You are on page 1of 37

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256817942

Spiraling motion of underwater gliders: Modeling, analysis, and experimental


results

Article  in  Ocean Engineering · March 2013


DOI: 10.1016/j.oceaneng.2012.12.023

CITATIONS READS

93 737

4 authors:

Shaowei Zhang Jiancheng Yu


sidsse Chinese Academy of Sciences
13 PUBLICATIONS   132 CITATIONS    60 PUBLICATIONS   393 CITATIONS   

SEE PROFILE SEE PROFILE

Aiqun Zhang Fumin Zhang


Chinese Academy of Sciences Georgia Institute of Technology
71 PUBLICATIONS   600 CITATIONS    190 PUBLICATIONS   2,889 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cyber Physical Systems Journal (Taylor & Francis) View project

Wave Glider View project

All content following this page was uploaded by Shaowei Zhang on 21 January 2018.

The user has requested enhancement of the downloaded file.


Spiraling Motion of Underwater Gliders: Modeling,

Analysis, and Experimental Results

Shaowei Zhang 1 , Jiancheng Yu 2 , and Aiqun Zhang3


Underwater Robot Center, No.114, Nanta Street, Shenyang, Liaoning,China, 110016.

Fumin Zhang4
School of Electrical and Computer Engineering,

Georgia Institute of Technology, Atlanta, GA 30332, USA

We present a thorough approach to characterize the spiraling motion of underwater gliders. The

dynamic model for underwater gliders steered by a single internal movable and rotatable mass is es-

tablished. Equations for the spiraling motion as equilibria of the dynamics are derived, and then solved

by a recursive algorithm with fast convergence. We apply the theoretical method to the Seawing under-

water glider whose hydrodynamic coefficients are computed using the computational fluid dynamics

(CFD) software packages. The glider produced a spiraling motion against strong ocean current in a

recent test at the South China Sea, with experimental results agreed with our theoretical predictions.

The recursive algorithm allows us to compute control input to achieve desired spiraling motion in

practice.

1 State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, 110016, China;
Graduate School of Chinese Academy of Sciences, Beijing, 100049, China. Email:zswsia@126.com
2 State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, 110016, China.
Email:yjc@sia.cn
3 State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, 110016, China.
Email:zaq@sia.cn
4 Electrical and Computer Engineering, Georgia Institute of Technology. Email:fumin@ece.gatech.edu

1
Nomenclature
e0 : (x, y, z) = the body frame

E0 : (i, j, k) = the inertial frame

π1, π 2, π 3)
π 0 : (π = the flow frame

V = [V1 , V2 , V3 ]T = translational velocity in the body frame

Ω = [p, q, r]T = angular velocity in the body frame

b = [x, y, z]T = glider position in the inertial frame

θ = [ϕ, θ, ψ]T = glider attitude in the inertial frame

ms = mass of the static block

rs = position of the static block in the body frame

Is = inertia matrix of the static block in the body frame

mr = mass of the semi-cylindrical movable block

rr = position of the semi-cylindrical movable block

in the body frame

Rr = eccentric offset of the semi-cylindrical movable block

γ = rotation angle of the semi-cylindrical movable block

around the x-axis

Ir (γ) = inertia matrix of the semi-cylindrical movable block

in the body frame

mb = mass of the adjustable net buoyancy

rb = position of the net buoyancy in the body frame

m = mass of the displaced fluid by a glider

ub = rate of change of the mass of the net buoyancy

qsE , qbE , qrE = positions of the center of mass of the static block, the net buoyancy,

and the movable block in the inertial frame, respectively

2
fext = total external force generated by the wings in the inertial frame

τ ext = total external moment generated by the wings

in the inertial frame

MA = added mass terms in the body frame

IA = added inertia terms in the body frame

CA = added coupling terms in the body frame

Tf = total kinetic energy of surrounding fluid

Lglider = glider hull length

Dglider = glider hull diameter

p = translational momentum of the glider in the inertial frame

π = angular momentum of the glider in the inertial frame

P = translational momentum of the glider in the body frame

Π = angular momentum of the glider in the body frame

η = [PT , Π T ]T = the generalized momentum of the glider in the body frame

ν = [VT , Ω T ]T = the generalized velocity of the glider in the body frame

M = the generalized inertia matrix

Vs , Ω s = translational and angular velocity of the static block

Ts = kinetic energy of the static block

Vrx = translational speed of the movable block

along x with respect to e0 in the body frame

ωrx = angular speed of the movable block

around x with respect to e0 in the body frame

Vr , Ω r = translational and angular velocity of the movable block

in the body frame

3
Tr = kinetic energy of the movable block

T = the total energy of the glider system

Fh = hydrodynamic forces in the flow frame

Th = hydrodynamic moments in the flow frame

F = hydrodynamic forces in the body frame

T = [T1 T2 T3 ] = hydrodynamic moments in the body frame

α = the attack angle

β = the flip angle

Re = the Reynolds number of the glider

Vc = velocity of the buoyancy center in CFX simulation

Rc = turning radius of the buoyancy center in CFX simulation

Vinlet = velocity of a fluid particle in the inlet surface

in CFX simulation

Rinlet = turning radius of a fluid particle in the inlet surface

in CFX simulation

ω = the angular speed of fluid in CFX simulation

g = the gravitational constant

µ = the fluid viscosity coefficient

ρ = the fluid density

KD0 , KD = coefficients of drag force D with respect to V 2 , α2 V 2

Kβ = coefficients of side force SF with respect to βV 2

KL0 , Kα = coefficients of lift force L with respect to V 2 , αV 2

KMR , Kp = coefficients of roll moment TDL1 with respect to βV 2 , pV 2

KM0 , KM , Kq = coefficients of pith moment TDL2 with respect to V 2 , αV 2 , qV 2

KMY , Kr = coefficients of yaw moment TDL3 with respect to βV 2 , rV 2

4
I. Introduction

Underwater Gliders are a class of underwater vehicles without external active propulsion systems. Com-

mercially available gliders such as the Slocum [1], the Spray [2] and the Seaglider [3], are capable of

long range missions with low energy consumption, low cost, and long endurance, which have found broad

applications in physical and biological oceanography.

By changing its buoyancy using an internal pumping system, an underwater glider moves up and down

in a water column. The forward gliding motion is generated by the hydrodynamic lift forces exerted on a

pair of wings attached to the glider hull, hence do not require extra energy consumption other than adjusting

the buoyancy. As a result, the glider usually follows a sawtooth motion pattern underwater, and progresses

along a straight line in the horizontal plane. Hence, a glider can be controlled to follow piecewise linear

paths, which has enabled various mission designs in real world applications where either a single or a group

of gliders are deployed [4–8].

The sawtooth pattern can be viewed as a steady state motion of the glider dynamics. The dynamic model

of gliders steered by a linearly movable internal mass are established in [9] and [10]. In [11] and [12], the

stability of the sawtooth gliding motion is analyzed, and a model-based feedback control method is developed

to stabilize the motion. A dynamic model is derived for USM’s underwater glider in [13, 14].

Other than the sawtooth motion, an underwater glider can perform another steady state motion whose

underwater trajectory is a spiral. Similar to the sawtooth motion, the spiraling motion is a result of hydro-

dynamic forces, hence is also energy efficient. The spiraling motion can be used to change the direction of

the glider movement underwater. The spiraling motion can be adjusted using a rotating internal mass e.g.

the battery pack, or a rudder attached to the tail section of the glider. Even though spiraling motions have

been observed in most glider systems, its underlying mechanism is much less understood than the linearly

sawtooth motion. This is because the complex dynamic model and the coupled hydrodynamic effects make

it very challenging to compute the relationship between the spiraling motion and the control inputs. In [10],

a first attempt is made to find numerical solutions for the equations that characterize the spiraling motion.

Some simulation results about spiraling motion are presented in [15]. In [16] and [17], an approximate

analytical solution for steady spiraling motion is derived by applying perturbation theory. Our previous con-

ference paper [18] simplified the equations, which leads to a recursive algorithm that produces fast solutions

5
for spiraling motion.

In order to apply the theoretical methods to real gliders, a complete characterization of the necessary

hydrodynamic coefficients in the dynamic models are necessary. Compared to the towing tank experiments,

computational fluid dynamics (CFD) softwares offer a less expensive alternative to estimate hydrodynamic

coefficients for underwater vehicles in [19, 20]. In [21], the added mass and added inertia terms of underwa-

ter vehicles are computed using the software package U SAERO. In [22], the drag and lift forces generated

by the wings of a fin-actuated underwater vehicle are computed. In [23] and [24], the hydrodynamic coeffi-

cients of underwater vehicles are verified by comparing CFD results with experiments. Glider shape design

has a significant impact on the performance. The hydrodynamic characteristic of the submarine launched

underwater gliders is presented in [25], and the glider shape is optimized to obtain a higher lift to drag ratio

in a large range of attack angle. The effect of using actuated wings on a glider to enhance maneuverability

are verified by experiments in a tank in [26] and [27].

This paper establishes a thorough approach to characterize the spiraling motion of underwater gliders,

including modeling, analysis, and experiments. We extend the theocratical effort initialized in [18] to es-

tablish an analytical model for spiraling motion of underwater gliders steered by an internal movable mass

block. Such theoretical model is then applied to the Seawing glider developed by the Shenyang Institute of

Automation, a subdivision of the Chinese Academy of Sciences. The Seawing glider, illustrated in Fig.1,

was prototyped in 2003 and has gone through several iterations of improvements. It has successfully accom-

plished a series of tests in the ocean in 2011. We then compute all the necessary hydrodynamic coefficients

using CFD software, which allows us to make theoretical predictions on the spiraling motion given a control

input e.g. position of the battery pack. Furthermore, we present results from an ocean experiment performed

at the South China Sea, where the glider produced a spiraling motion against strong ocean current. The

experimental results agree with our theoretical predictions. To our knowledge, the experimental results on

spiraling motion in the ocean have not been previously presented or analyzed in the literature.

The study of the spiraling motion is relevant to new vehicle designs, for example, hybrid underwater

gliders that combine the long endurance of an underwater glider with high maneuverability of a propeller

or jet driven vehicle [28–30]. The spiraling motion will offer a low energy turning behavior, which serves

as an alternative to fast turning behaviors that involve active propulsion. Path planning methods developed

6
Fig. 1 Seawing underwater glider

E0 : Inertial Frame
e0 : Body Frame
j
: Flow Frame i E0
0 T1 , p,

T2 , q,
V2
e0 k
y V2
0
V3 2
3

V1
V V1 V3
T3 , r ,

x z 1 V

Fig. 2 A 3D illustration of the Seawing underwater glider

in [31, 32] may also benefit from understandings of the spiraling motion so that a glider can be temporarily

parked at a spot against strong ocean current.

The paper is organized as follows. Section II derives the dynamic model for underwater gliders steered

by internal movable mass. Section III describes the procedure to compute the hydrodynamic coefficients

used by the dynamic model. Section IV presents a recursive algorithm to produce fast solutions for the

nonlinear equations that establish the relationship between the spiraling motion and the control input. Section

V presents the experimental results collected during an experiment at the South China Sea. Section VI

provides some discussions and conclusions.

7
II. Motion Model for an Underwater Glider

The Seawing glider in [33] is driven by an internal pumping system and is steered by a movable and

rotatable battery pack. The two tail wings aligned in the vertical plane are utilized to steady the vertical

gliding motion. Detailed specifications of the Seawing gliders are given in Table 1.

In this section, we establish the dynamic model for Seawing glider steered by a sliding and rotating

movable mass block. The glider has the following external structures (Fig.2): the prolate ellipsoid as the

rigid hull, the CTD sensor module mounted on top of the hull, two main wings located in the horizontal

symmetry plane e0 − xy near the middle of the hull, the tail wings in the vertical symmetry plane e0 − yz,

and the tail mast serving as the satellite communication antenna.

A. Coordinate Frames

Fig.2 shows the three coordinate frames: the inertial frame, the body frame, and the flow frame, es-

tablished to describe motion of the underwater glider. The body frame e0 : (x, y, z) is established at the

buoyancy center of the glider. The x axis coincides with the longitudinal axis of the glider, and the y axis lies

in the wing plane, pointing to the right when viewed along the direction of x. The z axis is selected as x × y,

as shown in Fig.2. The inertial frame is described by E0 : (i, j, k). In the body frame, the translational ve-

locity and the angular velocity of the underwater glider are defined as V = [V1 , V2 , V3 ]T and Ω = [p, q, r]T ,

respectively. In the inertial frame, the position and the attitude of the underwater glider are described by

b = [x, y, z]T and θ = [ϕ, θ, ψ]T , respectively. A rotation matrix REB maps V in the body frame to the rate

of change of b in the inertial frame as

ḃ = REB V (1)

Using the simplified notation c· = cos(·) and s· = sin(·), REB has the form of

 
cθcψ sϕsθcψ − cϕsψ cϕsθcψ + sϕsψ 
 
 
REB = cθsψ cϕcψ + sϕsθsψ −sϕcψ + cϕsθsψ 

 (2)
 
 
−sθ sϕcθ cϕcθ

8
According to [34], the relations between Ω and θ̇θ is given by,
 
1 sϕ tan θ cϕ tan θ
 
 

θ̇θ = 0 −sϕ 
cϕ Ω (3)
 
 
0 sϕ sec θ cϕ sec θ

Other than the inertial and the body frames, it is usually more convenient to compute and analyze hydro-

π 1 , π 2 , π 3 ) in Fig.2. In order to define the flow frame, we define


dynamics in a flow frame depicted as π 0 : (π

the attack angle α and the slip angle β as

V3
tan α =
V1
V2
sin β = √
V12 + V23 + V33

The flow frame is defined relative to the body frame as follows, first we rotate the body frame around the y

axis for an angle −α, as a result the z axis is rotated to a new position which is now defined as axis π 3 . We

then rotate the new frame around π 3 for an angle β, and the x becomes the flow frame axis π 1 , the y becomes

the flow frame axis π 2 .

B. Model of Mechanics

j
i E0

x Rr qrE
b k
mr qsE y qbE
rr r
rs e0 b
ms mb

Fig. 3 Mass distribution in the body frame and the inertial frame

Fig.3 shows a model of mechanics as a system of mass blocks: the static block with mass ms , the

sliding and rotating movable block with mass mr , and the adjustable net buoyancy mb , which represents the

9
difference between the total buoyancy and the total mass of the glider. We model the movable block as a

semi-cylinder with eccentric offset Rr . Its center of mass is located at position rrx along x, and is rotated

with an angle γ around x. The vectors qrE , qbE , and qsE denote the positions of the centers of mass for the

movable block, the net buoyancy, and the static block in the inertial frame, respectively.

' / PE
PU
7 /

PE! '
PE 

(a) Generate linear sawtooth motion (b) Generate turning spiraling motion

Fig. 4 Mechanisms for the linear sawtooth motion and the spiraling motion

When mb = 0, the displaced fluid mass of the glider is equal to the total mass of the movable block

and the static block, then the glider is neutrally buoyant. If the movable mass block mr stays at the initial

position, then the glider floats horizontally without motion relative to the surrounding fluid. When mb > 0,

the movable mass block mr slides forward to generate a negative pitch angle to drive the glider to dive down;

and when mb < 0, the movable mass block mr slides backward to generate a positive pitch angle to drive

the glider to climb up. The sawtooth motion is shown in Fig.4(a).

Other than the linear motion, the movable mass block mr can rotate around the x axis to achieve turning

motion shown in Fig.4(b): when mr rotates for an angle γ, the glider hull generates a roll angle in the

opposite direction of γ, creating a rotational offset for the static block. As a result, the lift force generated

by the pair of wings attached to the hull yields a force component along j in the inertial frame serving as the

centripetal force for the glider to turn.

C. Mathematical Model of Dynamics

We define p as the translational momentum and π as the angular momentum of the glider in the inertial

frame. We define P and Π as the translational momentum and angular momentum relative to the glider body

frame. The related translational velocity and angular velocity of the movable mass block mr relative to the

10
body frame, caused by the driving motor, are assumed to satisfy Vrx = 0 and ωrx = 0 all the time. These

assumptions are reasonable in practice since the movable mass block moves slowly. As in [12], according to

the Newton’s laws, we know that

ṗ = fext + (ms + mr + mb − m)gk

π = qrE × mr gk + qbE × mb gk + qsE × ms gk + τ ext


π̇ (4)

where k is the unit vector pointing to the direction of gravity, fext is the external force generated by the wings

attached to the glider, and τ ext is the external moment.


[ ]T
We map the generalized momentum pT π T from the inertial frame to the generalized momentum
[ ]T
η = PT Π T in the body frame as

p = REB P

π = REBΠ + b × p (5)

Then we combine the generalized translational velocity V and the angular velocity Ω of the glider in the body
[ ]T
T
frame as ν = V Ω T . The relationship between η and ν is

η = Mνν (6)

where M is the generalized inertia matrix of the glider system. By differentiating Eq.(6) with respect to time,

we get the relationship among the generalized force, velocity and acceleration as

η̇η = Ṁνν + Mν̇ν (7)

Then we need to derive the total kinetic energy of the glider system to find equations for M, Ṁ, ν , and ν̇ν .

We now derive the total kinetic energy of the glider system. Since the center of the movable block has

a constant offset from the center of the body frame e0 , we express the generalized translational velocity and

angular velocity in the body frame as

Vs = V − r̂sΩ

Ωs = Ω

We define the operator ˆ· as the skew-symmetric matrix constructed from a vector ·, for three dimensional

11
column vectors x and y, we have x̂y = x × y. Then the kinetic energy of the static block is

1 2 1
Ts = ms ∥Vs ∥ + Ω s · IsΩ s
2 2
1
= ν T Ms ν (8)
2

The net buoyancy is generated by the oil chamber variation of the pump system, which is located at the center

of glider buoyancy. So the net buoyancy does not contribute to the inertial terms, and also does not contribute

to the kinetic energy of glider dynamic systems. The velocities and kinetic energy of mr vary as rrx and γ

change. The position of the movable block rr is

( )
rr = rrx x + Rr cos(γ + π/2)y + sin(γ + π/2)z (9)

We express the translational velocity and the angular velocity of the movable block in the body frame as

Vr = V − r̂rΩ

Ωr = Ω (10)

So the kinetic energy of the movable block is expressed as:

1 2 1
Tr = mr ∥Vr ∥ + Ω Tr · Ir (γ)Ω
Ωr
2 2
1
= ν T Mrν (11)
2

The inertia matrix of the movable block is a function of γ as

Ir (γ) = RTx (γ)I0r Rx (γ) (12)

where
 
1 0 0 
 
 

Rx = 0 cos γ − sin γ 

 
 
0 sin γ cos γ

and I0r represents the principal inertia matrix of mr computed in the stationary state with γ = 0. Eq.(9)

and Eq.(12) models the effect of the movable/rotatable mass in both the sawtooth motion and the spiraling

motion. When the movable mass rotates, the roll angle changes. From Eq.(12), we know that the inertia of

Ir (γ) changes as γ varies.

12
When a glider accelerates in the flow, the surrounding fluid is accelerated resulting in an apparent in-

crease in mass and inertia. These forms of acceleration-dependent hydrodynamic effects are represented by

the added mass, added inertia, and cross terms. Following [12], we express the kinetic energy generated by

the added mass MA , the added inertia matrix IA and the cross term CA due to the fluid effect as

1
Tf = ν T Mf ν (13)
2

where
 
MA CA 
Mf = 


 (14)
T
CA IA

At last, we get the total energy of the glider/fluid system as:

T = Tr + Ts + Tf

1 T
= ν Mνν (15)
2

The generalized inertia matrix is then


 
Mt Ct 
M=



CTt It

where Mt = (mr + ms )I3 + MA , Ct = CA − ms r̂s − mr r̂r , and It = Is + Ir (γ) + IA − mr r̂r r̂r − ms r̂s r̂s .

We apply the Legendre transform to derive the equations for linear momentum and angular momentum

as Eq.(6). Then we take the time derivative of η and obtain


 
V̇
ν̇ν =   −1
  = M (η̇ − Ṁνν ) (16)

Ω̇

Differentiating Eq.(5) on both sides and applying the kinematic matrix in Eq.(1) and Eq.(3), we get the

relationship among η , ν , ν̇ν , and η̇η as

ΩP)
ṗ = REB (Ṗ + Ω̂

π = REB (Π̇
π̇ ΩΠ) + REB V × p + b × ṗ
Π + Ω̂ (17)

Substituting Eq.(17) into Eq.(4) will give

Ṗ = P × Ω + mb g(RTEB k) + RTEB fext

Π = Π × Ω + P × V + (ms rs + mr rr + mb rb )g
Π̇

×(RTEB k) + RTEBτ ext (18)

13
Then substituting Eq.(18) into Eq.(16), finally we get the dynamic model as in Eq.(19), where F = REB fext ,

and T = REBτ ext are the hydrodynamic forces and moments in the body frame.
  


  P×Ω 
ν̇ν = M−1 −Ṁνν +  




 Π ×Ω +P×V
   


 F
T
 mb g(REB k)
+
+ 
  
(m r + m r + m r )g × (RT k) 
T 
r r s s b b EB

ṁb = ub (19)

D. Hydrodynamic Forces
[ ]T
In the flow frame, the hydrodynamic force Fh = −D SF −L and hydrodynamic moment Th =
[ ]T
TDL1 TDL2 TDL3 are usually expressed as

D = (KD0 + KD α2 )V 2

SF = Kβ βV 2

L = (KL0 + Kα α)V 2

TDL1 = (KMR β + Kp p)V 2

TDL2 = (KM0 + KM α + Kq q)V 2

TDL3 = (KMY β + Kr r)V 2 (20)

A rotation matrix RBC is defined as


 
 cos α cos β − cos α sin β − sin α 
 
 
RBC =
 sin β cos β 0 
 (21)
 
 
sin α cos β − sin α sin β cos α

RBC maps the hydrodynamic force and moment from the flow frame to the body frame as

F = RBC Fh , T = RBC Th (22)

E. Added Mass and Added Inertia Terms

The added terms in Eq.(14) are computed by applying strip theory. The glider hull is approximately as a

slender hull, we consider it as a prolated ellipsoid, and apply the strip theory [34] to compute the added mass

14
Dglider 0.22m
0.25m

1.1m
Lglider 1.99m

Fig. 5 Parameters for the external geometry of the glider

and inertia terms numerically. Since the glider has two symmetry planes as e0 − xz and e0 − xy, the added

mass matrix is approximately diagonal


 
 Xu̇ 0 0 
 
 
MA = 
 0 Yv̇ 0


 
 
0 0 Zẇ

where Xu̇ , Yv̇ and Zẇ are the acceleration fluid mass terms around the glider generated from the force along

the x, y, z axis, respectively. Also, due to the symmetry of the glider, the added inertia matrix IA has a

diagonal form as:


 
 Kṗ 0 0 
 
 
IA = 
 0 Mq̇ 0


 
 
0 0 Nṙ

where Kṗ , Mq̇ , Nṙ are the acceleration fluid inertia terms generated from the moments around x, y, z axis,

respectively.

The matrix CA of cross terms contains the coriolis and centripetal terms. Considering the symmetry

property of the glider, we utilize the strip theory to estimate the following four parameters in matrix CA :

Nv̇ , Mẇ , Yṙ , Zq̇ , where Mẇ and Nv̇ are the pitch moment with respect to the acceleration in the z direction,

and yaw moment with respect to acceleration in the y direction, respectively, and Yṙ and Zq̇ are the side force

due to yaw acceleration and the lift force due to pitch acceleration, respectively. So CA has the form as
   T
0 0 0  0 0 0 
   
   
CA =  0 0 Mẇ  =  0 0 Yṙ 
  

   
   
0 Nv̇ 0 0 Zq̇ 0

15
Table 1 Glider Mechanical Property

List of symbol Value

Static block ms 54.28kg

Position of the static block rs [−0.0814 0 0.0032]m

Inertia of the static block Is diag([0.60 15.27 15.32])kg.m2

Movable block mr 11kg


−π
Rr = 0.014m, 2
≤γ≤ π
2
;
Position of the movable block rr
0.3516m < rrx < 0.4516m

Inertia of the movable block I0r diag([0.02 10.16 0.17])kg.m2

Net buoyancy mb −0.5kg ≤ mb ≤ 0.5kg

Position of the Net buoyancy rb [0 0 0]

Displaced fluid mass m 65.28kg

Glider hull diameter Dglider 0.22m

Glider hull length Lglider 1.99m

The principle of the strip theory includes the following steps: first, we partition the glider into a number

of strips, and then we compute the two-dimensional hydrodynamic derivatives of the added terms for each

strip and integrate the two-dimensional hydrodynamic derivatives to get the tree-dimensional added terms.

Detailed dimension and property of the glider are shown in Fig.5 and Tab.1. Using the strip theory, we

numerically compute the added hydrodynamic terms as Xu̇ = 1.48kg, Yv̇ = 49.58kg, Zẇ = 65.92kg,

Kṗ = 0.53kg.m2 , Mq̇ = 7.88kg.m2 , Nṙ = 10.18kg.m2 , Nv̇ = 2.57kg.m, Zq̇ = 3.61kg.m.

III. Hydrodynamic Modeling

When fluid flows past underwater vehicles, viscous turbulent flow happens at the fluid vehicle boundary

layer. The Reynolds Averaged Navier-Stokes equations (RANS) is a popular method for hydrodynamic

simulations of underwater vehicles. In this study, we use computational fluid dynamics (CFD) software

package, the CFX, that implements the RANS to compute the hydrodynamic coefficients. Proper setup of

CFD simulation requires knowledge of the Reynolds number, proper selection of a turbulence model, and

16
proper settings for grid generation and boundary conditions.

A. Turbulence Models

The Reynolds number Re is the ratio between the inertial effects and the viscous effect in the fluid, and

the solution of viscous flow is affected by the Reynolds number. The glider has its velocity ranges from

0.257m/s (0.5knot) to 0.514m/s (1.0knot), with the Reynolds number computed as:

ρV Lglider
Re = (23)
µ

where ρ = 1025kg/m3 is the water density, V is the glider velocity, Lglider is the total length (not including

the tail), and µ is fluid viscosity coefficient. The Reynolds number ranges from 4.32 × 105 (V = 0.257m/s)

to 8.65 × 105 (V = 0.514m/s).

When 1×105 < Re < 1×106 , the low Reynolds turbulence models give good estimation for underwater

vehicle hydrodynamics [24]. Two popular low Reynolds turbulence models [24, 35, 36] are used in our work:

(1) the standard k −ε turbulence model and (2) the standard k −ω turbulence model. In this paper, we use the

k − ε model to simulate the surrounding flow when the glider is performing linear motion, and use the k − ω

turbulence model to simulate the two-dimensional turbulent flow when the glider is performing spiraling

motion.

B. Simulation Setup

For the k − ε simulation, the fluid volume is selected as 5Lglider × 18Dglider × 18Dglider . The glider

buoyancy center is 9Dglider from the ceiling and the floor of the fluid volume, and 9Dglider from the left side

and the right side of the fluid volume, respectively. The distance between the glider buoyancy center and

the inlet of the fluid volume is 1.5Lglider and the distance between the glider buoyancy center and the outlet

of the fluid volume is 3.5Lglider . The fluid volume in the k − ω simulation is ring-shaped to simulate the

turning motion. The dimension of the volume is 5Lglider × 18Dglider × 18Dglider . The total arc length of

the centerline of the ring is 5Lglider , the arc length along the centerline from the inlet to the buoyancy center

is 1.5Lglider . Other parameters of the fluid volume is the same as in the k − ε simulation.

We use Gridgen as the pre-process tool to create a mesh for the glider. The fluid volume is meshed as

two unstructured grid fields: a coarse mesh field with grid size 200mm and a refined mesh field with grid

17
size 80mm. The refined mesh field is confined to the fluid volume close to the glider hull to enhance the

mesh topology performance near the glider surface. The coarse mesh is utilized for the fluid field further

away from the glider hull to expedite computing process.

After the fluid volume is defined, we specify the fluid boundary conditions for the fluid volumes. The

glider surface boundary is set as a no-slip wall. In the k − ε simulation, the upstream inlet is set as a velocity-

inlet boundary with Vinlet = 0.257m/s, which is perpendicular to the inlet flow surface; the downstream

outlet satisfies the zero static pressure condition; and the other four boundaries of the fluid volume are set as

free-slip walls.

Fig. 6 A section of the ring shaped water volume

The configuration for the k − ω simulation is shown in Fig.6. The linear speed at the glider buoyancy

center is fixed at Vc = 0.257m/s. The angular speed of the fluid volume ω is specified by changing the

curvature of the ring. Let the radius Rc be the distance between the curvature center of the ring and the glider

buoyancy center, then the angular speed of the fluid volume is

Vc
ω= (24)
Rc

The speed of each point of the inlet surface varies as a linear function of the radius of curvature Rinlet .

However, the angular speed ω of the fluid volume is kept as constant. The speed distribution at the inlet

surface can be generated by the angular speed and the distance between any point on the inlet surface and

the curvature center of the ring. Let the position of the curvature center of the ring in the flow frame be
[ ]T
Rc = Rx Ry Rz , and let the position of a point on the inlet surface in the flow frame be r′ =

18
Table 2 Combinations of Parameter Values for CFD Simulation

simulation Vinlet , Vc , α, β

Vinlet = 0.257m/s,

k−ε (β = 0, α = −12, −11, ...12),

(α = 0, β = −12, −11, ...12)

k−ω Vc = 0.257m/s, Rc = 5, 10, 20, 30, 40m;

in horizontal plane (α = 0, β = −3, 0, 3, 6, 9, 12 ;

and vertical plane, separately β = 0, α = −3, 3, 6, 9)

[ ]T
′ ′ ′
. The radius of curvature for a point on the inlet surface Rinlet is obtained as Rinlet =
x y z

(x′ − Rx )2 + (y′ − Ry )2 + (z′ − Rz )2 . Therefore, the speed of flow at that point is Vinlet = ωRinlet .

Other boundary conditions in the k − ω simulation are identical to those used in the k − ε simulation.

In the k − ε simulation, the angular speed of the fluid is zero. The glider buoyancy center is fixed at the

origin of the flow frame. We rotate the fluid volume to achieve different values for the attack angle α and

the side slip angle β, then the CFX software computes the hydrodynamic force in each case. In the k − ω

simulation, we rotate the ring-shaped fluid volume with attack angle α and flip angle β, and change the radius

of the ring. The CFX software computes the hydrodynamic moments in each case, as shown in Figure 7. Tab

2 lists the combinations of the values of α, β, and Rc for all CFD simulations performed.

C. Data Analysis and Curve Fitting

From the boundary conditions Vc , Vinlet , α, β, we can determine the glider velocities V1 , V2 , V3 , p, q, r.

Then from the measured forces and moments, the least mean square method is used to find out the hydrody-

namic parameters based on Eq.(20). The glider velocity V = [V1 , V2 , V3 ]T are computed by converting the

inlet velocity from the flow frame to the body frame as


     
 V1   Vinlet   Vinlet cos αcos β 
     
     
V  = RBC ×  0 = V  (25)
 2     inlet sin β 
     
     
V3 0 Vinlet cos βsin α

19
When the glider glides with the angular velocity aligned with π 3 in the flow frame, we map the angular

velocity to the body frame to compute p, q, r as


     
p 0   −ω sin α 
     
     
 q  = RBC  0 =  (26)
     0 
     
     
r ω ω cos α

On the other hand, when the glider glides with the angular velocity aligned with π 2 in the flow frame, we get

the angular velocity of the glider in the body frame as


     
p 0   −ω cos α sin β 
     
     
 q  = RBC  ω =  (27)
     ω cos β 
     
     
r 0 −ω sin α sin β

Fig.7(a) gives a snapshot of the k − ω simulation when the glider turns left around π 3 axis, and Fig.7(b) the

glider turns down around π 2 axis.

(a) Turning around π 3 (b) Turning around π 2

Fig. 7 Fluid speed distribution in the k − ω simulations.

Fig.8 illustrates the curve fitting procedure to estimate the hydrodynamic coefficients. Data from the

k − ε simulations are used to estimate the hydrodynamic force coefficients based on Eq. (20). The drag force

D and lift force L are functions of α only. Hence KD0 , KD , KL0 , and Kα can be determined from the data

collected by letting β = 0 and α vary from −12◦ ∼ 12◦ , as shown in Fig.8(b) and 8(c). The side force SF

is a function of the slip angle β. Hence Kβ is determined from the data collected by letting α = 0 and β

vary from −12◦ ∼ 12◦ , as shown in Fig.8(f). The moment TDL2 around y axis in the k − ε simulation is a

function of α when q = 0. Hence we can determine KM and KM0 from the data collected by letting α vary

from −12◦ ∼ 12◦ , as shown in Fig.8(d).

20
Ratio of Lift and Drag Glider Drag Force with respect to α Glider Lift Force with respect to α
5 0.5 8
CFD Result
4 Curve Fitting Result
6
CFD Result
3
Curve Fitting Result
4
2
0

Drag Force (N)


2

Lift Force (N)


1

Lift/Drag
0 0

−1
−2
−0.5
−2
−4
−3

−4
−6

−5 −1 −8
−12−11−10−9 −8 −7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 −15 −10 −5 0 5 10 15 −15 −10 −5 0 5 10 15
°
α( ) α ( −12°,−11°,..12°) α ( −12°,−11°,..12°)

(a) Ratio of Lift and Drag (b) Drag Force D vs α (c) Lift Force L vs α

Glider Moment T with respect to α(p=0) °


Glider SideForce SF with respect to β
DL2 Gider moment TDL2 with respect to q (β=0 )
1 2
CFD Result 1
α=−3° CFD Result
0.8 Curve Fitting Result
α=−3° 1.5 Curve Fitting Result
0.5 α=0°
0.6 α=0°
α=3
° 1
0.4
Moment TDL2 (N.m)

α=3°

(N.m)
0 α=6
°

SideForce (N)
0.2 0.5
α=6°
α=9°
DL2
0 −0.5 α=9
°
0
Moment T

°
α=12
−0.2 α=12°
−0.5
−0.4 −1
−1
−0.6
−1.5
−0.8 −1.5

−1 −2 −2
−15 −10 −5 0 5 10 15 −3 −2 −1 0 1 2 3 −15 −10 −5 0 5 10 15
α ( −12°,−11°,..12°) Angular Velocity q °/s β ( −12°,−11°,..12°)

(d) Pitch Moment TDL2 vs α (e) Pitch Moment TDL2 vs q (f) Side Force SF vs β

Glider Steer Moment TDL3 with respect to r (β=−3,0,3,6,9°) Roll Moment TDL1 Curve Fitting (glider turns down) Roll Moment TDL1 Curve Fitting (glider turns up)
1.5
β=−3
β=−3 R=5m
1 R=5m
β=0 0.5 R=5m 0.5
R=5m
β=0 R=10m
Steer Moment TDL3(N.m)

0.5 R=10m
Roll Moment TDL1

Roll Moment TDL1


β=3 R=10m
R=10m
β=3 0 R=20m 0
R=20m
R=20m
0 β=6 R=20m
R=30m
β=6 R=30m
−0.5 R=30m −0.5
−0.5 β=9 R=40m
R=30m
β=9 R=40m
R=40m
R=40m
−1 −1 −1
0.5 0.5
−1.5 15 5
0 10 0 0
5 −3 −5 −3
−2 0 x 10 −10 x 10
−3 −2 −1 0 1 2 3 β −0.5 −5 β −0.5 −15
p p
Steer Angluar Velocity r (°/s)

(g) Steer Moment TDL3 vs r (h) Roll Moment TDL1 vs β and p (i) Roll Moment TDL1 vs β and p

(Descending) (Ascending)

Fig. 8 CFD Simulation Results

Similarly, data from the k −ω simulation are used to find the coefficients for the hydrodynamic moments

based on Eq. (20). We know that the moment TDL3 is a function of β and r. Hence KMY and Kr can be

determined from data collected when glider turns around the π 3 axis, as shown in Fig.8(g). The roll moment

TDL1 in Eq.(20) is a function of β and p. Hence KMR and Kp are determined from data collected when

glider turns around the π2 axis, as shown in Fig.8(h) and 8(i). TDL2 is a function of α and q . Since KM

and KM0 are already determined from the k − ε simulations, the coefficient Kq can be determined from

the data collected when glider turns around the π2 axis, as shown in Fig.8(e). We list all the hydrodynamic

coefficients, added mass and added inertia terms in Table 3.

21
Table 3 Hydrodynamic Coefficients of the Seawing Glider

List of symbol Value

Added mass terms MA diag([1.48 49.58 65.92])kg

Added inertia terms IA diag([0.53 7.88 10.18])kg.m2

Added coupling terms CA Nv̇ = 2.57kg.m, Zq̇ = 3.61kg.m

KD0 = 7.19kg/m,
Coefficients of Drag Force D
KD = 386.29kg/m/rad2

Coefficients of Side Force SF Kβ = −115.65kg/m/rad

KL0 = −0.36kg/m,
Coefficients of Lift Force L
Kα = 440.99kg/m/rad

KMR = −58.27kg/rad,
Coefficients of TDL1
Kp = −19.83kg.s/rad

KM0 = 0.28kg,

Coefficients of TDL2 KM = −65.84kg/rad,

Kq = −205.64kg.s/rad2

KMY = 34.10kg/rad,
Coefficients of TDL3
Kr = −389.30kg.s/rad2

In order to enhance glider performance, we need to determine the attack angle to generate the maximum

lift-drag ratio. From Fig.8(a), we see that the optimal attack angle is approximately ±7◦ .

IV. Characteristics and Analysis of Spiraling Motion

Since we have determined all the coefficients in the dynamic model for the underwater glider, we can

fully characterize its steady state motion. Glider steady motion can be characterized as steady linear motion

and steady spiraling motion. When the movable block slides from initial position along x, the gravitational

center shifts away from the buoyancy center e0 along x, and the moment caused by this offset drives glider

up and down in the vertical plane. When the movable block rotates around x, the gravitational center deviates

from buoyancy center e0 along y. The moment generated by the offset of the movable block causes the glider

to roll around x until it is balanced by T1 and the static block. Since glider wing-forces are not collinear with

22
z, the lift force generates a vertical component to balance glider net buoyancy and a horizonal component as

the centripetal force for the glider to enter a spiraling motion.

With the assumption that the movable block is fixed at rr and with a constant net buoyancy, which imply

that ṁb = 0, glider equilibria equations in three dimension are

0 = P × Ω + mb g(RTEB k) + F (28)

0 = Π × Ω + P × V + (mr rr + ms rs ) g × (RTEB k) + T (29)

And P and Π are simplified as

P = Mt V − (ms r̂s + mr r̂r ) Ω (30)

Π = (ms r̂s + mr r̂r ) V + ItΩ (31)

The steady state linear motion has been investigated in [12]. In this paper, we analyze the steady state

spiraling motion where yaw angle changes at a constant rate while the pitch and roll angles are constant. This

implies that RTEB k are constant. By taking the time derivative of RTEB k, we get

Ω × (RTEB k) = 0 (32)

From Eq.(32), we know that the glider moves with constant speed along a circular helix which is aligned with

gravity, hence the angular velocity is Ω = RTEB kω3 . The spiraling motion can be projected into a rotational

motion around k and an linear motion along k. We project the total velocity V = V12 + V22 + V32 into the

turning direction and the vertical direction as

V cos (θ − α) = ω3 R (33)

V sin (θ − α) = Vvertical (34)

From Eq.(34), we get the vertical velocity in the inertial frame when the glider glides in 3D spiraling motion.

With the fact that Ω = RTEB kω3 , we expand the force equation along x and y in Eq.(28), and moment

23
equation around y in Eq.(29), then we get Eq.(35), Eq.(36) and Eq.(37).

0 = V ω3 (mt2 sin β cos ϕ cos θ − mt3 sin α cos β sin ϕ cos θ)


( )
sin 2θ
2 2
+mr ω3 rrx cos θ + Rr cos(ϕ + γ) − mb g sin θ
2
( )
2 2 sin 2θ
+ms ω3 cos θrs1 + cos ϕrs3
2

−D cos α cos β − SF cos α sin β + L sin α (35)

0 = V ω3 (−mt3 sin α cos β sin θ − mt1 cos α cos β cos ϕ cos θ)

+mb g sin ϕ cos θ − D sin β + SF cos β


( )
sin ϕ sin 2θ sin 2ϕ cos2 θ
+ms ω32
rs1 − rs3
2 2
( )
rrx sin 2θ sin ϕ Rr sin 2ϕ cos2 θ cos γ
+mr ω32

2 2
( )
+mr ω32 −Rr sin γ(cos2 ϕ cos2 θ + sin2 θ) (36)

sin 2γ sin ϕ sin 2θ


0 = (Iry − Irz − mr Rr2 )ω32
4
sin 2α sin 2ϕ cos2 θ sin γ
+(mA3 − mA1 )V 2 cos2 β − mr Rr rrx ω32
2 2

−mr V ω3 rrx (cos α cos β sin ϕ cos θ + sin θ sin β)


 
 Isx − Isz + Irx + IA1 − IA3  2
+  ω3 cos ϕ sin 2θ
  2
−Iry sin2 γ + (mr Rr2 − Irz ) cos2 γ − mr rrx
2

−mr grrx cos ϕ cos θ + TDL1 sin β + TDL2 cos β


( )
sin 2θ ( )
+ms ω3 (rs3 − rs1 ) cos ϕ
2 2 2
+ rs3 rs1 cos ϕ cos θ − sin θ
2 2 2
2

+ms ω3 V sin β (cos ϕ cos θrs3 − sin θrs1 ) − ms g (rs3 sin θ + rs1 cos ϕ cos θ)

−ms V ω3 sin ϕ cos θ cos β (rs3 sin α − rs1 cos α)

+mr V Rr ω3 cos γ cos θ(sin β cos ϕ − sin ϕ sin α cos β)


( )
+ cos γ −mr gRr sin θ + mr Rr rrx ω32 (cos2 ϕ cos2 θ − sin2 θ) (37)

Where mt1 , mt2 , mt3 are the diagonal terms of Mt , respectively, and mA1 , mA2 , mA3 , IA1 , IA2 , IA3 are

the diagonal terms of MA , IA , respectively.

The steady spiraling motion can be determined by ten parameters: V , α, β which describe the velocity

of the glider, ω3 , ϕ, θ which describe the glider angular velocity, net buoyancy mb , rrx and γ of the movable

block, and turning radius R. By combing Eq.(28), Eq.(29) and Eq.(33), we obtain seven equations of 3D

24
spiraling motion. Since these equations may not be independent, we need to specify at least three parameters

out of ten to solve for the other seven parameters. Solving the seven nonlinear equations are challenging for

the computer on-board the glider. Hence we derive a recursive algorithm to obtain fast solutions.

We consider the situation where V , α, β are known, from which we know the expression of T, F, and

RBC , then we solve for the other seven parameters. In order to make Eq.(28) and Eq.(29) simpler to solve,

we take the inner products with respect to Ω on both sides of Eq.(28) and Eq.(29), then we get

mb g
0 = Ω •Ω +F•Ω (38)
ω3

0 = (P × V + T) • Ω (39)

For the Seawing glider, the vector ms rs + mr rr is approximately aligned with the vector rr , therefore, we

take the inner products on both sides of Eq.(29) with respect to rr , and then we get the following equation:

0 = ms rs g × (RTEB k) • rr + (ItΩ × Ω ) • rr + (Mt V × V) • rr + T • rr (40)

From Eq.(38), we can solve for mb as

−F • (RTEB k)
mb = (41)
g

Using Eq.(39) and Eq.(28), we can get the angular velocity ω3 as


( )
F + mb gRTEB k • V
ω3 = − (42)
T • (RTEB k)

Note that T is also a function of ω3 . Hence Eq.(42) is a quadratic equation for ω3 , which can be solved

analytically if θ, ϕ are known. Therefore, supposing solutions of θ, ϕ are found, we can find mb and ω3 from

Eq.(41) and Eq.(42). After that, we can solve for the turning radius from Eq.(33) as

V cos(θ − α)
R= (43)
ω3

With the assumption that we know V , α, β, we can solve for mb , ω3 and R as functions of θ and ϕ. This

process has reduced the unknown parameters from seven to four. We still need to solve for θ, ϕ, γ, and rrx .

When ω3 is small, Eq.(35) and Eq.(36) can be used to solve for θ and ϕ. Then Eq.(37) and Eq.(40) can be

used to solve for γ and rrx . When ω3 is not small, Eq.(35), Eq.(36), Eq.(37), Eq.(40), Eq.(41), and Eq.(42)

have to be solved altogether. We define a recursive vector as ∆ = [θ, R, ϕ, mb , γ, ω3 , rrx ], and establish a

25
recursive equation as

( )
∆ k = F ∆ k−1 (44)

to solve for the recursive vector ∆ . Equations for mb , ω3 , and R has been derived as Eq.(41), Eq.(42), and

Eq.(43). We now derive the equations for θ, ϕ, γ, and rrx .

From Eq.(35), we solve for θ as


θ = arcsin √ 2 2
− λθ (45)
fθ1 + fθ2

where

fθ1 = V ω3 (mt2 sin β cos ϕ − mt3 sin α cos β sin ϕ)

fθ2 = −mb g

fθ = D cos α cos β + SF cos α sin β − L sin α


( )
sin 2θ
−mr ω32 rrx cos2 θ + Rr cos(ϕ + γ)
2
( )
sin 2θ
−ms ω32 cos2 θrs1 + cos ϕrs3
2
fθ1 fθ2
sin λθ = √ 2 2
, cos λθ = √ 2 2
fθ1 + fθ2 fθ1 + fθ2

Similarly, from Eq.(36), we express ϕ as


ϕ = arcsin √ − λϕ (46)
2 + f2
fϕ1 ϕ2

where

sin 2θ sin 2θ
fϕ1 = −mb g cos θ − ms ω32 rs1 − mr ω32 rrx
2 2

fϕ2 = V ω3 mt1 cos α cos β cos θ

fϕ = SF cos β − D sin β

sin 2ϕ cos2 θ
−V ω3 mt 3 sin α cos β sin θ − ms ω32 rs3
2
( )
Rr sin 2ϕ cos2 θ cos γ
+mr ω3 −
2
− Rr sin γ(cos ϕ cos θ + sin θ)
2 2 2
2
fϕ2 fϕ1
sin λϕ = √ , cos λϕ = √
2 2
fϕ1 + fϕ2 2 2
fϕ1 + fϕ2

When ω32 is quite small, terms that contains rrx and γ disappear from Eq.(45) and Eq.(46). Hence after

we replace mb by Eq.(41), the equations Eq.(42), Eq.(45), and Eq.(46) can be solved recursively for θ and ϕ.

26
We next derive the recursive equations from rrx and γ. From Eq.(37), we express γ as


γ = arcsin √ − λγ (47)
fγ21 + fγ21

Where sin λγ = √f 2 γ+f , cos λγ = √f 2 γ+f


f 1 f 2
2 2
, and fγ1 , fγ2 , fγ are defined in Eq.(48). From Eq.(40), we
γ1 γ2 γ1 γ2

get rrx as Eq.(49). Therefore, we have shown that the solution of steady spiraling motion can be derived with

a recursive algorithm given by Eq.(44).

fγ1 = mr V Rr ω3 cos θ(sin ϕ sin α cos β − sin β cos ϕ) + mr gRr sin θ

−mr Rr rrx ω32 (cos2 ϕ cos2 θ − sin2 θ)

fγ2 = mr Rr rrx ω32 cos ϕ cos2 θ sin ϕ

sin 2γ sin ϕ sin 2θ sin 2α


fγ = (Iry − Irz − mr Rr2 )ω32 + (mA3 − mA1 )V 2 cos2 β
4 2

−mr rrx ω3 V (cos α cos β sin ϕ cos θ + sin θ sin β)


 
 Isx − Isz + Irx + IA1 − IA3  2
+  ω3 cos ϕ sin 2θ
  2
−Iry sin2 γ + (mr Rr2 − Irz ) cos2 γ − mr rrx
2

( )
sin 2θ ( )
+ms ω3 (rs3 − rs1 ) cos ϕ
2 2 2
+ rs3 rs1 cos ϕ cos θ − sin θ
2 2 2
2

+ms ω3 V sin β (cos ϕ cos θrs3 − sin θrs1 ) + TDL1 sin β + TDL2 cos β

−ms V ω3 sin ϕ cos θ cos β (rs3 sin α − rs1 cos α)

−ms g (rs3 sin θ + rs1 cos ϕ cos θ) − mr grrx cos ϕ cos θ (48)

27
Table 4 Signs for initial values for the recursive vector

Gliding Motion V α β mb rrx γ ω3 θ ϕ

DL + + + + + - - + -

DR + + - + + + + - -

AR + - + - - + + - +

AL + - - - - - - + +

 
( )
 V (mt1 −
2
mt3 )Rr sin22αcos β sin γ + (mt1 −
2
mt2 )Rr sin22β cos α cos γ 
   
 
 
  sin γ cos ϕ(Isx + IA1 − Isz − IA3 )  
   
   
  + sin ϕ cos γ(I + I − I − I )  
  sx A1 sy A2  
   
   
  +I sin(γ + ϕ) − I sin 2γ cos(γ − ϕ)  
  rx rz  
 −ω32 Rr sin 2θ   
 2   
  +I cos 2γ sin(γ − ϕ)  
  ry  
   
  ( ( )  
  +m R sin γω 2 sin 2θ cos ϕ r2 − r2  
  s r 3 s1 s3  
  2
 
  ( ))  
 +rs3 rs1 sin θ − cos2 ϕ cos2 θ
2 
 
 
 
 −T 
 DL2 Rr sin γ + TDL3 Rr cos γ 
 
 ( ) 
 −m R ω 2 cos γr sin 2ϕ cos2 θrs3
− sin 2θ sin ϕrs1 
 s r 3 s1 
 2 2

 
+ms g ((rs3 sin θ + rs1 cos ϕ cos θ) Rr sin γ + rs1 Rr cos γ sin ϕ cos θ)
rrx =   (49)
( )
sin 2(ϕ−γ)
 ω3 cos θ (Isz + IA3 − Isy − IA2 ) 2 + (Irz − Iry )
2 2 sin 2ϕ

 2

 
 −T 
DL1 + (mt3 − mt2 )V sin α 2
2 sin 2β
 
 
 ( ) 
2
+ms ω3 rs3 sin 2θ
2 sin ϕr s1 − sin 2ϕ
2 cos 2
θr s3 + m s gr s3 sin ϕ cos θ

In order to get a rapid convergence in the recursive algorithm to the desired spiraling motion, proper

initialization for the recursive algorithm is necessary. As we know, the three dimensional spiraling motion

consists of four gliding situations: DL) glider descends and turns left; DR) glider descends and turns right;

AL) glider ascends and turns left; AR) glider ascends and turns right. The signs of the parameters in the

recursive vector are different with respect to different gliding state, which is shown in Table.(4).

We demonstrate a glider 3D spiraling motion equilibrium simulated by MATLAB in Fig.9. The pa-

rameters are: γ = 45◦ , rrx = 0.4216m, mb = 0.3kg, V = 0.490m/s, α = 1.267◦ , β = −1.283◦ ,

ω3 = 0.0039rad/s, ϕ = −13.703◦ , θ = −35.641◦ . The radius of the spiraling motion is R ≈ 100.83. We

28
3D Underwater Glider Simulation
(mb=0.3 kg, rrx=0.4216 m,γ =45°)
Glider Turning Radius R related to Battery Block Rotate Angle
(30°< γ < 60°, mb=0.3kg,rrx=0.4216m)

140
2000

Glider Turning Radius R (m)


130
1500
120
Z (m) 1000
110
500
100
0
200
90
100 100
0
0 −100 80

Y (m) −100 −200 30 35 40 45 50 55 60


X (m)
Battery Block Rotate Angle γ (°)

(a) Spiraling motion (b) Relationship between R and γ

Fig. 9 Left: Simulated trajectory of a glider spiraling downwards. Right: The nonlinear relationship between the

turning radius R and the angle of the battery pack γ

have also computed the relationship between the battery angle γ and the glider turning radius R as shown in

Fig.9. Other groups of parameters are also selected for simulation. Compared to the simulation results based

on the full glider dynamics, the results given by the recursive algorithm has an error of within 5%.

V. Experimental Results

In July, 2011, the Seawing glider was tested in Western Pacific Ocean ( longitude: 130◦ 1.26′ ,

latitude:3◦ 11.44′ ) near Mindanao, Philippines. A series of gliding experiments have been performed. We

present results collected for a downward spiraling motion during a 1.5 hour time window from 12:28 BJT to

14:10 BJT on 20th, July, 2011. The nominal net buoyancy is set to neutral e.g. mb = 0, and the nominal po-

sition of the weight block is at rrx = 0.4016m and γ = 0. When glider dove, the net buoyancy was set to be

mb = 0.5kg, and the movable block are controlled near the nominal position with offset | △rrx |≈ 0.02m.

Our main goal is to verify whether the theoretical model developed for the Seawing glider agrees with the

experimental observations. If the model is consistent with real experiments, then the recursive algorithm we

developed can be used to control the control input for the Seawing glider in practice.

During the experiments, the depth of the glider information is obtained from the CTD (the Conductivity-

Temperature-Depth profiler), and the attitude angles are measured by a TCM3 digital compass. The sampling

periods of TCM3 and CTD are approximately 6 seconds. Sea water density depends on temperature, depth,

and salinity. The density at different depth is estimated from the CTD data, as shown in Fig.10. The rate

29
1032

1030

Density(kg/m3)
1028

1026

1024

1022

1020
0 200 400 600 800 1000
Depth(m)

Fig. 10 Measured sea water density versus depth

0.8
VEast
0.6 VNorth
Max Velocity(m/s)

VVertical
0.4

0.2

−0.2
0 200 400 600 800
depth(m)

Fig. 11 Measured maximum velocity of ocean current versus depth. VEast is the strength of the latitudinal cur-

rent with positive direction eastward. VNorth is the strength of the longitudinal current with positive direction

northward. VVertical is the strength of the up and down current with positive direction aligned with gravity.

of change for density becomes smaller as the depth becomes larger. Sea water density increases about 0.7%

from the surface to 800m depth. The glider reached the maximum gliding depth approximately 800m before

it became neutrally buoyant, and then ascended to the surface by adjusting the buoyancy.

At the test site, there exists ocean current with velocity Vcurrent , An advanced doppler current profiler

(ADCP) (not installed on the glider) is utilized to measure the ocean current Vcurrent on a 8m depth interval

from the ocean surface to the depth of 800m. Fig.11 plots the maximum current measured in 24 hours during

the experiment day. The current measured by the ADCP at the depth where the glider has penetrated can

then we used to calculate the velocity of the glider relative to water as

Vr = V − R−1
EB Vcurrent (50)

30
In the dynamic model (Eq.(19)) for the glider, the term V in the Eq.(19) should now be replaced by Vr to

generate prediction of the motion of the glider based on computer simulation.

Depth Pitch Angle θ(°) Roll Angle φ (°)


0 0
Experiment
Experiment 40
−5 Simulation
100 Experiment Simulation
Simulation −10 30
200
−15
Glider Depth(m)

300 20
−20

θ(°)

φ (°)
400 −25
10
−30
500
−35 0

600
−40
−10
700 −45
1000 2000 3000 4000 5000 6000 500 1000 1500 2000 500 1000 1500 2000
time(s) time(s) times(s)

(a) Glider Depth (b) Glider Pitch Angle θ (c) Glider Roll angle ϕ

rrx of mr γ of mr
Yaw Angle ψ (°)
0.43 10
Experiment
60 Experiment Experiment Simulation
Simulation 0
Simulation
40
0.425 −10
20
Yaw Angle ψ (°)

−20
0
rrx(m)

γ(°)
−20 0.42 −30

−40 −40

−60 0.415 −50


−80
−60
−100
0.41 −70
1000 2000 3000 4000 5000 6000 500 1000 1500 2000 500 1000 1500 2000
time(s) time(s) time(s)

(d) Glider Yaw Angle ψ (e) Position of mr Along x (f) Rotational Angle γ of mr

Fig. 12 Comparing experimental results with simulation results.

The experimental data collected and the simulation results are compared in Figure 12. The simulation

results are generated by MATLAB simulation with the same parameters and added ocean currents measured

in the experiments. Fig.12(e) and Fig.12(f) plot the rrx and γ for the movable block. The position of the

movable block rrx is adjusted to control the pitch angle θ. The angle γ stays at −60◦ most time, but are

moved to 0◦ intermittently. The purpose for this intermittent switching is to counteract the strong ocean

currents. Since when γ = −60◦ , the vertical distance between the center of buoyancy and the center of

mass of the glider decreases comparing to γ = 0◦ , which makes the attitude of the glider less stable. By

intermittently switching the angle γ from −60◦ to 0◦ , we increase the tolerance of strong current. However,

when γ varies, that mass distribution of the glider changes, which further leads to variations of roll angle ψ

and small variations in pitch angle θ to balance the change of γ.

As shown in Fig. 12(a), the glider reached the depth of 800m. The pitch angle is approximately −40◦

shown in Fig.12(b). The roll angle varies around 28◦ as shown in Fig.12(c). The yaw angle changes from

−50◦ to 50◦ at relatively constant rate as shown in Fig.12(d). These results imply that a downward spiraling

31
motion has been successfully produced against the ocean current. From Fig.12(b),12(c), we do see small

oscillations caused by the intermittent rotation of the movable mass. We did not test uncontrolled spiraling

motion because of the strong ocean current that often exceeds glider speed, which may carry the glider out

of the range of detection from the supporting ship.

There are other factors that may have contributed to the difference between experimental data and sim-

ulation results. These include water density variation, change of mass distribution, and the compression of

glider hull at greater depth. In the glider dynamic model, the static mass and movable mass are treated as

rigid mass blocks. In the actual system, however, the net buoyancy is controlled by pumping oil between

different chambers inside the glider, hence the block distribution is difficult to determine. The uncertainty

affects the moment around y axis, which may account for the disagreements between the simulation and the

experiment in Fig.12(b). The Seawing glider has two main wings aligned in the horizontal plane and two tail

wings aligned in the vertical plane of the body frame. During a spiraling motion, the tail wings produce lift

forces that are orthogonal to the lift forces produced by the main wings. The combined force provides the

centrifugal force for turning. The effect of the tail wing is omitted during the theoretical analysis in this paper

since the size of tail wings is much smaller than the main wings. However, we suspect that the effect of the

tail wings on turning may not be neglected in certain circumstances. The optimal ratio between the size of

the tail wings and the size of the main wings is a topic for future research. But nevertheless, in these figures,

we see a consistent match between the simulations and the experimental results, which provides evidence

that the motion model captures the glider dynamics to a satisfactory accuracy.

VI. Conclusions

In summary, we have demonstrated that the dynamic model of underwater gliders can be simplified to

solve for the spiraling motion using a recursive algorithm. All hydrodynamic coefficients in the dynamic

model can be computed using CFD softwares. Because the theoretical glider model is consistent with exper-

imental results, we conclude that the proposed recursive algorithm, which is based on the theoretical model,

may be used to compute the desired control input to produce a circular helical motion in practice.

32
Acknowledgments

This work is supported by State Key Laboratory of Robotics, Grant No.2009-Z05, and Knowledge In-

novation Program of Chinese Academy of Sciences, Grant No.KZCX2-YW-JS205. The authors would like

to thank Zhier Chen, Wenming Jin and Yan Huang for the cooperating work in the glider experiment during

July, 2011. And thank Zhiqiang Hu and Haitao Gu for their help on CFD simulation.

References

[1] Douglas C. Webb, Paul J. Simonetti, and Clayton P. Jones. SLOCUM, an underwater glider propelled by environ-

mental energy. IEEE Journal of Oceanic Engineering, 26(4):447–452, 2001. doi: 10.1109/48.972077.

[2] Jeff Sherman, Russ E. Davis, W. B. Owens, and J. Valdes. The autonomous underwater glider “spray". IEEE

Journal of Oceanic Engineering, 26(4):437–446, 2001. doi: 10.1109/48.972076.

[3] Charles C. Eriksen, T. James Osse, Russell D. Light, Timothy Wen, Thomas W. Lehman, Peter L. Sabin, John W.

Ballard, and Andrew M. Chiodi. Seaglider: A long range autonomous underwater vehicle for oceanographic re-

search. IEEE Journal of Oceanic Engineering, 26(4):424–436, 2001. doi: 10.1109/48.972073.

[4] F. Zhang, D. M. Fratantoni, D. Paley, J. Lund, and N. E. Leonard. Control of coordinated patterns for ocean

sampling. International Journal of Control, 80(7):1186–1199, 2007. doi: 10.1080/00207170701222947.

[5] D. A. Paley, F. Zhang, D. M. Fratantoni, and N. E. Leonard. Cooperative control for ocean sampling: The glider

coordinated control system. IEEE Transaction on Control System Technology, 16(4):735–744, 2008. doi: 10.1109/

TCST.2007.912238.

[6] N. E. Leonard, D. A. Paley, R. E. Davis, D. M. Fratantoni, F. Lekien, and F. Zhang. Coordinated control of an

underwater glider fleet in an adaptive ocean sampling field experiment in monterey bay. Journal of Field Robotics,

27(6):718–740, 2010. doi: 10.1002/ROB.20366.

[7] Ryan N. Smith, Yi Chao, Burton H. Jones, David A. Caron, Peggy P. Li, and Gaurav S. Sukhatme. Trajectory

design for autonomous underwater vehicles based on ocean model predictions for feature tracking. In The 7th

International Conference on Field and Service Robots (FSR 2009), pages 263–273, MIT Campus, Cambridge, MA,

2009.

[8] Ryan N. Smith, Mac Schwager, Stephen L. Smith, Daniela Rus, and Gaurav S. Sukhatme. Persistent ocean mon-

itoring with underwater gliders: Adapting sampling resolution. Journal of Field Robotics, 28(5):714–741, 2011.

doi: 10.1002/ROB.20405.

[9] J. Graver and N. E. Leonard. Underwater glider dynamics and control. In 12th International Symposium on

Unmanned Untethered Submersible Technology, pages 1–14, Durham, 2001.

33
[10] Pradeep Bhatta and Naomi Ehrich Leonard. Nonlinear gliding stability and control for vehicles with hydrodynamic

forcing. Automatica, 44(5):1240–1250, 2008. doi: 10.1016/J.AUTOMATICA.2007.10.006.

[11] Asher Bender, Daniel Matthew Steinberg, Ariell Lee Friedman, and Stefan B. Williams. Analysis of an autonomous

underwater glider. In Australasian Conference on Robotics and Automation, pages 1–10, Canberra, Australia, 2008.

[12] N. E. Leonard and J. Graver. Model-based feedback control of autonomous underwater gliders. IEEE Journal of

Oceanic Engineering, 26(4):633–645, 2001. doi: 10.1109/48.972106.

[13] Nur Afande Ali Hussain, Mohd Rizal Arshad, and Rosmiwati Mohd-Mokhtar. Underwater glider modelling and

analysis for net buoyancy, depth and pitch angle control. Ocean Engineering, 38(16):1782–1791, 2011. doi:

10.1016/J.OCEANENG.2011.09.001.

[14] Khalid Isa and Mohd Rizal Arshad. Dynamic modeling and characteristics estimation for USM underwater glider.

In Proc. 2011 IEEE Control and System Graduate Research Colloquium, pages 12–17, Shah Alam, 2011.

[15] Lei Kan, Yuwen Zhang, Hui Fan, Wugang Yang, and Zhikun Chen. MATLAB-based simulation of buoyancy-

driven underwater glider motion. Journal of Ocean University of China (English Edition), 7:113–118, 2008. doi:

10.1007/S11802-008-0113-2.

[16] Nina Mahmoudian, Jesse Geisbert, and Craig Woolsey. Approximate analytical turning conditions for underwater

gliders: Implications for motion control and path planning. IEEE Journal of Oceanic Engineering, 35(1):131–143,

2010. doi: 10.1109/JOE.2009.2039655.

[17] Nina Mahmoudian and Craig Woolsey. Underwater glider motion control. In Proceedings of the 47th IEEE Con-

ference on Decision and Control, pages 552–557, Cancun, Mexico, December 2008.

[18] Shaowei Zhang, Jiancheng Yu, Aiqun Zhang, and Fumin Zhang. Steady three dimensional gliding motion of an

underwater glider. In Proc. 2011 IEEE Conference on Robotics and Automation, pages 2392–2397, Shanghai,

China, 2011.

[19] Benjamin J. Racine and Eric G. Paterson. CFD-based method for simulation of marine-vehicle maneuvering. In

Proceeding of 35th AIAA Fluid Dynamics Conference and Exhibit, pages 1–22, Toronto, Ontario, Canada, June

2005.

[20] David A. Boger and James J. Dreyer. Prediction of hydrodynamic forces and moments for underwater vehicles

using overset grids. In Proceeding of 44th AIAA Aerospace Sciences Meeting and Exhibit, pages 1–13, Reno, NV,

USA, 2006.

[21] Jesse Stuart Geisbert. Hydrodynamic modeling for autonomous underwater vehicles using computational and semi-

empirical methods. Master’s thesis, Virginia Polytechnic Institute and State University, Virginia, 2007.

[22] Kristi A. Morgansen, Benjamin I. Triplett, and Daniel J. Klein. Geometric methods for modeling and control of

free-swimming fin-actuated underwater vehicles. IEEE Transactions on Robotics, 23(6):1184–1199, 2007. doi:

34
10.1109/TRO.2007.911625.

[23] Sulin Tang, Tamaki Ura, Takeshi Nakatani, Blair Thornton, and Tao Jiang. Estimation of the hydrodynamic co-

efficients of the complex-shaped autonomous underwater vehicle TUNA-SAND. Journal of Marine Science and

Technology, 14(3):373–386, 2009. doi: 10.1007/S00773-009-0055-4.

[24] P. Jagadeesh, K. Murali, and V. G. Idichandy. Experimental investigation of hydrodynamic force coefficients over

AUV hull form. Ocean Engineering, 36(4):113–118, 2009. doi: 10.1016/J.OCEANENG.2008.11.008.

[25] Joshua D. Rodgers and John M. Wharington. Hydrodynamic implications for submarine launched underwater

gliders. In IEEE OCEANS 2010, pages 1–8, Sydney, May 2010.

[26] M. Arima, N. Ichihashi, and T. Ikebuchi. Motion characteristics of an underwater glider with independently con-

trollable main wings. In IEEE OCEANS 2008, pages 1–7, Kobe, Japna, 2008.

[27] M. Arima, N. Ichihashi, and Y. Miwa. Modelling and motion simulation of an underwater glider with independently

controllable main wings. In IEEE OCEANS 2009, pages 1–6, Bremen, Germany, May 2009.

[28] A. Alvarez, A. Caffaz, A. Caiti, G. Casalino, L. Gualdesi, A. Turetta, and R. Viviani. Fòlaga: A low-cost au-

tonomous underwater vehicle combining glider and AUV capabilities. Ocean Engineering, 36(1):24–38, 2008. doi:

10.1016/J.OCEANENG.2008.08.014.

[29] Shuxin Wang, Chungang Xie, Yanhui Wang, Lianhong Zhang, Weiping Jie, and S. Jack Hu. Harvesting of PEM

fuel cell heat energy for a thermal engine in an underwater glider. Journal of Power Sources, 169(2):338–346, 2007.

doi: 10.1016/J.JPOWSOUR.2007.03.043.

[30] Shuxin Wang, Xiujun Sun, Yanhui Wang, Jianguo Wu, and Xiaoming Wang. Dynamic modeling and motion

simulation for a winged hybrid-driven underwater glider. China Ocean Engineering, 25(1):97–112, 2011. doi:

10.1007/S13344-011-0008-7.

[31] J. Isern-González, Daniel. Hernández Sosa, Enrique Fernández-Perdomo, Jorge Cabrera-Gámez, and Antonio C.

Domínguez-Brito. Path planning for underwater gliders using iterative optimization. In Proceedings of 2011 IEEE

International Conference on Robotics and Automation, pages 1538–1543, Shanghai, China, 2011.

[32] Arvind A. Pereira, Jonathan Binney, Burton H. Jones, Matthew Ragan, and Gaurav S. Sukhatme. Toward risk

aware mission planning for autonomous underwater vehicles. In Proc. 2011 IEEE/RSJ International Conference on

Intelligent Robots and Systems, pages 3147–3153, San Francisco, CA, USA, 2011.

[33] J. Yu, F. Zhang, W. Jin, and Y. Tian. Motion parameter optimization and sensor scheduling for the sea-wing

underwater glider. IEEE Journal of Oceanic Engineering,conditionally accepted, 2012.

[34] T. I. Fossen. Handbook of Marine Craft Hydrodynamics and Motion Control. Wiley, UK, 2011.

[35] Hyeon Kyu Yoon, Nam Sun Son, and Gyeong Joong Lee. Estimation of the roll hydrodynamic moment model of a

ship by using the system identification method and a free running model test. Ocean Engineering, 32(4):798–806,

35
2007. doi: 10.1109/JOE.2007.909840.

[36] P. Jagadeesh and K. Murali. Application of low-Re turbulence models for flow simulations past underwater vehicle

hull forms. Journal of Naval Architecture and Marine Engineering, 1(2):41–54, 2005. doi: 10.3329/JNAME.V2I1.

2029.

36

View publication stats

You might also like