Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 65
SECTION 3
DYNAMIC SYSTEMS
& CONTROL
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 66
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 67
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
ORBITAL MECHANIC SIMULATION OF A 1U CUBESAT
Alicia Ratcliffe
University of South Alabama Mobile, Alabama, USA
Carlos Montalvo
University of South Alabama Mobile, Alabama, USA
ABSTRACT
The overall goal of this program is to investigate sending
multiple CubeSats discretely into orbit and connecting them to
form a monolithic system. Composite satellites have greater
abilities in areas such as data collection, mission, and
surveillance capabilities. The work investigated here will be a
start for a much larger program to build the tools necessary to
perform research in this area. Initial work will be done on
analyzing the orbit of a single satellite with the potential to
expand the simulation to multiple CubeSats.
INTRODUCTION CubeSats, small satellites found in low or medium Earth
orbit (LEO/MEO), typically collect data such as atmospheric,
radiation, and geographic readings. With a volume of one liter
and a mass up to 1.33 kilograms, they were originally designed
to set a standard for future picosatellites. Since these satellites
have a comparatively low price tag, there has been an increase
in demand for new technologies and research. Multiple
universities and institutes around the world have sent their own
CubeSats into space for development, research, earth science,
and other uses. For example, Auburn University’s CubeSat,
AubieSat -1, tested how well different solar panels endured
against the environment as well as transmitting Auburn
University’s motto, “War Eagle” back to Earth in Morse code
[1]. CubeSats were not originally designed to a set standard.
The concept began in 1999 as a design challenge for graduate
students and, as the program evolves, now building on sixteen
years of direct experience and 58 years of small satellite
technology [2], it is updated regularly to improve efficiency and
availability [3].There have been 341 missions identified by
Saint Louis University, including the first CubeSat sent into
space by the military just a year after the CubeSat concept was
proposed. PicoSAT 1 and 2, weighing 250 grams each, were
launched into orbit from the rocket Minotaur-1 on February 6,
2000. The PicoSATs’ primary mission was to demonstrate the
feasibility of deployable space technology and in addition, to
act as an amateur radio transmitter. Their primary mission was
a success, and as a result Universities deploy their own
satellites to facilitate research instead of attaching devices to
other spacecraft or designing systems for another agency [4].
Government agencies such as the National Aeronautics and
Space Administration as well as the European Space Agency
are also interested in these picosatellites, assisting various
teams’ CubeSats into orbit around Earth
[5]. In particular
NASA Ames Research Center’s project, PhoneSat, makes it
possible to control a CubeSat from a generic smartphone [6].
Looking to push farther afield, NASA also issued a challenge in
2013 to design a CubeSat which operates in Mars’ environment
and furthers our knowledge. They provided a list of potential
science or infrastructure objectives [7]. So far, little work has
investigated the possibility of physically connecting, not
tethering, multiple CubeSats while in orbit, but quite a bit of
work has been accomplished regarding defining and modeling
discrete flight and formation flight.
Although little work has been done on physical connection,
except tethers, quite a bit of research and development is being
accomplished. The Quakesat was sent into orbit by Quake
Finder, a company based in California. It provided a “proof of
concept,” collecting ULF (earthquake precursor) signals from
space. The NEE- 01 Pegaso was launched by the Ecuadorian
Space Agency and was the first CubeSat to transmit real time
video from orbit to be broadcast over the internet [8]. Due to
launch in 2015, multiple CubeSats are being developed to
demonstrate the utility of multiple small spacecraft working
together. The specific goal of this mission, the Edison
Demonstration of Smallsat Networks (EDSN), is to acquire
data from multiple points on Earth simultaneously [5].
Distributed satellite systems like the EDSN mission present a
faster and less expensive method of retrieving data while
allowing greater flexibility in formation. The ability to dock
two autonomous satellites for them to work jointly would
increase the likelihood that monolithic systems previously
unsuited to formation flight could be assembled in space. The
Hawaii Space Flight Laboratory has developed flight-like
software that may be used in real time simulation environments
to dock small satellites in a cooperative manner [9]. To help
further the goal of cost effective and capable satellites; orbital
mechanics, control law and simulation will be implemented.
Orbital Mechanics is a broad subject encompassing many
topics including the orbital environment, flight mechanics, type
of orbit, and constellation architecture [10]. The orbital
environment in LEO or MEO provides high bandwidth (data
transfer rate) and low communication lag, making it an ideal
area to maintain a satellite; however, it is becoming congested
with space debris, which increases the possibility of collision
[11]. Such matters may be avoided using a control system to
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 68
help the satellite system behave in a desired manner [12]. As a
subject, control is concerned with operating a dynamic system
at minimal cost. Operating the system requires an output
(typically pointing direction or relative position) that is sent to
an onboard controller, which dynamically compensates for
unwanted disruptions such a floating missiles of space junk.
Attitude orbit and control systems (AOCS) are required for all
space missions, and the more specialized functions on a control
system may include formation flying or orbital rendezvous
[13]. To solve this problem, one must specify which
performance function should be optimized within specified
parameters [14]. The general solutions to the optimal control
problem have been thoroughly documented in [15] and AOCS
in [14]. Some common ways the motion of a picosatellite is
controlled is using whip antennas, deployable arrays, tethers,
reaction wheels and moment gyros. Since CubeSats are
relatively new, there is no common qualification standard and
these mechanisms are largely custom made for a single mission
[16].
This paper will attempt to analyze a single 1U CubeSat
with the goal of simulating multiple satellites in orbit. Once a
model is built, control laws will be implemented to connect
each satellite in orbit. In order to analyze a 1U CubeSat, the
MATLAB programming language will be used to create a
sophisticated three degree of freedom simulation tool. A
satellite dynamic model will be placed in first order form for
successful implementation to be carried out by the software.
NOMENCLATURE Ek Kinetic energy
Ep Potential energy
mi Mass of body i
First derivative of position
G Gravitational constant
Lagrangian set
MATHEMATICAL MODEL
I. CIRCULAR ORBIT
Johannes Kepler was a pioneer of elliptical orbits and came
up with “three empirical laws of planetary motion:
1. The planets orbit the sun in elliptical orbits, with the
sun at one focus of the ellipse.
2. The radius vector sweeps out equal areas in equal
times.
3. The period of the orbit is proportional to the
semimajor axis cubed.
These laws constitute a complete solution to the two-body
problem of orbital motion [17]. The problem is that of
mathematically depicting the motion of two separate bodies in
space, such as a planet and a satellite. It begins with the
equations for kinetic and potential energy:
(1)
(2)
The center of gravity is defined as the point about which
all the gravitational torques vanish regardless of the orientation
of the body with respect to the gravitational field. The
equations shown above are a linearly independent set of
coordinates that can be used to formulate the Lagrangian
equations of motion [17]. Conventional Lagrange equations of
motion are defined in equations (3)-(5):
(3)
(4)
(5)
where r is defined as:
(6)
Performing this calculation for both bodies leads to the
following equations:
(7)
(8)
(9)
Equation (9) is simply the result of combining equations
(7) and (8). Integrating equation (9) twice yields two constants.
The significance of these two constants is that the system as a
whole is fixed in an inertial frame that follows a straight path
where the center of mass of the system is not accelerating.
Thus, the two body problem can be reduced to a single system
by combining equations (7) and (8):
(10)
Substituting the variable leads to the much
more compact form for the two body problem.
(11)
It is further noted in [14] that it is much easier to accurately
determine than it is to determine G (Earth’s gravitational
constant), so it may be preferable to use . This system is
modeled with two degrees of freedom such that
(12)
The equations of motion in vector form are then
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 69
(13)
Software has been implemented to allow the simulation of
multiple satellites. Constellation architecture and formation
flight are commonly used in communication systems and GPS
to provide full coverage of Earth’s surface. However, a limit of
a monolithic satellite of multiple CubeSats would only be
capable of scanning part Earth’s surface at any one moment and
as such is unsuited to act as a positioning system.
II. CONTROL SYSTEM
In order to perform a docking maneuver, the satellites must
be brought close together in orbit without damaging the
spacecraft. For the purpose of this simulation, velocity is
assumed to change instantaneously. The most efficient method
of bringing two spacecraft close in circular orbit is by
accelerating tangent to orbit, thus entering a faster, elliptical
orbit. This new orbit has a larger period ( ). Equation (14)
implies that the two satellites will rendezvous after one orbit
has occurred for the control satellite denoted .
(14)
represents the time it takes to traverse the distance at the
specified radius shown in Figure 1- Sample phase angle
between two representative satellites.
Figure 1- Sample phase angle between two representative
satellites
This time may be calculated using Equation (15).
(15)
The craft uses two engine impulses, which require two changes
in velocity: and to match orbits. The first impulse
moves the craft into the transfer orbit, and the second redirects
into the final, matched orbit. To calculate these changes in
velocity:
where the first change in velocity is the difference of the
maximum elliptical velocity and the lead satellite’s
initial velocity.
(16)
(17)
The semi-major axis, , is defined as:
(18)
Once has decreased for rendezvous, a burn is performed by
to match the velocity and direction of ,
entering into circular orbit.
SIMULATION SET UP: A candidate case has been run with two satellites of mass
and Due to the magnitude of the
mass, radius, and gravitational constant of Earth, the equations
of motion in [14] have been nondimensionalized using the
following parameters: and
This leads to being roughly equivalent to
which, when nondimensionalized, is denoted , and in this case
is equivalent to [18]. Geostationary orbit (GSO)
has the chosen orbital period of one day, intentionally matching
Earth’s sidereal rotation period. The initial conditions of the
satellite are set in GSO with an average orbital velocity of
.
RESULTS:
An example simulation has been run using the parameters
defined in the section above. Figure 2 shows the
nondimensional orbit of two separate CubeSats orbiting the
Earth. The CubeSats have been enlarged to show detail and are
in their initial positions.
Since the radius of is constant, a graph depicting
the radius or velocity over time would be a straight line with
components sinusoidal in nature and a phase difference of .
The radius of however is not a straight line; it is
parabolic in nature with the maximum taking place once the
satellite takes a phase angle of with respect to the rendezvous
point. As seen in Figure 3, the velocity decreases parabolically
over time until it has traveled half of its period. The velocity
then increases at the same rate along the transfer orbit’s larger
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 70
circuit and period. As the phase angle decreases the control
system is initiated and the velocities of the two spacecraft take
the same value.
Figure 2- Unit Orbit for a two satellite system
Figure 3- Velocity of
Figure 4 shows the phase angle over the course of one
orbit. Initially, the absolute value of the angle increases due to
the burn performed that places the CubeSat into the transfer
orbit. The operation causes the secondary CubeSat to lose
velocity so that the distance and phase angle between the two
satellites decrease.
CONCLUSION: This work verifies that multiple CubeSats may be docked
in orbit. A case study is generated of the rendezvous and
docking of two satellites in geostationary orbit. In future work,
more satellites may be added and a more sophisticated control
system and docking maneuver implemented. Four more degrees
of freedom should also be incorporated to model the tumbling
and altitude of a spacecraft as it is deployed and a control
system created to orient the structure towards Earth correctly.
Figure 4- as the control system is implemented and
rendezvous occurs.
ACKNOWLEDGMENTS This research was supported by the University Committee
on Undergraduate Research as well as the Mechanical
Engineering department at the University of South Alabama.
REFERENCES [1] “Auburn University Student Space Program,” Auburn
Univ. Stud. Space Program [Online]. Available:
http://www.space.auburn.edu/. [Accessed: 17-Jun-2015].
[2] Kramer, H. J., and Cracknell, A. P., 2008, “An overview of
small satellites in remote sensing,” Int. J. Remote Sens., 29(15),
pp. 4285–4337.
[3] 2015, “CubeSat Design Specification Revision 13.”
[4] “CubeSat Database - swartwout” [Online]. Available:
https://sites.google.com/a/slu.edu/swartwout/home/cubesat-
database. [Accessed: 17-Jun-2015].
[5] Mahoney, E., 2013, “CubeSats Initiative,” NASA [Online].
Available:
http://www.nasa.gov/directorates/heo/home/CubeSats_initiative
.html. [Accessed: 17-Jun-2015].
[6] Wall, M., “Tiny Cubesats Set to Explore Deep Space,”
Space.com [Online]. Available: http://www.space.com/29306-
cubesats-deep-space-exploration.html. [Accessed: 17-Jun-
2015].
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 71
[7] “Home | 2015 SpaceApps Challenge” [Online]. Available:
https://2015.spaceappschallenge.org/. [Accessed: 17-Jun-2015].
[8] WiSP Information Update, 1993, “WiSP Information
Update” [Online]. Available:
http://www.ne.jp/asahi/hamradio/je9pel/.
[9] Nunes, M. A., Sorensen, T. C., and Pilger, E. J., “On the
development of a 6DoF GNC framework for docking multiple
small satellites,” AIAA Guidance, Navigation, and Control
Conference, American Institute of Aeronautics and
Astronautics.
[10] Logsdon, T., 1998, Orbital Mechanics: Theory and
Applications, John Wiley & Sons.
[11] Working Group 2, 2013, Stability of the Future LEO
Environment, Inter-Agency Space Debris Coordination
Committee.
[12] Nicholas, A. K., 2013, “Attitude and formation control
design and system simulation for a three-satellite CubeSat
mission,” Thesis, Massachusetts Institute of Technology.
[13] “Control Systems,” Eur. Space Agency [Online].
Available:
http://www.esa.int/Our_Activities/Space_Engineering_Technol
ogy/Control_Systems. [Accessed: 17-Jun-2015].
[14] Graham C. Goodwin, S. F. G., and Mario E. Salgado,
2001, “Linear Quadratic Regulator,” Control System Design,
Prentice Hall.
[15] Palacios, L., Ceriotti, M., and Radice, G., 2013,
“Autonomous distributed LQR/APF control algorithms for
CubeSat swarms manoeuvring in eccentric orbits” [Online].
Available: http://www.iac2013.org/dct/page/1. [Accessed: 17-
Jun-2015].
[16] Paul Oppenheimer, 2009, “CubeSat Mechanisms
Workgroup.”
[17] George W. Collins, II, 2004, The Foundations of Celestial
Mechanics, Pachart Foundation dba Pachart Publishing House.
[18] California Institute of Technology, “Astrodynamic
Constants” [Online]. Available:
http://ssd.jpl.nasa.gov/?constants.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 72
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
A BEGINNER’S GUIDE TO CONTROLLER AREA NETWORK BUS ACCESS IN MODERN VEHICLES
Timothy D. Fisher
Kennesaw State University Kennesaw, GA, USA
Kevin McFall
Kennesaw State University Kennesaw, GA, USA
ABSTRACT
A controller area network (CAN) is a communication
system designed so that a microcontroller governing multiple
systems (nodes) on the network can function effectively. This
network can be accessed, interpreted, and manipulated by an
external computer. The purpose of this paper is to provide a
basic working knowledge of CAN architecture and protocols,
as well as show how to connect to and decode a vehicular
controller area network.
INTRODUCTION CAN protocol was introduced by the Society of
Automotive Engineers (SAE) in February of 1986 [1] as a
multiplexed system for sending messages between devices in
an automobile. The first production vehicle implementing the
system was the BMW 8 Series line, starting in 1989. In 1992,
CAN 2.0 was published with provisions for devices with both
11-bit and 29-bit identifiers (often referred to as CAN 2.0A and
CAN 2.0B, respectively). CAN 2.0 remains the foundation of
present day CAN architecture. The most recent update to the
CAN protocol, published in 2012, is CAN FD 1.0, or CAN with
Flexible Data-Rate. [2] This new format allows variation in
both the size of the data package and in the bit rate. CAN FD
1.0 is fully back-compatible with CAN 2.0.
CAN 2.0 is also used in the on-board diagnostics (OBD-II)
system, which allows an external device to communicate with
an automobile and assess its functionality. Starting in 1996, the
OBD-II system was made mandatory in many types of vehicles
throughout the world, and specifically those sold in the United
States and the European Union (which uses a marginally
different standard called EOBD).
Though CAN FD 1.0 was the last major update to the CAN
protocol, the technology continues both to improve and to
provide a basis for other improvements. In 2007, the New
Jersey State Police integrated the pursuit light package in its
patrol cars into the CAN bus in order to reduce extraneous
wiring in the cockpit, and to integrate more subsystems so that
maintenance time could be reduced. [3] In 2014, General
Motors introduced a wireless connection system called ViCAN
with the goals of making it possible to connect devices to the
network without physically wiring them into the bus, and to
further reduce the amount of wiring needed in the vehicle by
removing the need for a common wire pair connecting every
device. [4] Earlier this year, a new safety system that
automatically dims headlights for oncoming traffic, checks for
short circuits in vehicle wiring that can cause fires and rapid
battery drain, detects flammable fumes near the engine, and
monitors engine temperature was introduced at a conference at
Hindustan University. [5] Most recently, Kennesaw State
University began research into autonomously controlling a
2012 Kia Optima via manipulation of the CAN bus. This is the
project that generated the information presented in this paper.
BACKGROUND The CAN bus itself is really very simple. It consists of two
wires connected to every CAN device in the vehicle. Instead of
one wire carrying a signal and the other functioning as ground,
one carries high voltage and one carries low voltage. A signal
on the low voltage line is designated as a logical 1 (recessive),
and a signal on the high voltage line is designated as a logical 0
(dominant).
An individual message that a CAN node sends is called a
frame. Every frame begins with an identification number
unique to each node. The ID numbers also set message priority
on the CAN bus, with lower ID numbers taking precedence
over higher numbers. For instance, a high-priority system such
as the Engine Control Unit would have a very low ID number,
and low-priority systems such as power locks would have a
higher ID number. This precedence is established through the
following protocol, known as arbitration.
When one or more CAN nodes transmits a logical 0, all
nodes receive a logical 0. When one or more CAN nodes
transmit a logical 1 and at least one other node transmits a
logical 0, all nodes receive a logical 0, including the ones
transmitting a logical 1. When a node transmits a 1 and sees a 0
during identification, this tells the node that there is a higher
priority message being transmitted by a different node, and it
will wait until it receives an End-of-Frame signal before
reattempting transmission.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 73
Priority on the CAN bus is indicated by a node’s
identification number, as noted above. A lower identification
number will have a lower binary value, and it will win
arbitration over nodes with higher ID numbers. If a node
completes its ID transmission without interruption, it will
transmit the rest of its message.
Take, for instance, a conflict between two hypothetical
nodes with IDs of 47 and 11, respectively, which starts on the
same clock cycle. The CAN bus arbitration for their 11-bit ID
transmission is displayed in Table 1. ID Arbitration.
Table 1. ID Arbitration
Bit: 10 9 8 7 6 5 4 3 2 1 0
Node
11 0 0 0 0 0 0 0 1 0 1 1
Node
47 0 0 0 0 0 1 Not transmitting
CAN 0 0 0 0 0 0 0 1 0 1 1
Table 2. Can Base Frame Format
0 0 0 0 0 0 0 0 1 0 1 1 0 0
0 0 0 0 1 0 1 1 1 0 0 1 0 0
1 1 0 1 0 1 0 0 1 0 1 0 0 0
1 1 1 1 1 1 1 1 1 1 Inter-frame space
Table 3. CAN Extended Frame Format
0 0 0 0 0 0 0 0 1 0 1 1 1 1
1 0 0 0 0 1 1 0 0 1 1 1 0 0
0 1 0 1 0 0 0 0 0 0 1 0 1 1
1 0 0 1 0 0 1 1 0 1 0 1 0 0
1 0 1 0 0 0 1 1 1 1 1 1 1 1
1 1 Inter-frame space
ID
IDE
DLC
DATA
CRC
ACK
EOF
Misc. Figure 1. Legend for Shading in Tables 2 and 3
A standard frame can have between 55 and 131 data bits in it,
depending on whether it uses a base frame with 11 ID bits or an
extended frame with 29 ID bits. How much DATA is included
in the frame (between 1 and 8 bytes, or 8 and 64 bits) also has
an effect. Aside from the ID and DATA bits, the frames are
mostly similar.
The first bit of a frame is always a dominant 0 to cut
through the recessive background and alert other nodes that a
frame is beginning. The next 11 bits are the node ID.
The subsequent bit changes depending on whether the base
frame format (BFF, seen in Table 2) or extended frame format
(EFF, seen in Table 3) is used. The shading key for Table 2 and
Table 3 can be found in Figure 1. In BFF, this bit after the node
ID is the remote transmission request (RTR) bit, which is
dominant for data frames and recessive for remote requests.
Normally nodes transmit information as a matter of routine, but
if a node wishes to query a different node for information, it
will set the RTR bit to 1 as an alert to other nodes that it is
requesting information. In EFF, this bit instead becomes the
substitute remote request (SRR), and will always be recessive.
After the RTR/SRR bit comes the identifier extension bit
(IDE). This will be dominant if using BFF, and recessive if
using EFF. If the IDE bit is recessive, the next 18 bits will be
the remainder of the ID, and the 19th will be the RTR. In BFF,
the IDE will be followed by 1 dominant bit, and in EFF, the
RTR will be followed by 2 dominant bits.
From this point, BFF and EFF are formatted the same way.
The next 4 bits are the data length code (DLC), which describe
how many bytes the following DATA field will be (up to 8).
The DATA field contains the message that the rest of the frame
acts as an envelope for, ensuring its integrity and that it goes to
the correct destination. The contents of the DATA frame will
change depending on which node it was generated by or
addressed to (e.g. lock all doors, activate right turn signal, set
tachometer to 4500 RPM).
Following this is a 15 bit cyclic redundancy check (CRC)1
to test data integrity, and 1 recessive CRC delimiter bit. The
frame then drops to a recessive acknowledgement (ACK) bit,
during which any other node which has been paying attention
will transmit a dominant bit if it has read the current frame as
valid. A recessive ACK delimiter bit follows this, followed by a
total of 7 recessive end-of-frame (EOF) bits to indicate that the
frame is complete and that another node may begin
transmitting.
While a normal frame will typically contain between 55
and 131 useful bits, it may occasionally contain more due to an
error-checking method known as “bit stuffing.” Six sequential
bits of the same type (dominant or recessive) will be read as an
error by any receiving nodes, so when this occurs naturally, an
opposing bit will be inserted between the fifth and sixth
repeated bit. This does not occur with the CRC delimiter, either
ACK bit, or the end-of-frame bits, all of which are of fixed size
and value.
None of this information is strictly necessary in order to
interpret the CAN bus using the methods described below, but
it can help to understand why the network behaves as it does.
1 See Appendix for computation
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 74
PROCEDURE The most reliable way to get started with the CAN system
is to dismantle part of a car and hook up to the CAN bus that
way. However, sometimes an expendable vehicle may not be
available. In such cases, there are commercial CAN simulators
available that are useful for training, troubleshooting, and to
some degree, debugging. [6]
The easiest way to access the CAN bus in a modern
vehicle is via the On-Board Diagnostics port (OBD-II) under
the driver’s side dashboard. This port is installed in most
modern vehicles as a way of directly communicating with the
microcontroller running the network, and includes both the high
and low (i.e. CAN-H, CAN-L) terminals needed to access the
CAN network.
However, on occasion this method may not work. During
the course of the research that led to this paper, it was
discovered that the OBD-II port of the project’s Kia Optima
was unresponsive. In such a situation, the next course of action
is to find a CAN-connected device whose wire harness can be
used to read the CAN bus. Qualifying devices will vary from
vehicle to vehicle, but the instrument cluster (speedometer,
odometer, etc.) will typically have a CAN line if it is not
mechanically actuated.
Once the CAN bus has been located, it will be necessary to
connect it to a computer. A PEAK-System OBD-II to D-sub
adapter was used with accompanying software for this research,
but any compatible combination of adapters and software
would function adequately. After a physical connection has
been established, the bit rate must be synchronized. A modern
computer will often have a clock speed of several GHz,
whereas a CAN microcontroller will typically have a
transmission rate in the kHz to MHz range.
Upon accessing the CAN bus and synchronizing the bitrate
with the computer, decoding the bus is as simple as
manipulating equipment in the vehicle and recording the
changes in the data values for whatever addresses are present
on the bus. This process of experimentation is necessary
because the DATA field is so densely packed with information
that it has to be reduced to machine code in order to fit inside a
standard CAN frame. The formatting and meaning of this code
will vary from node to node, so the only way to interpret it
without proprietary information or software from the
manufacturer is via trial and error.
RESULTS Connection to the car was made via the CAN ports on pins
31 and 32 of the instrument cluster harness at a bit rate of 100
kHz. The general state of the CAN bus immediately after
ignition can be seen below in Table 4.
Table 4. Vehicle After Ignition
ID DATA
100h 31 1C 00 00 08 20 00 00
101h 80 40 11 00 00 00 00 00
104h 05 6E 00 00 00 00 00 00
10Ch 00 00 00 12 00 00 00 00
10Dh 16 83 C8 20 06 00 00 00
400h 01 02 00 00 00 00 FF FF
401h 0F 02 00 00 00 00 FF FF
40Fh 00 02 00 00 00 00 FF FF
501h 00 03 48 00 00 00 00 00
The first observation that was made when the car had
finished its power-on procedures was that bytes 2 and 3 in word
501h fluctuated slightly when the vehicle was completely idle
and parked. It is hypothesized that these two bytes, and
possibly byte 1, are tachometer readings. Converting 0348 from
hexadecimal to decimal yields 840, which is believable for an
RPM reading from an engine recently started. Further research
will be made when a splitter has been fashioned that will allow
the computer and the instrument cluster to read from the cluster
harness at the same time. This will confirm whether the raw
number corresponds to the RPM.
When the brake pedal was pressed, byte 4 of word 10Dh
changed from 00 to 02, setting bit 3 of the word high. This is
likely the bit that controls the brake lights, or the sensor
indicating that the brakes are engaged. Bytes 1 and 2 of the
same word appear to keep track of the current state of the
power system and the motor, respectively.
Word 104h seems to deal mainly with safety features such
as locks and flashers. Word 1 changes states depending on the
state of the driver’s side back door (open or closed, locked or
unlocked), and word 2 performs the same function for both
front doors. No data was collected on word 3, but logically, it
might relate to the passenger side back door. Word 4 controls
the hazard lights, and likely the turn signals.
CONCLUSION Preliminary results on this project are encouraging. By
mapping some of the more easily-accessible CAN codes, less
important data generated by normal operation of devices in the
car can be filtered out, and more useful codes such as those for
cruise control or ABS can be more easily isolated. The 2012
Kia Optima is lacking in what would traditionally be
considered drive-by-wire systems, but it may yet become
unmanned via more unconventional means.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 75
APPENDIX A CRC check is performed on the data in every CAN
frame to ensure data integrity. This is accomplished by adding a
number of 0s to the end of the DATA field equal to the size of
the CRC field (15 bits) and performing an XOR operation on
the DATA field with a predetermined divisor (for CAN
applications, 0x4599), aligning with the leftmost 1 and working
to the right until the remainder is 0. Each progression aligns the
first 1 in the divisor with the first 1 in the dividend. The process
for the CRC field in Table 2 and Table 3 is shown below in
Table 5.
Table 5. CRC Calculations
0 1 1 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 1 0 1 1 0 0 1 1 0 0 1
0 0 1 1 0 1 1 1 1 0 0 1 1 0 0 1 0 0 0 0 0 0 0
0 0 1 1 0 1 1 1 1 0 0 1 1 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 1 0 1 1 0 0 1 1 0 0 1
0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 1 0 0 0 0 0 0
0 0 0 1 0 1 0 1 0 1 0 1 0 1 0 1 1 0 0 0 0 0 0
0 1 0 0 0 1 0 1 1 0 0 1 1 0 0 1
0 0 0 0 0 1 0 0 0 0 1 1 0 0 1 1 1 1 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 1 1 0 0 1 1 1 1 0 0 0 0 0
0 1 0 0 0 1 0 1 1 0 0 1 1 0 0 1
0 0 0 0 0 0 0 0 0 1 1 0 1 0 1 0 0 1 0 1 0 0 0
Since the remainder of the DATA field is 0, the process is
now complete, and the result becomes the CRC field seen in
Table 2 and Table 3.
REFERENCES [1] Embedded Systems Academy, "Almost 25 Years of
CAN," [Online]. Available: http://www.canopen.us/history.htm.
[Accessed 12 August 2015].
[2] Robert Bosch GmbH, "CAN with Flexible Data-Rate,"
in 13th International CAN Conference, Neustadt an der Haardt,
Rhineland-Palatinate, Germany, 2012.
[3] A. Marino and J. Schmalzel, "Controller Area
Network for In-Vehicle Law Enforcement Applications," in
IEEE Sensors Applications Symposium, San Diego, California,
USA, 2007.
[4] M. Laifenfeld and T. Philosof, "Wireless Controller
Area Network For In-Vehicle Communication," in IEEE 28-th
Convention of Electrical and Electronics Engineers in Israel,
Hertzelia, Israel, 2014.
[5] K. Kalaiyarasu and C. Carthikeyan, "Design of an
Automotive Safety System using Controller Area Network," in
International Conference on Robotics, Automation, Control and
Embedded Systems, Kelambakam, Chennai, India, 2015.
[6] H. Guo, J. J. Ang and Y. Wu, "Extracting Controller
Area Network Data for Reliable Car Communications," IEEE,
Singapore, 2009.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 76
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
DYNAMIC SYSTEM ANALYSIS USING RECURRENCE QUANTIFICATION ANALYSIS AND RECURRENCE PLOTS IN MATHEMATICA
Serkan Güldal
Interdisciplinary Engineering University of Alabama at Birmingham
Birmingham, AL, USA
Veronica L. Baugh Electrical and Computer Engineering University of Alabama at Birmingham
Birmingham, AL, USA
ABSTRACT Recurrence quantification analysis and recurrence plots
are techniques of non-linear data analysis used for various
mechanical engineering analyses. The recurrence plot reveals
all the times when the phase space trajectory of the dynamical
system visits roughly the same area in the phase space. The
Euclidean distance between each state is used as a measure of
in n-dimensional space. The Recurrence Plot measures that are
considered are Percent Recurrence, Percent Determinism, and
Deterministic Ratio of the state matrix. These derived
measures are used to better understand the states of the sys-
tem. There are relatively few software options available for
creating recurrence plots and recurrence quantification analy-
sis. We proposed and developed a new algorithm in
Mathematica for creating recurrence plots and recurrence
quantification measures (see Appendix).
INTRODUCTION Recurrence plots are used as a technique of non-linear da-
ta analysis. In 1987, Eckmann et al. introduced the method of
recurrence plots to visualize the recurrences of dynamical sys-
tems [1]. Later in order to quantify the lines in a recurrence
plot, a way of measuring quantitative information was intro-
duced [2] to define measures of complexity including percent
recurrence, percent determinism, and deterministic ratio. The-
se measures have been used in various engineering disciplines
including mechanical engineering. Recurrence quantification
analysis has been used to compare the machinability of steels
[3], to assess the damage of mechanical systems [4], detecting
flank wear in face milling [5], and detection of two-phase flow
patterns [6].
EUCLIDEAN DISTANCE Consider two n-dimensional states, points, p1 = (x1,1, x 1,2,
…, x 1,n) and p2 = (x 2,1, x 2,2, .,.., x 2,n). The Euclidean distance
between these two points corresponds to the length of the
straight line drawn from one point to the other and is defined
in Equation (1).
(1)
Every particular state can be representing in vector form
as a point in n-dimensional space. Equation (1) can be used to
calculate the distance between any state and the origin or the
distance between any pair of solutions.
For each given state of size n and m number of states,
each row represents a point in n-dimensional space. The corre-
sponding state matrix of size m x n is given by Equation (2).
(2)
We define the Euclidean distance, d, between points
and as
(3)
where and The matrix of dis-
tances is of the form as shown in Equation (4).
(4)
The difference of two n-dimensional states is defined as
(5)
where and . The matrix of differ-
ences between a point and the next point has a size of .
RECURRENCE QUANTIFICATION ANALYSIS We consider a state space matrix representing the time
evolution of a dynamical system. The recurrence plot reveals
all the times when the phase space trajectory of the dynamical
system visits roughly the same area in the phase space.
Suppose we have a trajectory {xk}k=1…m of a system in its
phase space where m is the number of states with each state
being represented as an n-dimensional vector. The develop-
ment of the system is then described by a series of these vec-
tors representing each state and thus the trajectory in an ab-
stract mathematical n dimensional space. Each component of
the recurrence matrix RP = (RPij) is one if xj is contained in a
“ball” Bε(xi) of radius ε >0 centered at state and zero other-
wise as shown in [7]. The method of selecting the threshold,
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 77
the radius of the n-dimensional sphere, is based on the meth-
odology explained in [7]. The threshold, ε, is set to r x
std(data), where r = 0.25 is a common usage. We set r = 3 due
to the nature of our data and since our data is already “embed-
ded” and we use the maximum column-wise standard devia-
tion of the state matrix.
Figure 1. of radius centered at state used to determine value of
The recurrence plot shows the spatial dependence of the
system in its state space and it is not affected by the non-
stationarities in the signal. The lines formed by points that are
parallel to the main diagonal line represent the determinism of
the system.
The recurrence plot measures derived from recurrence
plots that we will discuss are Percent Recurrence (PR), Per-
cent Determinism (PD), and Deterministic Ratio (DR). Per-
cent recurrence is defined as the number of points x(i) and x(j)
closer to one another than the threshold, ε, over the total num-
ber of points in the recurrence plot. Percent determinism is
defined as the number of diagonal lines of a minimum length
of two points divided by the total number of points in the re-
currence plot. The deterministic ratio is defined as the ratio of
the percent determinism to the percent recurrence. The calcu-
lation of these parameters is based on the methodology by
Webber [8].
TESTING USING N-QUEENS DATA The N-Queens problem has been studied for over a centu-
ry and originated from the 8-Queens problem which was first
posed in 1848 by a German chess player, Max Bazzel [9, 10,
11] . The problem was proposed in Illustrirte Zeitung in 1850
[10, 11]. The goal of the 8-Queens problem is to achieve an
arrangement of 8 queens such that there is only one queen in
each column, row, diagonal, and anti-diagonal1. The problem
was eventually extended to the N-Queens problem for placing
queens on the general × board size. The problem has
been studied extensively and has attracted the attention of
many mathematicians, including Gauss [9, 11], Pólya [12],
and Lucas [13]. During the last five decades, the problem has
been discussed in the context of computer science and used as
an example of backtracking algorithms, permutation genera-
tion, divide and conquer paradigm, program development
methodology, constraint satisfaction problems, integer pro-
gramming, specification, and neural networks [14].
The N-Queens problem is classified as non-deterministic
polynomial time hard (NP-hard) [15]. This is a class of prob-
lems that are “at least as hard as the hardest problems in NP”.
Currently, no closed-form formula exists for the number of
solutions of the N-Queens problem as a function of the grid
size n. The number of solutions is determined through various
algorithms that are process intensive for large .
The solutions to the N-Queens problem form a group. The
notations of isomorphic and fundamental solutions follow the
methodology of Erbas [16]. From a particular solution, one
can generate other solutions by simple rotations. Two solu-
tions are said to be equivalent or isomorphic if either can be
transformed into the other by rotations of 90, 180, or 270 de-
grees and/or reflections about the horizontal, vertical, diago-
nal, or anti-diagonal axes. These rotations partition the solu-
tions into a set of equivalence classes. The set of solutions
consisting of exactly one representative from each equivalence
class is called a set of fundamental solutions. The number of
fundamental solutions and the number of all solutions for the
N-Queens problem are known for grid sizes less than or equal
to 26 and are listed in are listed in Table 2. These solutions are
obtained by backtracking algorithm separately, and the solu-
tion matrices from 8-Queens to 12-Queens are used in the fol-
lowing analysis. For example, 8-Queens solution matrix is 8
by 92 matrix, so it produces 8-dimentional 92 states. Figure 1
demonstrates the polynomial growth of the number of solu-
tions as a function of board-size in logarithmic scale.
The N-Queens solution matrices were used to test the re-
currence quantification analysis. Table 1 shows the recurrence
plot parameters calculated for all states and for the fundamen-
tal states respectively for grid sizes 8 to 12. For n smaller than
8, there is not a sufficient number of states to achieve reliable
results. The fundamental solutions have larger parameter val-
ues (PR, PD, and DR) compared to all states. This shows that
determinism of the fundamental states is higher than that of all
the states. In both the fundamental and all of the states, as the
grid size increases, determinism decreases. Hence, we can
claim that the states for higher grid sizes behave less determin-
istically. In other words, determinism is inversely proportion-
al to the grid size.
Table 1. Recurrence Plot Parameters
i) States of Grid Size, n Grid Size
n
Percent
Recurrence
Percent
Determinism
Deterministic
Ratio
8 0.2580 0.0484 0.1877
9 0.1590 0.0269 0.1691
10 0.1058 0.0150 0.1414
11 0.0683 0.0081 0.1181
12 0.0400 0.0042 0.1047
ii) Fundamental Solutions of Grid Size, n
Grid Size n Percent
Recurrence Percent
Determinism Deterministic
Ratio
8 .5972 .1528 .2558
9 .3875 .0907 .2341
10 .2540 .0477 .1879
11 .1473 .0221 .1497
12 .1012 .0133 .1310
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 78
Table 2. Number of Solutions for the N-Queens Problem [17, 18].
n Number of
Fundamental
Number of
All Solutions
1 1 1
2 01 0
1
3 01 0
1
4 1 2
5 2 10
6 12 4
2
7 6 40
8 12 92
9 46 352
10 92 724
11 341 2,680
12 1,787 14,200
13 9,233 73,712
14 45,752 365,596
15 285,053 2,279,184
16 1,846,955 14,772,512
17 11,977,939 95,815,104
18 83,263,591 666,090,624
19 621,012,754 4,968,057,848
20 4,878,666,808 39,029,188,884
21 39,333,324,973 314,666,222,712
22 336,376,244,042 2,691,008,701,644
23 3,029,242,658,210 24,233,937,684,440
24 28,439,272,956,934 227,514,171,973,736
25 275,986,683,743,434 2,207,893,435,808,350
26 2,789,712,466,510,280 22,317,699,616,364,000
Figure 2 shows the relationship between the grid-size and
the Percent Recurrence.
1 There are no solutions for board-sizes of 2x2 or 3x3.
2 The 6-Queens problem has fewer solutions than the 5-
Queens problem.
Figure 2. Percent Recurrence by Grid-size, n
Figure 3 and Figure 4 show the relationship between the
grid-size and the Percent Determinism and the Deterministic
Ratio respectively.
Figure 3. Percent Determinism by Grid-size, n
Figure 4. Deterministic Ratio by Grid-size, n.
RECURRENCE PLOTS A recurrence plot is a two-dimensional representation
technique to visualize distance correlations in a time series.
Recurrence plots easily demonstrate whether a system is peri-
odic or chaotic.
The recurrence plots for the 8-Queens through 10-Queens
solutions are shown in Figure 5. The recurrence plots for the
8-Queens through 10-Queens fundamental solutions are
shown in Figure 6.
0
0.5
1
13 9 10 11 12
Percent Recurrence
PR (All) PR (Fundamental)
0
0.1
0.2
8 9 10 11 12
Percent Determinism
PD (All) PD (Fundamental)
0
0.1
0.2
0.3
8 9 10 11 12
Deterministic Ratio
DR (All) DR (Fundamental)
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 79
i. 8-Queens All Solutions
ii. 9-Queens All Solutions
iii. 10-Queens All Solutions
Figure 5. Recurrence plot of from 8, 9 and 10-Queens’ solutions
i. 8-Queens Fundamental Solutions
ii. 9-Queens Fundamental Solutions
iii. 10-Queens Fundamental Solutions
Figure 6. Recurrence plot of fundamental solutions of 8, 9, and 10-Queens
OBSERVATIONS
Analysis of the recurrence plot of the N-Queens solution
matrix reveals a characteristic which is known as drift and also
reveals a checkerboard texture. Drift can be seen by the paling
of the recurrence plot away from the central diagonal. The
checker-board texture which is found in the N-Queens recur-
rence plot is a distinguishing trait of a Lorenz system in which
the points move in a spiral around one of the two symmetric
fixed points of the system [8]. A plot of an arbitrary Lorenz
System trajectory and the associated recurrence plot with a
checkerboard texture is shown in Figure 7 [19].
Figure 7. Arbitrary Lorenz System Trajectory and Asso-ciated Recurrence Plot [20].
The results obtained from the recurrence plot derived
measures suggest that the system’s complexity increases as the
grid size increases. This implies that it becomes more difficult
to speculate the position of the next solution as the grid size
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 80
increases. These results confirm observations over many dec-
ades in trying to solve the N-Queens problem.
FUTURE WORK A new area of interest would include creation of an asso-
ciated Lorenz System Trajectory.
ACKNOWLEDGEMENT We are grateful to Dr. Tanik for his encouragement and
inspiration.
REFERENCES
[1] N. Marwan, M. C. Romano, M. Thiel and j. Kurths,
"Recurrence plots for the analysis of complex systems,"
Science Direct, pp. 237-306, 3 November 2006.
[2] J. P. Zbilut and C. L. Webber, "Embeddings and delays
as derived from quantification of recurrence plots," Ph, vol.
171, pp. 199-203, 1992.
[3] Ravish, K. S. Umashankar, A. Abhinav, K. V.
Gangadharan and D. Vijay, "Recurrence Quantification
Analysis to Compare the Machinability of Steels," ARPN
Journal of Engineering and Applied Sciences, vol. 6, no. 1, pp.
8-13, 2011.
[4] Y. Qian, R. Yan and M. Shan, "Damage Assessment of
Mechanical Systems Based on Recurrence Quantification
Analysis," in Prognostics & System Health Management
Conference, Beijing, 2012.
[5] S. D. Mhalsekar, S. S. Rao and K. V. Gangadharan,
"Investigation of feasibility of recurrence quantficiation
analysis for detecting flank wear in face milling,"
International Journal of Enginering, Science, and Technology,
vol. 2, no. 5, pp. 22-38, 2010.
[6] R. Mosdorf and G. Gorski, "Detection of Two-Phase
Flow Patterns in a Vertical Minichannel using the Recurrence
Quantficiation Analysis," AMA, vol. 9, no. 2, pp. 99-102,
2015.
[7] R. Seker, M. M. Tanik and D. Callahan, "Analysis of N-
Queens Solutions: A Dynamical System Approach," in
Integrated Design and Process Technology, 2002.
[8] C. L. Webber, Jr. and J. P. Zbilut, "Dynamical
assessment of physiological systems and states using
recurrence plot strategies," American Physiological Society,
pp. 965-973, 1994.
[9] C. F. Gauss and H. C. Schumacher, Briefwechsel
zwischen, pp. 105-122, 1865.
[10] J. Gingsburg, "Gauss's arithmetrization of the problem of
n queens," Scripta Math. 5, pp. 63-66, 1939.
[11] S. Gunther, "Zur mathematisches theorie des
Schachbretts," Archiv der Mathematik und Physik, vol. 56, pp.
281-292, 1874.
[12] G. Polya, "Uber die 'doppelt-periodischen' losungen des
n-damen-problems," Mathematische Unterhaltungen und
Spiele, pp. 364-374, 1918.
[13] E. Lucas, Recreations mathematiques, 1891.
[14] C. Erbas, S. Sarkeshik and M. M. Tanik, "Different
Perspectives of the N-Queens Problem," ACM, pp. 99-108,
1992.
[15] Z. Wang, D. Huang, J. Tan, T. Liu, K. Zhao and L. Li,
"A parallel algorithm for solving the n-queens problem based
on inspired computational model," BioSystems, pp. 22-29,
2013.
[16] C. Erbas, S. Sarkeshik and M. M. Tanik, "Different
Perspectives of the N-Queens Problem," ACM, pp. 99-108,
1992.
[17] N. J. A. Sloane and S. Plouffe, "Number of ways of
placing n nonattacking queens on nxn board (symmetric
solutions count only once)," Academic Press, 17 April 2015.
[Online]. Available: http://oeis.org/A002562. [Accessed 17
June 2015].
[18] N. J. A. Sloane and S. Plouffe, "Number of ways of
placing n nonattacking queens on nxn board.," 17 June 2015.
[Online]. Available: http://oeis.org/A000170. [Accessed 17
April 2015].
[19] J. P. Eckmann, S. O. Kamphorst and D. Ruelle,
"Recurrence Plots of Dynamical Systems," Europhysics
Letters, pp. 973-977, November 1987.
[20] "Recurrence Plots and Dynamical System Analysis,"
[Online]. Available:
http://www.cs.colorado.edu/~lizb/rps.html. [Accessed 17 June
2015].
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 81
APPENDIX Recurrence plot code in Mathematica language
SetDirectory[$UserDocumentsDirectory]; data=Import["8.dat","CSV"]; n =Length[data]; size = Length[data[[1,All]]]; (* Threshold = R x the max standard deviation of the columns of the solution matrix*) R=3; stdev = {}; maxstdev = 0; For [i=1, i<= size, i++, AppendTo[stdev,N[StandardDeviation[data[[All,i]]]]]; ]; maxstdev = Max[stdev]; epsilon = R*maxstdev; (* Euclidean Distance and Recurrence Plot*) eucdis=Table[0,{n},{n}]; rp=Table[0,{n},{n}]; For[i=1, i<= n, i++, For[j=1, j< i, j++, eucdis[[i,j]]=N[EuclideanDistance[Take[data,{i,i}],Take[data, {j,j}]]]; If [eucdis[[i,j]]<epsilon , rp[[i,j]]=1;rp[[j,i]]=1 ] ]; rp[[i,i]]=1 ]; MatrixPlot[rp] (*Recurrence Plot Measures*) countlines=0; countpoints = 0; For[j=0, j< n ,j++, linelength = 0; For[i=1, i+j< n,i++, If[rp[[i,i+j+1]]==1, linelength++, If[linelength >=2,countlines++, If[linelength == 1, countpoints++ ]; ]; linelength = 0(* reset for a diagonal*) ]; ]; If[linelength >=2,countlines++, If[linelength == 1, countpoints++ ]; ]; ]; pr = N[(Total[Total[rp]])/(n*n)]; pd = N[2*countlines/(n*n)]; dr = N[pd/pr]; Print[StringJoin["Percent Recurrence (PR): ", ToString[pr]]]; Print[StringJoin["Percent Determinism (PD): ", ToString[pd]]]; Print[StringJoin["Deterministic Ratio (DR): ", ToString[dr]]];
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 82
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
SIMULATED LVAD PUMP MIMIC DEVICE FOR ANALYZING SAFETY, RISK AND RELIABILITY OF DESIGNS
Jacob King Department of Mechanical Engineering
University of Louisiana at Lafayette Lafayette, Louisiana
Clint Bergeron Department of Mechanical Engineering
University of Louisiana at Lafayette Lafayette, Louisiana
Krishna Chaitanya Manthripragada
Department of Mechanical Engineering University of Louisiana at Lafayette
Lafayette, Louisiana
Charles Taylor, Ph.D.
Department of Mechanical Engineering University of Louisiana at Lafayette
Lafayette, Louisiana
ABSTRACT
Congestive heart failure (CHF) is one of the most common
forms of debilitating illness that threatens the health and safety
of countless victims worldwide. Due to the severity of this
disease, victims must seek alternatives in hopes of improving
their quality of life. One such alternative is the support from a
left ventricular assist device (LVAD) to directly assist with
meeting the blood flow demands of the body.
The medical device design challenge answered within this
research lies in the inaccessibility of LVAD designs currently
being used on the market. It is very difficult to determine the
effectiveness of other medical device designs that may impact
the performance and reliability of an LVAD when the
functionality of these devices are so proprietary. As a result, in
order to assess potentially correlated medical device
performance and safety, it would be beneficial to replicate the
functionality of an LVAD, both numerically and empirically.
Thus, in order to accurately replicate the functionality of an
LVAD, for the purposes of analyzing the possible impacts of
external factors on the performance and reliability of the
device, an LVAD simulator or “mimic” would be critical for in
vitro use within a mock circulatory loop (MCL).
INTRODUCTION
Congestive heart failure (CHF) is the medical condition
that occurs when the heart is no longer capable of pumping
enough blood to maintain the adequate amount of blood flow
needed to meet the demands of the body. According to the
2013 heart disease and stroke statistical report by the American
Heart Association, approximately 5.1 million people in the
United States suffer from varying degrees of CHF and nearly
half of the people who develop CHF will die within 5 years of
being diagnosed [1-2]. A procedure developed in an effort to
combat this common and deadly condition is the replacement of
the damaged heart by means of a total heart transplant. It is
estimated that approximately 70,000 patients may benefit from
a heart transplant; however, only 2,000 hearts become available
per year [3]. Due to the severity of this condition, patients must
seek other alternatives in hopes of improving their quality of
life. One such alternative is to receive support from a left
ventricular assist device (LVAD) to directly assist with the
blood flow demands of systemic circulation, and consequently
pulmonary circulation. An LVAD’s primary function is to
effectively pump blood directly to the aorta, from the impaired
left ventricle, essentially bypassing the aortic valve. This is
typically due to the aortic valve being weakened or damaged,
which directly contributes to CHF.
To comply with the Food and Drug Administration’s
(FDA) regulations and standards, in vitro testing is utilized in
order to accurately and safely analyze device reliability and
performance, prior to in vivo trials. In vitro testing is primarily
utilized prior to in vivo in order to limit the complications and
coupled risks associated with experimentation on a living
organism.
Although a circulatory loop model cannot replace in vivo
testing, an LVAD’s design may be efficiently refined
beforehand by determining its effect on circulatory
hemodynamics. This is achieved through the use of a mock
circulatory loop (MCL), which simulates the human circulatory
system by accurately representing cardiovascular conditions in
a bench-top hydraulic circuit. Due to the complexity of the
circulatory system and the dynamics associated with cardiac
function, the MCL is an essential in vitro tool for testing
ventricular assist devices and other cardiac assist technologies.
In 1998, an article was published in the ASAIO Journal
from a joint collaboration with the FDA, The American Society
for Artificial Internal Organs (ASAIO), the Society of Thoracic
Surgeons (STS), and the National Heart, Lung and Blood
Institute entitled, “Long-Term Mechanical Circulatory Support
System Reliability Recommendation” [4]. In this article,
suggested conditions that must be accounted for in a circulatory
model are outlined. Additional guidance documents used for
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 83
assessment reaffirm this article’s recommendation: ISO 5198,
ISO 14791, and ISO 14708. Given the state of technology
when these documents were written and the many
advancements since then, these conditions need to be re-
evaluated for increased safety. Coupled with the increasing
need for these medical devices, a more rigorous and robust in
vitro analysis should be necessary before an LVAD design
begins in vivo trials.
The challenge lies in the inaccessibility of the LVAD
designs currently on the market. It is difficult to analyze
LVAD functionality on a broad range of circulatory conditions
and determine the effectiveness of other medical device designs
that may intentionally or unintentionally interact with an
implanted LVAD, when the information on LVAD
functionality is very proprietary.
For the empirical V&V aspect of this work, a pump mimic
device that could be included in a hydraulic test apparatus
would be necessary. This pump mimic would have to replicate
the hydraulic performance dictated by the affinity curves of the
chosen LVAD design being simulated. A higher performance
pump (the mimicking device), one with an operating range
encompassing that of the LVAD, could be controlled to
replicate the pressure-flow relationships of a constant RPM
LVAD through manipulation of the mimicking pump’s own
operating speed. Essentially, once the LVAD mimic is
implemented into the hydraulic system, the MCL, with all its
supplemental elements, will function as if it is operating with
the assistance of an LVAD directly from the manufacturer.
The purpose of this report is to outline the process and
methods developed in order to accurately simulate LVAD
performance for use within a MCL, to describe the
development of the in silico nested loop hydraulic testing
system used to determine the effectiveness of the LVAD
simulator or “mimic”, and to outline the simulated data to
substantiate that this verification and validation (V&V)
methodology is an effective means of addressing a key medical
device design challenge.
MATERIALS AND METHODS
The published affinity curves, following ISO 5198
specifications, for various pump designs enable the hydraulic
performance to be replicated through the use of a parameterized
pump model. These pumping curves communicate the
hydraulic performance of the LVAD without divulging the
driveline characteristics, which implies that the LVAD must
maintain constant RPM in its function. The SimHydraulics™
toolbox, in the Simulink® Simscape™ product line, contains a
pump-modeling block element that could replicate the pumping
profiles of many different types of pumps, including LVADs.
DEVELOPMENT OF THE HYDRAULIC SIMULATION
MODEL USING MATHWORKS’ MATLAB® AND
SIMULINK®
The most dominant pumping methods in the ventricular
assist research space are axial and centrifugal; the scope of this
research was limited to those two types of rotary blood pumps.
Using Simulink® Simscape™, computational models were
developed for the HeartMate II and the Jarvik 2000 using a
parameterized pump block. These two LVAD designs were
chosen for their well-documented use and performance
capabilities; the affinity curves of both are publically accessible
[5-6]. An MP Pumps FRX-50 centrifugal pump was chosen as
the pump mimicking device for its operating range, driveline
response, and published affinity curves [7].
Figure 1 is a compilation of all published pump
performance affinity curve data documented by the
manufacturer at each published operating speed for the
HeartMate II, Jarvik 2000, and the MP Pumps FRX-50. This
pump performance data was digitized for use within the model
using the program DataThief. Noted is that each pump was
tested under varying conditions, such as different fluid
properties, described by the manufacturer. Thus careful
consideration had to be made when developing a parameterized
pump model that replicates the exact conditions under which
the data was collected.
The manufacturers of the Jarvik 2000, the Heartmate II,
and the FRX-50 published data on the parameters of the fluid
used in the performance trials of each respective pump. For the
Jarvik 2000, the manufacturer specified that the fluid used was
a glycerol/water mixture with a kinematic viscosity of 3.3
centistokes (cSt). For the Heartmate II, the manufacturer
specified that the fluid used was bovine blood with a dynamic
viscosity of 3.8 centipoise (cp); the dynamic viscosity and
density of the fluid was calculated based on an average specific
gravity of 1.05 for bovine blood [8]. For the FRX-50, the
manufacturer specified that the trials were performed using
water; it was assumed that the trials were completed at room
temperature. Table 1 displays the manufacturer’s data along
with the calculated data of the aforementioned fluids at room
temperature (22° C).
A control architecture was created using Simulink®
Simscape™ based on a nested loop control structure. Figure 2
shows the hydraulic loop created for testing the performance of
the FRX-50 and the two selected LVADs in silico.
Through simulation, it was found that the Simscape™
modeling tools were able to precisely replicate the affinity
curves of the chosen LVAD designs and the MP Pumps FRX-
50 centrifugal pump as shown in Figure 3. The control
architecture, based on a nested loop control structure, enabled
the FRX-50 to successfully reproduce the function described by
the LVAD’s affinity curves. This forces the assumption that
the FRX-50 pump mimic’s differential pressure output can be
controlled to replicate any pump’s affinity curve that is below
the characteristic affinity curve of the FRX-50.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 84
Figure 1. Affinity curves for (a) Heartmate II, (b) JARVIK 2000, (c) MP PUMPS FRX-50 at given operating speeds
Table 1. Pump fluid properties
Pumps (22° C) Kinematic
Viscosity
Density
Jarvik 2000 3.3 cst 1107.1 kg/m3
Heartmate II 3.627 cst 1047.66 kg/m3
FRX-50 0.9565 cst 997.8 kg/m3
Implementation of a discrete-time PID controller into the
nested loop control architecture was necessary for the FRX-50
to effectively track and replicate the performance of an LVAD.
Using the control elements in Simulink®, a PID control
structure was developed that enabled a computational model of
the FRX-50 to vary its operating RPM such that it could
reproduce an affinity curve of one of the LVADs, given an
operating speed. This controller setup is illustrated in Figure 4.
The PID controller that was utilized for this application is
an ideal, discrete-time, forward Euler, PID controller with a
sampling time of 0.01 seconds [9]. Table 2 displays the gains
for the PID controller. Table 3 displays the controller
performance based on the gains:
Table 2. PID controller gains
Table 3. PID controller performance
After tuning the PID controller to produce the desired
response believed to be effective at controlling the plant (Figure
2), the implementation of a Lookup Table from the Simulink®
library was necessary for matching a specific flow rate with the
corresponding differential pressure, given a specified LVAD
and operating speed.
An operating structure written in Matlab® was created in
order to construct and organize the sequence of events
necessary to simulate the FRX-50 pump mimicking the affinity
curve of a selected LVAD at a particular operating speed.
RESULTS The operating structure and the PID control architecture
was successful at simulating the FRX-50 pump mimicking both
LVADs at the eight different operating speeds published by
each manufacturer. As can be seen from Figure 5, the LVAD
mimic very closely tracks the actual LVAD’s affinity curve.
PID Controller Parameters Gain Value
Proportional (P) 0.8689
Integral (I) 200
Derivative (D) 0
PID Controller Specifications Performance Value
Rise Time 0.16 s
Settling Time 0.28 s
Overshoot 0 %
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 85
Figure 2 - SIMULINK® SIMSCAPE™ created nested loop model for simulating pump performance. The nested loop reflects a simple hydraulic circuit with three subassemblies: Reservoir, Resistance Valve and Hydraulic pump. The SIMSCAPE™ block library provided all components necessary to create a realistically functioning model, such as
pressure sensors, flow rate sensors, hydraulic pipes, reservoirs, valves, pumps, etc.
Figure 3. SIMULINK® SIMSCAPE™ simulated affinity curves for (a) Heartmate II AND (b) JARVIK 2000. Values published by the manufacturer. The valves are specified in Figure 1. Simulations performed with 60/40 glycerol / water
moisture at 22° C to represent blood flow at room temperature. The FRX-50’s performance curves (simulated from 200 to 900 rpm) completely encompass the curves of the specified LVADS, verifying that the operating speed capabilities of the
FRX-50 cover the necessary differential pressure and flow rate ranges needed to mimic LVAD’s affinity curve.
(a) (b)
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 86
Figure 4. PID control architecture featuring flow rate vs differential pressure lookup table block (to deliver a specific LVAD differential pressure, given a particular flow rate), an initial condition block, a discrete-time PID controller block, a ramp
input block connected to a needle valve (to increase flow rate over time), and two feedback loops for differential pressure and flow rate (to provide the control system with real-time sensor data). The control architecture was created using
Mathworks SIMULINK®.
Figure 5. Comparison of the simulated Heartmate II affinity curve and the MP PUMPS FRX-50 mimicking the affinity curve of the Heartmate II at 15,000 RPM. The FRX-50’s performance data reflects the PID controller adjusting to track the
target affinity curve based on a specified sampling time. This is a reflection of the PID control architecture logic illustrated in Figure 4.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 87
Figure 6. (a) Differential pressure vs time, (b) Flow rate vs time, (c) PID-Controlled pump speed vs time, and (d) Controller
signal error vs time for the FRX-50 mimicking the Heartmate II at 15,000 rpm
The HeartMate II operating at a speed of 15,000 RPM was
chosen in this example; however, every condition tested
behaved exactly in the same manner, implying that this
heuristic model does produce results that are concurrent with
the theory that a higher performance pump can be programmed
to mimic an LVAD’s affinity curve using MathWorks’
Matlab® and Simulink®.
Figure 6 reflects the performance data of the FRX-50
mimicking the HeartMate II at 15000 RPMs. Provided is the
differential pressure, flow rate, PID-controlled pump speed, and
the controller signal error versus time for the FRX-50
mimicking the HeartMate II at 15000 RPM.
In silico verification (of the FRX-50 performance) and
validation (of the control architecture) support this modeling
and control method as an effective means of reproducing pump
performance and mimicking functionality. It was found that
the Simscape™ modeling tools were able to precisely replicate
the affinity curves of the chosen LVAD pumps and the FRX-50
pump as shown in Figure 3. The control architecture, based on
a nested loop PID control structure, enabled the FRX-50 pump
(a) (b)
(c) (d)
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 88
to successfully reproduce the function described by the
LVAD’s affinity curves in silico, Figure 5.
DISCUSSION AND CONCLUSION
The final product of this study is a hydraulic pump that can
mimic the functionality of any lower performance pump with
known affinity curves, e.g. an LVAD, by producing the
requisite flow rate against a prescribed differential pressure,
given a fixed rotational speed. Utilizing established V&V
techniques from the aerospace and automotive industries, these
in silico system modeling tools were shown to be effective at
establishing baseline performance specifications for subsequent
in vivo testing, while assisting in confirming proof of concept
prior to spending time, money, and effort developing a
hydraulic testing system. Furthermore, this medically novel
V&V methodology successfully assisted in addressing a key
medical device design challenge, proving this to be a viable
method for use within the medical device design industry.
The limitations of this methodology are restricted to the
accuracy and precision of the information necessary for
Simulink® SimscapeTM
to perform realistically. The
parameters which define the system, such as pumping profiles,
tubing length and inner diameter, etc., were all meticulously
investigated in order to provide this in silico system analysis
with the most realistic and mechanically practical design for
future in vitro studies.
FUTURE WORK
The success met in this study provides a useful model of
how to apply the outlined V&V methodology on future medical
device research, while illuminating a path to further
development in this area. Further research of this topic may
result in advances in the V&V methodology for evaluating
medical device designs prior to in vitro and in vivo trials, due to
the quick and affordable ease of application of this in silico
method.
This V&V methodology could be implemented as an
effective means of analyzing the safety and performance of
many different types of hydraulic systems, such as those used
in the manufacturing industry. These studies would lead to
further knowledge regarding LVAD performance and reliability
under conditions that have never been properly tested in vitro,
but have caused serious injury and death in vivo. Additionally,
this research will assist in producing a testing methodology and
MCL design that will advance the V&V process for LVAD
research. The overall intent of this work is to decrease the
number of device recalls in future LVAD designs through more
thorough in silico tools and in vitro methods.
REFERENCES
[1] A. S. Go, et al., and on behalf of the American Heart
Association Statistics Committee and Stroke Statistics
Subcommittee, “Heart Disease and Stroke Statistics--2013
Update: A Report From the American Heart Association,”
Circulation, vol. 127, no. 1, pp. e6–e245, Jan. 2013.
[2] K. D. Kochanek, J. Xu, S. L. Murphy, A. M. Miniño, and
H.-C. Kung, “National vital statistics reports,” National Vital
Statistics Reports, vol. 59, no. 4, p. 1, 2011.
[3] “New Heart, New Hope.” [Online]. Available:
http://www.newsweek.com/new-heart-new-hope-153439.
[Accessed: 16-Feb-2015].
[4] F. Altieri, A. Berson, H. Borovetz, K. Butler, G. Byrd, A.
A. Ciarkowski, R. Dunn, B. Griffith, D. W. Hoeppner, and J. S.
Jassawalla, “Long-Term Mechanical Circulatory Support
System Reliability Recommendation American Society for
Artificial Internal Organs and Society of Thoracic Surgeons:
Long-Term Mechanical Circulatory Support System Reliability
Recommendation,” ASAIO journal, vol. 44, no. 1, p. 108, 1998.
[5] M. D. Macris, P. Michael, S. M. Parnis, M. D. Frazier, J.
M. Fuqua Jr, M. D. Jarvik, and K. Robert, “Development of an
implantable ventricular assist system,” The Annals of thoracic
surgery, vol. 63, no. 2, pp. 367–370, 1997.
[6] Thoratec® Corporation, “HeartMate® II Left Ventricular
Assist System (LVAS),” Food & Drug Administration,
Circulatory System Devices Panel, FDA EXECUTIVE
SUMMARY MEMORANDUM, Nov. 2007.
[7] “FRX 50 1/2‘ x 1/2’ End Suction Centrifugal Pump.”
[Online]. Available: http://mppumps.com/Product/FRX-50-
End-Suction-Centrifugal-Pump. [Accessed: 16-Oct-2013].
[8] R. K. Chaplin, D. E. Waldern, and O. L. Frost, “Specific
gravity of bovine blood as affected by breed and age,”
American Journal of Veterinary Research, vol. 31, pp. 1887 –
1888, 1970.
[9] R. C. Dorf and R. H. Bishop, Modern control systems, 12th
ed. Prentice Hall: Pearson, 2010.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 89
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
A LASER SENSOR FOR TIRE REAL-TIME PARAMETER ESTIMATION
Mostafa Salama (PhD Candidate) Mechanical Engineering Department University of Alabama at Birmingham
Birmingham, Alabama, USA
Marc A. Parker Mechanical Engineering Department University of Alabama at Birmingham
Birmingham, Alabama, USA
Vladimir V. Vantsevich
Mechanical Engineering Department University of Alabama at Birmingham
Birmingham, Alabama, USA
ABSTRACT As more demands are made on the automotive industry for
energy efficiency, more physical conditions must be
continuously monitored on vehicles in order to find areas to
make efficiency gains. One such dynamic condition that is not
yet monitored is the changing of wheel rolling radius while in
driving mode. This project demonstrates the feasibility of
continuously monitoring dynamic wheel radius in a single tire
by experimentally measuring the tire’s deflection from inside a
tire. A laser displacement sensor mounted to the rim of the
wheel measured the rubber’s deflection. A housing structure is
also designed to mount the sensor, amplifier, myRIO data
acquisition device, and power source.
The goal of this project is to monitor tire deflection from
inside the tire, transmit the information to the software, and
allow the user to monitor the tire deflection. Operators will be
able to monitor their tires continuously in real-time and make
informed decisions to improve their vehicle performance.
Measuring axial tire deflection is chosen specifically because it
has a direct measurement of the tire’s actual shape, while tire
pressure only is a rough indicator of shape, which is ultimately
the property that influences tire performance and reliability.
There are a number of challenges inherent in this project.
Since the sensor unit is inside the tire, it will be unserviceable.
The proposed designed device will be durable enough to go
without maintenance for extended periods of time, and its
battery able to be recharged through the wheel’s valve stem. It
must also be wirelessly connected to the controller due to the
rotating nature of the wheel.
Benefits of device are, but are not limited to, allowing for
easy measuring of tire deflections, allowing for monitoring of
deflections while on the road (on-line measurement), and
facilitating the detection of the length of the contact patch of
the tire.
INTRODUCTION The kinematic and force factors of a pneumatic wheel and
their functional mathematical description are determined by the
tire’s ability to deform in the normal, tangential and lateral
directions. To understand the manner in which the driveline
system affects the vehicle performance, it is important to
compile functional relationships that relate the kinematic and
force factors of the wheel. This is, in fact, important, because
the force factors of the wheel depend on the torque that is fed to
the wheel by the driveline system.
The kinematic, together with the force factors, determine
the power that is supplied to the wheel and its part that is lost in
the wheel. It should be emphasized that these factors are highly
affected by the properties of the surface of motion: the type and
state of the solid cover; presence of a wet layer (water and
snow) on a solid road and the physical and mechanical
properties of the terrain under off-road conditions.
Benefits of properly inflated tires
Improved vehicle energy efficiency
Longer tire life
Superior traction
Less compaction on soil (when moving on deformable
surfaces)
Features
Device will increase the life of the tire
Energy Efficient
System displays the percentage of tire deflection
This project is to continuously monitor the deflection of a
tire wall in an operating vehicle and to continuously
comprehend the surface underneath the vehicle through the
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 90
acquired data. The scope of this paper only involves a proof of
concept and some design work for the overall project. The first
goal is to determine, acquire, assemble, and test the necessary
hardware to accomplish this goal. Second is to create the
necessary software tool using LabVIEW to operate the data
acquisition hardware and record the results in a format that can
be analyzed. Third is to design a housing structure that can
protect and hold the electronic hardware in place inside of a
tire. The final goal of this project is to temporarily mount the
sensor inside of a test tire, deflect the tire, and analyze the
numerical results to identify the deflection within the readings
in order to verify the entire system up to this point.
Figures 1 and 2 demonstrate a layout of a proposed
intelligent tire system and components. A laser displacement
sensor is mounted inside the tire to continuously monitor the
tire deflection and connected to a data acquisition system which
has the capability to be connected to a computer using a WIFI.
Figure 1. Layout of intelligent tire system
Figure 2. Intelligent tire system components
Once the MyRIO detects laser sensor readings it sends
these data wirelessly to a computer host in the vehicle to
monitor and save tire deflection data.
LITERATURE REVIEW Armstrong et al. used low-cost piezoelectric film sensors
between the inner tube and the tire to characterize the surface of
motion. They suggested that the sensors were capable of
detecting normal pressure, deflection, and/or longitudinal
strain. However, their results were less consistent for larger
tires [1]. Erdogan et al. introduced a simple approach for the
analysis of tire deformation and proposed a new piezoelectric
tire sensor to physically measure tire deformation to estimate
slip angle [2]. Tuononen used a tire laser sensor based on a
laser triangulation technology, which can measure the carcass
deflections of a rolling tire. The results were shown for several
wheel loads and inflation pressures, which have linear influence
on both tire radius mean value and radius amplitude [3]. Xiong
et al. presented a laser-based sensor system to measure tire-
tread block deformation. Validation experiments were
conducted on a chassis dynamometer, and an asymmetric tire
tread deformation along the contact patch was observed. They
proposed that asymmetric tread deformation is due to rolling
resistance [4]. Erdogan et al. proposed a tire-road friction
coefficient estimation approach, which makes use of the
uncoupled lateral deflection profile of the tire carcass measured
from inside the tire through the entire contact patch [5].
EFFECTIVE ROLLING RADIUS AND SLIP RATIO The rolling radius of a wheel with a pneumatic tire is one
of the most important concepts of wheel kinematics. The rolling
radius that links the actual velocity of the center of the wheel
to the wheel angular velocity is given by
(1)
Since the rolling radius of a pneumatic wheel is not
equal to its free radius , a pneumatic wheel and a rigid wheel
of the same radius with no slip or skid will make different
numbers of revolutions for the same traveled distance [6].
The rolling radius of a pneumatic wheel is not constant
because the wheel will make different numbers of revolutions
on the same travel distance when loaded by different
combinations of moments and longitudinal forces at the wheel
axis and at the tire/terrain contact patch [6].
Any change in the wheel loads changes the rolling radius
and the angular velocity of the wheel. This leads to a definition
of the slip ratio,
(2)
where
is the theoretical linear velocity of the wheel
is the actual linear velocity of the wheel
When a wheel moves in the free mode and the slip ratio is
zero, the actual and theoretical velocities are equal and the
rolling radius is . The rolling radius in free mode can be
determined analytically [6]. However, it is more convenient to
use the rolling radius in the driven mode instead for the
following reasons:
The difference between radii and
can be neglected
for hard-surface roads, and the radius can be used as the
reference for zero slip ratio [6].
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 91
DESCRIPTION OF THE SYSTEM COMPONENTS 1. Wheel and Tire
The wheel chosen for study is a passenger car with
Michelin MXV4 tires. The tire size is 195/60R15 with load
index 88 and speed symbol H. This means that the maximum
load on the tire is 560 kg (1235 lbs.), maximum inflation
pressure is 300 kPa (44 psi) and maximum speed is 210 km/h
(130 mph). See Figures 3 and 4.
Figure 3. Wheel and tire
Figure 4. Tire dimensions
The most important factor in tire dimensions is the aspect
ratio where higher aspect ratio means more space to mount all
system equipment (data acquisition system, sensor, wires,
batteries, etc.) on the wheel.
2. Laser Displacement Sensor (LDS)
A laser tire sensor can measure the carcass deflection of a
rolling tire. The sensor’s output is red using one of the
controller’s analog pins. This sensor gives a voltage output
related to the distance of the tire from the sensor’s location.
The sensor used to measure deflections is a Keyence IL-
065 analog laser sensor with an IL-1000 amplifier. The sensor
head produces a single laser and also intercepts its reflection off
of whatever surface is being measured. Before purchase it was
evaluated for compatibility with the equipment already in hand.
It was evaluated for output signal and power requirements to
make sure that a myRIO could intercept the sensor’s analog
output and run on the same power source. See Figure 5.
Figure 5. Keyence IL-065 Laser sensor [7]
The IL-1000 sensor amplifier provides an analog output
signal between -5 V and 5 V, but it can be limited to a range of
0 V to 5 V. The myRIO system can process 0 V to 5 V analog
signals. The sensor also requires a power source between 10
VDC to 30 VDC and up to 18 W. The myRIO device requires
between 6 VDC and 16VDC and up to 14 W. Therefore, the
Keyence sensor and amplifier are compatible with the NI
myRIO, and any shared power source must be within 10 V and
16 V. See Figure 6.
Figure 6. Laser sensor amplifier IL-1000 connections [7]
Table 1 shows laser sensor and amplifier specifications.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 92
Table 1: Laser sensor and amplifier specifications [7]
Model IL-065
Mounting distance 65 mm
Measurement range 55 to 105 mm
Mass 75 g
repeatability 2 m
Linearity 0.1%
Light source output 560 W
Model IL-1000
Amplifier type DIN rail mount
Analog voltage output 0 to 5 V
Analog current output 4 to 20 mA
Power voltage 10 to 30 VDC
Power consumption (no
load) 2300 mW
Mass 150 g
3. Data Acquisition System
A signal transmitter will take the information sent by the
sensor and send it to an external receiver connected to a
MyRIO and input to a PC station equipped with LabVIEW
software. Up to this moment, laser sensor, sensor amplifier, and
MyRIO are installed inside the tire. The MyRIO is already
equipped with built-in wireless characteristics with ISM 2.4
GHz frequency band and 20 MHz channel width while the
outdoor range is up to 150 m. The laser sensor and amplifier are
connected directly to the MyRIO mounted inside the tire.
The National Instruments (NI) myRIO was selected to be
the data acquisition system to be used in this project. It is a new
product designed with students in mind that makes creating
systems embedded with sensors and actuators more accessible.
It can also be controlled using NI’s LabVIEW, a graphical
programming software system licensed to UAB’s School of
Engineering. LabVIEW uses a block-diagram design that
should be familiar to mechanical engineering students who
have studied system modeling and controls.
4. Power Source
A self-contained power source such as a battery pack will
eventually be needed to carry out a tire deflection experiment in
a moving vehicle. Because lithium-ion and lithium-polymer
batteries have a relatively high ratio of energy storage over
mass, those two types of battery packs were considered. The
minimum numbers of hours were calculated based on each
component consuming the maximum amount of power
possible, but that would not be the actual case since the
majority of myRIO functions would not be utilized.
Instead of using a battery pack at this stage in the tire
deflection monitoring project, a DC power supply provides
energy to the sensor and amplifier, while the myRIO is plugged
into an AC wall socket. The battery pack is not necessary until
the housing structure is built and the entire system can be
integrated in the tire. Figure 7 shows all of the electronic
hardware utilized in this project except for a personal computer
used to run LabVIEW, perform research, and analyze results.
5. User Interface (LabVIEW)
LabVIEW software installed on the host computer will
display the output deflection of the tire signal for several wheel
loads and inflation pressure and terrain conditions. LabVIEW
has the capability to display, save, and signal condition the
sensor readings using the host computer mounted in the
vehicle.
Figure 7. Electronic equipment used in tire deflection monitoring proof of concept including myRIO (1), DC
power supply (2), laser sensor (3), and sensor amplifier (4).
RING STRUCTURE DESIGN Before designing the ring structure, which will hold all
system components inside the tire, measurement of the wheel
rim dimension is essential. Three dimensional scanner is used
to scan the rim and create a 3D model. See Figures 8 and 9.
Figure 8. 3D scanner in Enabling Technology Lab
The rim was scanned from three different perspectives. The
results of this process are multiple three-dimensional point
clouds representing the measured displacements across the
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 93
object. The three perspective views were stitched together from
the same points on the object for different images from
different perspectives by the scanner’s software. These point
cloud representations were then converted into a mesh as
shown in Figure 9. The mesh file was then exported to Creo
CAD software to get wheel rim dimensions.
Figure 9. Mesh generation of wheel rim
Finally, the dimensions of the middle section of the tire
rim’s edge were obtained and a solid model was developed in
Creo Parametric 2.0 of a structure that can hold the laser sensor,
amplifier, power source, and myRIO. A wire tunnel between the
separate chambers was used to connect them. The modeled
object is in Figure 10 with a wire frame version in Figure 11.
The ring structure was designed to support the sensor,
amplifier, myRIO, and battery pack in a stationary position with
the laser oriented downward through a radial line from the axis
of rotation of the wheel. The device is intended to keep the
laser stationary when the wheel is turning at slow speeds.
It can be seen that there is a “cap” that spans the upper
component. This will be held on by the straps which will fit
into holes on the upper and lower sections. This cap allows for
components to be placed within the envelope. Notice that there
is a hole near the top of Figure 10. This is for the laser sensor to
take readings of the inner wall of the tire. Another point of
interest is that the bottom section houses none of the
equipment. It exists only to facilitate strapping the structure to
the rim and to balance the tire.
Figure 10. Ring 3D CAD design
Figure 11. Ring 3D dimensions
In the wire frame version the places where the straps hook
in are apparent. Also, the chambers which hold the different
electronic system components are visible. Starting with the
parallelogram at the far left is the chamber for the myRIO. Next
is the sensor head, the amplifier, and finally the power source.
Note that the myRIO will sit on its thin edge. This keeps the
entire structure shorter and helps to maintain its structural
integrity. Figure 11 is the orthographic mechanical drawings
created from this model.
COST ANALYSIS The desired objectives of the personnel implementing the
research and progression of the Tire Deflection Monitor project
are impeded by the remaining components required for the
correctly validated completion of the application. The grant that
is generously provided will assist the VREL and its aspiring
undergraduate researchers to obtain the necessary equipment to
further advance the research, application, and testing of the
proposed project. The research will allow the direct observation
and demonstration of the effects of proper tire deflection on the
tire slippage and vehicle energy efficiency.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 94
Table 2: Cost Analysis
Part Quantity Price
MyRIO Controller 1 $499
Laser Displacement Sensor 1 $1750
Rechargeable NiMH Battery Back 1 $35
Charger Adapter 1 $40
Materials (Aluminum 18”x18”) 1 $115
Machining and Manufacturing 1 $300
17” 5 bolt pattern rim 1 $150
17” high profile tire 1 $100
total $2989
HARDWARE TESTING After choosing and acquiring components, it is always
necessary to test their functionality as well as any other
qualities related to fulfilling objectives such as durability. In
this case functionality and range of the sensor and amplifier
was needed. Exploring the functionality of the amplifier
showed that displacement is measured in millimeters and that
any sensible point can be set to equal zero so that the user can
more easily read changes instead of absolute distance. In
addition, it was discovered that when the amplifier creates an
analog output signal between 0 V and 5 V, the signal voltage is
scaled to one-tenth of the displacement when measured in
millimeters. Figure 12 shows a displacement reading on the
sensor amplifier. The voltage of the output signal is equal to
one tenth of that number.
Figure 12. Sensor amplifier (center) displaying displacement measured by sensor head (left).
SENSOR RANGE The setup in Figure 12 is actually the arrangement used for
determining the range in which the laser can measure distance.
The Keyence IL-065 can measure distances between 55 mm
and 105 mm away from the face of the sensor [7]; as shown in
Figure 12. In order to test this range a simple experiment was
conducted. The sensor head was taped to a piece of paper and a
line drawn at the edge of its face to mark its position. Next a
straight ruler was placed perpendicular to the sensor head with
0 at the edge. The ruler was taped down for stability. A box was
placed directly in front of the sensor head and slowly moved
away so that it was moving up the ruler. Once a reading
appeared on the amplifier, a line was drawn in front of the box
so to mark the beginning of the sensor’s sensible range. The
box continued to be slid down the ruler until readings stopped.
Another line was drawn to mark the end of the sensible range.
Then the lengths along the ruler were measured where the lines
were drawn to see how far and how close to the sensor head it
can read the distance. Figure 14 shows how the sensor and ruler
were arranged as well as the final lines had drawn demarcating
the sensible range.
Figure 13. IL-065 laser sensor operating range [7]
Figure 14. Sensor head and ruler setup for range test. Range demarcation is also shown
DATA ACQUISITION SYSTEM In order for voltage values that represent displacement
detected by the sensor to be received by the myRIO and for it to
be displayed and recorded, a software application must be
created. National Instruments’ LabVIEW was developed to
facilitate control and data acquisition projects such as this.
After connecting the myRIO device to a computer in the
laboratory, LabVIEW was made to detect it.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 95
LabVIEW uses a block diagram structure to build control
and instrumentation software. The block diagram built for this
particular project is shown in Figure 15. LabVIEW
automatically creates a front panel that allows user
manipulation of the software components that correlate with
actions signified in the block diagram. This front panel can and
was further customized for clarity and aesthetics. The final
front panel is shown in Figure 16.
Figure 15. Block diagram of LabVIEW data acquisition, interpretation, display, and recording application
The block diagram is entirely closed in a loop so that the
program will run repeatedly. It does not run continuously,
however. There is a constant value connected to a watch
symbol in the top, left corner. The program will delay by the
number of milliseconds equal to the constant before executing
the loop again. This way, the user can control the rate at which
data is sampled. By not providing a stop condition as part of the
diagram itself, it will run until the user ends the program.
The block diagram begins with an analog input from the
sensor as a numerical value. This value will equal the voltage
output of the sensor. Next, a value specified by the user is
subtracted from the analog input to represent an offset, in
millimeters. This is where the user can define the distance
between the laser sensor head and the original position of the
tire rubber if the user wants an output that represents change in
displacement of the point on the tire rather than an absolute
value. Next, the information is scaled by multiplying the
numerical datum by 10. This is done because the numerical
value for the voltage output of the sensor is one-tenth of the
numerical value of the measured displacement.
Finally, the measured displacement value is displayed in
four different forms, which is how the system output is
communicated to the user on the front panel. One display
shows the current value in a box, and it is also shown in the
form of a circular gauge. The last ten values are plotted over
time, as well. Finally, every value is stored in a vertical table
which can then be exported to a spreadsheet or delimited file
and analyzed.
Figure 16. Front panel of LabVIEW data acquisition, interpretation, display, and recording application
UNCERTAINTY ANALYSIS The uncertainty of deflection measurements can fall into
four categories. The first source of uncertainty is due to sensor
calibration. This is provided by the vendor, Keyence, in the IL
Series User Manual and is ±2 µm. There is also uncertainty due
to measurement resolution, taken to be ± the last digit of the
readout. If using the amplifier this would mean 0.001 mm, but
the recorded measurements actually come from the LabVIEW
program. This results in an uncertainty of 0.01 mm.
An important source of uncertainty is due to the angle in
placement. Ideally there would be an exact right angle between
the laser and a line tangent to the curvature of the tire wall at
the point where the laser comes in contact with it. However,
that was seen to not be the case when analyzed before setting
the tire back on the rim. Therefore, this uncertainty must be
quantified. If real deflection is ΔS, the measured deflection is
ΔS*, and the angle between the actual placement of the laser
and the ideal placement is , then
(3)
It is assumed that the difference between the actual angle and
90º is less than 10º. If 10º is assumed then this means an
uncertainty of ±0.015 of the sensor’s measurements, or 1.5 %.
Table 3 is repeated again, taking all three of these uncertainties
into account.
The fourth source of uncertainty only applies in the cases
where the initial reading of the sensor is set to zero, and it is
virtually unquantifiable human error. This is because the tire
was held in place by a person, not a rigid structure. The
movement it took to set the initial value to zero could have
caused a slightly different deflection from the actual rest
position before loads were applied. This is why data sets begin
with a number other than zero for an unloaded state.
EXPERIMENTAL RESULTS Experimental results presented in this section provide
preliminary results for the developed data acquisition system
mounted inside the wheel with no inflation pressure. System
response is tested for several types of loads (steady and high
impact) and different sampling rate is examined in order to test
the developed system. For instance, a high velocity impact does
not allow for an adequate function to be derived that describes
deflection over the time domain when the sampling rate is only
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 96
0.1 sec. Radii were calculated using the undeflected tire radius
of 302.5 mm. The pressure inside of the tire was atmospheric
pressure in all cases. This means that it was not inflated. The
tire was not inflated because measurable deflections would
have required loads too large to safely apply given the available
equipment and resources. As a result, the tests indicate basic
properties of the tire material, but this is not as practical as if it
were inflated to normal operating pressure.
Figure 17 demonstrates wheel radius changes when
deflecting the tire with slow steady load while the sampling
time is 0.1 sec.
Figure 17. Wheel radius at slow steady load with sampling time of 0.1 sec
Figures 18, 19 and 20 show tire radius values in real-time
when high impact load is applied to the wheel using different
sampling time. Three consecutive high impact loads are applied
to the wheel with about 1 sec in between in Figure 18, with
sampling time of 0.01 sec. It is apparent that 0.01 sampling
time is not sufficient to capture all high impact loads
representing bumps in the road, as shown in the first and third
impact loads.
Figure 18. Wheel radius at high impact load to simulate bump with sampling time of 0.01 sec
Using sampling time of 0.1 sec to capture tire high impact
load is not suitable, as shown in Figure 19, when a tire is
subjected to six consecutive high impact loads with 1 second
interval time. It can be seen that only one impact load is
captured by the data acquisition system.
Figure 19. Wheel radius at high impact load with 0.1 sec sampling time
Figure 20 displays tire radius with two consecutive high
impact loads with less than 0.5 sec in between. The sampling
time is set to be 0.001 sec. It was found that 0.001 sec sampling
time is the best sampling time to capture tire deflection for high
impact loads.
Figure 20. Wheel radius at high impact load with 0.001 sec sampling time
From the results it is evident that quick, forceful impacts
must be monitored with more considerations than smoothly
applied forces. The sampling rate should be higher, ideally at
least 1000 samples per second. In the second final case with a
high velocity impact and sampling rate of 0.1 per second,
almost no impact was detected. In a real situation when a tire is
rolling, the only type of impact that can be read will likely be of
high velocity because the sensor will be rotating along with the
tire.
CONCLUSION In conclusion, a continuous tire monitoring system was
achieved as realistic and practical, but still much work needs to
be done. A beneficial next step will be to manufacture the ring
structure for component housing inside a tire. This will allow
for more robust testing, because the monitoring system will be
secured and protected. That will allow for the tire to be
professionally worked on, which will then allow for its internal
296
297
298
299
300
301
302
303
0 2 4 6 8 10
Tire
Rad
ius
(mm
)
Time (s)
275
280
285
290
295
300
305
0 2 4 6 8 10
Tire
Rad
ius
(mm
)
Time (s)
297
298
299
300
301
302
303
0 2 4 6 8 10
Tire
Rad
ius
(mm
)
Time (s)
280
285
290
295
300
305
0 0.5 1 1.5 2
Tire
Rad
ius
(mm
)
Time (s)
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 97
pressure to be increased. It will also ensure less error in
readings.
In addition, testing equipment should be acquired that can
apply well-controlled forces in order to establish a functional
relationship between deflection and load. It may also be worth
acquiring access to more recent three-dimensional printing
software or other expertise to create a more accurate model of
the tire rim’s dimensions.
REFERENCES [1] Armstrong, E. G., Sandu, C., and Taheri, S., "A Study on
Using Piezoelectric Sensors in a Wheeled Robot Tire For
Surface Characterization," Proc. 18th International Conference
of the ISTVS.
[2] Erdogan, G., Alexander, L., and Rajamani, R., 2010, "A
novel wireless piezoelectric tire sensor for the estimation of slip
angle," Measurement of Science and Technology, 21.
[3] Tuononen, A. J., 2011, "Laser triangulation to measure the
carcass deflections of a rolling tire," Measurement of Science
and Technology, 22.
[4] Xiong, Y., and Tuononen, A., 2014, "A laser-based sensor
system for tire tread deformation measurement," Measurement
Science and Technology, 25.
[5] Erdogan, G., Alexander, L., and Rajamani, R., 2011,
"Estimation of Tire-Road Friction Coefficient Using a Novel
Wireless Piezoelectric Tire Sensor," IEEE Sensors Journal,
11(2), pp. 267-279.
[6] Andreev, A. F., Kabanau, V. I., and Vantsevich, V. V., 2010,
Driveline Systems of Ground Vehicles Theory and Design,
Taylor and Francis Group/CRC Press.
[7] 2015, "Keyence Laser Sensor,"
http://www.keyence.com/products/measure/laser-
1d/il/index.jsp.
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 98
Proceedings of the Fifteenth Annual Early Career Technical Conference The University of Alabama, Birmingham ECTC 2015
November 7, 2015 - Birmingham, Alabama USA
KINEMATIC ANALYSIS AND SIMULATION OF THEO JANSEN MECHANISM
Mehrdad Mohsenizadeh, Jenny Zhou
Department of Mechanical Engineering Lamar University
Beaumont, Texas, USA
ABSTRACT
Legged robots have come to researchers’ attention due to
their abilities to surmount or deal with complex obstacles, high
levels of mobility (changing direction while moving), active
suspension using force-controllable actuators in confronting
irregular terrains, maneuverability and terrain adaptability.
Within a few decades, Theo Jansen’s wind-powered beach
animals found their place among the other types of legged
machines. These autonomous creatures are able to live by their
own, storing wind energy to move their PVC-made legs,
detecting wet sands by the help of implanted nose feelers and
escaping from danger by anchoring themselves to the ground.
The advantage of these animals is that the required power of the
engine must be sufficient to transfer only the legs, and not the
transported elements. In this paper we employed the loop
closure equations method using MATLAB for kinematic
analysis of a Theo Jansen mechanism. Moreover, we performed
simulations in SolidWorks to validate the method.
INTRODUCTION Lately, there has been much work done on the synthesis,
analysis and optimization of Theo Jansen mechanisms using
programming software including Mathematica, MaTX and
Borland’s Delphi. Ghassaei [1] uses Mathematica for the design
of the Theo Jansen mechanism using the Chebyshev replication
and an approximation approach, where the latter simulates the
links’ movement as sine and cosine functions in order to
produce desired loci, and performs an optimization to extract
the optimal links length. However, no systematic approach for
the kinematic analysis is provided in [1]. Nansi. et al. [2]
implements a comprehensive dynamic analysis based on the
projection method using the MaTX, which is a continuation of
the proposed work by Blajer [3]. Ingram [4] examines a
genetic algorithm to optimize foot trajectory. Ingram [5] also
employs an elementary numerical algorithm for kinematic as
well as kinetic analysis using Borland’s Delphi software.
In this work, we present a complete, straightforward, yet
precise kinematic analysis of the Theo Jansen mechanism based
on the loop closure equations method, using MATLAB to
provide the preliminary kinematic data as a basis for future
examinations of this novel walking mechanism such as
dynamic analysis and optimization. To validate the analytical
data obtained in MATLAB, we performed simulations of the
paired legs using the SolidWorks Motion Study package.
DESCRIPTION OF THEO JANSEN MECHANISM Figure 1 shows the Theo Jansen mechanism, which is a
single degree of freedom mechanism composed of eight or
more legs. Applications of this marvelous legged mechanism go
beyond human-powered machines such as multi terrain
personal transport [6], multi terrain wheel chair, beach vendor
carts, robotic house pets, and steampunk walking ship [7].
Figure 2 depicts a schematic of one leg of the Theo Jansen
mechanism which has a crank (m), two oscillating rockers
(b,c), and two couplers (j,k), all connected by pivot joints. This
linkage is very similar to a Klann linkage [8]. Each leg consists
of six parts (see Fig. 2): 1) two three-bar linkages (triangles)
“bde” and “ghi”, which are rigid bodies while the crank rotates;
2) an upper and lower four-bar linkages (crank-rocker) “nm-bj”
and “nm-ck”, respectively, whose governing equations of
motion are similar to any other crank-rocker mechanisms; 3) an
open four-bar linkage “dc-fg” called parallel-like linkage
because of its figurative resemblance to parallelogram; 4) a
rigid three-bar linkage “ghi” called foot; 5) a ground, which is
the link between the two pivoted fixed points ( ),
represented by the dashed vector “n”; and 6) the
interconnection point “P” of links “h” and “i”, which is called
the toe. Each leg is fixed at two points, O2 and O4 in Fig. 2.
Point O4 is where the crank is pivoted to the ground and point
O2 is located at the connection of the upper rocker link “b”, the
lower rocker link “c”, and the link “d” of the linkage “bde”.
Figure 1. A CAD of the Theo Jansen mechanism [9]
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 99
Figure 2. Links and linkages of the Theo Jansen mechanism
Moreover, each two legs are joined together at the point
where the crank is fixed to the ground (point O4), and they are
out of phase with each other for one-half cycle of rotation of
the crank.
The length of each link in the mechanism is defined to
make the foot movement (approximately) linear for one-half of
the rotation of the crank. The remaining rotation of the crank
allows the foot to raise to a predetermined height, which is
called the height of step (see Fig. 3) before returning to the
starting position, which is the beginning of the next cycle.
During each movement cycle of the Theo Jansen mechanism,
each leg experiences four different movement phases which
make a closed curve. These phases are called stride phase, lift
phase, return phase and lower phase as shown in Fig. 3. The
straight line “AB” is called the stride phase in which the leg is
in touch with the ground where it should move at a constant
velocity. The curve “BCA” consists of two portions, “BC” and
“CA”. The first portion “BC” is called the lift phase. During
this phase the leg moves toward its maximum height. As the leg
passes the point “C”, it enters into the second portion of the
“BCA” which is called the return phase. During the return
phase, the leg moves in the same direction as the whole
mechanism. The shape of this phase affects the maximum
acceleration of the leg. Any attempts to make the curve “CA”
as a straight line will ideally minimize the acceleration of the
leg. Lastly, the leg descends to the ground until it reaches to the
point “A” to complete the movement cycle.
Fig. 3 Movement phases of one leg for each cycle [10]
Figure 4. Numbering points for kinematic analysis
KINEMATIC ANALYSIS In this section we studied the kinematics of one leg of the
Theo Jansen mechanism. For this kinematic analysis, we are
interested in the position, velocity and acceleration of points 1,
2, 4, 5, 6 and 7 which are shown in Fig. 4. We obtained the
general equations of motion for the linkages “nm-bj”, “nm-ck”,
“dc-fg” and “cip” (see Fig. 2) using the loop closure method, in
which links are represented as position vectors (see Fig. 5), and
then we solved these analytical equations for one full rotation
of the crank, by numerical analysis at each degree of rotation.
The upper and lower linkages are categorized as crank-
rocker mechanisms because of conforming to the Grashof
condition [11]. As in the kinematic analysis we need the
dimensions of the links, we used the dimensions that are
provided in Theo Jansen’s book [12]. To write the equations of
motion for each linkage, we used complex numbers and the
Euler identity in polar coordinates, which simplifies the
calculations.
POSITION ANALYSIS For position analysis [13] we put the origin of the general
coordinate system (GCS) at point O2 as illustrated in Fig. 5.
The vector loop equation for a general four-bar linkage “dc-ab”
as shown in Fig. 6 will be:
using complex numbers Eq. 1 can be rewritten as:
(2)
where present the link lengths. Equation 2 can be
solved for two unknown parameters,the angular positions of the
rocker and coupler . These unknowns are functions of the
links length and the angle of the crank :
where means either or . As the parameters
and represent the parameters of the
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 100
(a)
(b)
(c)
(d)
Figure 5. Vector loops of the Theo Jansen mechanism;
(a) nm-bj (b) nm-ck (c) dc-fg (d) cip
general linkage “dc-ab”, they should be replaced by the
corresponding parameters in the linkages under study.
(5)
(6)
(7)
(8)
(9)
for example, Eq. 2 can be written in terms of the real and
imaginary parts for the upper linkage:
(11)
for the set of Eq. 10 and Eq. 11, the unknowns are .
From the Eq. 10 and Eq. 11 one can find:
where the constants “x”, “y”, “R”, “ ” and “ ” are:
(14)
(15)
In Eq. 12 and Eq. 15, only one subscript inside the curly
brackets will be considered at a time. Solving Eq. 12 will
determine the angular position of the coupler and rocker of the
upper linkage. A similar approach can be used to find the
unknown parameters for the lower linkage.
Figure 6. Vector loop of a four-bar linkage
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 101
(a)
(b)
Figure 7. Position of points 1, 2, 4, 5, 6 and 7 for one rotation of the crank; (a) x-components (b) y-components
Since the triangle “bde” has rigid body behavior, the angle
of can be obtained using the Law of Cosines where is
known and has obtained from solving Eq.12:
The outputs of the kinematic analysis of the upper and
lower linkages are used as the inputs of the parallel-like
linkage. In the parallel-like linkage “dc-fg”, link “d” is
presumed to play the same role as link “n” does in the analysis
of the upper and lower linkages. However, its angle is not
constant as the crank rotates. Similarly, applying the same
method to the parallel-like linkage; separating equations into
real and imaginary parts will give unknown parameters of this
linkage which are .
The angular position is the foot input. To depict the
trace path of the toe (see Fig. 8(f)), it is required to find angular
position of the variable dashed vector “p” for each increment of
rotation of the crank. This vector together with vectors “c” and
“i” creates the triangle “cip” on which the toe is located (see
Fig. 5 (d)). The angle can be found in a similar way as .
The vector loop equation for this triangle gives the angular
position of the toe as:
(a) (b)
(c) (d)
(e)
(f)
Figure 8. Loci of all points; (a) Point 1 (b) Point 2 (c) Point 4 (d) Point 5 (e) Point 6 (f) Point 7
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 102
The x and y components of the position vector for each
point for one full rotation of the crank are illustrated in Fig. 7.
Moreover, the locus for each point of interest is shown in Fig.
8.
VELOCITY ANALYSIS Taking the time derivative of Eq. 2 gives the velocity
equation:
For the upper and lower linkages, the first and second time
derivatives of the ground link “n” are zero. Moreover, the
angular velocity of the crank is assumed to be known and
constant at 1 rad/s. Writing the velocity equations in terms of
complex numbers for the upper and lower linkages and solving
them for unknown parameters and gives:
(20)
Correspondingly, the linear velocity for each link will be:
(21)
(22)
(23)
To find vectors of angular velocities for the parallel-like
linkage, Eq. 18 can be written in the form of known and
unknown matrices:
where:
(25)
As mentioned before, the triangle “bde” acts kinematically as a
rigid body, hence . Similar rigid body behavior is also
obtained for the linkage “ghi” where . The
components of linear velocity of the toe can be written as
(a)
(b)
Figure 9. Linear velocities of points 1, 2, 4, 5, 6 and 7 for one rotation of the crank;
(a) x-components (b) y-components
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 103
(a) (b)
Figure 10. Linear accelerations of points 1, 2, 4, 5, 6 and 7 for one rotation of the crank;
(a) x-components (b) y-components
the summation of the x and y components of the linear
velocities of the link “c” and link “i” (see Fig. 5(d)):
The x and y components of the linear velocity for all points
(see Fig. 4) for one full rotation of the crank are shown in Fig.
9.
ACCELERATION ANALYSIS Acceleration equations are derived from differentiating of
the velocity equations. Simplifying along with grouping terms
gives:
All the links lengths, angles, and angular velocities have been
determined so far. If the input angular acceleration of the crank
is known, then the angular acceleration of the coupler and
rocker of the upper and lower linkages will be functions of only
aforementioned parameters:
(29)
Expanding the Eulerian form of Eq. 27 in terms of complex
numbers; rearranging real and imaginary parts solely to solve
for angular accelerations of the upper and lower linkages leads
to:
where:
Figure 11. Trace path (loci) of all points of a paired leg using SolidWorks Motion Study Package
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 104
(32)
(33)
(34)
(35)
(36)
Now the components of linear acceleration for each link will
be:
(37)
(38)
(39)
We also can employ the same method used in the velocity
analysis section to write the acceleration equations for the
parallel-like linkage and solve them for the unknown
parameters :
(40)
where:
(41)
(42)
(a)
(b)
Figure 12. Linear velocities of points 1, 2, 4, 5, 6 and 7
for the equivalent time for one rotation of the crank using SolidWorks Motion Study Package;
(a) x-components (b) y-components
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 105
(a)
(b)
Figure 13. Linear accelerations of points 1, 2, 4, 5, 6 and 7 for the equivalent time for one rotation of the crank
using SolidWorks Motion Study Package; (a) x-components (b) y-components
Since the triangle “bde” moves rigidly, the angular
accelerations of all its links will be identical, meaning .
Likewise, for the foot linkage “ghi”, we have
Then linear acceleration of the point “P” can be simply
written as (see Fig. 5 (d)):
The x and y components of the linear acceleration for all
the points (see Fig. 4) for one full rotation of the crank are
shown in Fig. 10.
SOLIDWORKS SIMULATION In order to measure the precision of the analytical results,
we simulated the motion of the Theo Jansen mechanism using
SolidWorks Motion Study package. The position (loci), velocity
and acceleration results for the points 1, 2, 4, 5, 6 and 7 are
represented in Fig. 11, Fig. 12 and Fig. 13, respectively. The
linear velocities and the linear accelerations plots obtained from
SolidWorks are shown for the time of 6.28 seconds which is
equal to one full rotation of the crank at constant angular
velocity of 1 rad/s.
CONCLUSIONS In this paper we employed the loop closure method for
kinematic analysis of the Theo Jansen mechanism. Each leg of
this mechanism is divided into four linkages, which are called
upper, lower, parallel-like and “cip” linkages. Furthermore, the
kinematics is studied for six points of the leg for one full
rotation of the crank, using MATLAB. A simulation was
performed to evaluate the analytical results for each point of
interest using SolidWorks Motion Study package for the
equivalent time for one full rotation of the crank. The position
(loci), linear velocity and linear acceleration results for the
desired points obtained in SolidWorks were in good agreement
with those obtained from the analytical approach in MATLAB.
REFERENCES [1] Ghassaei, A., 2011, “The Design and Optimization of a
Crank-Based Leg Mechanism,” M.S. thesis, Pomona College,
Claremont, CA, www.amandaghassaei.com/files/thesis.pdf
[2] Estremera, J., J. A. Cobano, and Pablo Gonzalez de Santos.
”Continuous Free-Crab Gaits for Hexapod Robots on a Natural
Terrain with Forbidden Zones: An application to humanitarian
Demining.” Robotics and Autonomous Systems 58, no. 5
(2010): 700-711.
[3] Blajer, Wojciech. ”A Projection Method Approach to
Constrained Dynamic Analysis.” Journal of applied Mechanics
59, no. 3 (1992): 643-649.
[4] Ingram, A. J., 2006, “A New Type of Mechanical Walking
Machine,” Ph.D. thesis, University of Johannesburg, South
Africa,https://ujdigispace.uj.ac.za/bitstream/handle/10210/598/
A%20new%20type%20of%20mechanical%20walking%20mac
hine.pdf
[5] Ingram, A. J., 2004, “Numerical Kinematic and Kinetic
Analysis of a New Class of Twelve Bar Linkage for Walking
Machines,” M.S. thesis, Raand Afrikaans University, South
Africa,https://ujdigispace.uj.ac.za/bitstream/handle/10210/1738
/Numerical.pdf?sequence=39
[6] https://www.youtube.com/watch?v=JCPlczI3k-c
[7] https://www.youtube.com/watch?v=b5E-VyQOfNM
[8] https://en.wikipedia.org/wiki/Klann_linkage
[9]www.shapeways.com/product/YHNUXCWDB/grosman-
strandbeest
UAB School of Engineering – Mechanical Engineering - ECTC 2015 Proceedings – Vol. 14 Page 106
[10] Shigley, J. E., 1960, “The Mechanics of Walking Vehicles:
A Feasibility Study: Final Report”, The University of Michigan
Research Institute, Ann Arbor, MI.
[11] https://en.wikipedia.org/wiki/Four-bar_linkage
[12] Jansen, T., 2007, Theo Jansen: The Great Pretender, 010
Publishers, Rotterdam, Netherlands.
[13] Norton, R. L., 2011, Design of Machinery: An Introduction
to the Synthesis and Analysis of Mechanisms and Machines,
5th ed., McGraw-Hill, New York, USA, Chap. 4.