3.3.1. Wave spectrum method

This method estimates the wave spectrum of the encountered short-term sea state from the measured ship responses and associated linear transfer functions. Once the wave spectrum is identified, the spectrum of the unmeasured responses at any location in the ship structure can be estimated (Figure 3(a)). Several studies on sea-state estimation from measured ship responses have already been proposed. Iseki and Ohtsu (Reference Iseki and Ohtsu2000) developed a Bayesian modeling procedure to estimate directional wave spectra based on ship motion data. Nielsen (Reference Nielsen2006) compared the Bayesian method and the so-called parametric method and discussed their applicability. In the former method, the directional wave spectrum is estimated at a number of discretized points of the wave field, while in the latter, the wave field is expressed by the summation of a number of parameterized wave spectra.

In these previous studies, ship motions such as heave, pitch, and roll were primarily used as the measured responses. However, for a large ship like a large container ship, the ship motions caused by high frequency waves are relatively small, while the hydro-elastic structural response becomes more significant. From this viewpoint, Chen et al. (Reference Chen, Okada and Kawamura2018) proposed a new parametric method using the longitudinal hull girder bending stresses as one of the measured inputs, and showed that the wave spectrum could be predicted more accurately by this method. Chen et al. (Reference Chen, Okada, Kawamura and Mitsuyuki2021) further extended this method so that the multi-peaked directional wave spectrum could be estimated. In this DTSS project, the wave spectrum was discretized into 36 directions, as shown in Figure 4, and the Ochi–Hubble’s six-parameter spectrum $ {S}^{(q)}\left(\omega \right) $ given in equation (1) was applied to each direction. $ \omega $ is the wave circular frequency, and $ \chi $ is the relative angle between the ship course and wave direction. The subscript $ j\;\left(=1,2\right) $ in $ {S}^{(q)}\left(\omega \right) $ represents the case when bimodal directional waves are considered, as will be explained in Section 3.3.4. The spectrum parameters, $ {T}_j^{(q)},{H}_j^{(q)},{\lambda}_j^{(q)} $ , were determined so that the squared difference between the measured and estimated response spectra $ {R}_{mn}\left(\omega \right) $ could be minimized by using GA:

(1) $$ {R}_{mn}\left(\omega \right)={\int}_{-\pi}^{\pi }{\left[{A}_m\left(\omega, \chi +\theta \right)\right]\left[{A}_n\left(\omega, \chi +\theta \right)\right]}^{\ast}\times S\left(\omega, \theta \right) d\theta, $$

Figure 4. Image of assumed directional wave spectrum.

where $ {A}_m\left(\omega, \chi +\theta \right) $ : RAO of the mth response variable,

$ S\left(\omega, \theta \right)={S}^{(q)}\left(\omega \right) $ : (q = 1, …, 36; Discrete variables for wave direction),

$$ {S}^{(q)}\left(\omega \right)=\frac{1}{4}\sum_{j=1}^2\frac{{\left(\frac{4{\lambda}_j^{(q)}+1}{4}{\left(\frac{2\pi }{T_j^{(q)}}\right)}^4\right)}^{\lambda_j^{(q)}}}{\Gamma \left({\lambda}_j^{(q)}\right)}\frac{{H_j^{(q)}}^2}{\omega^{4{\lambda}_j^{(q)}+1}}\exp \left\{-\left(\frac{4{\lambda}_j^{(q)}+1}{4}\right){\left(\frac{2\pi }{\omega {T}_j^{(q)}}\right)}^4\right\}. $$

3.3.2. Kalman filter (KF) method

The KF method is widely used for response prediction and control of dynamic systems, such as robots and vehicles, based on observation data. Miyake et al. (Reference Miyake, Iijima, Tatsumi and Fujikubo2023) applied the KF method to data assimilation in DTSS to predict the real-time response of a ship hull structure using the real-time measured strains (Figure 3(b)). The basic idea is to represent the deformation of a whole ship by a linear summation of the natural vibration modes. The equation of motion considering the elastic vibrations as well as rigid-body motions can be expressed as

(2) $$ {\displaystyle \begin{array}{c}\boldsymbol{M}\ddot{\boldsymbol{q}}+\boldsymbol{C}\dot{\boldsymbol{q}}+\boldsymbol{Kq}=\boldsymbol{F},\end{array}} $$

where $ \boldsymbol{q} $ , $ \dot{\boldsymbol{q}} $ , and $ \ddot{\boldsymbol{q}} $ are modal displacement, velocity, and acceleration vectors, respectively, and $ \boldsymbol{M},\boldsymbol{C},\boldsymbol{K} $ and $ \boldsymbol{F} $ are modal mass, damping, and stiffness matrices and the modal exciting force vector, respectively. Equation (2) can be discretized in time as

(3) $$ {\displaystyle \begin{array}{c}{\boldsymbol{x}}_{\boldsymbol{k}}=\boldsymbol{G}{\boldsymbol{x}}_{\boldsymbol{k}-\mathbf{1}}+\boldsymbol{v},\end{array}} $$

where $ {\boldsymbol{x}}_{\boldsymbol{k}}={\left[{\boldsymbol{q}}_{\boldsymbol{k}},{\dot{\boldsymbol{q}}}_{\boldsymbol{k}},{\boldsymbol{F}}_{\boldsymbol{k}}\right]}^T $ is a state vector and G is a system matrix. v is a noise vector based on the assumption of the Gaussian process of loads and responses. On the other hand, the observation equation is given in the form

(4) $$ {\displaystyle \begin{array}{c}{\boldsymbol{z}}_{\boldsymbol{k}}=\boldsymbol{H}{\boldsymbol{x}}_{\boldsymbol{k}}+\boldsymbol{w},\end{array}} $$

where H is a matrix that relates the state vector $ {\boldsymbol{x}}_{\boldsymbol{k}} $ with the observation data $ {\boldsymbol{z}}_{\boldsymbol{k}} $ , and w is a noise vector which follows a normal distribution. Using measured responses including strains as $ {\boldsymbol{z}}_{\boldsymbol{k}} $ and applying the standard procedure of the KF, the time history of modal displacement $ \boldsymbol{q} $ is obtained. Then, the time history of the structural responses at any location in the structure can be obtained by a linear summation of the modal vectors. Since the KF method gives a real-time response to the operator, it can be used, for example, in a real-time alert system in the extreme sea condition. Recently, Komoriyama et al. (Reference Komoriyama, Iijima, Tatsumi and Fujikubo2023) applied the KF method to prediction of real-time wave profiles using measured responses. This is expected to make it possible to predict not only structural responses but also any ship performance in waves, including cargo responses, added resistance, and so forth.

3.3.3. Inverse FEM (iFEM)

iFEM is an algorithm that reconstructs full-field structural displacements based on the strains at a limited number of finite elements as an inverse problem (Tessler and Spangler, Reference Tessler and Spangler2003). In FEM, the element strain $ {\boldsymbol{\varepsilon}}^{\boldsymbol{e}} $ is calculated from the nodal displacement u ^e , whereas in iFEM, the nodal displacement u ^e is inversely estimated from the element strain $ {\boldsymbol{\varepsilon}}^{\boldsymbol{e}} $ so as to satisfy the nodal displacement-element strain relationship:

(5) $$ {\displaystyle \begin{array}{c}\boldsymbol{B}{\boldsymbol{u}}^{\boldsymbol{e}}={\boldsymbol{\varepsilon}}^{\boldsymbol{e}}\end{array}} $$

by solving the following minimization problem:

(6) $$ {\displaystyle \begin{array}{c}\operatorname{minimize}\hskip0.5em \Phi \left({\boldsymbol{u}}^{\boldsymbol{e}}\right)=\frac{1}{2}{\int}_{\Omega^e}{\left({\boldsymbol{Bu}}^{\boldsymbol{e}}-{\boldsymbol{\varepsilon}}^{\boldsymbol{e}}\right)}^T\left({\boldsymbol{Bu}}^{\boldsymbol{e}}-{\boldsymbol{\varepsilon}}^{\boldsymbol{e}}\right)d{\Omega}^e,\end{array}} $$

where B is the nodal displacement-element strain matrix of the element. From the condition that

(7) $$ {\displaystyle \begin{array}{c}\frac{\mathrm{\partial \Phi}\left({\boldsymbol{u}}^{\boldsymbol{e}}\right)}{{\partial \boldsymbol{u}}^{\boldsymbol{e}}}={\int}_{\Omega^e}\left({\boldsymbol{B}}^T\boldsymbol{B}{\boldsymbol{u}}^{\boldsymbol{e}}-{\boldsymbol{B}}^T{\boldsymbol{\varepsilon}}^{\boldsymbol{e}}\right)d{\Omega}^e=0,\end{array}} $$

we have

(8) $$ {\displaystyle \begin{array}{c}{\boldsymbol{K}}^{\boldsymbol{e}}{\boldsymbol{u}}^{\boldsymbol{e}}={\boldsymbol{F}}^{\boldsymbol{e}}\end{array}} $$

(9) $$ {\displaystyle \begin{array}{c}{\boldsymbol{K}}^{\boldsymbol{e}}={\int}_{\hskip-4pt {\Omega}^e}{\boldsymbol{B}}^T\boldsymbol{B}d{\Omega}^e,\hskip2pt {\boldsymbol{F}}^{\boldsymbol{e}}={\int}_{\hskip-4pt {\Omega}^e}{\boldsymbol{B}}^T{\boldsymbol{\varepsilon}}^{\boldsymbol{e}}d{\Omega}^e.\end{array}} $$

Using the measured strains as $ {\boldsymbol{\varepsilon}}^{\boldsymbol{e}} $ for the elements fitted with strain sensors, the element equations of equation (8) are assembled into a global system of equations which can be readily solved to obtain real-time deformed structural shapes, including the strains at unmeasured locations (Figure 3(c)). The more detailed procedure of iFEM is given in Kefal et al. (Reference Kefal, Mayang, Oterkus and Yidis2018) and Mikami et al. (Reference Mikami, Jumonji, Kobayashi, Toh, Komoriyama, Ma and Murayama2020).

Among these three data assimilation methods, the wave spectrum method is useful for short-term statistical assessments of structural responses and the KF method and iFEM are suitable for real-time structural response assessments. The three methods can be used either independently or in conjunction. For instance, the output time history from the KF method or iFEM may be used for the input to the wave spectrum method, and iFEM may be utilized in combination with other methods for the analysis of structural details. These remain as future extensions.

3.3.4. Validation by wave tank test

A wave tank test was performed for validation of the data assimilation methods. Table 1 shows the wave conditions and ship speed used in the wave tank test (Houtani et al., Reference Houtani, Mikami, Komoriyama, Chen, Miyake, Tatsumi, Kuroda, Okada, Iijima, Murayama and Oka2022). Bimodal short-crested irregular waves as a combination of swell and wind waves were considered. The values in the parenthesis in Table 1 are the full-scale values corresponding to the prototype scale. Figure 5 shows the locations of the strain sensors. The strains at the sensors marked by stars were estimated from those at the sensors marked by circles. Four deck strains were used as the measured strains for the wave spectrum method, and six deck strains were used for the KF method. For iFEM, 56 strains were used as the measured strains, including those at the side and bottom parts.

Table 1. Wave conditions for wave tank test (1/72 scale)

Figure 5. Locations of strain sensors on model ship used for validation of data assimilation methods.

Figure 6 shows a comparison of the time history of the strains obtained by the KF method, iFEM, and measurement. To quantify the estimation error, the following two metrics, RMSE (root-mean-square error) and ME (maximum-value error), are considered:

(10) $$ {\displaystyle \begin{array}{c}\mathrm{RMSE}=\frac{\sqrt{\frac{1}{N}{\sum}_{i=1}^N{\left({\varepsilon}_i^{est}-{\varepsilon}_i^{meas}\right)}^2}}{2\sqrt{\frac{1}{N}{\sum}_{i=1}^N{\varepsilon_i^{meas}}^2}}\times 100\;\left[\%\right],\end{array}} $$

(11) $$ {\displaystyle \begin{array}{c}\mathrm{M}\mathrm{E}\hskip2pt =\hskip2pt \frac{\max (|{\varepsilon}^{est}|)-\max (|{\varepsilon}^{meas}|)}{\max (|{\varepsilon}^{meas}|)}\times 100\hskip2pt [\%].\end{array}} $$

Figure 6. Comparison of time histories of longitudinal strain by KF method, iFEM, and measurement.

The denominator of RMSE corresponds to half of the significant value of the total strain amplitude. As given in Table 2, both iFEM and the KF method can estimate the strain time histories within an error of about 20% RMSE and 10% ME. iFEM also gives estimates with similar accuracy, but generally requires a larger number of measured strains than the KF method, because the KF method estimates the full-field deformation as a summation of the natural vibration modes, whereas iFEM gives direct estimates from the strains at the measured locations based on the displacement–strain relationship.

Table 2. Comparison of root-mean-square error (RMSE) and maximum-value error (ME) of strain time histories between iFEM and KF method in irregular wave test

Figure 7 shows a comparison of the wave spectrum obtained by the direct wave measurement (Figure 7(a)) and spectra estimated by the wave spectrum method (Figure 7(b,c)). The radial coordinate of each figure represents the circular frequency of the wave components approaching from each direction. Two cases were considered for the wave spectrum method. In the method A, the heave, pitch, and roll motion data were used as the observed response, whereas in the method B, the four deck strains were used. Better agreement with the direct wave measurement was observed with method B than with method A, including the energy peak period and wave directions, because higher frequency wave components are filtered in method A but can be captured in method B through the hull structural response data. Table 3 is a comparison of the difference, defined by equation (12), in the standard deviation of the strains at some locations by the wave spectrum method and measurement.

(12) $$ {\displaystyle \begin{array}{c}\mathrm{Difference}=\frac{\mathrm{Estimation}-\mathrm{Measurement}}{\mathrm{Measurement}}\times 100\;\left[\%\right].\end{array}} $$

Figure 7. Comparison of wave spectrum by direct measurement and data assimilation using wave spectrum method.

Table 3. Comparison of difference in standard deviation of strain by wave spectrum method in irregular wave test

As shown in Figure 7 and Table 3, estimation accuracy is significantly improved by considering strains as the observed response.

3.3.5. Validation using existing hull monitoring data of actual ships

Validation of the data assimilation methods was performed for actual ships using existing hull monitoring data. Regarding the wave spectrum method, the hull monitoring data for ten 14,000 TEU container ships were used as reference data, and good accuracy of the stress estimates was confirmed (Chen et al., Reference Chen, Okada, Kawamura and Mitsuyuki2021).

For the validation of the KF method, the hull monitoring data for an 8,600 TEU container ship were used. Figure 8(a) shows the typical vertical bending (VB), horizontal bending (HB), and torsional (T) natural vibration modes. Four vertical bending, four horizontal bending, and two torsional vibration modes were considered for data assimilation. The results are shown in Figure 8(c). The vertical hull girder response could be reproduced satisfactorily by using six deck strains, and reproduction of the horizontal and torsional hull girder responses was also possible by adding a further six bottom strains. Improvement is necessary to obtain more accurate estimates of the local deformation of transverse cross sections, in particular, for the bottom parts, as found at location No. 8 in Figure 8(c). It should be noted that all the data assimilation in this project is based on the response function assuming linear responses. To estimate strongly nonlinear responses, such as a whipping response, improvement is necessary.

Figure 8. Result of data assimilation by KF-method for 8,600 TEU container ship.

3.5.1. Prediction of maximum-value distribution of wave bending moment in encountered sea state

The maximum response of ships in a short-term sea state is usually estimated by the following steps:

1. i) Estimate the significant wave height $ {H}_S $ and the mean wave period $ {T}_m $ in the encountered short-term sea state based on wave forecasting and the navigation route.

2. ii) Determine the wave spectrum for the given $ {H}_S $ and $ {T}_m $ using simple formulae as given in design rules.

3. iii) Obtain the response spectrum from the wave spectrum and the RAO of the target response by applying linear spectrum theory.

4. iv) Calculate the variance R ² of the response from the response spectrum.

5. v) Obtain the maximum value distribution using the variance R ².

The green dashed line in Figure 10 shows the time history of the standard deviation R of the vertical wave bending moment (VBM) amidship of the 8,600 TEU container ship, which was estimated by steps i) to v) using the wave spectrum based on the wave nowcast database provided by JWA (Japan Weather Association), and the red line shows the time history of the standard deviation R of VBM observed by hull monitoring. It was found that the R based on the nowcast data over-predicts the observed values. The possible reasons for this error include uncertainties in the wave data, assumed spectrum shape, relative heading angle model, RAO, nonlinear effects, and so forth.

Figure 10. Example of comparison of variation of VBM estimated by linear theory $ {r}_t $ and measured data $ {R}_t $ .

Tatsumi proposed a method for reducing the error in the predicted R ² time-series using the observation data by applying a time-series stochastic model (Osawa et al., Reference Osawa, Tatsumi and Takeuchi2022). The following log series function was used as an observation equation:

(13) $$ {\displaystyle \begin{array}{c}\log \left({R}_t^2\right)={a}_t\log \left({r}_t^2\right)+{b}_t+{w}_t,\end{array}} $$

where $ {r}_t^2 $ is the response R ² obtained from the wave spectrum, $ {R}_t^2 $ is the R ² obtained from the observation data, and $ {w}_t $ is observation noise, which is assumed to follow a normal distribution. $ {a}_t $ and $ {b}_t $ are the time-varying regression coefficients, and were developed by the random work model. The variance of the observation noise is also included in the state variables. Because the state-space model is nonlinear, a particle filter was applied to determine the state variables. The blue line in Figure 10 shows the median of the probabilistic distribution of the corrected R time-series predicted based on the observation data up to a time instance of 93 h. The upper bound of the blue area corresponds to the 75 percentile line, and almost coincides with the observed R. It was confirmed that the accuracy of the prediction of the R of VBM is significantly improved by using the observation data, leading to improved accuracy of structural reliability assessments for severe sea states.

3.5.2. Prediction of long-term cumulative fatigue damage

The cumulative fatigue damage of critical details of ship structures subjected to long-term wave fluctuation is normally calculated from the joint probability distribution of $ {H}_S $ and $ {T}_m $ based on the wave scatter data and the assumed probability distribution of the relative heading angle $ \chi $ (Figure 11). The method of tracking a ship’s automatic identification system (AIS) data and relating that data with MetOcean data is also widely applied to construct the joint probability distribution of $ {H}_S $ , $ {T}_m $ , and $ \chi $ (e.g., Oka et al., Reference Oka, Takami, Ma, Okada, Suzuki and Kawamura2021). Linear spectrum theory is generally assumed for calculation of the response spectrum, but there may be significant error in the cumulative fatigue damage between the prediction by this conventional method and the results of observation by hull monitoring. As in the case of the maximum value prediction in Section 3.5.1, this is primarily because of the uncertainties in the assumed wave parameters, spectrum shape, relative heading angle, RAO, nonlinear effects, and so forth. Figure 12 shows a comparison of the measured stress variance R ² and median of fatigue damage per hour and the values estimated by the conventional method. The conventional linear spectrum theory provides conservative predictions of R ² and fatigue damage, but both are significantly over-predicted compared to the observed values.

Figure 11. Long-term fatigue damage assessment based on linear spectrum theory.

Figure 12. Comparison of measured wave stress variance R ² and median of fatigue damage per hour and those estimated by linear spectrum theory for No. 6 sensor in 8,600 TEU container ship.

In order to improve the prediction of long-term cumulative fatigue damage by employing observed stress data, Osawa et al. (Reference Osawa, Tatsumi and Takeuchi2022) proposed an “equivalent wave probability (EWP)” concept. Although the formulae, probabilistic models and structural characteristics (e.g., RAO) of the conventional fatigue analysis method are unchanged in this concept, the joint probability distribution of the wave parameters $ {H}_S $ and $ {T}_m $ is approximated by a simple mathematical function, and its population parameters are determined so that the estimated probabilistic distribution of stress R ² (not a wave R ²) agrees with the observed distribution. The additional concrete steps are as follows:

1. i) The EWP equation is represented by six parameters $ \left\{m,\eta, {a}_0,{a}_1,\hskip0.4em ,{b}_0,\hskip0.4em ,{b}_1\right\} $ as

$$ p({H}_S,{T}_m)=p({H}_S)p({T}_m|{H}_S), $$

(14) $$ p\left({H}_S\right)=\mathrm{Weibull}\hskip0.4em \left(\mathrm{shape}\ \mathrm{parameter},=,m,,,\hskip0.4em ,\mathrm{scale}\ \mathrm{parameter},=,\eta \right) $$

$$ p({T}_m|{H}_S)=\mathrm{W}\mathrm{e}\mathrm{i}\mathrm{b}\mathrm{u}\mathrm{l}\mathrm{l}\hskip0.4em (\mathrm{s}\mathrm{h}\mathrm{a}\mathrm{p}\mathrm{e}\ \mathrm{p}\mathrm{a}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{r}=\alpha, \hskip2pt \mathrm{s}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{e}\ \mathrm{p}\mathrm{a}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{r}=\beta ) $$

$$ \alpha \left({H}_S\right)={a}_0+{a}_1{H}_S,\hskip0.96em \ln \left\{\beta \left({H}_S\right)\right\}={b}_1+{b}_2\ln \left({H}_S\right). $$

1. ii) The probability distribution of stress R ² is represented by the generalized gamma distribution with three population parameters $ \left\{\mu, \sigma, Q\right\} $ .

2. iii) The correlation between the wave state parameters $ \left\{m,\eta, {a}_0,{a}_1,{b}_0,{b}_1\right\} $ and the stress R ² parameters $ \left\{\mu, \sigma, Q\right\} $ is established by performing stress response analyses for all sea areas of global wave statistics (GWS).

3. iv) Using the statistical model established in step iii) and applying MCMC Bayesian inference, $ \left\{m,\eta, {a}_0,{a}_1,{b}_0,{b}_1\right\} $ and $ \left\{\mu, \sigma, Q\right\} $ are determined so that the likelihood of the observed stress R ² is maximized, as shown in Figure 13.

Figure 13. Schematic flow of EWP determination.

The EWP can be determined at diverse locations where stress is measured. In the event that the stress R ² at a position distinct from the reference position, estimated from the EWP at the reference position, agrees with that evaluated from the EWP at that measurement point, the probabilistic distribution of the stress R ² and the fatigue damage can be estimated at any position throughout the entirety of the vessel from the EWP at the reference position. This requirement is achieved by assimilating the statistical model determined in step iii) before the EWP determination process.

Figure 14 shows a comparison of the measured and estimated fatigue damages calculated for the 8,600 TEU container ship in Figure 8(b) (Takeuchi et al., Reference Takeuchi, Osawa, Tatsumi, Inoue, Hirakawa, Seki, Yoshida, Miratsu and Ikeda2023). The statistical model was assimilated in two cases of a voyage A of Far East–Suez–Europe–North America–Far East and a voyage B of Far East–Suez–Europe, which were applied to the estimation of voyage A, respectively. The EWP was determined from the observed stress R ² at one reference position (No. 6 position in Figure 8(b)), and was then applied to all other positions. As shown, the estimated fatigue damages obtained by the EWP are in good agreement with the measured results for all positions. The difference of the fatigue damage defined by equation (12) is −8.51 – 25.5% at the deck (sensor no. 1, 2, 5, 6, 9, 10), and −14.0 – 58.9% at the bottom (sensor no. 3, 4, 7, 8, 11, 12). The difference is relatively larger for the bottom than for the deck, but the damage is smaller and the estimate generally falls on the safe side.

Figure 14. Comparison of median of fatigue damage per hour for 8,600 TEU container ship by measurement and estimation by EWP identified by measured stress R ² at No. 6 position.

The EWP parameters are determined from the hull monitoring data for a mid-term period (one season to one year). The time-series of the EWP and operation parameters are made available by repeating the EWP parameter determination and recording the operating condition in service, and the state-space model for the EWP and the operational parameters can then be developed based on these data. This model enables forecasting of whole-ship fatigue damage considering the future operating condition. More details about the EWP approach can be found in Takeuchi et al. (Reference Takeuchi, Osawa, Tatsumi, Inoue, Hirakawa, Seki, Yoshida, Miratsu and Ikeda2023).

4.1.1. Navigation support

When the KF method or iFEM is employed, DTSS can output the real-time global and local structural responses of any position of the ship hull in encountered waves. When the wave spectrum method is employed, DTSS can provide short-term statistics for the responses, for example, for a period of a few hours. Since this method essentially estimates the short-term wave state, any wave-induced response may be estimated, including ship motion, cargo behavior and even the wave effect on propulsion resistance. The KF method has this possibility in the time domain. These short-term or real-time capabilities of DTSS, together with the prediction methods described in Section 3.5.1 and visualization techniques, can be used for ship navigation support, such as issuing alerts and suggesting alternative routes to operators when the responses reach unacceptable levels, or selecting routes with better punctuality or fuel efficiency within the allowable safety level.

4.1.2. Maintenance support

The stress history and its variance obtained by DTSS enable calculation of the long-term cumulative fatigue damage experienced in actually encountered waves. By applying the EWP concept described in Section 3.5.2, multi-point long-term fatigue assessments can be made based on hull monitoring data captured at a small number of reference sensors. This kind of DTSS output can be utilized for rational and efficient inspection and maintenance schemes enhanced to focus on positions with a truly high risk of damage. This means that hull structural maintenance can be shifted from periodic maintenance to risk-based maintenance. Moreover, this may be possible not only for individual ships but also at the fleet level, so that maintenance is performed when the required fleet capacity is low and is not performed when it is high. Although relevant rule changes will be needed in order to implement risk-based maintenance of ships, the introduction of DTSS definitely expands the possibilities.

4.1.3. Rule improvement

From information obtained through DTSS, such as how a ship was operated, what loads acted on the ship hull, what stresses were induced in each member, and so forth, it is possible to grasp what was actually happening in the ship structure in service. Sharing this information within the maritime cluster of shipyards, shipping companies, ship owners, and classification societies will enable clear identification and quantification of various uncertain factors underlying the present ship structural rules. This is expected to result in a shift from the conventional rules, which comprehensively and prescriptively cover the uncertain factors of various ships, to a more data-driven risk-based rule framework, enabling more cost-effective structural design while also ensuring greater safety. This can be a trigger to breaking away from the commoditization of conventional ship structural design.

4.1.4. Product value improvement