POLISH MARITIME RESEARCH 4 (108) 2020 Vol. 27; pp.

157-169
10.2478/pomr-2020-0076

AUTOMATIC CONTROL OF SHIP MOTION CONDUCTING

SEARCH IN OPEN WATERS

Krzysztof Stefan Kula

Gdynia Maritime University, Poland

ABSTRACT

In this work a search operation support system is presented, which is an additional autopilot function block. Its task
is to lead a vessel included in the search by the SAR services to the datum position on the basis of the data entered
by the operator into the system and then automatically to search the indicated area according to the newly defined
search pattern. The set goal can be achieved thanks to the autopilot that guides the ship along a given trajectory
consisting of straight lines and arcs. High control accuracy is provided by the IMC control system using a relatively
simple non-linear ship model. The simulation tests of the tanker model confirmed that the indicated search area can
be precisely checked in a shorter time than when using the expanding square search pattern.

Keywords: : computer control, IMC control, safety and risk analysis

INTRODUCTION The planning process should be carried out in accordance

with scientific criteria from both probability theory and
Oceans and seas provide convenient opportunities for the technical aspects in order to increase its effectiveness. SAR
transport of people and goods. Unfortunately, ships involved operations planned in order to detect a missing object at sea
in sea transport are exposed not only to typical transport in the minimum possible time are nowadays supported by
accidents, most often occurring in areas where the density computational methods. The science of search theory was
of shipping is high, but also due to the impact of powerful described by Koopman [13, 14] and Stone [24]. Mathematical
forces of nature occurring in this environment. In order to models based on this theory were formulated by Frost for
provide assistance to the injured in such cases, appropriately this purpose using tidal current prediction and leeway
trained services are required. The International Convention simulations [11]. The first computer-based SAR planning
on Maritime Search and Rescue (SAR) was adopted at an system was developed by the United States Coast Guard [12].
international conference held in Hamburg, Germany, in Visualisation of ocean drift models was introduced as
April 1979. Primarily, it was designed to improve the existing a helpful tool in preparing actions by the SAR coordinator
arrangements and provide a framework for carrying out [7, 30]. Li [18] and Azofra et al. [2] solved the problem of
search and rescue operations following accidents at sea. optimising the deployment of rescue units so that they could
Many states have accepted the obligation to provide maritime cover the largest possible area of coverage while maintaining
SAR coordination and services on a 24-hour basis for their the required safety standards. Burciu [5] proposed a Bayesian
territories, territorial sea and where appropriate, the high Network model to assess the reliability of SAR operations.
seas. To fulfill these responsibilities, they established national Bayesian networks, which were used in probabilistic inference
SAR organisations or joined one or more other states to form concerning the causes of maritime accidents [3], were also
a regional SAR organisation associated with a certain sea area. used to determine the reliability of the search and rescue

POLISH MARITIME RESEARCH, No 4/2020 157

operations (SAR) in coordination centres [22]. A simulation usually be divided into certain sectors, inside each of which
model to evaluate the performance of a SAR system was the POD is constant, though their POC may be different.
constructed by Goerlandt et al. [6]. The SAR handbook [17] recommends the use of expanding
As is often the case with road traffic accidents, the vessels square search, sector search track, line search or parallel
closest to the scene of the incident are fundamental to the sweep search in an open sea.
effectiveness of the assistance provided. Under long-standing
traditions of the sea and various provisions of international DIRECT SEARCH ACTION
law, ship masters are obliged to assist others in distress at sea
whenever they can safely do so. In the case of SAR operations, Based on information from the SMC, including estimation
when the distance from the location of an event is greater than of the most probable position of a distressed craft or survivors
150 Mm, units from outside the SAR resources are included and determination of the search area, taking a drift effect
in the action. The Coordination Centre recommends the into consideration, the assisting ship or, if there are more,
involvement of the vessel in the search action and outlines the on-scene coordinator, can choose the search pattern and
the proposal for its framework. However, the final decision develop a search plan. The SAR headquarters have adopted
in this regard is taken by the master of the vessel, who has certain rules of search action planning. They are largely based
the necessary knowledge of the characteristics and condition on the theory presented in Koopman’s works. Determination
of the ship. of the optimal route of search requires consideration of the
Bearing this kind of SAR operation in mind, the concept of state of the sea, the time of day and weather conditions as well
extending the autopilot function with a system that will allow as steering of the ship. The classical search planning method
for more effective participation of these units in the search (CSPM) assumes that the detection (vs. coverage) function is
action is considered in this article. The aim of this work was to based upon the inverse cube law of visual detection:
develop a system whose purpose is to facilitate merchant ships
(incorporated into the SAR action) in carrying out an optimal γ ≈ kh / r3 (1)
search of the designated area. This tool, referred to as a search
operation support system, SOSS, is an extension of modern where k – constant of proportionality, r – lateral range from
digital autopilots with an additional function block. Its tasks the observer to the search object, h – height above the ocean’s
include planning a search route and automatic control of the surface, being applied under uniform search conditions by
ship motion that guides it along the set trajectory of the route using search patterns consisting of long, straight, equally
determined in the first stage. spaced, parallel tracks relative to the search object.
This method also assumes that the optimal search area is
an area centred on the datum that is searched with uniform
COORDINATION OF SEARCH density of the search effort.
AND RESCUE OPERATION
EXPANDING SQUARE SEARCH
The SAR system in each country, besides the coordinator
representing the highest level of SAR management, has Most search patterns consist of parallel tracks covering
a Search and Rescue Mission Coordinator (SMC) and, if a rectangular area. Non-SAR units near the scene of the
necessary, on-scene coordinators. The SMC typically provides accident usually use the so-called expanding square method
the search or rescue action plan, but may also be required to for their search operations. The shape of the search route
perform other operations. recommended by SAR is shown in Fig. 1.
The elements of the basic problem of an optimal search
are defined by Koopman [13]. They are as follows:
– an a priori probability density distribution on the search
object location, with which it is possible to estimate the
probability of containment (POC) for any subset of the
possible area,
– a detection function relating to the search effort density
or coverage C and the probability of detecting (POD) the
object if it is in a searched area,
– maximising the probability of finding the object, often
referred to as the probability of success (POS), which is
subject to constraints on the effort involved.
In the case of searching for survivors, the data provided
by the SMC contains a brief description of the incident and
a recommended search area. It includes primarily the centre
point, Ro – the radius of the circle in which it is located, and
sometimes size corner points. The region of the incident will Fig. 1. Expanding square search pattern

158 POLISH MARITIME RESEARCH, No 4/2020

 The commence search point CSP is specified by the SMC. This is the case if the course changes at a right angle as
Usually it is the most likely location of the missing objects. illustrated in Fig. 2.
The recommended route consists of pairs of segments running
perpendicularly to each other, whose length gradually increases
by a segment equal to the sweep band width. The distance
between adjacent tracks is called the track spacing S.
Suppose that we are searching uniformly throughout
a region of area As with the effectiveness of “random” search.
The ship is moving at speed V and its sweep width is W.
A certain measure of the effectiveness of the sweeping process
is called, in SAR terminology, coverage C, which expresses the
density of the search effort in the region:

C = Z / AS (2)

where Z – area effectively swept.

This may be determined by multiplying the effective
search width and the effort defined as the total length of the Fig. 2. Changing the sweep width in the corner of the square
searchers’ tracks while searching: As shown, there will be areas farther from the observer on
the vessel than the assumed track spacing. In other words,
Z=W·L (3) the maximum lateral distance between the observer and the
sensor (survivor) r can be greater than S/2 by ΔS, whose value
The effective search width, if a searcher passes through can be determined from the following expression:
a swarm of identical stationary objects uniformly distributed
over a large area, is defined by the equation [12] ΔS = – = 0.2071S (8)

W= (4)
SPIRAL SEARCH
where ΣO/[t] number of objects detected per unit time,
ΣO/[A] number of objects per unit area, V searcher speed and Of course, light and manoeuvrable SAR units can make
where all the values are averages over a statistically significant such a return at approximately right angles. It is different in
sampling period. If the lateral range function is known for the case of large merchant ships, whose minimum turning
a given search situation, then the area under the lateral range radius is much bigger. The new search pattern proposed here
curve equals the sweep width W for that situation. That is, takes into account the real manoeuvring properties of larger
if the detection probability is expressed as a function dr of ships. Its diagram is shown in Fig. 3.
lateral range r from a sensor’s single straight track through
the swarm of objects,

dr(r) = 1 – e (5)
then
W= dr dr (6)

This effective sweep width is also twice the maximum

detection range of an “equivalent” definite range detection
profile (one that is 100% effective out to some definite lateral
range either side of its track and completely ineffective beyond
that range). Here, “equivalent” means that the definite range
detection profile and the actual detection profile both detect,
on average, the same number of objects per unit time under
the same conditions of object density and the searcher’s speed.
Fig. 3. Spiral search
The total length of the route covered by the ship during
the search in time Ts is equal to According to this pattern, apart from the immediate
neighbourhood of commence point search which avoids
L= Vdt (7) duplication of the searched area, the ship sails in a circle
with a radius increasing after a 180-degree change of course.
As we consider this search pattern in detail, it can be The planned trajectory, therefore, consists of half-circles with
seen that the adopted track spacing is by no means constant. a radius equal to the multiple of track spacing.

POLISH MARITIME RESEARCH, No 4/2020 159

COMPARISON OF TWO SEARCH METHODS In the case of the spiral method, taking into account the
dependence S = 2Ro /n and the closing of the last circle, the
In order to compare this method with the expanding total route can be computed from the expression
square search pattern, the length of the route L the ship has
to travel to search the area indicated by the SMC and the time Ltsp = π(n + 2)Ro / 2 (16)
ts needed to complete it were considered.
Let’s assume that the distance sailed by a ship using the Tab. 1. Comparison of lengths of routes needed to sweep through
a given area with a radius Ro
expanding square search pattern is the sum of straight
sections, with adjacent sections having the same length, then n 3 4 5 6 8

Ltsq 10Ro 12Ro 14Ro 16Ro 20Ro

Lsq = 2Σ i · S (9)
Ltsp 7.85Ro 9.42Ro 11Ro 12.57 15.7Ro
where n · S – length of the side of the largest square.
Referring to the expression for the sum of the n-first As can be seen from Table 1, the amount of search effort
elements of the partial sum of this series, we will obtain required to cover the area recommended by the Coordination
Centre using the expanding square pattern is greater.
Σ i·S=S (10) However, it should be taken into account that, after crossing
the Ltsq route, the swept area, although partly outside the area
As the next two consecutive straight sections are of the indicated by the SAR services, is in fact 21.5% larger than it.
same length, the whole length of the route will be double, so The second comparison criterion is the time ts needed to
cover the area indicated by the SAR services, assuming the
Lsq = n(n + 1)S (11) same propulsion of the vessel. Since the average speed of
the vessel, which depends on the rudder angle and changes
In order for the ship to close the square and reach e.g. in time, plays a role in this criterion, the comparison of the
position F (Fig. 1), she must additionally travel a distance two methods was carried out on the basis of simulation tests,
equal to the length of the last side, i.e. n · S. Then, the total which are described in one of the next chapters.
distance travelled Lt while sweeping the square is
PROBABILITY OF SUCCESS
Ltsq = n(n + 2)S (12)
The probability of success can be computed as POC×POD.
In the case of the new search pattern considered, the If the goal is to maximise it, this creates new opportunities
distance sailed by the vessel will be the sum of half-circles when planning a rescue operation, but to compare the
with an increasing radius value. The length of half-circles two methods, it is enough only to examine the POD vs.
with equal radii is equal to 0.5πS, πS, and 1.5πS, respectively. the effort density detection function. It should be noted that
Therefore, the total distance travelled by the vessel can be the following dependences were derived assuming that, in
determined on the basis of the following formula: order to conduct search operations in an area, it is necessary
to move the sensor along a series of long, straight, equally
Lsp = 0.5πS Σ i (13) spaced parallel tracks. Koopman derived the general POD vs.
the coverage detection function for this type of search using
where n – number of the largest half-circle. the simplified inverse cube model of visual detection and
Taking into account the expression for the partial sum of found [13] that
natural numbers, we will receive the expression for the ship
route consisting of semicircles: POD = erf ( ) (17)

Lsp = 0.25πn(n + 1)S (14) where erf is the error function

2
Considering that the search area for vessels does not erf (r) = e–t dt (18)
normally exceed 20 NM, we compare in Table 1 the f-indices
only from a small range. The area to be searched is given The classical search planning method assumes that, for all
by the SMC in the form of a circle with a radius Ro around searches, the track spacing will equal the sweep width, i.e. the
the datum. In the expanding square method, this circle is to coverage will be 1.0. If parallel tracks occur during searches
be inscribed in the largest square, thus its side nS will be equal (Fig. 4b), then the probability of detection will drop half-
to 2Ro. Assuming that it is a multiple of the track spacing, way between them and then increase again. If the tracks are
in the expanding square pattern the ship will pass the route spaced one sweep width apart, the average cumulative POD
should be about 79%. Let’s also consider the impact on the
Ltsq = 2(n + 2)Ro (15) POD that the lateral range in the expanding square pattern
can have. Although the conditions of observation in search

160 POLISH MARITIME RESEARCH, No 4/2020

square corners are slightly different from those of parallel
bands, there are some areas that are more distant from the
observer than half the track spacing distance. The increase in
the distance between two paths in this search area is shown
below in Fig. 4a, where the length of the route travelled by
the ship is plotted on the horizontal axis.

Fig. 5. POD dependence on coverage

For the coverage equal to unity, the POD is 79%. If the

distance between adjacent routes is in the extreme case equal
to 1.207W, the coverage falls to the value of 0.83 and the
POD to 70%.

DURATION OF THE SEARCH

The second comparison criterion is the time ts needed to

Fig. 4. a) Real track spacing varying along the side of the square,
b) Parallel tracks, c) Square search cover the area indicated by the SAR services, assuming the
same propulsion of the vessel.
The lateral range near the vertices of each square increases When planning a search, the objective is often to maximise
by ΔS (8) regardless of the ship’s turning radius, as shown the probability of detecting the missing object by some
in Fig. 2. time. Because the survival time of a human without special
Suppose that track spacing S = W. On the basis of the above protective clothing rapidly reduces as the water temperature
dependences (17), we get that the POD = 0.7899. However, decreases, a special case of SAR action is to maximise the
if the lateral range from the observer to the sensor increases probability of detecting the survivor alive.
to half the maximum distance between adjacent routes and If we are searching uniformly throughout a region of
reaches the value of 1.207.W, the probability of detection area A with the effectiveness of “random” search, then the
decreases to a value of 0.7009. The probability that the distance probability of detecting the search object by time t can be
between the observer and the missing object near the vertices expressed by the so-called random search formula
of the square will be greater than 0.5 S is due to the ratio of
the area, in which this condition is met for the total surface POD = 1 – e (22)
of the sweep band
If we take into account Table 1, it follows from this
P(r > 0.5S) = (19) dependence that the spiral method can be a useful tool for
arranging a plan of the SAR action.
This can be written as Determining the optimal sweep width on the basis of
factors influencing its value due to their fuzziness is difficult
P(r > 0.5S) = (20) to explain. Interpreting them has a significant impact on
the risk. On the one hand, by reducing the track spacing,
which means that the longer the sweep bands are, the smaller the probability of overlooking the missing object decreases.
will be the drop of the POD caused by a temporary increase On the other hand, it may increase the time frame for
in the distance between tracks. finding the survivors. This may not matter much when they
The average value of the range extension in corner squares are in a well-stocked, unsinkable boat. Otherwise, when
of the sweep band is people are directly affected by the environment at a low
water temperature, a change of the strategy to “T-optimal”
m= δdl = 1.1478 S (21) maximises the probability of finding the missing object by
a limited time T. However, a higher speed will increase A but
Fig. 5 shows the graphical dependence of the POD on the reduce the search efficiency at the same time. This could
search effort density, which is expressed here as the ratio W/S. increase the probability of detecting the missing object faster.

POLISH MARITIME RESEARCH, No 4/2020 161

 CONTROLLING THE SHIP
MOTION ALONG AN ARC R = 0.5c / cosα (24)

To conduct a spiral search along the arc-shaped trajectory, where α a is half of the so-called central angle corresponding
a ship motion control system was designed. This innovative to the change of the course angle by Δψ from ψi into ψi+1, and
approach to ship control was described in [16]. This chapter c means the chord, which is the length of the straight line
presents the concept of using this system, extended by a case joining the beginning and end of a desired arc and can be
of turning a complete 180 degrees. developed from the equation
Let’s assume that the considered autopilot can control the
ship to a given way-point. The set course ψref is not a fixed value c= (25)
and depends on the current bearing to the destination point.
If we assume that G(xs,ys) and the offset from the next way- where Δxi+1 = xi+1 – xi Δyi+1 = yi+1 – yi .
point is at the moment t in the meridian line Δxi(t) = xi – xs(t)
and in the parallel line Δyi(t) = yi – ys(t), then the reference The angle α is equal to
course can be determined from the dependence
α = 90° – β (26)
ψref(t) = · arctg + n · 180 [deg] (23)
β = ψi – ψi→i+1 (27)
where x, y – the coordinates of the centre of gravity of the ship,
xi, yi – the specified coordinates for the successive turning where ψi→i+1 is the bearing from wpi to wpi+1.
point,
n = 0 if Δxi > 0 and Δyi > 0,
n = 1 if Δxi < 0, CONTROL SYSTEM OF
n = 2 if Δxi > 0 and Δyi < 0 THE SHIP MOTION
In the proposed layout, the transition from wpi to wpi+1 is
to take place on an arc with a given radius R, as illustrated Most autohelm systems are mainly formulated to follow
in Fig. 6. a desired course under constant speed settings. Because the
ship, as a control plant of its motion, is highly nonlinear,
the object places high demands on the regulation system.
The course control is still an active field of research. Nowadays,
considering structural and model parametric uncertainties
in marine systems, some nonlinear adaptive techniques are
being tested. In recent years, systems such as artificial neural
networks [4, 23], sliding mode control [9, 19], predictive
control [8, 25], fuzzy control [29], backstepping control [26]
and control applying a nonlinear observer [31] have been
tested to achieve high quality regulation. The advantages
of the IMC for the control of the ship motion are presented
in [28]. Because it is not suitable for controlling integral
processes, the structure of the system has been modified to
one in which the angular velocity is regulated. The IMC [20]
is generally an open-loop control system, in which the input
signal is corrected according to the difference between the
c-chord, s-sagitta, R-radius, M-middle of the circle, l-advance section, plant and the model. The diagram of this control system is
B-starting of the manoeuvre
Fig. 6. Turning along the set arc to a fixed way-point
shown in Fig. 7.

The determination of a reference trajectory in the shape of

an arc, which should be kept by the ship from the waypoint
wpi to wpi+1, requires giving their coordinates (xi, yi), (xi+1, yi+1)
and the ship’s course ψ1, ψi+1 at these points. This will make it
possible to determine the required turning radius R, assuming
that the turn is made by an angle of no more than 180 degrees.
There is only one arc, whose slope of tangents in the points
common to neighbouring straight lines will coincide with the
slope of these sections. The radius of this arc can be calculated
on the basis of the trigonometric dependence Fig. 7. Structure of Internal Model Control IMC

162 POLISH MARITIME RESEARCH, No 4/2020

 The signal of mistuning of the model and the object ε may be However, the change in longitudinal and lateral velocities
the result of the influence of disturbances, model uncertainty or during the manoeuvre leads to changes in the dynamics of
model incorrectness. The error signal can adjust the set point the ship and thus increases the incorrectness of the linear
in such a way that the value, adjustable in the steady state, is model. To prevent this, in the proposed control structure, such
equal to the desired value, even when the model differs from a model with parameters depending on propeller revolutions
the object. The transfer function of this system is equal to n and the rudder deflection d was introduced. Eq. (31), after
rearranging to the new form,
G(s) = = (28)
(t) = · (T3(n,δ) · (t) + δ(t)) –
Let’s assume that the dynamics of the linearised plant is
presented in the form of the transfer function Gp(s) and that the –( + ) (t) – r(t) (33)
model fully reflects the properties of the object M(s) = Gp(s).
Then, the controller which could ensure the perfect control can be directly used to create a model, whose operator transfer
in this system should have the transfer function equal to the function with variable parameters better reflects the nonlinear
inverse of the object properties of the ship. A schema of this second order model
is shown in Fig. 8.
Goc(s) = Minv(s) (29)

where Minv(s) is an inverse model of the process taking into

account rules of Q-parametrisation.
To make this controller realisable, it is necessary to add
into the inverse model the forming filter so that its transfer
function should be proper

Q(s) = F(s) · Minv(s) (30)

Fig. 8. Nomoto’s model of the ship with variable parameters k and time
constants T1,T2,T3
where F(s) – the transfer function of the filter.
The dashed lines indicate the influence of the above-
MODEL OF THE SHIP mentioned variables on the model parameters. As can be
seen, deflection of the rudder is not only a control value, but
It can be seen from Eq. (28) that, in the absence of also affects the value of the time constants.
uncertainties and plant modelling errors Eq. (29), the control The dependence of the steady state gain, time constants
system works as an open-loop system. The performance of T1, T2, T3 and the rudder angle at shaft velocity n = 200, are
the control system will depend on the accuracy of the model. shown in Fig. 9.
Unfortunately, the dependences describing the ship’s motion
are nonlinear. The movement of a ship can be described using
six nonlinear differential equations [10], which are based on
the equations of moments and forces acting on a ship in three
dimensions. The first step to obtain a simplified model is to omit
some coupling between the oscillating motions of the vessel
around the axes. Then, after linearisation around a selected
point of work and after elimination of the sway velocity, we get
the following simplified linear differential equation:

(t) + ( + ) · (t) + r(t) =

= · (T3 · (t) + δ(t)) (31)

where δ – rudder angle.

The time constants T1, T2 , T3 and gain k depend on
derivatives of hydrodynamic forces and momentums with
respect to the sway and surge velocity and the yaw rate r. Fig. 9. Model parameters as functions of rudder deflection
Assuming constant speed of the ship, the transfer function
of the model [21] is equal to The steady state gain of the model k at a different number
of propeller revolutions n is calculated from the dependence
M(s) = = (32)
k(δ, n) = k(δ, 200) · n/200 (34)

POLISH MARITIME RESEARCH, No 4/2020 163

 This will ensure better quality of the model in a wider range of the RSB in the section providing control along an arc shape
of changes of the state variables of the system. trajectory is shown in Fig. 11.
If the shaft velocity is equal to 200 revolutions per minute,
then, after a 1 degree rudder deflection, the responses of the 3
DOF nonlinear model and Nomoto’s model are almost the
same, as presented in Fig. 10. However, with larger rudder
deflections, these differences will be bigger.

Fig. 11. Block schema of the Reference Signal Builder M-manual,

A-automatic

The most common use of arc control is a course change

when the curve radius is limited to the specified R max (switch
position M). Arc control from wpi to wpi+1 requires their
coordinates to be entered into the database. The set value of
the turning radius R is calculated in the function block “Turn
radius determination” on the basis of way-points coordinates
and the course to wpi, from Eqs. (24)–(27). The yaw rate
transformer sets on-line for the entire duration of the planned
ship manoeuvre and the reference value of the yaw rate rref
Fig. 10. Model responses of 1 deg rudder deflection
n = 200, d = 1(t = 600), T1= 15 s, T2= 21.6 s, T3= 29.7, k = 0.028 according to the measured vessel speed u.
In order for this system to be able to automatically search
In order to determine the transfer function of the IMC the SAR-derived area around the datum position using the
regulator Goc(s), an inverse model of the plant is required. spiral method, the RSB has been expanded with the way-
It must be compatible with the model of the ship used, so that points generator. Before the automatic search mode can be
Minv(s).M(s) = 1. Then the transfer function of the controller activated, the following data: spacing S, acceptable deviation
Goc(s) will be equal using inertia of the first order (n = 1) from consecutive way-points ε 2 and the coordinates of
as a filter the commence search point (xd, yd) as the centre of a circle
with radius Ro typed by the SMC as the most likely area to be
Goc(s) = (35) searched, must be entered into the system. The initial value
of its radius is usually taken to be 10 Mm.
It should be noted at this point that such an approach as Spiral search is a sequence of turns that constitute
the IMC may cause arbitrary determination of the parameter semicircles with increasing radii. Their radius at the time
Tf to limit the system’s ability to maximise the full power of of transition to a new arc is determined by the dependence
the actuators. However, it is extremely useful in implementing
harmonious control of nonlinear systems. The use of a linear Ri+1 = 0.5(|x’i+1| + |x’i|)| (36)
model to control the nonlinear plant is acceptable; however,
to improve the quality of control, a nonlinear model in the where x’i is the coordinate of the vertical axis, normal to the
structure of the IMC controller could be included, causing direction of approach to the starting point, which is the centre
a better reflection of the actual controlled object than a simple of this reference frame.
linear model. The advantages of such a model in controlling The following radius reference values are determined until
a nonlinear object such as a ship are presented in [15]. the next way-point is reached. The horizontal axis Y of the
Such autopilots as described above cannot be directly used coordinates system of the search area coincides with the
to automatically control a ship during spiral search. In order direction of the ship’s approach to the successive waypoints
to be able to control the ship along the desired path consisting that are equal to 0. However, in order for the entry into the
of straight lines and circular arcs, it is necessary to couple the spiral to actually start near the datum (starting point), the
control system of the ship motion with a device that, according manoeuvre should start some time in advance. This is because
to the designated tasks, will set the current reference value when the rudder is turned, the ship does not immediately enter
of the controlled variable. The task of this device, called the the turn. In order to increase the accuracy of the path-keeping,
Reference Signal Builder (RSB), is to determine the set angular the rudder deflection command must precede the way-point
velocity rref (t) according to the selected operating mode such by a certain distance l. The control system determines the
as the control to a reference course or to a chosen way-point length of advance of manoeuvre l, which depends on the
along a straight line or an arc. The functional block schema ship’s inertia based on the following dependence:

164 POLISH MARITIME RESEARCH, No 4/2020

 l = u(t) · T(u) (37)

where T – resultant time constant of Nomoto’s model

T = T1 + T2 – T3, which also depends on the ship’s velocity u.
The coordinates of the point of starting turn B (x’1, y’1) that
goes beyond the way-point wp1 can be estimated as

y’1= –l x’1 = 0 (38)

Accordingly, the control is redirected to this mode by the

“Manoeuvre activating” block after confirming that it has
reached the designated position, i.e. even before the way-point.
The function block “Position detection” tracks the current
position of the ship and relates it to the set target. Fig. 13. Blue Lady
Transition to the next way-point occurs when one of the lake taken from Tomera’s work [27] was used. Simulation tests
following conditions is met: were performed using the Matlab/Simulink software. The aim
of the study was to verify the operation of the ship control
ε1= ψi+1 – ψ(t) ≤ ± 2° (39) system after the specified arc and to assess the suitability
of the system to automatically perform a search of the area
designated by the SAR service to rescue the survivors. For
ε2= < rt (40) the synthesis of the IMC controller to test its robustness
for model inaccuracy, a model adequate for the number of
where rt – distance representing an acceptable deviation from revolutions n = 200 and rudder deflection δ = 4 deg was used.
the way-point wpi+1. As the filter time constant Tf the value of 30 s was adopted.
Figs. 12.a and 12.b illustrate the importance of anticipating Matlab’s Guide User Interface GUI facilitated communication
the start of a manoeuvre. During the first, the command in the system (Fig. 14).
of the rudder was given when the ship reached wpi and
during the second one at point B, a distance l meters earlier
(l = 11.9 m), which allowed the vessel in the absence of
disturbances to pass the planned trajectory along the arc of
a circle without a shift.

Fig. 14. Graphical User Interface GUI

Simulations of the search vessel movement according

Fig. 12. Route of the ship without (a) and with (b) to the square search method and to the spiral search were
a shift of the manoeuvre beginning
proposed in this work. Based on the results obtained, these
methods were compared in terms of the time consumed to
SIMULATION TESTS achieve the goal. Of course, the aim is to rescue the survivors,
but for the purpose of testing the methods, it was assumed
Simulation tests were carried on Lake Silm out using the that the aim would be to sweep all the points of the desired
model of the ship “Blue Lady” belonging to the Foundation area as quickly as possible from a distance not greater than
for Safety of Navigation and Environment Protection in Iława. the specified effective sweep width. It was assumed that
It provides training for masters and merchant navy officers to the area designated by the SAR mission coordinator to search
practise and improve their skills through trial manoeuvres in for a small boat (less than 5 m) would include a circle with
a safe environment. It is a copy of the tanker made on a scale a radius of 8.3 NM. The weather was windless and visibility was
of 1:24. For the purpose of computer simulation of the ship rated at 15 NM. After scaling, an area with radius Ro = 640 m
motion, a mathematical model verified during trials on the was tested. The ship movement at full load was registered

POLISH MARITIME RESEARCH, No 4/2020 165

in a perpendicular coordinate system, where the datum In the spiral method, taking into account the closing of the
position coincided with the commence search point and last circle, as shown in Fig. 16, the total routes were 62.83.S
had the coordinates (x = 0, y = 0). The ship speed was half- and 64.55.S, respectively. Figs. 17 and 18 showed the ship’s
forward at the specified number of revolutions of the propeller forward speed during the search. As seen in the proposed
n = 200 rpm. With regard to the SAR recommendations, a track method, the hydrodynamic resistance of the hull is lower,
spacing value of 2.1 NM was adopted. After rescaling, this which allows a higher average speed to be obtained when
corresponded to a distance of 160 m. In Figs. 15 and 16, the limiting the maximum speed. However, the optimal speed
trajectory in which the vessel arrived at the starting point was should in any case be determined by the shipmaster.
shown during automatic execution of the square and spiral
search. In the other method, the same turning radius was used
to change the course by 90 degrees as for the first semicircle
in the square method.

Fig. 17. Time series of vessel speed u(t) and average velocity ua(t)
during the square search

Fig. 15. Vessel trajectory during the expanded square search

Fig. 18. Time series of vessel speed during the spiral search
Comparing the probabilities of detection of a missing
object after completing the spiral test, that is, t = 19450 s,
the tested area As is equal to the sum of Ao and Ad, where Ad
doubled the search, including the closing of the last full circle
and sailing to the CSP, because

Ao = πR2o = 16πS2 (41)

Ad = S2 (42)

On the basis of Eq. (22) it is possible to determine

the detection probability for the tested spiral search
PODsp = 0.6955. In the expanded square search, after the
same time, the ship reaches a point with the coordinates
Fig. 16. Vessel trajectory during the spiral search
(–640, 142) and the examined surface reaches the value of
Performing a task using the square method requires 57.86.S2 and the PODsq = 0.5951.
a longer distance. According to Eq. (12), taking into account Figs. 15 and 16 show the trajectories of the ship without
that n = 8, the total distance Ltsq should be equal to 80.S. external disturbances. However, waves, wind and sea currents
However, 79.06.S was measured, because the ship was sailing may affect the ship. On the basis of the relative wind speed
on an arc with the radius of 80 m when changing course. and wind direction, as well as the ship’s response to this

166 POLISH MARITIME RESEARCH, No 4/2020

disturbance measured during the tests on the lake, a model SUMMARY
of the wind force generator was prepared. A fragment of its
recording showing changes in the relative wind speed and The presented innovative method of automatic ship guidance
direction is shown in Fig. 19. The obtained wind model was during the search for a missing object in a spiral is intended
then applied to simulate the ship movement [27]. The effect to shorten the time of sweeping the area around the starting
of wind on the ship will change the shape of the trajectory point indicated by the SAR service. However, carrying out
along which it will move. This depends on several factors. a rescue operation according to this pattern requires additional
First of all is whether the next half circles of the planned computer support. For its execution, a control system was
designed using a technique for guiding the ship along a set
trajectory in the shape of an arc. It allows a constant track
spacing to be kept, which increases its reliability.
The spiral search process was checked and compared with
the usual expanding square method by computer simulation
of the movement of a surface unit with verified properties.
The comparison was made on the basis of the detection
probability and the previously defined performance indicators.
The computer simulations confirmed that the open water area
indicated by the SAR coordinator, determined by means of
a circle with a given radius around the commence search point
with the same main motor drive power, can be swept with
a higher probability of success than would be the case after using
the expanding square method. Simulation tests also showed that
the proposed control system is capable of performing the test
automatically and with high accuracy. Despite the promising
results of the computer simulations, there is a need to conduct
research on Lake Silm using a physical model to confirm the
thesis. The spiral search method extends the time of wave action
in sensitive sectors of its angle of attack, which may limit its
application for smaller vessels without anti-roll stabilisers and
Fig. 19. Time series of the relative wind speed Vw [m/s] and wind direction γ [deg]
therefore requires further research in this regard.
route will have sequential set rays, or the set rays will be
determined successively based on knowledge of the ship’s ACKNOWLEDGEMENTS
position and coordinates of the set waypoints. The SAR
services prefer the first of these approaches, because the same This research was supported by the Ministry of Science
disturbances also affect the missing object and the search and Higher Education, within the framework of funds for
should be carried out around a certain starting point, which research activity No. WE/2020/PZ/03.
will change its position when the same disturbances occur.
However, if the search is to be carried out around a fixed
reference point, then their simulated path is shown in Fig. 20. REFERENCES
1. Ai Y., J. Lu, L.-L. Zhang L-L (2015): The optimization model
for the location of maritime emergency supplies reserve bases
and the configuration of salvage vessel. Transportation
Research Part Vol. 83,170–188.

2. Azofra M., Pérez-Labajos C.A., Blanco B. (2007): Optimum

placement of sea resources. Safety Sciencen Vol. 45 (9),
941–951.

3. Banda O.A.V. (2015): A Bayesian network for assessing

the collision induced risk of an oil accident in the Gulf
of Finland. Environmental Science & Technology. 49,
5301−5309, doi: 10.1021/es501777g.

4. Burns R., Richter R. (1996): Neural-network approach to

the control of surface ships. Control Eng. Practice, Vol. 4,
Fig. 20. Ship’s trajectory during the search in the presence of north or west winds No. 3, 411–416.

POLISH MARITIME RESEARCH, No 4/2020 167

5. Z. Burciu Z. (2010): Bayesian methods in reliability of search 17. IAMSAR Manual, IMO/ICAO-International Aeronautical
and rescue action. Polish Maritime Research. 4(67), Vol. 17, and Maritime Search and Rescue Manual, Vol. III, Mission
72–78, 2010, doi: 10.2478/v10012-010-0039-76. Co-ordination, London/Montreal. 2008.

6. Goerlandt F, Torabihaghighi, Kujala P.(2013): A model 18. Li L. (2006): Rescue vessel location modeling. MSc thesis.
for evaluating performance and reliability of the voluntary Halifax, Canada, Dalhousie University, Department of
maritime rescue system in the Gulf of Finland. 11th Industrial Engineering.
International Conference Probabilistic Safety, Reliability and
Risk Analysis: Beyond the Horizon, Amsterdam, Netherlands. 19. Liu Y., Bu R., Gao X. (2018): Ship Trajectory Tracking Control
2013, 1351–1356. System Design Based on Sliding Mode Control Algorithm,
Polish Maritime Research. No 3,Vol. 25, 26–34, doi: 10.2478/
7. Guoxiang L., Maofeng L., (2010): SARGIS: a GIS-based pomr-2018-0093.
decision-making support system for maritime search and
rescue’, International Conference on E-Business and 20. Morari M., Zafiriou E (1989): Robust process control. Englewood
E-Government, doi: 10.1109/ICEE.2010.398. Cliffs, NJ Prentice Hall, 1989, ISBN-13:978-0137821532.

8. H. N. Esfahani H.N., Szlapczynski R. (2019): Model Predictive 21. Nomoto K., Taguchi T., Honda K., Hirano S. (1957): On
Super-Twisting Sliding Mode Control for an Autonomous the steering qualities of ships. International Shipbuilding
Surface Vehicle. Polish Maritime Research. No. 3, Vol. 26, Progress 4, 354–370.
163–171, doi: 10.2478/pomr-2019-0057.
22. Norrington L., Quigley J., Russel A., Van der Meer R. (2008):
9. Fang M.-C., Luo J.-H. (2007): On the track keeping and roll Modeling the reliability of search and rescue operations with
reduction of the ship in random waves using different sliding Bayesian Networks. Reliability Engineering & System Safety
mode controllers. Ocean Engineering 34, 479–488. Vol. 93(7), 940–949.

10. Fossen T.I. (2002): Marine control systems: guidance, 23. Richter R., Burns R. (1993): An artificial neural network
navigation and control of ships, rigs and underwater vehicles. autopilot for small vessels. Proceedings of the 1st Conference
Marine Cybernetics. Trondheim. ISBN 82-92356-00-2. of the UK Simulation Society, Edinburgh, 168–172.

11. Frost J.R., Cooper D.C., Robe Q. (2003): Compatibility of 24. Stone L.D.(1989): Theory of optimal search. Military
Land SAR Procedures with Search Theory. U.S. Department Applications Section. Operations Research Society of
of Homeland Security United States Coast Guard Operations America, Arlington, Virginia, ORSA Books.
(G-OPR) Washington, D.C. 20593-0001.
25. Velasco F.J., López E.(2000): Predictive Control of Ship
12. Frost J.R.,. Stone L.D. (2001): Review of search theory: Steering Autopilots. 2nd International Congress on Maritime
Advances and applications to search and rescue decision Technological Innovations and Research, Spain, 89–98.
support. US Coast Guard Research and Development Center,
Groton. 2001. 26. Witkowska A., R. Śmierzchalski (2012): Designing a ship
course controller by applying the adaptive backstepping
13. Koopman B.O. (1946): Search and screening. OEG Report method. Int. J. Appl. Math. Comput. Sci., Vol. 22, No. 4,
No. 56, The Summary Reports Group of the Columbia 985–997, doi: 10.2478/v10006-012-0073-y.
University Division of War Research, Alexandria, Virginia,
Center for Naval Analyses. 1946. 27. Tomera M. (2018): Switching-Based Multi-Operational
Control of Ship Motion. Akademicka Oficyna Wydawnicza
14. 14. Koopman B.O. (1980): Search and screening: general EXIT. Warszawa.
principles with historical applications. Revised. New York,
Pergamon Press. 28. Tzeng. C.-Y. (1999): An internal model control approach to
the design of yaw-rate-control ship-steering autopilot, IEEE
15. Kula K.S. ( 2015): Autopilot Using the Nonlinear Inverse Ship Journal of Oceanic Engineering, Vol. 24, No. 4, pp. 507–513.
Model. Marine Navigation and Safety of Sea Transportation,
CRC Press, ISBN: 978-1-138-02858-6, 101–108. 29. Yang Y.S., Ren J.S. (2003): Adaptive fuzzy robust tracking
controller design via small gain approach and its application.
16. Kula K.S.,Tomera M. (2017): Control system of training IEEE Transactions on Fuzzy Systems, 11, 783–795.
ship keeping the desired path consisting of straight-lines and
circular arcs. TransNav Journal Vol. 11 No. 4. 2017, 711–719, 30. Zeid I., Frost J.R. (2005) A decision support system for
doi: 10.12716/1001.11.04.19. Canadian search and rescue operations’, European Journal
of Operation Research. 162 3, 630–653.

168 POLISH MARITIME RESEARCH, No 4/2020

31. Zwierzewicz Z. (2015): The design of ship autopilot by applying
observer-based feedback linearization. Polish Maritime
Research 22(1), 16–21.

CONTACT WITH THE AUTHOR

Krzysztof Stefan Kula

e-mail: [email protected]

Gdynia Maritime University

Department of Ship Automation
Morska 83-87
81-225 Gdynia
Poland

POLISH MARITIME RESEARCH, No 4/2020 169