PhD thesis

Estimation of waves and ship responses

using ship-board measurements

by:
Najmeh Montazeri

Supervisors:
Associate professor: Ulrik Dam Nielsen
Professor: Jørgen Juncher Jensen

Technical University of Denmark

Department of Mechanical Enginnering

In collaboration with:

NK classification society

1
 Contents
 Introduction: decision support system

 Wave estimation using ship responses

 Model description

 Global optimization

 Wave estimation based on numerical simulations

 Wave estimation based on full-scale measurements

 Onboard prediction of waves and responses

 Automatic response selection

 Uncertainty analysis

 Summary and conclusion

 Future works
2
 Introduction: Decision Support System

Excessive wave-induced motions, accelerations and stresses, may increase the risk of:
• Capsizing;
• Large roll motion;
• Fatigue damage;
• Structural damage;
• Damage of equipment on deck;
• Loss or shift of cargo;
• Sea sickness;
• …
How to reduce the risks?
 Making decision on the heading and the speed of the ship by use of:

1. Wave data
2. Estimation of ship responses (sea keeping, structural loads, added resistance, …)
3. Statistical predictions of expected responses in a time horizon of 20-60 minutes

3
 Decision Support System
Tools for estimation of waves

• Wave rider buoys

• Marine radar
• Satellite measurements
• Measured ship responses

The advantage of the last method is that:

 the instrumentation is simple and inexpensive compared to

other means.
 The wave data can be provided real-time and at the actual
position of the ship.

4
 Wave estimation using ship responses

Sensor Data Processing

measurements & Analysis
Monitoring

Operational Hydrodynamic Sea state Motions/Loads

data data estimation predictions

Check with criteria

Polar plots
 Wave estimation using ship responses

• A ship can act as a wave buoy.

• The theoretical relationship between the spectral density of a ship response and
the directional wave spectrum is given by:

Ship Responses Optimization Waves

𝜙(𝜔) S(𝜔, 𝜃)

RAO
𝐻(𝜔, 𝜃)

2
2.5

+ 1.5

E( )
S

1.5
1
1

0.5 0.5

0
0 0.5 1 1.5
0
 0 0.5 1 1.5

6
 Model description
Optimization problem based on spectral moments

• The amount of energy of the responses should be conserved whether using

measured signals or theoretical calculations.

• Higher order moments can also be used as well

• A number of responses (N) are used to establish the set of objective functions.

7
 Model description
Parametric modeling approach
• There are different methods including parametric and non-parametric to solve the above
obtimizaion problem.
• In the parametric method, the wave spectrum, S, is assumed to follow a standard e.g.
JONSWAP model:

2𝜋
, 𝜔𝑝 =
𝑇𝑝

• Short-crested waves are considered:

• With a spreading function as:

µ : relative mean wave direction

𝜔: wave frequency
𝛳: relative wave direction
𝜔𝑝 : peak frequency
S: spreading parameter
𝛾: peakedness factor
8
 Model description
Spectral partitioning and inequality constraints
• Identification of swell and wind-sea components

𝑆 𝜔, 𝜃 = 𝑆𝑤 𝜔, 𝜃 + 𝑆𝑠𝑤 𝜔, 𝜃

• The current wind information can be used for partitioning:

• Separation frequency in the spectrum:
𝑔
𝜔𝑠 =
𝛽𝑈𝑤
• General wave steepness constraint:

• Lower bound for wind sea steepness:

• Additional constraint for wave direction estimation:

𝜃𝜔 − 𝜇 < 90°

𝑈𝑤 : wind speed
𝛽: empirical constant
g: gravity acceleration
𝜃𝜔 : relative wind direction
𝜇: relative mean wave direction
9
 Global Optimization

• The fitting process is set by the difference between the measured and the
theoretical spectral moments normalized by the measured moment
(variance) of the response.

• Optimization is based on some of squared residuals for all considered

responses.

• Vertical motion, pitch, sway and vertical bending moment are used
according to experience and literature.
• Global optimizaion is applied using Multi-Start and Genetic algorithm in
matlab.
• The wave parameters (Hs, Tp, 𝜇, ...) are then estimated.
10
Global Optimization
Flowchart

11
 Case study I: 9400 TEU container ship

Parameter Dimension
Overall Length 349.0 m
Beam 42.8 m

Maximum Draught 15.0 m

DWT 113,000 ton

Capacity 9415 TEU
Operational Draft 14.2 m
Operational Speed 21.0-23.5 kn

Other equipment for onboard wave estimation:

1. An X-band radar from WAMOS® (Wave and current Monitoring System)
2. Wave radar from RADAC® called wave guide system

12
 Wave estimation based on numerical simulations

For various wave scenarios, the responses were generated using JONSWAP spectra
and the corresponding transfer functions.

We used 2 different sets of RAOs in wave estimation to consider

hydrodynamic uncertainties. RAO1 is based on panel method whereas
RAO2 is based on linear strip theory

13
RAO1 RAO2

14
Wave estimation based on numerical simulations
Results for unimodal waves

15
Wave estimation based on numerical simulations
Results for bimodal waves (wind part)

16
Wave estimation based on numerical simulations
Results for bimodal waves (swell part)

17
Wave estimation based on numerical simulations
Example Contour plots of estimated wave spectra

18
Wave estimation based on numerical simulaions
Example contour plots of estimated wave spectra

19
 Wave estimation based on full-scale measurements
Ship No. I: 9400 TEU container ship

Ship characterisitcs:
Parameter Dimension
Overall Length 349.0 m
Beam 42.8 m
Maximum Draught 15.0 m
DWT 113,000 ton
Capacity 9415 TEU

Operational conditions:

20
Wave estimation based on full-scale measurements
Example results for ship No. I

21
Wave estimation based on full-scale measurements
Example results for ship No. I

22
 Wave estimation based on full-scale measurements
Ship No. II: 6800 TEU container ship

Ship characterisitcs:

Operational conditions:

23
Wave estimation based on full-scale measurements
Example results for ship No. II

24
Wave estimation based on full-scale measurements
Example results for ship No. II

25
 Onboard prediction of waves
Trend modelling
• Using estimations, a local regression trend model is applied to track the evolution
of wave parameters during the voyage.
• Regression model:

is the wave parameter at time N+t

3rd order polynomial is used as
𝜽 is the vector of regression parameters: 𝜽 = (𝜃0 , 𝜃1 , 𝜃2 )
is is normally distributed random variable.

26
 Onboard prediction of waves
Trend modelling
• Weighted least square method is applied where the estimations in the
far past are given less weight than the recent estimations. This is
implemented by a forgetting factor 𝜆.

• Prediction of wave parameters in the next time step is then possible.

• The parameters are updated once new estimations are available.

27
Onboard prediction of wave parameters
Example results for ship No. I

28
Onboard prediction of wave parameters
Example results for ship No. I

29
 Onboard prediction of ship responses

Once, prediction of the wave parameters are available, any response can be calculated
using the JONSWAP model, the new operational condition and the corresponding
RAO. The variance of a response is calculated by:

𝑅𝑖 : 0𝑡ℎ order spectral moment (or the variance) of the 𝑖 𝑡ℎ response which represents
the energy amount of that response.
Statistics of responses are usually represented by this value. Therefore, this quantity is
evaluated for different responses and compared with the measurements.

30
Onboard prediction of ship responses
Examples for vertical acceleration

31
Onboard prediction of ship responses
Examples for vertical bending moment

32
 Automatic response selection
Sensitivity analysis
• Selection of the best combination of responses is very important. The optimum
selection of responses may not be identical for all ships and all operational
conditions. Therefore, this choice should be made for a particular ship in a typical
operational condition.
• As shown before, the basic cost function for wave estimation is

∞ 𝜋
2
𝑅𝑖 = 𝐻𝑖 𝜔, 𝜃 𝑆 𝑯𝒔, 𝑻𝒑, 𝝁, 𝜔, 𝜃 𝑑𝜃𝑑𝜔 𝑖 = 1, … , 𝑛
0 −𝜋

• Variability of different responses with respect to the main wave parameters can be
compared.

33
Effect of wave parameters on individual response variances

34
 Automatic response selection
Sensitivity analysis

• In order to decide which responses to use for wave estimation,

sensitivity analysis can be implemented.

• The effect of each response variance on a specific parameter estimate can be

calculated using the derivatives. A normalized sensitivity factor is defined as:

p: parameter (𝑯𝒔, 𝑻𝒑, 𝝁)

The derivatives are calculated using the JONSWAP model.

35
Automatic response selection
Example results

36
 Summary and conclusion

 In the proposed method, the sea state is estimated using measured ship
responses and a parametric wave spectrum.
 The optimization is set based on spectral moments of responses.
 The method is applied on numerical data and full scale measurements of in-
service container ships.
 Comparisons are made between the current method and the previous
methods in the literature.
 The results show that the method is efficient and promising.
 A local regression trend model is proposed for onboard prediction of sea state
parameters.
 The trend model provides a smooth and consistent evolution of wave
parameters.
 The predictions are then used to estimate the responses of the ship in 20
minutes.
 The prediction results show a good agreement with the actual measurements.
 The procedure can be applied onboard ships to provide the actual safe and
risky areas in terms of different ship speeds and courses.
 This information can be shown as polar plots.

37
 Future works

• Consider a wide set of wave scenarios numerically to exract the limiting wave
characteristics that can be estimated by the current method.
• Consider Larger ships e.g. 18000 TEU container ships as case study to see if the
method is still efficient for wave estimation.
• Improving initializations and constraints of the optimization by using e.g. hindcast
data and trend analysis.
• The proposed local sensitivity analysis in this thesis can be employed to select the
response combination in specific sea sates. The results should be compared with
the originals.
• More full-scale tests/experiments are required to validate the robustness of wave
buoy analogy in general.
• Further studies are recommended to do a comparative evaluation of the different
methods.
• Ways to combine the parametric moment-based method with other methods, e.g.
Bayesian method can be proposed to improve the efficiency of wave buoy analogy.
• Uncertainty evaluation of all methods should be developed.

38
Thank you!

39