ESTIMATION OF A TANK SLOOP RESPONSE USING A MACHINE LEARNING TRAINED STATISTICAL MODEL

Description

Technical field

[0001] The invention relates to the estimation of a sloshing response of a sealed and thermally insulating tank for the transport of liquefied gas. More particularly, the invention relates to methods for obtaining a statistical model capable of estimating a sloshing response of such a tank, to methods for obtaining a database usable for estimating a sloshing response of such a tank, and to management systems for a ship comprising at least one such tank. In this document, the term "ship" designates a means of transporting and / or using, for its own propulsion, liquefied gas between two points on the globe or a floating unit for processing and / or storing one or more liquefied gases. Conventionally, for ships using liquefied gas as fuel, mention may be made of LNG-powered ships, container ships, cruise ships or bulk carriers. Technological background

[0002] Sealed and thermally insulated tanks are commonly used for the storage and / or transport of low-temperature liquefied gas, such as tanks for the transport of Liquefied Petroleum Gas (also known as LPG) with a temperature between -50°C and 0°C, for example, or for the transport of Liquefied Natural Gas (LNG) at approximately -162°C at atmospheric pressure. These tanks can be intended for the transport of liquefied gas and / or to receive liquefied gas used as fuel for the propulsion of the floating structure. Many liquefied gases can also be considered, including methane, ethane, propane, butane, ammonia, dihydrogen or ethylene.

[0003] Ship tanks can be single- or double-sealed membrane tanks that allow transport at atmospheric pressure. The membranes are usually made of thin stainless steel sheet or Invar. One membrane is usually in direct contact with the liquefied gas.

[0004] During transport, the liquid contained in a tank is subject to various movements. In particular, the movements of a ship at sea, for example due to climatic conditions such as sea state or wind, cause agitation of the liquid in the tank. The agitation of the liquid, generally referred to as "sloshing" or tossing, creates stresses on the walls of the tank that can harm the integrity of the tank.

[0005] These sloshing phenomena appear on carrier and / or user vessels (often referred to as "LNG as Fuel" vessels) of natural gas, hereinafter LNG, or methane carriers as well as on anchored storage vessels called FPSO (Floating Production Storage & Offloading) such as for example a natural gas extraction platform and liquefaction plant, commonly called FLNG (Floating Liquefied Natural Gas) or a floating storage and regasification unit, commonly called FSRU (Floating Storage and Regaseification Unit), i.e. more generally a floating production, storage and export support.Sloshing phenomena occur in both rough and almost calm sea conditions, as soon as the liquefied gas cargo resonates with the excitation created by a swell, even a small one, to which the ship is subjected. In these cases of resonance, the sloshing can become extremely violent, particularly when breaking on vertical walls or in corners, thus risking damaging the liquefied gas containment system, or the insulation system present just behind said containment system.

[0006] However, the integrity of the tank is particularly important in the case of a liquefied gas tank, such as LNG, due to the flammable or explosive nature of the liquid transported and the risk of a cold spot on the steel hull of the floating unit in the event of a leak.

[0007] US Patent 8,643,509 discloses a method for reducing the risk associated with sloshing of a liquefied gas cargo. In this document, a resonant frequency of the liquid in a tank is estimated based on the tank and its fill level. During transport, a ship's motion frequency is evaluated based on climatic and maritime conditions and the ship's speed. Predicted motion frequencies are also evaluated on the ship's route to be followed. When one of the motion frequencies gets too close to the resonant frequency of the liquid in a tank, an alarm is signaled to change course and / or change the ship's speed to avoid finding itself in a risky situation.

[0008] Despite the implementation of sloshing reduction measures, the sloshing of liquids in the tanks, in particular due to resonance phenomena, can lead to occasional risks of deformation of the primary waterproofing membrane, damage to the underlying structures present in the primary and / or secondary spaces on which the primary waterproofing membrane rests, falling objects, in particular from static equipment, likely to damage the primary waterproofing membrane in the short or medium term, or more generally deformations of the primary waterproofing membrane beyond its structural tolerances.

[0009] The publication by Yangjun Ahn, Yonghwan Kim, Sang-Yeob Kim (2019). Database of model-scale sloshing experiment for LNG tank and application of artificial neural network for sloshing load prediction, Marine Structures, Vol. 66, pp. 66-82, discloses a method for estimating a sloshing response of tanks during ship navigation. This paper describes the use of test data for training an artificial neural network. The publication does not describe obtaining a database that can be used to estimate a sloshing response of a ship's tank, as the database is obtained from a statistical model.

[0010] There is therefore still a need for methods to estimate the sloshing response of the tanks during the navigation of the vessel and, if necessary, to take the necessary measures to avoid the occurrence of excessive sloshing which could damage the primary sealing membrane of the tank. Summary

[0011] An idea underlying the invention is to train, by a supervised machine learning method, a statistical model capable of estimating a sloshing response of the tank as a function of a filling level of the tank and a current sea state and optionally at least one of a draft of the ship, a speed of the ship and a heading of the ship. The statistical model is trained on a set of test data obtained from results of a plurality of tests each consisting of subjecting a test tank having a given filling level to movements and measuring a pressure on at least one point of a wall of the test tank and / or a number of impacts on at least one wall of the test tank.The statistical model can then be used to estimate a tank sloshing response as part of a management system for a vessel, for example by building a database that can be used for real-time consultation.

[0012] The invention is defined by the appended claims.

[0013] According to a first variant, the invention involves a method for obtaining a statistical model capable of estimating a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas, the method comprising a step consisting of: training a statistical model by a supervised machine learning method on a test data set, the statistical model being capable of estimating a sloshing response of the tank as a function of a filling level of the tank and a current sea state and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel, and the test data set being obtained from results of a plurality of tests each consisting of subjecting a test tank having a given filling level to movements and measuring a pressure on at least one point of a wall of the test tank and / or a number of impacts on at least one wall of the test tank.

[0014] According to a second variant, the invention also involves a method for obtaining a statistical model capable of estimating a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas, the method comprising a step consisting of: training a statistical model by a supervised machine learning method on a test data set, the statistical model being capable of estimating a sloshing response of the tank as a function of a filling level of the tank, a current state of movement of the vessel and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel, and the test data set being obtained from results of a plurality of tests each consisting of subjecting a test tank having a given filling level to movements and measuring a pressure on at least one point of a wall of the test tank and / or a number of impacts on at least one wall of the test tank.

[0015] A "supervised learning method" is a machine learning method (also referred to in French as artificial learning or statistical learning) that consists of learning a prediction function from annotated examples. In other words, a supervised learning method allows a model capable of forecasting to be built from a plurality of examples for which the response to be predicted is known. A supervised learning method is typically implemented by computer; the step of training the statistical model is thus typically implemented by computer.

[0016] The statistical model to be trained within the framework of the present invention is capable of estimating a sloshing response, which comprises one or more quantitative variables, at least as a function of a filling level of the tank and a current sea state or a current state of movement of the ship. The statistical model to be trained is therefore capable of responding to a regression problem.

[0017] By means of the step of training the statistical model using a supervised machine learning method on the test data set, the statistical model is capable of calculatingly estimating a sloshing response of a tank at least as a function of a tank filling level and a current sea state or a current ship movement state, including for values ​​of the tank filling level and for current sea states or ship movement states for which no test has been carried out. The statistical model can therefore be used to estimate a sloshing response of a tank under real conditions of use on a ship.

[0018] The test tank may be of reduced dimensions compared to the tank whose sloshing response is to be estimated by the statistical model. It may have a geometry representative of the tank whose sloshing response is to be estimated by the statistical model. It is also understood that in the phrase or characteristic "train a statistical model by a supervised method on a set of test data", the "set of test data" may also include or consist of data from so-called "real" campaigns, i.e. obtained or recorded on ships operating as carriers and / or users of liquefied gas.

[0019] Depending on the embodiments, the methods described above may include one or more of the following characteristics.

[0020] Sloshing response refers to any qualitative and / or quantitative parameter and set of parameters likely to represent mechanical stresses to which the tank is exposed during a cargo sloshing phenomenon.

[0021] According to one embodiment, the sloshing response includes at least one of a rate of fluid impacts on the walls of the tank, a maximum pressure on the walls of the tank, and a probability of damage to the tank.

[0022] Thus, the statistical model is able to estimate a probability of damage to the tank and / or a parameter making it possible to estimate such a probability of damage.

[0023] According to one execution mode, the supervised machine learning method is a Gaussian process regression method.

[0024] A Gaussian process regression method is well suited to training the statistical model because it allows producing a statistical model capable of addressing a regression problem, for any input data set, by training from a relatively limited number of data. However, it is possible to retain other supervised machine learning methods without departing from the scope of the present invention.

[0025] According to one execution mode, at least one constraint is imposed on the statistical model during its training by the supervised machine learning method.

[0026] Thus, the training of the statistical model can be guided on the basis of elementary physical considerations, such as the absence of sloshing in the case where the tank filling level is zero, and / or on the basis of considerations obtained from practical experience, such as the fact that larger movements or larger tank dimensions can potentially lead to a larger sloshing response. As a result, the accuracy of the estimation of the sloshing response by the statistical model is increased.

[0027] According to one embodiment, the method further comprises a step of excluding from the test data set test results having a sloshing response below a threshold before the step of training the statistical model.

[0028] Thus, the statistical model is trained only from test data that have revealed significant sloshing, more particularly in terms of the number of impacts. Indeed, according to one aspect of the invention, the number of occurrences encountered is a more important factor, for statistical convergence, than the intensity of the impacts. As a result, the accuracy of the estimation of the sloshing response by the statistical model is further increased.

[0029] According to one embodiment, the statistical model considers a plurality of tanks, the statistical model being able to estimate a sloshing response of each tank as a function of its position within the vessel.

[0030] According to an embodiment in accordance with the first variant, the invention also provides a method for obtaining a database which can be used to estimate a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas, the method comprising the steps consisting of: generating a plurality of input data vectors each comprising a tank fill level and a current sea state; and for each input data vector thus generated: obtaining an estimated sloshing response of the tank using the statistical model obtained by the method described above in accordance with the first variant; and storing in a database the estimated sloshing response of the tank in association with the input data vector.

[0031] According to an embodiment in accordance with the second variant, the invention also provides a method for obtaining a database which can be used to estimate a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas, the method comprising the steps consisting of: generating a plurality of input data vectors each comprising a tank fill level and a current state of movement of the vessel; and for each input data vector thus generated: obtaining an estimated sloshing response of the tank using the statistical model obtained by the method described above in accordance with the second variant; and storing in a database the estimated sloshing response of the tank in association with the input data vector.

[0032] Although the statistical model is capable of calculating a sloshing response of a tank for tank fill level values ​​and for common sea or ship motion states for which no tests have been carried out, the calculation required to do so may be too long and / or require too many computing resources to be implemented on board a ship, for which it is important to obtain an estimate of the sloshing response as quickly as possible and with the least expensive on-board system possible.An idea behind these methods is therefore to perform the majority of these calculations in advance based on a plurality of input data vectors, which can be suitably chosen to cover an entire operational operating domain of the vessel, and to store in a database the estimated sloshing response of the tank for each of these input data vectors in association with the input data vector. The estimated sloshing response of the tank can then be obtained either by simply reading the database when the input data vector is present in the database, or by interpolation from the database otherwise. This requires much less computational time and computational resources than estimation from the statistical model itself.As a result, the statistical model itself is no longer even necessary to perform the estimation of the statistical model on board the ship, the database alone being sufficient. The estimation from the database can then be carried out by a system on board the ship, or even by a ground station that communicates with the ship, for example by radio or satellite.

[0033] According to an embodiment in accordance with the first variant, the invention also provides a method for estimating a sloshing response of a sealed and thermally insulating tank for the transport of liquefied gas on board a ship, the method comprising the steps of: determining a current filling level of the tank; determining a current sea state; generating an input data vector comprising the current filling level of the tank and the current sea state thus determined; and estimating a sloshing response of the tank from the input data vector thus generated and the database obtained by the method according to the first variant.

[0034] According to an embodiment in accordance with the second variant, the invention also provides a method for estimating a sloshing response of a sealed and thermally insulating tank for the transport of liquefied gas on board a ship, the method comprising the steps of: determining a current filling level of the tank; determining a current state of movement of the vessel; generating an input data vector comprising the current filling level of the tank, and the current state of movement of the vessel thus determined; and estimating a sloshing response of the tank from the input data vector thus generated and from the database obtained by the method according to the second variant.

[0035] Using these methods, it is possible to estimate a tank sloshing response using the statistical model previously trained on test data, through the database. As mentioned above, this estimation requires much less computational time and resources than the estimation from the statistical model itself, and can be performed by a system on board the ship or by a ground station that communicates with the ship.

[0036] According to one embodiment, a plurality of tanks are considered and the method comprises a prior step of defining the position of each of the tanks within the ship.

[0037] According to one embodiment, the method further comprises a step of providing an alarm to a user if the estimated sloshing response of the tank exceeds an alert threshold, and preferably a decision-making assistance step intended to reduce the sloshing. This decision-making assistance step may consist of a proposal for a change in direction or trajectory of the vessel, a change of course particularly suitable for stationary floating structures, a modification of the speed of the vessel or a change in the filling rate of the tank(s) (between the tanks or between a tank and a reservoir external to the vessel in the case of a stationary floating structure).

[0038] Furthermore, the alarm may consist of signalling a problem to be corrected immediately or in the short term, if possible an alarm designating the tank(s) among a plurality of tanks of a ship which require a control and maintenance operation with a view to possible repair.

[0039] Thus, a user such as a crew member can take any necessary action to limit sloshing in the tank if necessary, such as slowing down or stopping the vessel or changing the vessel's course, and therefore reduce the risk of damage to the tank.

[0040] According to an embodiment in accordance with the first variant, the invention also provides a management system for a ship comprising at least one sealed and thermally insulating tank for transporting liquefied gas, the system comprising: at least one fill level sensor for measuring a current fill level of the tank; a sea state evaluation device capable of evaluating a current sea state; and processing means configured to generate an input data vector comprising a current fill level of the tank and a current sea state evaluated by the sea state evaluation device, and to estimate a sloshing response of the tank from the input data vector thus generated and from the database obtained by the method according to the first variant.

[0041] According to an embodiment in accordance with the second variant, the invention also provides a management system for a ship comprising at least one sealed and thermally insulating tank for transporting liquefied gas, the system comprising: at least one filling level sensor for measuring a current filling level of the tank; a device for evaluating the current state of movement of the ship capable of evaluating a current state of movement of the ship; and a processing means configured to generate an input data vector comprising a current filling level of the tank and a current state of movement of the ship, and to estimate a sloshing response of the tank from the input data vector thus generated and from the database obtained by the method according to the second variant.

[0042] Such systems provide the same advantages as the methods described above.

[0043] According to one embodiment, the processing means is further configured to provide an alarm to a user if the estimated sloshing response of the tank exceeds an alert threshold and is preferably configured to provide the user with decision support intended to reduce the sloshing.

[0044] According to another embodiment, the invention also provides a method for estimating a sloshing response of a sealed and thermally insulating tank for the transport of liquefied gas on board a ship, the method comprising the steps of: determining a current tank filling level; estimating future sea states from meteorological information and a ship's route; generating a plurality of input data vectors each comprising a current tank filling level and an estimated future sea state; and estimating a future sloshing response of the tank from the input data vectors thus generated and the database obtained by the method according to the first variant.

[0045] With this method, it is possible to estimate an upcoming sloshing response of the tank from meteorological information and a ship's route, thanks to the statistical model previously trained on the basis of test data, through the database. As mentioned above, this estimation requires much less computational time and computing resources than the estimation from the statistical model itself, and can be performed by a system on board the ship or by a ground station that communicates with the ship.

[0046] According to yet another embodiment, the invention also provides a method for estimating a sloshing response of a sealed and thermally insulating tank for the transport of liquefied gas on board a ship, the method comprising the steps of: determining a current tank fill level; estimating future ship motion states from meteorological information and a ship course; generating a plurality of input data vectors each comprising a current tank fill level and an estimated future ship motion state; and estimating a future sloshing response of the tank from the input data vectors thus generated and the database obtained by the method according to the second variant.

[0047] According to one embodiment, the method further comprises a step consisting of determining a route of the ship and / or a modification of the filling level of the tank making it possible to reduce the future sloshing response of the tank. The expression "route of the ship" means a heading of this ship, the speed of the latter or a simple avoidance of a geographical area. For stationary floating structures (ships, barges), that is to say having a fixed position, the change of heading results in a change of angle between the North direction and the longitudinal axis of the structure so as to orient or bring the floating structure to conventionally reduce the negative consequences of swell and waves on the floating structure.

[0048] Thus, a user such as a crew member can make the decision to have the vessel follow a route that reduces the upcoming sloshing response of the tank, and therefore reduces the risk of damage to the tank.

[0049] According to another embodiment, the invention also provides a management system for a ship comprising at least one sealed and thermally insulating tank for transporting liquefied gas, the system comprising: at least one fill level sensor for measuring a current fill level of the tank; a sea state estimation device capable of estimating future sea states from meteorological information and a route of the ship; and a processing means configured to generate a plurality of input data vectors each comprising a current fill level of the tank and a future sea state estimated by the sea state estimation device, and to estimate a future sloshing response of the tank from the input data vectors thus generated and the database obtained by the method according to the first variant.

[0050] According to yet another embodiment, the invention also provides a management system for a ship comprising at least one sealed and thermally insulating tank for transporting liquefied gas, the system comprising: at least one fill level sensor for measuring a current fill level of the tank; a motion state estimation device capable of estimating upcoming motion states of the ship from meteorological information and a ship's route; and a processing means configured to generate a plurality of input data vectors each comprising a current fill level of the tank and an upcoming motion state of the ship estimated by the motion state estimation device, and to estimate an upcoming sloshing response of the tank from the input data vectors thus generated and the database obtained by the method according to the second variant.

[0051] According to one embodiment, the processing means is further configured to determine a route of the vessel making it possible to reduce the upcoming sloshing response of the tank. Brief description of the figures

[0052] The invention will be better understood, and other aims, details, characteristics and advantages thereof will appear more clearly during the following description of several particular embodiments of the invention, given solely for illustrative and non-limiting purposes, with reference to the accompanying drawings. [ Fig. 1 ] There [ Fig. 1 ] is a schematic representation of a liquefied gas transport vessel. [ Fig.2 ] There [ Fig.2 ] represents an integrated management system in the ship of the [ Fig. 1 ]. [ Fig. 3 ] There [ Fig. 3 ] represents a management system according to another mode of execution. [ Fig.4 ] There [ Fig.4 ] is a schematic representation of a test vessel sloshing response testing device. [ Fig.5 ] There [ Fig.5 ] is a flowchart representing a method of obtaining a database usable for estimating a sloshing response of a tank. [ Fig.6 ] There [ Fig.6 ] is a flowchart representing a method for estimating a tank sloshing response. [ Fig.7 ] There [ Fig.7 ] is a flowchart representing another method of estimating a tank sloshing response [ Fig.8 ] There [ Fig.8 ] is a flowchart representing yet another method of estimating a tank's sloshing response. Description of the embodiments

[0053] The following embodiments are described in relation to a ship comprising a double hull forming a supporting structure in which a plurality of sealed and thermally insulating tanks are arranged. In such a supporting structure, the tanks have, for example, a polyhedral geometry, for example a prismatic shape.

[0054] Such sealed and thermally insulating tanks are intended, for example, for the transport of liquefied gas. Liquefied gas is stored and transported in such tanks at a low temperature, which requires thermally insulating tank walls in order to maintain the liquefied gas at this temperature. It is therefore particularly important to maintain the integrity of the tank walls, including the thermal insulation spaces located under the sealing membrane, on the one hand to maintain the tank's tightness and prevent leaks of liquefied gas from the tanks and, on the other hand, to avoid degradation of the insulating characteristics of the tank in order to maintain the gas in its liquefied form.

[0055] Such watertight and thermally insulating tanks also comprise an insulating barrier anchored to the double hull of the ship and carrying at least one watertight membrane. For example, such tanks may be produced using technologies marketed under the applicant's Mark III ®< or NO96 ®< trademarks, or others.

[0056] There [ Fig. 1] illustrates a ship 1 comprising four watertight and thermally insulating tanks 2. The four tanks 2 may have identical or different filling states. When at sea, the ship 1 is subject to numerous movements linked to navigation conditions. These movements of the ship 1 are reflected in the liquid contained in the tanks 3, 4, 5, 6 which, consequently, is subject to movements in the tanks 3, 4, 5, 6. These movements of the liquid in the tanks 3, 4, 5, 6 generate impacts on the walls of the tanks 3, 4, 5, 6 which can damage the tanks immediately if they are too violent. In addition, repeated hammering of the walls of the tanks 3, 4, 5, 6 at high and non-destructive levels can lead to fatigue wear degradation of said walls. It is important to maintain the integrity of the walls of tanks 3, 4, 5, 6 to maintain the sealing and insulation characteristics of tanks 3, 4, 5, 6.

[0057] It is known to avoid critical navigation conditions to avoid liquid movements that may immediately damage the tank. However, there is still a need for methods to estimate the sloshing response of tanks during the navigation of the vessel and, if necessary, to take the necessary measures to avoid the occurrence of excessive sloshing that may damage the primary sealing membrane of the tank.

[0058] There [ Fig.2] illustrates an example of a management system 100 on board the ship 1. This management system 100 comprises a central unit 110 connected to a plurality of on-board sensors 120 making it possible to obtain different parameter measurements. Thus, the sensors 120 comprise, for example and in a non-exhaustive manner, at least one filling level sensor 121, for each tank, different movement sensors 122 of the ship and maritime condition sensors 123. The management system 100 also comprises a communication interface 130 allowing the central unit 110 to communicate with remote devices, for example to obtain meteorological data, ship position data or other.

[0059] The ship's motion sensors 122 determine measured motions of the ship, for example by measuring the accelerations experienced by the ship along three perpendicular axes in translation and rotation. To assess the ship's motions, it will be possible to advantageously use an Inertial Measurement Unit, hereinafter UMI, which consists of one or more accelerometers and / or one or more gyroscopes, for example a mechanical gyroscope, and / or one or more magnetometers. These measurement units, in the event that several are used (of the same type or two different types), are advantageously distributed over the ship so as to record a precise measurement of the ship's motion. It will be noted that an UMI is sometimes commonly referred to as an MRU (from the English Motion Reference Unit).

[0060] The sea condition sensors 123 obtain a current sea condition in the vessel environment, such as a wave height and frequency in the vessel environment. For example, in one embodiment, the wave height and / or frequency are provided from a visual observation by the crew.

[0061] The management system 100 further comprises a human-machine interface 140. This human-machine interface 140 comprises a display means 41. This display means 41 allows the operator to obtain the various management information calculated by the system or the measurements obtained by the sensors 120 or even a current tossing state, which can be estimated as will be detailed later.

[0062] The human-machine interface 140 further comprises an acquisition means 42 allowing the operator to manually provide quantities to the central unit 110, typically to provide the central unit 110 with data that cannot be obtained by sensors because the vessel does not have the necessary sensor or the latter is damaged. For example, in one embodiment, the acquisition means allows the operator to enter information on the height and / or frequency of the waves from a visual observation and / or to manually enter a heading and / or a speed of the vessel.

[0063] The management system 100 further comprises a database 150. This database can be used to estimate a sloshing response of a tank as will be detailed later.

[0064] There [ Fig. 3] illustrates an example of a management system 200 located on land and communicating with the ship 1. The ship comprises the central unit 110, the sensors 120 and a communication interface 130. The management system 200 comprises a central unit 210, a communication interface 230, a human-machine interface 240 and a database 250. The operation of the management system 200 is similar to the operation of the management system 100 and differs only in that the information measured by the sensors 120 of the ship 1 is sent to the management system 200 located on land via the communication interfaces 130 and 230. For example, the communication interfaces may use data transmission by terrestrial or satellite radio frequency.

[0065] We will now describe, using the Figures 4 and 5 , obtaining the database 150.

[0066] There [ Fig.4] schematically represents an example of a test device 1000 for carrying out tests on a test tank 1010. The tests consist of subjecting the test tank 1010 to movements, the test tank 1010 having a given filling level of fluid 1011, and measuring a pressure on at least one point of a wall 1010a of the test tank 1010 using a pressure sensor 1012 and / or a number of impacts on at least one wall of the test tank 1010.

[0067] The test tank 1010 may be of reduced dimensions compared to a tank whose sloshing response is to be estimated, and / or have a geometry representative of the tank whose sloshing response is to be estimated.

[0068] Of course, the fluid 1011 is preferably of the same nature, and ideally at the same temperature, density, viscosity, as that transported by the tank whose sloshing response is to be estimated; it may in particular be Liquefied Petroleum Gas (also called LPG) having for example a temperature between - 50°C and 0°C, or Liquefied Natural Gas (LNG) at approximately -162°C at atmospheric pressure. Many liquefied gases may also be considered, in particular methane, ethane, propane, butane, ammonia, dihydrogen or ethylene.

[0069] Furthermore, it is possible to measure a pressure on a plurality of points of a wall 1010a of the test tank 1010, or even on several or all of these walls, the number and arrangement of the pressure sensors 1012 being adapted accordingly. When measuring a number of impacts on at least one wall of the test tank 1010, this measurement is carried out using a plurality of pressure sensors 1012 suitably arranged on this wall. It is possible to measure a number of impacts on several of the walls of the test tank 1010 or on all of the walls of the test tank 1010.

[0070] As mentioned above, the test tank 1010 is subjected to movements during the tests. In the example shown, the device 1000 thus comprises a platform 1013 to which the test tank 1010 is secured. The platform 1013 is driven into movement by the action of six hydraulic cylinders 1015 connected at one of their ends to the platform at three fixing points 1014 and at the other end to a frame or to a floor 1001. This makes it possible to drive the test tank 1010 into movement according to six degrees of freedom, in translation and in rotation. Of course, the test tank 1010 can be driven into movement by different means.

[0071] The device 1000 further comprises a test control unit 1020. The test control unit 1020 is configured to control the hydraulic cylinders 1015 to subject the test tank 1010 to movements determined in advance in a test program. In one exemplary embodiment, these movements are movements representative of a given movement of the vessel, which preferably take into account the position of the tank on the vessel and / or the geometry of the tank. In another exemplary embodiment, these movements are movements representative of a given sea state, which are converted into corresponding movements of the vessel, preferably taking into account the position of the tank on the vessel and / or the geometry of the tank. The evaluation of the corresponding movements of the vessel from a given sea state is a common task in the evaluation of the seaworthiness of a vessel.Furthermore, the test control unit 1020 records the values ​​read during the test by the at least one pressure sensor 1012.

[0072] The test control unit 1020 communicates with a test data processing unit 1030. The test data processing unit 1030 comprises a communication interface 1031 for receiving from the test control unit 1020 the values ​​recorded during the test by the at least one pressure sensor 1012 as well as the movements imposed on the test tank 1010 during the test. The test data processing unit 1030 further comprises a memory 1033 and a central unit 1032.

[0073] The test data processing unit 1030 is configured to train, on the central unit 1032 communicating with the memory 1033, a statistical model by a machine learning method. The statistical model is capable of estimating a sloshing response of the tank as a function of a filling level of the tank and a current sea state and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel. The sloshing response comprises at least one of a rate of fluid impacts on the walls of the tank, a maximum pressure on the walls of the tank, and a probability of damage to the tank. In a variant, the statistical model considers a plurality of tanks, the statistical model being capable of estimating a sloshing response of each tank as a function of its position within the vessel.

[0074] More specifically, the training of the statistical model is carried out by a supervised machine learning method. For example, the supervised machine learning method can be a Gaussian process regression method. Gaussian process regression methods are well known as such; they are well suited to the training of the statistical model because they allow producing a statistical model capable of answering a regression problem, for any input data set, by training from a relatively limited number of data. However, it is possible to retain other supervised machine learning methods.

[0075] The training of the statistical model is carried out from the results of the tests carried out on the test tank 1010. More particularly, in a preferred example, the training of the statistical model is carried out from the sloshing response of the test tank 1010 during each test, this sloshing response being calculated beforehand from the values ​​recorded during the test by the at least one pressure sensor 1012. The sloshing response of the test tank 1010 may comprise at least one of a rate of fluid impacts on one or some of the walls 1010a of the test tank 1010 and a maximum pressure on the walls 1010a of the test tank 1010 over a given period.In one variant, the training of the statistical model is carried out both from the results of the tests carried out on the test tank 1010 and from test data obtained or recorded on ships operating as carriers and / or users of liquefied gas, one or more tanks of these ships playing the role of the test tank 1010. In another variant, the training of the statistical model can be carried out solely from test data obtained or recorded on ships operating as carriers and / or users of liquefied gas, one or more tanks of these ships playing the role of the test tank 1010.

[0076] We will now describe, using the [ Fig.5 ], a method 300 for obtaining the database 150. Steps 301 to 305 can be performed by the central unit 1032 communicating with the memory 1033.

[0077] Optionally, the method 300 may comprise a step 301 consisting of excluding from the test data set used for training the statistical model test results having a sloshing response of the test tank 1010 below a certain threshold. Thus, the statistical model is trained only from the test data having revealed a significant sloshing of the test tank 1010, which improves the accuracy of the estimation of the sloshing response by the statistical model.

[0078] The method 300 comprises, after the possible step 301, a step 302 consisting of training the statistical model as already described above.

[0079] Optionally, at least one constraint is imposed on the statistical model during its training by the supervised machine learning method during step 302. These constraints can be defined on the basis of elementary physical considerations, such as for example the absence of sloshing in the case where the filling level of the tank is zero, and / or on the basis of considerations obtained by practical experience, such as for example the fact that larger movements or larger dimensions of the tank can potentially result in a larger sloshing response. As a result, the accuracy of the estimation of the sloshing response by the statistical model is increased.

[0080] At the end of step 302, a statistical model is available which is capable of estimating a sloshing response of the tank as a function of a filling level of the tank and a current sea state and optionally at least one of a draft of the ship, a speed of the ship and a heading of the ship, and this for any values ​​of these quantities, including for which no test has been carried out on the test tank 1010. However, the calculation necessary to do this may be too long and / or require too many computing resources to be able to be implemented on board a ship, for which it is important to obtain an estimate of the sloshing response as quickly as possible and with the least expensive on-board system possible.Therefore, after step 302, a step 303 is implemented consisting of generating a plurality of input data vectors each comprising a filling level of the tank and a current sea state, followed by a step 304 consisting of, for each input data vector generated in step 303: obtaining an estimated sloshing response of the tank using the statistical model in step 302; and storing in a database the estimated sloshing response of the tank in association with the input data vector.

[0081] Optionally, in a step 305, the database obtained in step 304 is transmitted to the management system 100, or stored on a computer-readable recording medium. The database 150 is also obtained, the use of which will be described later.

[0082] So far, a case has been described where the statistical model is capable of estimating a sloshing response of the tank as a function of at least a filling level of the tank and a current sea state. However, as a variant, the statistical model is capable of estimating a sloshing response of the tank as a function of a filling level of the tank, a current state of movement of the ship and optionally at least one of a draft of the ship, a speed of the ship and a heading of the ship. Steps 302, 303, 304 are then modified accordingly.

[0083] We will now describe, using the [ Fig.6 ], a method 400 for estimating a sloshing response of a tank using the database 150. The method 400 may consider a plurality of tanks instead of a single tank. In this case, before implementing the method 400, a preliminary step of defining the position of each of the tanks within the ship may be carried out.

[0084] According to a first embodiment, the organizational chart of the [ Fig.6 ] is implemented entirely on the central unit 110 forming a single processing means. According to a second embodiment, the flowchart of the [ Fig.6 ] is partially carried out on the land-based management system 200 which communicates with the ship. According to this second embodiment, the ship 1 transmits to the land-based station all the information coming from the sensors 120, and the central unit 110 and the central unit 210 together form a shared processing means.

[0085] The method 400 comprises a first step 401 consisting of determining a current filling level of the tank and a current sea state. The current filling level of the tank is typically determined from a filling indication provided by the tank filling level sensor 121. The current sea state can be determined from indications provided by the sea condition sensors 123 and / or by communication, by terrestrial radio frequency or by satellite, with a network of weather stations.

[0086] Optionally, in step 401, a draft of the vessel and / or a heading of the vessel are also determined, typically from indications provided by the ship's onboard systems. The draft of the vessel is typically provided to the ship's onboard systems by one or more float-type and / or hydrostatic pressure sensors. The heading of the vessel is typically provided to the ship's onboard systems by one or more navigation compasses.

[0087] The method 400 further comprises a second step 402 of generating an input data vector comprising the data determined in step 401.

[0088] The method 400 further comprises a third step 403 consisting of estimating a sloshing response of the tank from the input data vector generated in step 402 and from the database 150. More concretely, if the input data vector is present in the database 150, the sloshing response is obtained by simply reading the database 150. However, more typically, the database 150 will not contain the input data vector, but input data close to those contained in the input data vector. In this case, the sloshing response will be obtained by interpolation of the sloshing response associated with two or more neighboring input data vectors present in the database 150.

[0089] After step 403, the sloshing response obtained can be compared to an alert threshold, and an alarm can be displayed to a user, for example on the display means 41, if the sloshing response exceeds the alert threshold. Preferably, after displaying this alarm, a decision-making assistance step can be carried out to reduce the sloshing. This decision-making assistance step can consist of a proposal to change the direction or trajectory of the ship, a change of course particularly suitable for stationary floating structures, a modification of the speed of the ship or a change in the filling rate of the tank(s) (between the tanks or between a tank and a reservoir external to the ship in the case of a stationary floating structure).Furthermore, the alarm may consist of signaling a problem to be corrected immediately or in the short term, if possible an alarm designating the tank(s) which require a check and maintenance operation with a view to possible repair.

[0090] We will now describe, using the [ Fig.7 ], another method 500 for estimating a sloshing response of a tank using the database 150. In this variant, the database 150 is obtained from a statistical model which is capable of estimating a sloshing response of the tank as a function of a filling level of the tank, a current state of movement of the vessel and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel as described above in connection with the [ Fig.4]. The method 500 may consider a plurality of tanks instead of a single tank. In this case, before implementing the method 500, a preliminary step of defining the position of each of the tanks within the ship may be carried out.

[0091] The method 500 comprises a first step 501 of determining a current filling level of the tank and a current state of movement of the vessel. The current filling level of the tank is typically determined from a filling indication provided by the tank filling level sensor 121. The current state of movement of the vessel may be determined from indications provided by the vessel movement sensors 122.

[0092] Optionally, in step 501, a draft of the vessel and / or a heading of the vessel are also determined, typically from indications provided by the ship's onboard systems. The draft of the vessel is typically provided to the ship's onboard systems by one or more float-type and / or hydrostatic pressure sensors. The heading of the vessel is typically provided to the ship's onboard systems by one or more navigation compasses.

[0093] The movement sensors of the ship 122 typically having an acquisition frequency much higher than the typical duration of the sloshing of the tank, the indications provided by the movement sensors of the ship 122 can be averaged over an acquisition period, the other data determined in step 501 then being averaged over this same acquisition period.

[0094] The method 500 further comprises a second step 502, analogous to step 402, consisting of generating an input data vector comprising the data determined in step 501.

[0095] The method 500 further comprises a third step 503 of estimating a sloshing response of the tank from the input data vector generated in step 502 and the database 150. Step 503 is analogous to step 403 and is therefore not explained in detail again.

[0096] After step 503, the sloshing response obtained can be compared to an alert threshold, and an alarm can be displayed to a user, for example on the display means 41, if the sloshing response exceeds the alert threshold. Preferably, after displaying this alarm, a decision-making assistance step can be carried out to reduce the sloshing. This decision-making assistance step can consist of a proposal for a change in direction or trajectory of the ship, a change of course particularly suitable for stationary floating structures, a modification of the speed of the ship or a change in the filling rate of the tank(s) (between the tanks or between a tank and a reservoir external to the ship in the case of a stationary floating structure).Furthermore, the alarm may consist of signaling a problem to be corrected immediately or in the short term, if possible an alarm designating the tank(s) which require a check and maintenance operation with a view to possible repair.

[0097] We will now describe, using the [ Fig.8 ], another method 600 for estimating a sloshing response of a tank using the database 150. In this variant, the database 150 is obtained from a statistical model which is capable of estimating a sloshing response of the tank as a function of a filling level of the tank, a current sea state and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel.

[0098] The method 600 comprises a first step 601 of determining a current tank fill level and estimating future sea states. The current tank fill level is typically determined from a fill indication provided by the tank fill level sensor 121. Future sea states are estimated from meteorological information and a ship's course. The ship's course is typically obtained from indications provided by the ship's onboard systems, such as the ship's speed and the ship's heading. The meteorological information may be provided by the sea condition sensors 123 and / or by communication, by terrestrial radio frequency or by satellite, with a network of weather stations.

[0099] The method 600 further comprises a second step 602 of generating a plurality of input data vectors each comprising a current tank fill level and an estimated future sea state.

[0100] Optionally, in step 601, a draft of the vessel, a heading of the vessel and a speed of the vessel are also determined, typically from indications provided by the ship's onboard systems. The draft of the vessel is typically provided to the ship's onboard systems by one or more float and / or hydrostatic pressure sensors. The heading of the vessel is typically provided to the ship's onboard systems by one or more navigation compasses. The speed of the vessel is typically provided to the ship's onboard systems by an IMU and / or by a GPS-type satellite navigation receiver.

[0101] The method 600 further comprises a third step 603 of estimating a future sloshing response of the tank from each of the input data vectors generated in step 602 and the database 150. Step 603 is analogous to step 403 and is therefore not explained in detail again.

[0102] After step 603, a ship route can be determined to reduce the upcoming sloshing response of the tank compared to the sloshing response of the tank that would be expected if the ship maintained its current route. The expression "ship route" means a heading of this ship, the speed of the latter or a simple avoidance of a geographical area. For stationary floating structures (ships, barges), i.e. having a fixed position, the change of heading results in a change of angle between the North direction and the longitudinal axis of the structure so as to orient or bring the floating structure to conventionally reduce the negative consequences of swell and waves on the floating structure. In addition or as an alternative, a modification of the filling level of the tank can be determined to reduce the upcoming sloshing response of the tank.

[0103] A variation of the 600 process of the [ Fig.8 ] is described below. In this variant, the database 150 is obtained from a statistical model which is capable of estimating a sloshing response of the tank as a function of a filling level of the tank, a current state of movement of the vessel and optionally at least one of a draft of the vessel, a speed of the vessel and a heading of the vessel.

[0104] In this variant, step 601 consists of determining a current tank fill level and estimating future ship movement states. The current tank fill level is typically determined from a fill indication provided by the tank fill level sensor 121. Future ship movement states are estimated from meteorological information and a ship's course. The ship's course is typically obtained from indications provided by the ship's onboard systems, such as the ship's speed and the ship's heading. The meteorological information may be provided by the sea condition sensors 123 and / or by communication, by terrestrial radio frequency or by satellite, with a network of weather stations.According to one example, the future ship motion states can be estimated by first estimating future sea states estimated from meteorological information and a ship's route, and then by secondly estimating future ship motion states from the future sea states thus estimated. It is recalled that, as already mentioned above, the evaluation of the corresponding ship motions from a given sea state is a common task in the assessment of the seakeeping of a ship.

[0105] Step 602 then consists of generating a plurality of input data vectors each comprising a current filling level of the tank and an estimated future movement state of the ship.

[0106] Optionally, in step 601, a draft of the vessel, a heading of the vessel and a speed of the vessel are also determined, as already mentioned previously.

[0107] Step 603 is to estimate a future sloshing response of the tank from each of the input data vectors generated in step 602 and the database 150. Step 603 is analogous to step 403 and is therefore not explained in detail again.

[0108] After step 603, a ship's course may be determined that reduces the upcoming tank sloshing response relative to the tank sloshing response that would be expected if the ship maintained its current course, as previously mentioned.

[0109] Although the invention has been described in connection with several particular embodiments, it is quite obvious that it is in no way limited thereto and that it includes all technical equivalents of the means described as well as their combinations if these fall within the scope of the invention as defined by the appended claims.

[0110] Furthermore, it is quite obvious that a feature or combination of features described in connection with a method applies equally to a corresponding system, and vice versa.

[0111] The use of the verb "comprise", "comprise" or "include" and its conjugated forms does not exclude the presence of other elements or other steps than those set out in a claim.

[0112] In the claims, any reference sign in parentheses cannot be interpreted as a limitation of the claim.

Claims

1. Method (300) for obtaining a database (150) usable to estimate a sloshing response of at least one sealed thermally insulating tank for the transport of liquefied gas, the method comprising the steps consisting in: - training (302) a statistical model by a supervised machine learning method on a set of test data, the statistical model being able to estimate a sloshing response of the tank as a function of a level of filling of the tank and of a current sea state and optionally at least one of the following: a draft of the ship, a speed of the ship and a heading of the ship, and the set of test data being obtained from results of a plurality of tests each consisting in subjecting a test tank (1010) having a given level of filling to movements and measuring a pressure at at least one point on a wall (1010a) of the test tank (1010) and / or a number of impacts on at least one wall of the test tank (1010); - generating (303) a plurality of input data vectors each comprising a level of filling of the tank and a current sea state; and - for each input data vector generated in this way: obtaining (303) an estimated sloshing response of the tank with the aid of the statistical model; and storing (303) in a database the estimated sloshing response of the tank in association with the input data vector.

2. Method according to Claim 1, in which the sloshing response comprises at least one of a number of impacts of fluid on the walls of the tank, a maximum pressure on the walls of the tank, and a probability of damage to the tank.

3. Method according to any of Claims 1 to 2, in which at least one constraint is imposed on the statistical model during its training by the supervised machine learning method.

4. Method according to any of Claims 1 to 3, further comprising a step (301) consisting in excluding from the set of test data test results featuring a sloshing response below a threshold before the step (302) of training the statistical model.

5. Method (300) for obtaining a database (150) usable to estimate a sloshing response of at least one sealed thermally insulating tank for the transport of liquefied gas, the method comprising the steps consisting in: - training (302) a statistical model by a supervised machine learning method on a set of test data, the statistical model being able to estimate a sloshing response of the tank as a function of a level of filling of the tank, a current state of movement of the ship and optionally of at least one of the following: a draft of the ship, a speed of the ship and a heading of the ship, and the set of test data being obtained from results of a plurality of tests each consisting in subjecting a test tank (1010) having a given level of filling to movements and measuring a pressure at at least one point on a wall (1010a) of the test tank (1010) and / or a number of impacts on at least one wall of the test tank (1010); - generating (302) a plurality of input data vectors each comprising a level of filling of the tank and a current state of movement of the ship; and - for each input data vector generated in this way: obtaining (303) an estimated sloshing response of the tank with the aid of the statistical model; and storing (303) in a database the estimated sloshing response of the tank in association with the input data vector.

6. Method according to Claim 5, in which the sloshing response comprises at least one of a number of impacts of fluid on the walls of the tank, a maximum pressure on the walls of the tank, and a probability of damage to the tank.

7. Method according to any of Claims 5 to 6, in which at least one constraint is imposed on the statistical model during its training by the supervised machine learning method.

8. Method according to any of Claim 5 to 7, further comprising a step (301) consisting in excluding from the set of test data test results featuring a sloshing response below a threshold before the step (302) of training the statistical model.

9. Method (400) for estimating a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas onboard a ship, the method comprising the steps consisting in: - determining (401) a current level of filling of the tank; - determining (401) a current sea state; - generating (402) an input data vector comprising the current level of filling of the tank and the current sea state determined in this way; and - estimating (403) a sloshing response of the tank from the input data vector generated in this way and from the database (150) obtained by the method according to any of Claims 1 to 4.

10. Method (500) for estimating a sloshing response of at least one sealed and thermally insulating tank for the transport of liquefied gas onboard a ship, the method comprising the steps consisting in: - determining (501) a current level of filling of the tank; - determining (501) a current state of movement of the ship; - generating (502) an input data vector comprising the current level of filling of the tank and the current state of movement of the ship determined in this way; and - estimating (503) a sloshing response of the tank from the input data vector generated in this way and from the database (150) obtained by the method according to any of Claims 5 to 8.

11. Method according to Claim 9 or 10, in which a plurality of tanks are considered and the method comprises a previous step of definition of the position of each of the tanks of the ship.

12. Method (400, 500) according to any of Claim 9 to 11, further comprising a step consisting in furnishing an alarm to a user if the estimated sloshing response of the tank exceeds an alert threshold, and preferably a step of assisting the decision intended to reduce the sloshing.

13. Management system (100) for a ship (1) including at least one sealed and thermally insulating tank (2) for transporting liquefied gas, the system comprising: - at least one filling level sensor (121) for measuring a current state of filling of the tank (2); - a device (123) for evaluation of the sea state able to evaluate a current sea state; and - a processing means (110) configured to generate an input data vector comprising a current level of filling of the tank and a current sea state evaluated by the sea state evaluation device (123) and to estimate a sloshing response of the tank (2) from the input data vector generated in this way and from the database (150) obtained by the method (300) according to any of Claim 1 to 4.

14. Management system (100) for a ship (1) including at least one sealed and thermally insulating tank (2) for transporting liquefied gas, the system comprising: - at least one filling level sensor (121) for measuring a current state of filling of the tank (2); - a device (122) for evaluation of the current state of movement of the ship able to evaluate a current state of movement of the ship; and - a processing means (110) configured to generate an input data vector comprising a current level of filling of the tank and a current state of movement of the ship and to estimate a sloshing response of the tank from the input data vector generated in this way and from the database (150) obtained by the method according to any of Claim 5 to 8.

15. Method (600) of estimating a sloshing response of a sealed and thermally insulating tank (2) for the transport of liquefied gas onboard a ship (1), the method comprising the steps consisting in: - determining (601) a current level of filling of the tank; - estimating (601) future sea states from meteorological information and a course of the ship; - generating (602) a plurality of input data vectors each comprising a current level of filling of the tank and an estimated future sea state; and - estimating (603) a future sloshing response of the tank from the input data vectors generated in this way and from the database (150) obtained by the method according to any of Claims 1 to 4.

16. Method (600) of estimating a sloshing response of a sealed and thermally insulating tank (2) for the transport of liquefied gas onboard a ship (1), the method comprising the steps consisting in: - determining (601) a current level of filling of the tank; - estimating (601) future states of movement of the ship from meteorological information and a course of the ship; - generating (602) a plurality of input data vectors each comprising a current level of filling of the tank and an estimated future state of movement of the ship; and - estimating (603) a future sloshing response of the tank from the input data vectors generated in this way and from the database (150) obtained by the method according to any of Claim 5 to 8.

17. Method according to Claim 15 or 16, further comprising a step consisting in determining a course of the ship and / or a modification of the level of filling of the tank enabling reduction of the future sloshing response of the tank.

18. Management system (100) for a ship (1) including at least one sealed and thermally insulating tank (2) for transporting liquefied gas, the system comprising: - at least one level of filling sensor (121) for measuring a current level of filling of the tank (2); - a sea state estimation device (123) able to estimate future sea states from meteorological information and from a course of the ship (1); and - a processing means (110) configured to a generate a plurality of input data vectors each comprising a current level of filling of the tank and a future sea state estimated by the sea state estimation device (123), and to estimate a future sloshing response of the tank from the input data vectors generated in this way and from the database (150) obtained by the method according to any of Claims 1 to 4.

19. Management system (100) for a ship (1) including at least one sealed and thermally insulating tank (2) for transporting liquefied gas, the system comprising: - at least one level of filling sensor (121) for measuring a current level of filling of the tank (2); - a state of movement estimation device able to estimate future sea states from meteorological information and from a course of the ship (1); and - a processing means (110) configured to a generate a plurality of input data vectors each comprising a current level of filling of the tank and a future state of movement of the ship estimated by the state of movement of the ship estimation device, and to estimate a future sloshing response of the tank from the input data vectors generated in this way and from the database (150) obtained by the method according to any of Claims 5 to 8.

20. System according to Claim 18 or 19, in which the processing means (110) is further configured to determine a course of the ship enabling reduction of the future sloshing response of the tank.