International Conference on Control, Automation and Systems 2008
Oct. 14-17, 2008 in COEX, Seoul, Korea
Fuzzy Logic PID Based Control Design and Performance for a Pectoral Fin Propelled
Unmanned Underwater Vehicle
Jason D. Geder1, John Palmisano2, Ravi Ramamurti1, William C. Sandberg1*, and
Banahalli Ratna2
1
Laboratory for Computational Physics and Fluid Dynamics, U.S. Naval Research Laboratory, Washington, DC
(Tel : +1-202-***-****; E-mail: abqix7@r.postjobfree.com)
2
Center for Bio-molecular Science and Engineering, U.S. Naval Research Laboratory, Washington, DC
(Tel : +1-202-***-****; E-mail: abqix7@r.postjobfree.com)
(Current Address) Modeling and Analysis Division, Science Applications International Corporation, McLean, VA
(Tel : +1-703-***-****; E-mail: abqix7@r.postjobfree.com)
Abstract: This paper describes the modeling, simulation, and control of a UUV in six degree-of-freedom (6-DOF)
motion using two NRL actively controlled-curvature fins. Computational fluid dynamic (CFD) analysis and
experimental results are used in modeling the fin as part of the 6-DOF vehicle model. A fuzzy logic
proportional-integral-derivative (PID) based control system has been developed to smoothly transition between
preprogrammed sets of fin kinematics in order to create a stable and highly maneuverable UUV. Two different
approaches to a fuzzy logic PID controller are analyzed: weighted gait combination (WGC), and modification of mean
bulk angle bias (MBAB). Advantages and disadvantages of both methods at the vehicle level are discussed.
Simulation results show desirable system performance over a wide range of maneuvers.
Keywords: Biomimetic pectoral fin, UUV, unsteady CFD, fuzzy logic, PID, adaptive curvature, weighted gait
combination
1. INTRODUCTION 2. VEHICLE DESIGN AND MODELING
Low-speed and high-maneuverability performance, 2.1 Vehicle design
required in near-shore and littoral zone missions, is a Our team has completed the design and construction
major weakness of current unmanned underwater of a biomimetic controlled-curvature robotic pectoral fin
vehicle (UUV) technology. To address this issue, [3]. A test vehicle (Fig. 1) has been designed to
flapping fin mechanisms have been studied to demonstrate the speed and maneuverability enabled by a
understand how certain aquatic organisms achieve their pair of these pectoral fins [4]. The hull measures 1.3
high levels of controllability and how these mechanisms high, 13 long, and 7 wide, and has a dry weight of 2.6
can be adapted to UUVs [1]. pounds making it slightly negatively buoyant.
In our previous work, we concluded that flapping
pectoral fins were the solution for low-speed,
high-maneuverability operation [2]. We designed a
biomimetic fin propulsor with actively
controlled-curvature [3], and have designed a test
vehicle that utilizes two of these fins for propulsion and
control [4].
Because vehicle simulation results show a need for
more than bang-bang (or case-based) control, a fuzzy
logic proportional-integral-derivative (PID) based
control system is developed. This novel controller
commands changes in fin kinematics for vectoring
control forces necessary to enable a stable and highly
maneuverable UUV.
This paper studies two such non-finite pectoral fin
UUV control methods. The first method is weighted
gait combination (WGC) [5]. WGC is a control Fig. 1 Test vehicle carrying two NRL fins.
method that takes several preprogrammed fin gait
motions and recombines them in real-time to form The vehicle has been integrated with onboard sensors
intermediate hybrid motions. and a custom designed microcontroller [5]. Angular
The second UUV control method studied is mean rate gyroscopes and linear accelerometers are attached
bulk angle biasing (MBAB). When a pectoral fin flaps, to three independent axes, and a pressure sensor for
it produces both lift and thrust forces. By modifying depth measurement is mounted onboard as well.
the mean angle of the flapping fin, the lift force can be Combined with signal amplifiers for high resolution
controlled without affecting forward thrust. data collection, this system provides all necessary
onboard hardware for testing fin force production and provide a more accurate representation of the forces
vehicle maneuverability. generated by the fin because the parameters used in
CFD simulation do not take into account fluid-structure
interaction.
2.2 Equations of motion
A model of the physical system is formulated to aid
in control system development and vehicle performance
studies. Our unique UUV propulsion system makes
use of actively controlled elastic fin deformations that
produce forces at any desired vector [5, 6]. The rest of
the vehicle is rigid and can thus be defined by standard
rigid body equations of motion [7],
v f 0 = m 0 + v0 + rG + rG (1),
t
r
v ~ ~
m 0 = I 0 + I 0 + mrG 0 + v, (2)
t 0
r
where m is mass, 0 is the translational state vector, is
the rotational state vector, rG is the center of gravity Fig. 2 Mean fin generated thrust as a function of the
location, I0 is the inertia tensor, f0 is the external forces controllable parameters, bulk fin stroke amplitude and
vector, and m0 is external moments vector. fin curvature, both non-dimensional. Max. bulk angle:
From Eqs. (1)~(2) we have derived the complete 85 (blue) biased by +0.1N, 65 (green), 45 (red)
equations of motion and computed the prototype vehicle biased by -0.1N.
coefficients from CFD analysis, vehicle geometry, and
mass placement [4].
2.3 Fin Forces
The external forces and moments on the vehicle, f0
and m0, from Eqs. (1)~(2) include those generated by
the NRL fins. A combination of 3-D unsteady CFD
computations [2] and experimental measurements [6]
are used to determine force time-histories generated by
the fins for a range of fin kinematics.
Initial computational studies to optimize the
performance of the NRL fin [3, 8] provided a basis for
determining which fin parameters to analyze as
potential control variables. These controllable fin
parameters include bulk fin stroke amplitude, fin
mean stroke position ( mean), deflection of each of the
four actuated fin ribs ( 1-4), and flapping frequency.
Fig. 3 Mean fin generated lift as a function of the
Unsteady CFD simulations were completed across a
controllable parameters, bulk fin stroke amplitude and
wide span of fin motions across all controllable
fin curvature, both non-dimensional. Max. bulk angle:
parameters [2]. Non-dimensional scale factors, k and
85 (blue) biased by +0.1N, 65 (green), 45 (red)
k, were used to define the values of the controllable
biased by -0.1N.
parameters, and, as,
2.4 Preprogrammed Gaits
(radians) = 1.2514 k, (3)
A specific set of fin kinematics is called a gait, and
(radians) = 1 = 0.1152 k, (4)
ideally, an infinite number of these gaits would be
programmed onto the onboard vehicle microcontroller.
where is a measure of the fin curvature from leading
This would ensure an exact gait to match any force
to trailing edge, and is defined by the maximum tip
vector commanded by the vehicle controller. However,
deflection of the leading edge rib.
this being infeasible, instead we must carefully choose
To evaluate UUV controller performance in
only a few preprogrammed gaits that the vehicle
simulation, a full range of fin forces need to be known.
navigation control system can draw from [5].
Therefore, curve-fitting equations were derived to map
Therefore, the control system uses combinations of this
fin kinematics to stroke averaged forces (Figs. 2~3).
limited set of gaits to generate the infinite number of fin
After validating the experimental force measurement
force vectors needed to perform six degree-of-freedom
accuracy [3, 5], a library relating force data to specific
(6-DOF) maneuvers. The results from Figs. 2~3
fin gaits has been built [6]. These experimental results
identify trends in force output which help determine the certain onboard sensors, such as video cameras for
preprogrammed gaits. object imaging, blurring low-light imaging and adding
Four gaits have been chosen, including a maximum significant noise to the inertial measurement unit.
forward thrust gait (Kf), a maximum reverse thrust gait
(Kr), and maximum upward and downward thrust gaits
(Ku and Kd, respectively) (Fig. 4). Any desired force
vector (Kc) for each fin can be achieved through
weighted combinations of preprogrammed gaits [5], and
then maps of these combinations to forces through the
equation, Kc = f K f, Kr, Ku, Kd . (5)
Note that all gaits are chosen to be entirely uncoupled
allowing each gait to be independently optimized and
studied to improve the overall controller. Extensive
experimental testing has been carried out to determine
optimum gaits for generating maximum forward and
reverse thrust, and maximum positive and negative lift
[6].
Kr Ku
Kc
Fig. 5 Vehicle response to steady level flight command
with bang-bang control
Kf
Kd
Fig. 4 Preprogrammed gait forces (red vectors) used to
achieve desired force vectors (blue vectors).
3. VEHICLE CONTROL
3.1 Bang-bang (or case-based) control
Before designing the fuzzy logic PID based control
algorithm for the vehicle, we first analyzed the
performance of a simple bang-bang control algorithm.
This algorithm limits the fins to switching discretely
between only the four preprogrammed gaits (Figs. 6, 8).
Vehicle simulations employing the bang-bang control
approach exhibit undesirable highly oscillatory motions
dictating the need for development of a more Fig. 6 Control response to steady level flight command
sophisticated control technique. with bang-bang control
The four experimentally selected gaits, previously
mentioned for producing each of the major thrust and The limitations of bang-bang control are displayed on
lift vectors, have been modeled as baseline sets for both a broader scale in lateral turning maneuvers (Fig. 7)
of the NRL fins. where we see vehicle roll, pitch, and yaw (as
We can see in the z-position response (Fig. 5) that the defined in Fig. 1) oscillate with peak-to-peak excursions
vehicle exhibits undamped oscillatory behavior with in excess of 60 degrees. These results show that
bang-bang control. This is also apparent in the pitch bang-bang control is unable to provide the precise
angle (as defined in Fig. 1) response which is maneuvering required of the UUV.
marked by 60 degree peak-to-peak oscillations. This
high magnitude oscillation would preclude the use of
For example, we linearize about steady level flight at a
constant forward speed and determine control gains for
depth response using the algorithm,
u z = K zp e z + K w w + K zi e z, (6)
where uz is the control output, Kzp is the proportional
gain, Kw is the derivative gain, Kzi is the integral gain, ez
is the vertical position error, and w is the vertical
velocity. Similar control outputs are found for the
other states in the forms,
u x = K xp e x + K u u + K xi e x
u y = K yp e y + K v v + K yi e y
u = K p e + K q q + K i e
Fig. 7 Vehicle response to yaw command with
u = K p e + K r r + K i e
bang-bang control
(7)
3.2.1 Weighted Gait Combination (WGC) Method
The individual PID control outputs for various
vehicle states, Eqs. (6)~(7), are then combined to
compute the total control value for each of the fins. In
the WGC method, these control values are computed as
shown in Eq. (8).
u LEFT _ FIN = u x + u y + u z + u + u
(8)
u RIGHT _ FIN = u x u y + u z + u u
In this weighted gait method, the output control
values for the left and right fins are then mapped to
percentages of the four preprogrammed gaits using a
membership function. Using vertical position control
in forward flight as an example (Fig. 10), the
membership function determines weighting of three of
Fig. 8 Vehicle response to yaw command with
the sets of preprogrammed gaits that contribute to the
bang-bang control
control output kinematics for each fin.
3.2 Fuzzy Logic PID Based Control
To improve stability and damp the oscillatory
behavior of the vehicle, a continuous transition between
the four preprogrammed gaits is necessary. One way
to program this type of control is using a fuzzy logic
system in which two or more preprogrammed fin gaits,
each optimized for producing a desired force vector, are
combined to create an intermediate gait. Fig. 9
outlines the design steps necessary to the creation of a
controller for the NRL test vehicle.
Fig. 10 Membership function for vertical position
controller in forward flight.
Fig. 9 Representative control block diagram.
If the vehicle is very far below desired depth, the
maximum positive lift gait is 100% used for both fins,
Linearized equations of motion were computed from
thereby saturating the controller. Similarly, if the
the 6-DOF vehicle model to determine control gains for
vehicle is very far above desired depth, the maximum
various maneuvers using classical control techniques [4].
negative lift gait is 100% used for both fins. And if the a much smoother transition to steady state compared
vehicle is within some predetermined range of desired with bang-bang control response (Fig. 5). During the
depth the output kinematics is calculated as a maneuver, the full weight percentage is on the positive
combination of the optimal positive lift, forward thrust, lift gait briefly, and then smoothly levels out to a full
and negative lift gaits. forward thrust gait weighting (Fig. 13).
3.2.2 Mean Bulk Angle Bias (MBAB) Method
An alternate to the WGC method of fuzzy logic PID
control has also been investigated in which control over
fin lift force is achieved through biasing the fin mean
bulk rotation position ( mean) up or down (Fig. 11).
This propulsion control method entirely decouples lift
from thrust, simplifying the controller design.
right left
right left
(a) (b)
Fig. 11 NRL two fin vehicle showing wing mean
position at, (a) -30 which produces a positive lift force,
and (b) +30 which produces a negative lift force.
The effectiveness of this MBAB method was
suggested by the results of the CFD studies [2]
Fig. 12 Vehicle response to step command in vertical
illustrated in Fig. 3 where we see lift increase as the fin
position with fuzzy logic PID control using WGC.
stroke is biased downwards. Control over horizontal
plane motion is still achieved through combining the
preprogrammed forward and reverse gaits for the left
and right fins as in Eq. (9).
u LEFT _ HORIZ = u x + u y + u
u RIGHT _ HORIZ = u x u y u (9)
u BULK _ BIAS = u z + u
The final step in modeling the vehicle control law
progression, for both the WGC and MBAB methods, is
determining how each gait or combination of gaits maps
to a force vector generated by each fin. Based on CFD
and experimental testing, a mapping from membership
in the various gaits to these fin generated force vectors
is made as discussed in section 2.3 [2, 6].
4. SIMULATION RESULTS Fig. 13 Control response to step command in vertical
position with fuzzy logic PID control using WGC.
Simulation was highly effective for investigating
stability and maneuverability performance of the UUV One concern with the WGC method is the difficulty
using the WGC and MBAB control methods. in controlling multiple vehicle states independently. In
The fuzzy PID controller using the WGC method was the climb maneuver response, we see z-position
tested first in simulation on the 6-DOF model controlled, but this is at the expense of an accurate
previously developed [4], producing predictable results. forward speed control (Fig. 12). We can correct for
Vehicle response to a climb maneuver (Fig. 12) displays this by scaling down the overall weight on the gait
combinations, but this would reduce the envelope in
which the vehicle can operate. More specifically, we
cannot obtain maximum lift simultaneously with
maximum thrust.
The MBAB method solves this problem. By using
this method, we are almost completely decoupling lift
forces from thrust forces when we combine gaits. This
allows us to more easily create a linear map from
control output to fin force.
Vehicle response to a climb maneuver (Fig. 14) is
analyzed using the MBAB method, but in this case
vehicle speed is held constant at 0.5 m/s. During the
maneuver the mean bulk rotation position ( mean) is at
maximum position briefly, and then slowly returns to its
level cruise position for a smooth transition in the
z-response. Meanwhile, a combination of the forward
and reverse gaits is achieved to maintain desired
Fig. 15 Control response to step command in vertical
forward speed (Fig. 15).
position with fuzzy logic PID control using MBAB.
Vehicle response to a yaw maneuver (Fig. 16) using
mean stroke bias for lift control also displays a much
smoother transition to steady state, and less oscillation
in steady state than bang-bang control response. Roll
and pitch ( and ) are limited to 3-4 degrees of steady
state oscillation, while yaw displays even less at
~1-2 degrees. Rise time is also not adversely affected,
as the control limits are saturated throughout 80% of the
yaw angle change (Fig. 17), just as they are using the
WGC method.
Fig. 16 Vehicle response to yaw maneuver with fuzzy
logic PID control using MBAB.
Fig. 17 Control response to yaw maneuver with fuzzy
Fig. 14 Vehicle response to step command in vertical
logic PID control using MBAB.
position with fuzzy logic PID control using MBAB.
AIAA-2006-3658, San Francisco, CA, 2006.
4. DISCUSSION
[3] J. Palmisano, R. Ramamurti, K. Lu, J. Cohen, W.
Sandberg and B. Ratna, Design of a Biomimetic
While the benefit of decoupling lift from thrust is
Controlled-Curvature Robotic Pectoral Fin, In
seen in vehicle results using the MBAB controller, this
Proc. of the IEEE International Conference on
advantage only applies when the fin is at or near a zero
Robotics and Automation, Rome, IT, 2007.
degree angle of attack with respect to the incoming flow.
[4] J. D. Geder, J. Palmisano, R. Ramamurti, W. C.
At higher magnitude angles, the lift and thrust are no
Sandberg and B. Ratna, A New Hybrid Approach
longer decoupled when we bias the fin mean bulk angle
to Dynamic Modeling and Control Design for a
up or down. This becomes an important issue as we
Pectoral Fin Propelled Unmanned Underwater
look forward to redesigning the UUV.
Vehicle, In Proc. of the Fifteenth International
Other results indicate that turning radius and speed
Symposium on Unmanned Untethered Submersible
can be improved by one or a combination of several
Technology, Durham, NH, 2007.
methods. These methods include improving the
[5] J. S. Palmisano, J. Geder, R. Ramamurti, W.
negative thrust gait, changing the vehicle geometry (ie.
Sandberg, B. Ratna, Real-Time Robotic Pectoral
decreasing the size of the vertical tail would facilitate
Fin CPG Using Weighted Gait Combinations,
faster yaw maneuvers), and investigating the advantages
submitted to IEEE Transactions on Robotics,
of non-linear gait weighting.
2008.
Since an extensive search for reverse thrust
[6] J. S. Palmisano, J. Geder, R. Ramamurti, W.
kinematics has already been conducted [6], a mix of
Sandberg, B. Ratna, How to Optimize Efficiency
changing the vertical tail and investigating non-linear
and Propulsion of a Controlled-Curvature Robotic
control weighting need to be considered for future
Pectoral Fin, to be submitted for publication in
generations of the vehicle. Since vehicle pitch
October 2008.
oscillations are relatively small (~3-4 degrees),
[7] T. I. Fossen, Guidance and Control of Ocean
adjusting the hull geometry to increase the pitch
Vehicles, John Wiley & Sons, New York, 1994.
moment of inertia would decrease these excursions even
[8] R. Ramamurti, W. Sandberg, J. Geder,
more. However, this would be at the expense of
Computations of Flapping Fin Propulsion for
decreasing the maximum attainable vehicle pitch angle
UUV Design, submitted to 47th AIAA Aerospace
with current fin gaits, thereby slowing vertical position
Sciences Conference, January 2009.
response. A fine balance needs to be found.
5. CONCLUSIONS
A prototype vehicle to demonstrate the propulsion
and maneuvering performance of its actively controlled
surface curvature deforming pectoral fins has been
developed. Three control techniques were
quantitatively compared for their ability to enable
hovering and maneuvering of our UUV using two NRL
flapping fins. We have described the mathematical
development and maneuvering performance obtained
using this controller. Experimentally and
computationally demonstrated fin performance provided
adequate force production for vehicle propulsion and
control, as shown in simulation results.
While bang-bang control yielded poor results in
simulation, algorithms for combining preprogrammed
fin gaits using both WGC and MBAB methods were
successful in controlling vehicle speed, position and
orientation.
REFERENCES
[1] J. E. Colgate and K. M. Lynch, Mechanics and
Control of Swimming: A Review, IEEE Journal
of Oceanic Engineering, vol. 29, pp. 660-673, July
2004.
[2] R. Ramamurti and W. C. Sandberg,
Computational Fluid Dynamics Study for
Optimization of a Fin Design, In Proc. of the 24th
AIAA Applied Aerodynamics Conference