuation of an LD caused by the coupled cavity
may cause measurement errors. Therefore, the feedback loop
for the stabilization of an LD with a proportional-integral (PI)
controller was used. The end of the lensed fiber attached to the
mechanical modulator vibrates and is modulated with the
piezoelectric transducer (PZT). The interference signal is
detected at PD1. After phase demodulation by digital signal
processing in real time, the phase of the machined surface can
be extracted and the surface profile can be obtained.
Figure 8. Free-space KEI experiment setup of cutting tool damage detection
For comparison, the surface profile of the sample was
measured by a commercially available white-light
interferometer microscope (Zygo), which had an average
roughness Ra of 7nm. The surface profile was measured by the
proposed fiber-optic Fizeau interferometer and compared in
Fig. 7. This surface profile with a small dent in a narrow width
is normally difficult to measure, even for the commercially
available white-light interferometer. The proposed
measurement result showed results similar to those obtained
with a white light interferometer. In this experiment, it was seen
that the measured surface profile using the proposed phase-
locked interferometry showed a good agreement with that of
the white-light interferometer microscope and had an
observable flaw size depth of about 100nm.
CUTTING TOOL DAMAGE DETECTION METHOD Fundamental understanding of the cutting tool morphology
properties and its quantitative measurement and
characterization methods are not well-documented. A KEI
technique has recently been created for sharp knife-edge profile
characterization, but fundamental questions on the
interferometric effects of randomly-varying rough knife-edge
remain unanswered. This establishes a new paradigm for the
future study of damage detection and morphology
characterization of cutting tools. Unlike an amplitude-splitting
interferometer, such as a Michelson, the KEI utilizes
interference of a transmitted wavefront as a reference and a
diffracted (or scattered) wavefront under test at the knife-edge.
The interference fringes created by diffraction at the cutting
tool edge are highly sensitive not only to the cutting tool’s
geometry but also to its morphology.
In this study, KEI effects will be tested on various cutting
tools for turning, and then, we will quantify these edge
conditions by creating free-space and fiber-coupled OMM
systems for damage detection and morphology characterization
of the cutting tools. This research promises to create a new
transformative damage detection principle in cutting tools for
improving reliability and promoting automation of
manufacturing processes. The successful implementation of this
research will potentially lead to a fundamental shift in
measurement methodologies in structural health monitoring of
cutting tools such as morphology property characterization and
the life cycle expectancy of cutting tools. Additionally, this
research will lead to the widespread use of a new measurement
methodology that can be used for a variety of edge surface
sensing, monitoring, and inspection for AFM and STP probes,
grinding wheels, or milling tools. This work will focus on three
step-wise strategic aims:
1) Cutting tool edge morphology characterization: Characterize
the geometry of various cutting tools for turning by using
conventional microscopes. Determine the amplitude and power
of harmonic wave components for cutting tool edge
morphology modeling by a fast Fourier transform analysis.
Relate the cutting tool edge geometry model with knife-edge
diffraction characteristics according to various edge
morphology conditions.
2) Free-space KEI technique for cutting tool damage detection:
Create a KEI setup and calibrate the displacement of a knife-
edge with a capacitive type sensor. The displacement of the
knife edge can be considered as a wear effect of cutting tools.
Establish image correlation methods to quantify the cutting tool
wear, roughness, chipping, and even breakage by relating edge
diffraction fringe curves with cutting tool edge conditions.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 62
Also, investigate knife-edge effects of cutting tools with
different shapes in the same way above.
We will build free-space KEI test bench on an optical
isolator to verify the proposed knife-edge model as seen in Fig.
8. The solid-state laser (Verdi G2, Ti:Sapphire pumping, λ
532nm, Coherent Inc.) and silicon photodiode detector (918D,
Newport Inc.) with pinholes (10, 50, 100, 200, 500, 1000μm)
will be used. The modulated Ti: Sapphire laser will be incident
on the cutting tool edge through a polarizing filter and spatial
filter, and the PD can detect the interferogram created by
superposition of the transmitted and diffracted lights. For a
preliminary study, we will monitor changes in the
interferogram by scanning the cutting tool with picomotor PZT
actuator (30nm resolution). Nanopositioning control will be
implemented with a capacitive sensor (CS, 10nm resolution,
Lion Precision Inc.). A lock-in amplifier (LIA, Stanford
Research Systems Inc.) will collect the signals from modulator
and PD and allow high accuracy measurement through a
demodulation process. All data produced in the experiment will
be monitored in real-time in a LabView environment.
Figure 9. Auto correlation of normalized total power curves: (a) normalized total power curves according to
the cutting tool conditions and (b) cross correlation (abbreviated to Corr) results of two sets of curves.
Spatial cross correlation statistics will be used to measure
and analyze the cutting tool wear, especially for edge
morphology. As studied earlier, the more the edge morphology
is rough and dull, the more incoherent the diffracted field
becomes because the phase between the transmitted and
diffracted fields changes according to edge morphology
conditions. The interference of the transmitted and edge-
diffracted field can theoretically explain the general oscillatory
behavior. Note that as the edge roughness increases, the
amplitude oscillation decreases due to the destruction of the
phase interference between the transmitted and diffracted fields
as seen in Fig. 9. Assuming a nanoscale edge morphologic
change results in decrease in the oscillation amplitude and is
too small to cause a shift of the curves, the cross-correlation
technique as a measure of similarity of two curves with no lag
can again be performed. In a preliminary study, the cross
correlation coefficient of two curves (new f v.s. used g1, g2, or
g3) decreased as the edge became rough. Firstly, three new
diamond cutting tools with various pinhole sizes will be
measured and the most sensitive condition will be chosen.
Secondly, each cutting tool will be used to cut the metal
(Cu/Al) and an interferogram of each cutting tool on every 1
mile basis will be measured. Lastly, a correlation between
signal processing and each interferogram will be related with
the cutting tool wear condition. For reference, we will use AFM
and SEM images. The next objective is to apply a correlation
method to cutting tool condition monitoring, to establish a
quantitative relationship between correlation and cutting tool
micro/nanoscale morphology, and to provide a sensing limit of
free-space KEI and its signal processing for the cutting tool
damage detection in terms of accuracy, resolution, repeatability,
stability, and uncertainty.
Figure 10. Fiber-coupled OMM system setup.
3) Fiber-coupled OMM system for cutting tool damage
detection: Create an optical fiber-based delivery method for an
OMM configuration, implement fiber-coupled OMM system,
characterize sensing performance of cutting tool damage in the
same way done above, and compare it with free-space KEI
results. Determine the viability for characterizing and
quantifying the cutting tool damage for fiber-based OMM
configurations using KEI technique.
The chosen optical fiber-based probe has a pigtailed laser
diode (PLD) module that has the same wavelength, 532nm, as
the Verdi G2. This system is highly flexible to adjust optical
offset distances, and it is reliable, because it utilizes fiber-based
light delivery as seen Fig. 10. The best optical offset distance
between the fiber-end and cutting tool will be determined by
not only properly adjusting the distance but also theoretically
calculating the distance. The light from the fiber-based probe
has a divergence angle along the propagation direction and will
show a different interferometric behavior with the Ti-Sapphire
laser system. The fiber-coupled measurement system will be
characterized in terms of resolution, accuracy, repeatability and
stability and these outputs will be compared with those of a
free-space measurement system.
CONCLUSION Until recently, there was no way to quantitatively inspect
damage and/or breakage of cutting tools other than visually, and
these morphology properties and characterization methods have
not been well-documented. This research promises to create a
new and transformative damage detection principle in cutting
tools for improving reliability and promoting automation of
manufacturing processes. This research will also answer
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 63
fundamental questions regarding the performance boundaries of
the knife-edge interferometer and on-machine measurement
technologies for cutting tool damage detection applications, and
it will produce new knowledge on the measurement,
instrumentation and sensors.
The fundamental questions on the interferometric effects of
randomly-rough knife-edge will be answered. In addition, we
will determine the fundamental sensing limits in free-space and
fiber-coupled on-machine measurement systems by testing
performance, compatibility, and reliability. The successful
implementation of this research will potentially lead to a
fundamental shift in measurement methodologies in structural
health monitoring of cutting tools, such as morphology property
characterization and life cycle expectancy of cutting tools. In
addition, this research will lead to the widespread use of a new
measurement methodology that can be used for a variety of
edge surface sensing, monitoring, and inspection tasks for
atomic force microscopy or touch-triggered stylus probes,
diamond cutting tools, grinding wheels, or milling tools.
ACKNOWLEDGEMENT The research was supported by NSF (Award Number:
CMMI 1463502/1463458) through the Tennessee
Technological University. Also, the authors gratefully
acknowledge use of facilities and instruments of the Gwangju
Institute of Science and Technology.
REFERENCES [1] G Boothroyd and W A Knight, 2005, Fundamentals of
Machining and Machine Tools 3rd
Edition, CRC Press.
[2] W Grzesik, 2008, Advanced Machining Processes of
Metallic Materials, Elsevier.
[3] J Ahmad, 2009, Machining of Polymer Composites,
Springer.
[4] Fderick W Taylor, 1907, On the Art of Cutting Metals,
American Society of Mechanical Engineers.
[5] Product: Ultra-Precision cutting tool UPC, A.L.M.T. Corp.
(retrieved on Sep. 5, 2014) at http://www.allied-
material.co.jp/english/products/diamond/cutting/upc/
[6] R Hocken and P Pereira, 2012, Coordinate Measuring
Machines and Systems, Second Edition, CRC Press.
[7] F Waters, 2003, Fundamentals of Manufacturing for
Engineers, UCL Press.
[8] C Zhang, 2011, “Tool wear measurement of ball-end cutter
based on shape mapping,” Adv. Mater. Res., Vol. 291-294, pp.
1905-1908.
[9] Renishaw Inc. materials (retrieved on Aug. 30, 2015) at
http://resources.renishaw.com/gen/details/rlp40-touch-probe--
25514.
[10] A Devillez, S Lesko and W Mozer, 2004, “Cutting tool
crator wear measurement with white light interferometry”,
Wear, Vol. 356, Iss. 1-2, pp. 56-65.
[11] Veeco Inc. Materials (retrieved on Sep. 02, 2015) at
http://www.veeco.com/pdfs/database_pdfs/High_Speed_Machi
ning_28.pdf.
[12] C Lee, J-H Kang, J-Y Joo, and S-K Lee, 2011, “Phase-
locked loop based topography measurement of ultraprecision
machined surface using the ball lensed and tapered fiber,” The
10th Intl’ Symp. Meas. Technol. Intelligent Instrum., Jun. 29-
Jul. 2.
[13] J-H Kang, C Lee, J-Y Joo, and S-K Lee, 2011, “Phase-
locked loop based on machine surface topography measurement
using lensed fibers,” Appl. Opt., Vol. 50, No. 4, pp. 460-467.
[14] T J Ko, J W Park, H S Kim and S H Kim, 2007, “On-
machine measurement using a noncontact sensor based on a
CAD model, International Journal of Advanced Manufacturing
Technology,” Vol. 32, pp. 732-746.
[15] Y Takaya, 2014, “In-process and on-machine measurement
of machining accuracy for process and product quality
management: A review,” Intl’ J. Automation Technol., Vol. 8,
No. 1, pp. 4-19.
[16] W Gao, 2010, Precision Nanometrology: Sensors and
Measuring Systems for Nanomanufacturing, Springer.
[17] C Lee, S-K Lee and J A Tarbutton, 2014 “Novel design
and sensitivity analysis of displacement measurement system
utilizing knife edge diffraction for nanopositioning stages,” Rev.
Sci. Instrum., Vol. 85, 095113.
[18] C Lee, S-K Lee and J A Tarbutton, 2015, “Positioning
control effectiveness of optical knife edge displacement sensor-
embedded monolithic precision stage,” Sens. Actuators A:
Physical, Vol. 233, pp. 390-396.
[19] B A Davis and G S Brown, 2012, “Diffraction by a
randomly rough knife edge,” IEEE Trans. Antennas Propag.,
Vol. 50, No. 12.
[20] E Hecht, 1987, Optics 4th
Ed., Addison-Wesley Longman
Inc.
[21] J W Goodman, Introduction to Fourier Optics 3rd
Ed.,
2005, Roberts & Company Publishers.
[22] Y Ryu, Y Shin, D Lee, J Y Altarejos, E Chung and H-S
Kwon, 2015, “Fiber-optic probe design for efficient photon
collection in scattering media,” Biomedical Opt. Express, Vol.
6, Iss. 1, pp. 191-210.
[23] S Kato, S Chonan and T Aoki, 2014, “High-numerical-
aperture microlensed tip on an air-clad optical fiber,” Opt. Lett.,
Vol. 39, Iss. 4, pp. 773-776.