A method and system for dynamic modeling of a shipboard three-dimensional stockpile

Abstract

This application relates to the field of three-dimensional measurement technology and discloses a method and system for dynamic modeling of shipborne three-dimensional material stacks. The method includes: establishing a finite element mechanical model of the ship and calculating the theoretical strain distribution of the ship; determining the installation position of fiber optic strain sensors; matching the dynamic strain data collected in real time by the fiber optic strain sensors with the theoretical strain values ​​to generate an instantaneous displacement field; correcting the transformation matrix of the virtual global coordinate system; transforming point cloud data collected by multiple three-dimensional scanners into the virtual global coordinate system according to the corrected transformation matrix; fusing and registering the transformed point clouds to construct a three-dimensional model of the material stack inside the hold. This application effectively eliminates the interference of hull deformation on three-dimensional modeling by converting real-time strain sensing data into a continuous displacement field of the entire ship and dynamically correcting the transformation matrix of the multi-source scanners, thereby reducing the probability of accidents during ship unloading.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional measurement technology, and more specifically, to a method and system for dynamic modeling of shipborne three-dimensional material stacks. Background Technology

[0002] In the field of automated bulk cargo handling at ports, the construction of 3D stockpile models is a key technology for achieving unmanned unloading, stacking and reclaiming, and precise inventory management. By deploying multiple LiDARs or depth cameras at locations such as unloaders, hatch coamings, and quay cranes, point cloud data of the material surface inside the ship's hold can be acquired in real time. After registration and fusion, the 3D shape of the stockpile can be reconstructed, providing a basis for decision-making in subsequent grab bucket path planning, remaining cargo volume calculation, and ship balance adjustment.

[0003] The primary prerequisite for building this model is establishing a unified and stable spatial reference, accurately transforming the point cloud data collected by sensors distributed across different carriers to the same coordinate system. Current mainstream multi-sensor fusion methods typically assume all sensors are located on the same rigid reference, or use high-precision inertial measurement units and differential GPS to calculate the absolute pose of each sensor in real time, thereby unifying the point cloud to a fixed coordinate system. However, this crucial prerequisite faces severe challenges in large ship loading and unloading operations. On the one hand, modern Cape-size and very large ore carriers can reach lengths of over 300 meters. During unloading, as the cargo gradually decreases, the bending moment on the hull continuously changes, resulting in significant longitudinal bending deformation—that is, the midships sag when fully loaded and rebound after unloading, with the vertical displacement difference between the bow, stern, and midships reaching tens of centimeters. On the other hand, wave action and cargo eccentricity can also induce lateral torsional deformation, causing relative vertical displacement between the port and starboard sides. This dynamic elastic deformation leads to nonlinear changes in the relative positions and attitudes of sensors installed in different locations such as hatch coamings and hull decks over time, which cannot be described using traditional rigid body transformations.

[0004] Existing technologies attempt to compensate for deformation by increasing the number of shipboard attitude sensors, such as deploying differential GPS receivers and fiber optic gyroscopes at multiple key points on the hull to measure the absolute coordinates of each point in real time. However, such solutions are costly, and GPS signals are easily blocked by the metal structure of the unloading machine. Inertial measurement units also suffer from integral drift, making it difficult to maintain accuracy over long periods. Another approach is to use visual tracking of cooperative targets fixed to the hull and the dock, calculating relative displacement in real time through image processing. However, the high concentration of dust, drastic changes in lighting, and the ease with which targets are blocked by grabs or materials in the port environment significantly reduce the reliability of visual tracking. Furthermore, it can only acquire deformation information at a limited number of discrete points and cannot reconstruct a continuous deformation field across the entire area.

[0005] Due to the lack of effective compensation for the dynamic deformation of the hull, point clouds from different scanners cannot be accurately aligned, resulting in distortion, breakage, or systematic deviations in the reconstructed model of the cargo hold. This problem is particularly prominent at the end of the unloading process, when the distribution of remaining materials is crucial for the cleanup operation. Model distortion may cause collisions between the grab bucket and the bulkhead, or even lead to capsizing risks due to miscalculation of the ship's stability. In addition, inventory count errors can also cause trade settlement disputes and result in huge economic losses. Summary of the Invention

[0006] To overcome the challenges posed by the lack of effective compensation for dynamic hull deformation during the final stages of unloading, which leads to inaccurate alignment of point clouds from different scanners and results in distortion, breakage, or systematic deviations in the reconstructed cargo hold model, model distortion can cause collisions between the grab bucket and the bulkhead, and even capsize risks due to miscalculations of ship stability, this invention transforms real-time strain sensor data into a continuous displacement field across the entire ship. This dynamic correction of the transformation matrix from the multi-source scanners eliminates the interference of hull deformation on 3D modeling, reducing the probability of accidents during unloading. To achieve the above objectives, this invention provides the following technical solution:

[0007] In a first aspect, the present invention discloses a method for dynamic modeling of a shipborne three-dimensional stockpile, comprising:

[0008] A finite element mechanical model of the ship is established. The finite element mechanical model is constructed based on the ship design drawings and material properties. It is used to simulate the longitudinal bending and transverse torsional deformation of the ship under different load conditions. The theoretical strain value and theoretical displacement value of multiple preset monitoring points under different load conditions are calculated to obtain the theoretical strain distribution of the ship.

[0009] The installation location of the fiber optic strain sensor is determined based on the theoretical strain distribution, and multiple fiber optic strain sensors are installed. The fiber optic strain sensors are used to collect dynamic strain data of the ship hull during actual operation caused by cargo unloading in real time.

[0010] The dynamic strain data collected in real time by the fiber optic strain sensor is matched with the theoretical strain value. Based on the matching result, the actual three-dimensional displacement of each monitoring point at the current moment is determined from the correspondence between the theoretical strain value and the theoretical displacement value, and an instantaneous displacement field is generated.

[0011] Based on the instantaneous displacement field, the transformation matrix of each of the multiple 3D scanners installed on the hatch coaming, quay crane and unloader relative to the virtual global coordinate system is corrected so that the spatial pose of each 3D scanner relative to the virtual global coordinate system is updated in real time with the deformation of the hull.

[0012] Point cloud data collected by multiple 3D scanners are transformed into a virtual global coordinate system according to the corrected transformation matrix. The transformed point clouds are then fused and registered to construct a 3D model of the material stack inside the cabin.

[0013] Furthermore, when matching dynamic strain data with theoretical strain values, the distribution similarity between dynamic strain data and theoretical strain values ​​under each load condition is calculated. The load condition corresponding to the maximum distribution similarity is selected as the actual load condition of the current ship, and the theoretical displacement value under the corresponding load condition is used as the initial displacement estimate.

[0014] Furthermore, the triggering condition for correcting the transformation matrix is ​​that the actual three-dimensional displacement of any monitoring point in the instantaneous displacement field exceeds a preset displacement threshold, or the rate of change of the dynamic strain data collected by the fiber optic strain sensor exceeds a preset rate of change threshold.

[0015] Furthermore, the method also includes:

[0016] After establishing the correspondence between theoretical strain values ​​and theoretical displacement values, when the deviation between the dynamic strain data collected N times and the theoretical strain values ​​all exceed the preset deviation threshold, the finite element mechanical model is calibrated according to the actual loading records of the ship, and the correspondence is recalculated and updated, where N is a preset positive integer.

[0017] Furthermore, when determining the installation location of the fiber optic strain sensor based on the theoretical strain distribution, the region in the theoretical strain distribution where the strain gradient exceeds a preset strain gradient threshold is selected as the installation location. The strain gradient threshold is determined based on the statistical characteristics of the theoretical strain distribution output by the finite element mechanical model.

[0018] Furthermore, the virtual global coordinate system is established with the fixed point of the dock as the origin and the direction of the quay crane track as the coordinate axis. The transformation matrix includes translation and rotation components. The correction amount of the translation component is equal to the three-dimensional displacement at the corresponding installation position of the 3D scanner in the instantaneous displacement field, and the correction amount of the rotation component is equal to the local torsion angle at the corresponding installation position of the 3D scanner in the instantaneous displacement field.

[0019] Furthermore, when performing fusion registration on the transformed point clouds, the corrected transformation matrix is ​​used as the initial registration parameter, and a point cloud fine registration algorithm is used to register point clouds from different 3D scanners.

[0020] Secondly, this invention discloses a shipborne three-dimensional material stack dynamic modeling system for implementing the aforementioned shipborne three-dimensional material stack dynamic modeling method. The system includes: a finite element modeling module, a fiber optic strain sensing array, a displacement field calculation module, a transformation matrix correction module, and a point cloud fusion module.

[0021] The finite element modeling module is used to establish a finite element mechanical model of a ship, simulate the longitudinal bending and transverse torsional deformation of the ship under different load conditions, calculate the correspondence between the theoretical strain values ​​and theoretical displacement values ​​of multiple preset monitoring points under different load conditions, and output the theoretical strain distribution of the ship.

[0022] The fiber optic strain sensor array is installed at a location determined based on the theoretical strain distribution to collect dynamic strain data of the ship hull during actual operations caused by cargo unloading in real time.

[0023] The displacement field calculation module is used to match dynamic strain data with theoretical strain values, and determine the actual three-dimensional displacement of each monitoring point at the current moment based on the correspondence between theoretical strain values ​​and theoretical displacement values ​​according to the matching results, thereby generating an instantaneous displacement field.

[0024] The transformation matrix correction module is used to correct the transformation matrix of each of the multiple 3D scanners installed on the hatch coaming, quay crane and unloader relative to a virtual global coordinate system based on the instantaneous displacement field, so that the spatial pose of each 3D scanner relative to the virtual global coordinate system is updated in real time with the deformation of the hull.

[0025] The point cloud fusion module is used to transform point cloud data collected by multiple 3D scanners into a virtual global coordinate system according to the corrected transformation matrix, and then perform fusion and registration on the transformed point cloud to construct a 3D model of the material stack inside the cabin.

[0026] Furthermore, the fiber optic strain sensor array is continuously arranged along the longitudinal direction of the hull, and the spacing between adjacent fiber optic strain sensors is determined based on the strain distribution curvature output by the finite element mechanical model.

[0027] Furthermore, the finite element modeling module is also used to divide the finite element mechanical model into multiple sub-models based on the compartment division and compartment location in the ship design drawings. Each sub-model corresponds to the hull section between two adjacent transverse bulkheads, and the module calculates the correspondence between the theoretical strain value and the theoretical displacement value of each sub-model under different load conditions.

[0028] Compared with related technologies, the present invention has the following beneficial effects:

[0029] This invention introduces a finite element model as a physical knowledge engine, pre-revealing the deep correlation between strain and displacement determined by structural mechanics laws through simulation calculations. It transforms the continuous displacement field of the entire ship, which cannot be directly measured, into discrete strain values ​​at a few key points that are easy to measure with high precision. When fiber optic strain sensors capture the dynamic strain at these key points in real time, the measured strain information is fed back into the pre-established physical model through distribution similarity matching and linear interpolation. This model then deduces the precise three-dimensional displacement at every position on the ship at the current moment. This displacement is subsequently used to dynamically correct the transformation matrices of all scanners, ultimately ensuring that point clouds from different scanners are precisely aligned in the virtual global coordinate system. This effectively eliminates the interference of hull deformation on three-dimensional modeling and reduces the probability of accidents during ship unloading. Attached Figure Description

[0030] Figure 1 This invention provides a flowchart illustrating the steps of a dynamic modeling method for a shipborne three-dimensional stockpile.

[0031] Figure 2 This invention provides a data processing flowchart for a shipborne three-dimensional material stack dynamic modeling system. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1

[0034] Please see Figure 1 As shown, this embodiment provides a method for dynamic modeling of a shipborne three-dimensional stockpile. The complete process of this method includes steps one through five. Next, using a 320-meter-long Cape-size bulk carrier as the application object, each step will be described in detail with reference to specific implementation methods:

[0035] Step 1: Establish a finite element mechanical model of the ship. The finite element mechanical model is constructed based on the ship design drawings and material properties. It is used to simulate the longitudinal bending and transverse torsional deformation of the ship under different load conditions, and calculate the correspondence between the theoretical strain values ​​and theoretical displacement values ​​of multiple preset monitoring points under different load conditions to obtain the theoretical strain distribution of the ship.

[0036] First, it's important to clarify that a finite element mechanical model refers to a numerical simulation model of a ship structure established based on the finite element analysis method. It consists of a large number of discrete elements and monitoring points, capable of simulating the deformation behavior of a real ship under external loads. In this embodiment, the model is constructed based on the geometric parameters and material properties contained in the ship's design drawings. Specifically, the full structural drawings of the Cape-size bulk carrier are obtained from the ship design unit, extracting geometric information such as the distribution of compartments, bulkhead locations, deck thickness, and longitudinal girder layout. Simultaneously, the material parameters of the hull steel, including the elastic modulus, Poisson's ratio, and density, are obtained. These are then input into commercial finite element analysis software, such as ANSYS or Abaqus, to establish a three-dimensional shell element model of the entire ship. The element size is controlled between 500 mm and 800 mm, a size range that ensures both computational accuracy and computational efficiency for subsequent multi-condition simulations. To verify the model's accuracy, the empty ship weight is compared with the empty ship weight given in the design drawings, ensuring the error is within 3%.

[0037] After the model is built, multiple preset monitoring points need to be set. The selection principle of the preset monitoring points is to cover the parts of the hull with the most obvious deformation, while taking into account the feasibility of installing the strain measurement device. In this embodiment, monitoring points can be set along the length of the ship on the deck surface and the upper edge of the hatch coaming at each transverse bulkhead, for a total of thirty-six monitoring points. Each monitoring point records its corresponding three-dimensional spatial coordinates as the reference position for subsequent extraction of theoretical strain and theoretical displacement values. The reason for selecting the deck surface at the transverse bulkhead to set monitoring points is that the transverse bulkhead is the main load-bearing component of the hull structure. Observation shows that the strain change in the area near the transverse bulkhead is the most intense (obvious), so it can sensitively reflect the overall deformation of the hull.

[0038] After the monitoring points are set up, simulation calculations are then performed under different load conditions. The definition of the load conditions needs to fully cover the entire unloading process of the ship from fully loaded to empty, and the wave loads that the ship experiences in the actual marine environment must also be considered. In this embodiment, the construction of the load conditions includes a combination of two dimensions: calm water conditions and wave conditions.

[0039] Specifically, regarding the classification of still water conditions, this embodiment divides the still water load rating into 21 levels, from fully loaded to empty. The first level is the fully loaded draft state given in the ship design loading manual, where all compartments are 100% loaded. Each subsequent 5% reduction in cargo load is considered a monitoring point until the cargo load reaches zero, resulting in 21 still water load ratings. The 5% granularity was chosen based on the following considerations: For ultra-large ships of 300 meters or more, each 5% change in cargo load corresponds to a change in hull bending moment of approximately 6% to 8%. This change is sufficient to generate a measurable strain difference at the monitoring points. Furthermore, the simulation calculations for the 21 conditions can be completed within 24 hours under current computer hardware conditions, balancing engineering practicality and accuracy requirements. If the classification is too coarse, the interpolation error during subsequent real-time matching will increase; if the classification is too fine, the simulation calculation cost will increase exponentially while the benefits will decrease.

[0040] For each still water load class, it is necessary to determine the still water pressure distribution borne by the hull under the corresponding operating conditions. The still water pressure is determined by the ship's draft and includes the external water pressure borne by the hull plating and the internal pressure generated by the cargo in the holds. In this embodiment, the draft corresponding to each load class can be calculated based on the ship's still water curve. Specifically, after the ship is built, the design unit will generate a complete still water curve diagram or data table based on the ship's lines drawing through still water calculation. The ship's still water curve describes the still water parameters such as displacement, center of buoyancy position, and tons per centimeter of draft at different draft depths, and is a basic technical document for ship design and operation. In this embodiment, the still water curve data table of the Cape-size bulk carrier is obtained from the ship design unit. This data table uses the draft depth as the independent variable and records a set of parameters such as displacement and longitudinal coordinate of the center of buoyancy every 0.1 meters, recording a total of 148 sets of data points from the empty draft of 3.5 meters to the full load draft of 18.2 meters, forming a continuous still water characteristic curve.

[0041] Based on this, this embodiment takes the working condition with a load level of 60% as an example to illustrate the specific calculation process of hydrostatic pressure distribution. First, the draft values ​​at the beginning and end corresponding to the load level are determined. According to the loading manual, when the load is 60%, the draft at the beginning is 9.8 meters, the draft at the end is 11.2 meters, and the average draft is 10.5 meters. Because the hull is streamlined along its length, the actual draft at each longitudinal position is not a constant value and needs to be obtained through linear interpolation. Specifically, for a certain cross section position 40 meters from the bow, the draft can be calculated by linear interpolation of the bow and stern drafts: the draft is equal to the stern draft minus the difference between the stern and bow drafts multiplied by the distance between the cross section position and the stern, and then divided by the length of the ship. The draft at the cross section position is calculated to be 10.6 meters. The hydrostatic pressure on the outer plating at the cross section position is equal to the seawater density of 1025 kg / m³ multiplied by the gravitational acceleration of 9.8 m / s², and then multiplied by the water depth of 10.6 meters. The calculated hydrostatic pressure on the outer plating at that point is approximately 106.5 kPa.

[0042] This embodiment uses the second cargo hold as an example to illustrate the calculation of cargo pressure within the hold. The loading manual shows that at a load rating of 60%, the second cargo hold carries 18,000 tons of iron ore, with a hold floor area of ​​approximately 1,800 square meters and a cargo stacking height of 8 meters. Due to the fluidity of bulk materials, the pressure on the hold floor is approximately equal to the material density multiplied by the gravitational acceleration multiplied by the stacking height. Taking the density of iron ore as 2,500 kg / m³, the calculated cargo pressure on the hold floor is approximately 196 kPa. For the bulkhead, due to the slope formed by the material stacking, the pressure at different heights on the bulkhead exhibits a triangular distribution, with the highest pressure at the bottom and near zero at the top. In this embodiment, the bulkhead is divided into ten equal segments along the height direction. The pressure value at the midpoint of each segment is calculated, and the average pressure within the segment is taken as the uniformly distributed pressure on the segment surface and applied to the finite element model. Through the above method, the cargo pressure is calculated for all compartments in sequence, and the hydrostatic pressure of the outer plate is superimposed with the cargo pressure inside the compartment to form a complete hydrostatic load distribution that can be input into the finite element model for solution.

[0043] Regarding the superposition of wave conditions, this embodiment considers the wave environment that ships inevitably encounter when sailing at sea or moored at docks. Based on long-term statistical data of the operating sea area for this ship type, three typical wave conditions are selected for superposition: mid-sag wave condition, mid-bow wave condition, and transverse torsional load condition. The mid-sag wave condition corresponds to a wave pattern where the wave crest is located midway and the wave trough is located at the bow and stern. In this case, the midships are subjected to downward pressure, while the bow and stern are supported upwards, resulting in pressure on the deck and tension on the hull. The mid-bow wave condition is the opposite, with the wave trough located midway, resulting in tension on the deck and pressure on the hull. The transverse torsional load condition corresponds to oblique wave action, causing torsional deformation of the hull around its longitudinal axis. The wave parameters are selected according to the requirements of the International Association of Classification Societies (IACS) Joint Standard. For the operating sea area of ​​this ship type, a meaningful wave height of 4 meters and a spectral peak period of 8 seconds are taken as representative wave parameters. The wave load can be applied using the design wave method, which involves applying a wave dynamic pressure distribution corresponding to the wave parameters in the finite element model. The wave pressure distribution is calculated based on the linear wave theory and varies periodically along the ship's length and width. The pressure amplitude is proportional to the meaningful wave height.

[0044] For each combined load case—a product of 21 still water load levels and 3 wave load cases, totaling 63 combined load cases—the solutions were performed using a finite element model. The solution process first applied still water pressure to bring the model to static equilibrium. Then, wave dynamic pressure was superimposed on the still water pressure to calculate the structural response under the combined loads. The solver employed an implicit static analysis method, considering geometric nonlinearity but neglecting material nonlinearity because the hull steel operates within its elastic range. After solving each load case, six strain components were extracted from all 36 monitoring points, including three normal strain components and three shear strain components, as well as displacement components in three directions: longitudinal displacement, lateral displacement, and vertical displacement. The strain and displacement components were then output as a pair.

[0045] After the solution is obtained, a correspondence between theoretical strain values ​​and theoretical displacement values ​​is established. This correspondence is stored in the form of a multidimensional lookup table. The multidimensional lookup table can use load conditions, monitoring point locations, and strain components as input indices, and the corresponding three-dimensional displacements as output results. This embodiment provides a specific example after testing: For monitoring point number 12, under a load rating of 60% and mid-sag wave conditions, the six strain components calculated by finite element analysis are: longitudinal normal strain of 320 microstrain, transverse normal strain of -45 microstrain, vertical normal strain of 8 microstrain, longitudinal shear strain of 12 microstrain, transverse shear strain of -3 microstrain, and vertical shear strain of 2 microstrain. Simultaneously, the three displacement components of this point are output as follows: vertical displacement of 210 mm, longitudinal displacement of 18 mm, and transverse displacement of 5 mm. Finally, this set of data is stored in the multidimensional lookup table. By using this lookup table, all load conditions and monitoring points are traversed to form a complete strain-displacement database, and the theoretical strain distribution of the ship is output, which describes the strain variation law of each region under different load conditions throughout the ship. To ensure efficient lookup, the lookup table can be stored in the computer as a multidimensional array with dimensions of 21 load levels multiplied by 3 wave conditions multiplied by 36 monitoring points. Each array element stores a six-dimensional strain vector and a three-dimensional displacement vector.

[0046] In other embodiments, the finite element mechanical model can be constructed using different element types and simplification strategies. For example, for small ships with relatively conventional hull shapes, beam element models can be used instead of shell element models, significantly improving calculation speed while ensuring the main deformation characteristics. The division of load conditions can also be dynamically adjusted according to actual operational needs. If the unloading speed is slow and higher-precision intermediate process compensation is required, the load condition levels can be densified to one monitoring point for every 2%. Furthermore, the storage format of the multidimensional lookup table can be a multidimensional array, a tree structure, or a hash table; the specific choice depends on the hardware platform and computing resources of the subsequent lookup and interpolation module. The selection of wave conditions can also be adjusted according to the specific port environment. For example, wave loads can be ignored or only calm water conditions can be considered for inland river ports, while for open sea areas with significant swells, the number of wave condition types can be increased.

[0047] Existing methods typically measure ship displacement directly via GPS or visual tracking, which is a passive measurement method. Its accuracy is limited by signal obstruction and lighting conditions, and it can only obtain displacement information at a finite number of discrete points. Therefore, this embodiment establishes an intrinsic physical relationship between strain and displacement in advance using a finite element mechanical model. This allows for the subsequent measurement of real-time strain at only a few key points to reconstruct the continuous deformation field of the entire ship. This shift in approach means that the acquisition of the displacement field no longer relies on direct mutual observation between sensors, but rather on calculation through a unified physical model.

[0048] The theoretical strain distribution obtained through pre-calculation in Step 1 provides a scientific basis for the subsequent optimized placement of sensors, avoiding monitoring blind spots that may be caused by traditional experience-based sensor placement. The established multi-dimensional lookup table provides an efficient query method for real-time displacement field calculation, transforming the finite element solution process, which originally required several hours, into a millisecond-level table lookup and interpolation operation, meeting the real-time requirements of industrial sites. Furthermore, this correspondence serves as the benchmark for the entire compensation mechanism, and its accuracy directly determines the reliability of all subsequent correction steps. This embodiment ensures that the benchmark data error is controlled within the engineering allowable range through refined working condition division and model verification.

[0049] Step 2: Determine the installation location of the fiber optic strain sensor based on the theoretical strain distribution and install multiple fiber optic strain sensors. The fiber optic strain sensors are used to collect dynamic strain data of the ship hull during actual operation caused by cargo unloading in real time.

[0050] After obtaining the theoretical strain distribution of the ship, the installation positions of the fiber optic strain sensors are determined based on the theoretical strain distribution. Multiple fiber optic strain sensors are then installed at the determined installation positions. This transforms the theoretical strain distribution output from the first step of the finite element mechanical model into an actual sensor deployment scheme, ensuring that the subsequently collected dynamic strain data can accurately reflect the overall deformation characteristics of the ship.

[0051] First, it's important to clarify that a fiber optic strain sensor is a strain measurement device based on the principle of a fiber Bragg grating. It senses micro-strain along the fiber axis by detecting the drift of the reflected wavelength of the grating embedded in the fiber. In this embodiment, a fiber Bragg grating sensor with a center wavelength of 1550 nanometers is selected. Its strain measurement accuracy reaches ±1 micro-strain, and its response frequency is not less than 100 Hz, fully meeting the dynamic response requirements for hull deformation monitoring. The sensor is adhered to the hull structure surface using epoxy resin adhesive. Before adhesion, the steel surface must be polished and cleaned to ensure strain transfer efficiency. Each sensor is equipped with an independent temperature-compensated grating to eliminate thermal output interference caused by temperature changes.

[0052] Regarding the method for determining the installation locations, this embodiment uses the theoretical strain distribution obtained in the first step for scientific point placement. Firstly, the theoretical strain distribution describes the strain variation patterns in various regions of the entire ship under different load conditions. Its output includes the strain values ​​at each monitoring point and the strain gradient between adjacent regions. The strain gradient is defined as the rate of change of the strain value along the spatial direction, and its magnitude reflects the severity of strain change in that region. In this embodiment, the equivalent strain values ​​of all monitoring points on the entire ship under various load conditions are extracted from the results output by the finite element mechanical model. Then, the strain difference between each monitoring point and its surrounding monitoring points is calculated through post-processing, and then divided by the distance between the monitoring points to obtain the strain gradient value at the corresponding location. The strain gradient values ​​under all load conditions are statistically analyzed, and the average and maximum strain gradients of each monitoring point are calculated as the basis for determining the priority of sensor deployment.

[0053] In practical operation, the region where the strain gradient in the theoretical strain distribution exceeds a preset strain gradient threshold is selected as the sensor installation location. The preset strain gradient threshold can be determined based on the statistical characteristics of the strain distribution output by the finite element mechanical model. In this embodiment, the strain gradient values ​​of all monitoring points on the entire ship under all operating conditions are statistically analyzed, and the mean and standard deviation of the strain gradient values ​​are calculated. The sum of the mean strain gradient and one standard deviation is used as the preset strain gradient threshold. Regions where the strain gradient exceeds the preset strain gradient threshold are marked as high strain change areas, and sensors are preferentially deployed there. Regions where the strain gradient is lower than the preset strain gradient threshold are considered strain level areas, and the number of sensors can be reduced or no sensors can be deployed there. The reason for using the mean plus one standard deviation as the threshold is that, under the assumption of normal distribution, this threshold corresponds to approximately the 84th percentile, meaning that only about 16% of the region is marked as a high change area. This proportion ensures that key parts are fully monitored without leading to an excessive number of sensors, resulting in wasted costs and difficulties in installation and maintenance.

[0054] In other embodiments, distributed fiber optic sensing technology can be used instead of fiber Bragg gratings to select fiber strain sensors. Distributed fiber optic sensing uses ordinary communication optical fiber as the sensing medium and obtains strain information continuously distributed along the fiber by measuring the frequency change of backscattered light in the fiber. Its advantage is that it can continuously sense rather than measure discrete points, but it is more expensive and the demodulation equipment is complex. Regarding the method for determining the sensor installation location, the strain gradient threshold can also be set using other statistical methods, such as taking the 90th percentile of the strain gradient values ​​of all monitoring points, or directly specifying a fixed threshold based on engineering experience, such as 20 microstrain per meter. For different types of ships, such as oil tankers or container ships, their strain distribution characteristics are different, and the corresponding threshold setting can be adjusted according to the finite element analysis results of the specific ship type. In addition, besides surface bonding, for newly built ships, fiber optic sensors can also be pre-embedded inside the steel during the structural construction stage to achieve better protection and a longer service life.

[0055] Taking the 320-meter Cape-size bulk carrier in this embodiment as an example, the strain gradient statistics output by the finite element mechanical model show that the average strain gradient of the entire ship is 12 microstrains per meter, and the standard deviation is 6 microstrains per meter. Therefore, the preset strain gradient threshold is set to 18 microstrains per meter. Thus, the high strain variation areas selected according to this threshold are mainly concentrated in the following locations: the connection between each transverse bulkhead and the hatch coaming, the ends of the deck longitudinal girder, the corners of the hatch coaming, and the midship region. The strain gradient values ​​in these areas are generally between 20 and 35 microstrains per meter.

[0056] In terms of specific placement, this embodiment places one fiber optic strain sensor at the center of each high-strain-change zone, and a redundant fiber optic strain sensor at the edge of the zone to prevent data loss due to the failure of a single sensor. For particularly important areas, such as the four corners of the hatch coamings amidships, which experience the greatest bending moment changes during unloading, this embodiment places two backup fiber optic strain sensors at each corner, for a total of eight fiber optic strain sensors. Furthermore, to infer the continuous deformation field of the entire ship, a small number of fiber optic strain sensors are also placed in areas with gentle strain as reference sensors to verify the accuracy of displacement field interpolation. This embodiment places two reference sensors on the deck surfaces at the bow and stern, resulting in a total of 48 fiber optic strain sensors deployed throughout the ship, forming a fiber optic strain sensing array.

[0057] After the sensors are installed, initial calibration is required. This can be done by recording the initial wavelength values ​​of all sensors under no-load conditions, using them as a zero-point reference. Simultaneously, a static load test should be performed, where ballast water is injected into a specific compartment of the ship to simulate a certain load. The responses of each sensor are recorded and compared with the theoretical strain values ​​from the finite element model under this load condition to verify the sensor installation quality and system consistency. If the measured value of a sensor deviates from the theoretical value by more than 5%, the sensor's adhesion quality should be checked or it should be recalibrated.

[0058] Existing methods typically employ uniform or empirical placement of strain sensors at key locations, lacking quantitative basis and prone to issues like omissions or over-placement. Step two, however, uses the theoretical strain distribution output from the finite element model as a direct basis for sensor placement, employing strain gradient threshold quantification to achieve optimal sensor resource allocation. This physical model-based placement strategy ensures the highest density of monitoring data in regions of most dramatic strain changes, while maintaining appropriate sparseness in regions with moderate strain, maximizing the accuracy of deformation field reconstruction within a limited number of sensors.

[0059] Step three involves matching the dynamic strain data acquired in real time by the fiber optic strain sensor with the theoretical strain value. Based on the matching result, the actual three-dimensional displacement of each monitoring point at the current moment is determined from the correspondence between the theoretical strain value and the theoretical displacement value, generating an instantaneous displacement field. The core of this step lies in transforming the real-time sensed local strain information into a continuous deformation field covering the entire ship, realizing the transition from a physical model to real-time monitoring.

[0060] First, it needs to be clarified that dynamic strain data refers to the sequence of strain values ​​collected in real time by the fiber optic strain sensor array during the actual unloading process. In this embodiment, all 48 fiber optic strain sensors operate continuously at a sampling frequency of 20 times per second, outputting the current strain value of each sensor at each sampling moment. Since the sensors contain temperature-compensated gratings, the output data has been temperature-corrected and directly reflects the mechanical strain of the hull structure caused by load changes. The raw strain data is preprocessed after acquisition, including removing data that is significantly outside the reasonable range, such as abnormal data with strain values ​​exceeding ±1000 micro-strain, and using a moving average filter to suppress high-frequency noise. The filter window length is set to five sampling points, that is, the average value of five consecutive data acquisitions is taken as the current output value.

[0061] After data preprocessing, dynamic strain data can be matched with theoretical strain values. The purpose of this matching is to determine the actual load condition of the ship so that the corresponding theoretical displacement value can be extracted from a pre-established multidimensional lookup table. This embodiment uses distribution similarity as the matching criterion. The distribution similarity is calculated by constructing a strain vector from the dynamic strain data of all valid monitoring points at the current moment, with the dimension of this vector equal to the number of valid monitoring points. Simultaneously, the theoretical strain values ​​of the corresponding monitoring points under each load condition are extracted from the multidimensional lookup table established in the first step, forming the theoretical strain vector for that condition. For each load condition, the Euclidean distance between the current strain vector and the theoretical strain vector for that condition is calculated. A smaller Euclidean distance indicates a higher similarity between the two vectors. The Euclidean distance is calculated as the square root of the sum of the squares of the strain differences between the corresponding monitoring points. During the calculation, since different monitoring points may be located in different stress areas with significant differences in strain amplitude, data normalization is required. In this embodiment, the strain value of each monitoring point can be divided by its theoretical maximum strain value under all working conditions, so that the data of each monitoring point are on the same order of magnitude, avoiding the dominance of monitoring points with large amplitude in distance calculation.

[0062] After calculating the Euclidean distance for all operating conditions, the load condition corresponding to the minimum Euclidean distance is selected as the actual load condition of the current ship. To further improve the matching reliability, this embodiment sets up a confidence judgment mechanism: if the minimum Euclidean distance is less than the preset matching threshold, the matching is considered successful, and this operating condition is adopted as the current operating condition; if the minimum Euclidean distance is greater than or equal to the preset matching threshold, the current strain distribution is considered not close enough to any pre-stored operating condition, and an unmodeled load state may have occurred. In this case, interpolation is used. The preset matching threshold can be determined based on the noise level. In this embodiment, the noise standard deviation of the fiber optic strain sensor under static no-load conditions is statistically analyzed, and the Euclidean distance corresponding to three times the noise standard deviation is taken as the matching threshold. Specifically, under the static condition of the ship being unloaded and without operation, fiber optic strain sensor data is continuously collected for 60 seconds, obtaining a total of 1200 sampling points. The standard deviation of the strain value sequence measured by each fiber optic strain sensor is calculated. The standard deviations calculated from all 48 fiber optic strain sensors are distributed between 4 microstrain and 7 microstrain, and the average value of 5 microstrain is taken as the overall noise level. The Euclidean distance is a comprehensive measure of the difference between the measured strain vector and the theoretical strain vector. Its calculation involves the strain difference values ​​of all 48 sensors. When all sensors are at a noise level, the theoretical value of the Euclidean distance is approximately the square root of the sum of the squares of the noise, which is 5 microstrain multiplied by the square root of 48, approximately equal to 35 microstrain. Therefore, the preset matching threshold is set to 35 microstrain. When the Euclidean distance between the measured strain vector and the theoretical strain vector exceeds this threshold, it is determined to be an unmatched pre-stored condition.

[0063] After determining the current actual load condition, the actual three-dimensional displacement of each monitoring point at the current moment is determined from the correspondence between theoretical strain and theoretical displacement values ​​based on the matching results. Since the actual strain value is usually not exactly equal to the theoretical strain value, the theoretical displacement value cannot be directly used; instead, interpolation correction based on the current measured strain is required. This embodiment uses a linear interpolation method. For each monitoring point, the theoretical strain and theoretical displacement values ​​for that monitoring point under the current actual load condition are extracted from a multidimensional lookup table, and the theoretical strain and theoretical displacement values ​​for adjacent load conditions are also extracted. The selection of adjacent load conditions is based on the relative relationship between the current measured strain value and the theoretical strain value. For example, if the current measured strain value is between the theoretical strain values ​​for the 60% load condition and the 65% load condition, then these two load conditions are selected as interpolation benchmarks. Then, the interpolation coefficient is calculated. The interpolation coefficient is equal to the current measured strain value minus the theoretical strain value for the lower load condition, divided by the theoretical strain value for the higher load condition minus the theoretical strain value for the lower load condition. This interpolation coefficient is applied to the theoretical displacement value to obtain the actual three-dimensional displacement of the current monitoring point. Taking monitoring point number twelve as an example, the theoretical vertical displacement of this point under 60% load condition is 210 mm, and the theoretical vertical displacement under 65% load condition is 225 mm. The interpolation coefficient corresponding to the current measured strain value is 0.4. Therefore, the current actual vertical displacement is 210 mm plus 0.4 multiplied by 15 mm, which equals 216 mm. Similarly, the calculation method for longitudinal and lateral displacement is the same.

[0064] After performing the above interpolation calculations on all monitoring points, the actual three-dimensional displacement of each monitoring point at the current moment is obtained, i.e., the displacement values ​​at a set of discrete points. However, to generate an instantaneous displacement field reflecting the deformation of the entire ship, it is necessary to extend the displacements of these discrete points to the entire hull surface. This embodiment can use the radial basis function interpolation method to reconstruct the continuous displacement field from discrete monitoring points. The radial basis function is a function with distance as the independent variable, which can interpolate the value at any position based on the value of a known point. Specifically, the spatial coordinates of all monitoring points are used as known points, and the three-dimensional displacement calculated at each monitoring point is used as known values ​​to construct a radial basis function interpolation model. This embodiment selects the Gaussian function as the radial basis function, and its shape parameter is determined according to the average spacing of the monitoring points, so that the influence range of each monitoring point covers the area between adjacent monitoring points. By solving the linear equation system, the weight coefficient of each monitoring point is obtained, and then for any position on the hull, the displacement value at that position can be calculated by weighted summation. In the calculation, independent interpolation models were constructed for longitudinal, lateral, and vertical displacements. Taking vertical displacement as an example, the spatial coordinates of all 48 fiber optic sensors were used as known points, and the vertical displacement calculated by each sensor was used as known values. A Gaussian function was selected as the radial basis function, and the shape parameter was taken as the reciprocal of the average spacing between monitoring points. The weight coefficient of each monitoring point was obtained by solving a system of linear equations. The hull surface was discretized into grid points, with one point every 2 meters along the ship's length, for a total of 160 points; and one point every 2 meters along the ship's width, for a total of 25 points, forming 4000 grid points. For any given grid point, its coordinates are substituted into the radial basis function interpolation model to calculate the vertical displacement of that point. Taking a grid point in the middle of the hull as an example, with coordinates at 160 meters in length and 0 meters in width, the vertical displacement of this point is calculated to be 215 mm by weighted summation of the radial basis function values ​​of all monitoring points. Similarly, the longitudinal and lateral displacements are calculated in the same way, yielding a longitudinal displacement of 17 mm and a lateral displacement of 4 mm for this grid point. Finally, by traversing all 4000 grid points, the instantaneous displacement field of the entire ship at the current moment is obtained. To ensure the accuracy and stability of the radial basis function interpolation, a regularization parameter can be added when constructing the interpolation model. In this embodiment, the regularization parameter is set to 0.01. Finally, the displacement field is stored in the form of a three-dimensional array with dimensions of 160×25×3, and each array element stores the three displacement components of the grid point. The generation frequency of the instantaneous displacement field is consistent with the sensor sampling frequency, i.e., 20 times per second, to achieve real-time tracking of hull deformation. To verify the interpolation accuracy, a comparison can be made at the reference sensors at the bow and stern: the measured displacement of the reference sensor is compared with the predicted displacement of the interpolation model at that location, and if the deviation exceeds 5 mm, the model calibration process is triggered.

[0065] In this embodiment, the specific implementation of the model calibration process involves the displacement field calculation module continuously monitoring the deviation between the dynamic strain data and the theoretical strain value at each monitoring point, recording the deviation value after N consecutive acquisitions. N is set to ten times, corresponding to a 0.5-second time window. This value is selected based on the following considerations: ten consecutive acquisitions can effectively filter out single-shot noise interference, while the 0.5-second delay is acceptable during unloading. The preset deviation threshold is set to 5% of the theoretical strain value. The preset deviation threshold can be determined based on the statistical results of sensor measurement accuracy and finite element model calculation error, ensuring that deviations exceeding the preset deviation threshold have a greater than 95% probability of originating from real physical changes rather than measurement noise. When the deviation of ten consecutive acquisitions exceeds 5%, the displacement field calculation module triggers a calibration request. The actual loading record at the current moment is obtained from the ship's loading computer, including parameters such as cargo weight, draft, and ballast water distribution in each compartment. Using these actual loading records as input, the finite element mechanical model is rerun to calculate the theoretical strain and theoretical displacement values ​​under the current real load conditions, and the data for the corresponding working condition in the multidimensional lookup table is updated. After the update is complete, the new data will replace the old data for subsequent matching. The calibration process is completed within 30 seconds and will not affect the normal unloading operation.

[0066] In other embodiments, cosine similarity can be used instead of Euclidean distance to calculate distribution similarity. Cosine similarity focuses on the directional consistency of vectors rather than their magnitude, making it more suitable for cases where the magnitude is scaled but the distribution shape is similar. Cubic spline interpolation or kriging interpolation can be used instead of linear interpolation for interpolation, providing higher interpolation accuracy when the monitoring points are densely distributed. The type of radial basis function can also be selected based on the actual data characteristics, such as using multiple quadratic functions or thin-plate spline functions. For the meshing of points on the hull surface, the mesh density can be adjusted according to computational resources. If a higher resolution displacement field is required, the mesh can be densified to a one-meter spacing; if computational resources are limited, a five-meter spacing can be used. Furthermore, the update frequency of the instantaneous displacement field can be dynamically adjusted according to the unloading speed. The frequency can be reduced to decrease computational load when changes are slow in the early stages of unloading, and increased for finer tracking when changes are drastic at the end of unloading.

[0067] Step three achieves indirect yet continuous full-field displacement sensing through the physical conversion of strain to displacement. Using the finite element model as a bridge, the continuous deformation field of the entire hull can be deduced from a small number of discrete strain points, overcoming the limitations of direct measurement, which only obtains a limited number of points and is susceptible to environmental interference. Simultaneously, by combining distribution similarity matching with linear interpolation, the theoretical basis of the finite element model is utilized while preserving the dynamic information of real-time strain data, achieving an organic integration of theory and practice.

[0068] Step four involves correcting the transformation matrices of multiple 3D scanners installed on the hatch coaming, quay crane, and unloader relative to a virtual global coordinate system based on the instantaneous displacement field. This ensures that the spatial pose of each 3D scanner relative to the virtual global coordinate system updates in real time following the ship's deformation. The core of this step is to convert the instantaneous displacement field generated in step three into executable scanner reference correction instructions, eliminating the inconsistency in multi-sensor data references caused by ship deformation.

[0069] First, it's important to clarify that the virtual global coordinate system is an abstract spatial reference framework used to unify point cloud data from 3D scanners at different locations. In this embodiment, the virtual global coordinate system can be established using a fixed point at the dock as the origin. This fixed point is selected at the end of the quay crane's seaside track, a position that will not shift due to changes in ship load throughout the operation. The direction of the quay crane track is taken as the positive X-axis, pointing towards the bow of the ship; the direction perpendicular to the track pointing towards the seaside is taken as the positive Y-axis; and the vertically upward direction is taken as the positive Z-axis. The reference point for establishing this coordinate system remains unchanged throughout the entire operation cycle, ensuring that all corrected point cloud data can be fused within the same absolute spatial coordinate system.

[0070] In other embodiments, the method of establishing the virtual global coordinate system can be adjusted according to the specific layout of the wharf. For example, if the wharf has a permanent measurement reference pier, the reference pier can be used as the origin of the coordinate system. If the quay crane track has a long-distance gradient change, a dynamic reference surface that changes along the track can be established instead of a single fixed plane. The threshold setting of the correction trigger mechanism can also be adjusted according to the actual operation requirements. For unloading operations with higher precision requirements, the displacement threshold can be reduced to five millimeters; for rough operations, it can be increased to twenty millimeters to reduce the computational load. In addition, for the 3D scanner installed on the grab bucket of the unloader, its transformation matrix correction also needs to integrate the opening and closing state of the grab bucket and the swing angle information of the wire rope. This information can be obtained in real time from the unloader control system and superimposed on the displacement field correction in this step.

[0071] The transformation matrix is ​​a mathematical expression describing the spatial transformation relationship between the coordinate system of each 3D scanner and the virtual global coordinate system. After installation, each 3D scanner obtains its initial transformation matrix through on-site calibration. The calibration process can adopt the cooperative target method, in which multiple target points with known virtual global coordinates are arranged at a fixed position on the dock. The scanner to be calibrated scans these target points to obtain their coordinates in the scanner coordinate system, and the initial transformation matrix is ​​obtained by solving the point set registration problem. Taking a 3D scanner installed on the hatch coaming as an example, the initial transformation matrix of the scanner is denoted as T, which contains a 3x3 initial rotation matrix R0 and a 3D translation vector t. This initial transformation matrix indicates that, under the undeformed state of the hull, point P in the hatch coaming scanner coordinate system is transformed to point P1 in the virtual global coordinate system, satisfying P1 = R0 × P + t.

[0072] When the ship's hull deforms due to cargo unloading, the 3D scanner on the hatch coaming undergoes spatial displacement along with the ship's structure at its location. If the initial transformation matrix is ​​still used to transform the scanner's point cloud to the virtual global coordinate system, the transformed point cloud position will deviate from its true spatial position, causing it to misalign with the point cloud of the quay crane scanner. Therefore, it is necessary to dynamically correct the transformation matrix based on the instantaneous displacement field.

[0073] In this embodiment, the specific operation method for correction is as follows: The actual three-dimensional displacement and local torsion angle at each 3D scanner installation location are extracted from the instantaneous displacement field generated in step three. Taking the 3D scanner on the hatch coaming as an example, the installation location of this scanner corresponds to a specific spatial point in the finite element mechanical model, with coordinates (x, y, z). In the instantaneous displacement field, the three-dimensional displacement at this location is calculated using the radial basis function interpolation method, including longitudinal displacement Δx, lateral displacement Δy, and vertical displacement Δz. Simultaneously, the instantaneous displacement field also provides the local torsion angle at this location, including the torsion angle θx around the longitudinal axis, the bending angle θy around the transverse axis, and the deflection angle θz around the vertical axis. These torsion angles are obtained by taking the spatial derivative of the displacement field. Specifically, the torsion angle around the longitudinal axis is equal to the rate of change of the lateral displacement along the ship's width direction, and the bending angle around the transverse axis is equal to the rate of change of the vertical displacement along the ship's length direction.

[0074] After obtaining the displacement and torsion angles, the transformation matrix is ​​then corrected. The correction consists of two parts: translational component correction and rotational component correction. The translational component correction is directly equal to the three-dimensional displacement at the scanner's mounting position; that is, the new translation vector t1 equals the initial translation vector t plus the displacement vector (Δx, Δy, Δz). The rotational component correction is achieved by constructing an incremental rotation matrix R', which is built from the local torsion angles (θx, θy, θz) in the ZYX order of three-dimensional rotation. The new rotation matrix R1 equals the incremental rotation matrix R' multiplied by the initial rotation matrix R0. Finally, the corrected transformation matrix T1 consists of R1 and t1.

[0075] For 3D scanners mounted on quay cranes, since the quay cranes themselves are fixed to the dock, their installation position does not shift with hull deformation. Therefore, when the position of the quay crane relative to the virtual global coordinate system remains unchanged, the correction amount of its transformation matrix is ​​zero, which can be considered equivalent to requiring no additional correction. However, it should be noted that the point clouds of the quay crane scanner and the scanner on the hatch coaming will eventually be merged in the same virtual global coordinate system, so the point cloud transformation process of the quay crane scanner remains unchanged. For scanners mounted on unloader trolleys, the transformation matrix correction needs to consider two levels of motion: first, the movement of the trolley itself, whose position is provided in real time by the unloader control system; second, the reference displacement caused by hull deformation, which is also corrected by extracting the displacement and torsion angle corresponding to the current working position of the trolley from the instantaneous displacement field. The correction method is the same as that for the hatch coaming scanner, that is, based on the known spatial position of the trolley, the hull deformation displacement at that position is superimposed.

[0076] The triggering conditions for correction operations need to balance computational load and model following accuracy. This embodiment can set two triggering mechanisms. The first is based on a displacement threshold: when the 3D displacement of any monitoring point in the instantaneous displacement field exceeds a preset displacement threshold, the correction of all scanner transformation matrices is triggered. The preset displacement threshold is set to 10 mm, which can be determined based on the nominal accuracy of the 3D scanner. Typically, the point cloud accuracy of a scanner is five to eight millimeters; therefore, if the deformation exceeds 10 mm and no correction is made, the point cloud fusion error will exceed the scanner's own accuracy. The second is based on the rate of change: when the rate of change of the dynamic strain data collected by any fiber optic strain sensor exceeds a preset rate of change threshold, correction is triggered. The preset rate of change threshold is set to 50 microstrains per second, which can be determined based on the time of one material handling cycle of the unloader's grab bucket. One grab bucket cycle is approximately 40 to 60 seconds. A rate of change of 50 microstrains per second corresponds to a strain change of 2000 to 3000 microstrains within one cycle, sufficient to cause a sufficiently significant displacement change that needs to be followed promptly.

[0077] When the trigger condition is met, all transformation matrices of the scanners requiring correction are immediately updated. The update process typically takes five milliseconds to ensure that the next frame of point cloud data can be transformed using the latest transformation matrix. Furthermore, to avoid noise introduced by frequent corrections, the corrected transformation matrix remains in effect until the next trigger condition is met.

[0078] Existing methods typically assume that the transformation relationship of all scanners relative to the global coordinate system remains constant, or perform attitude compensation solely through the inertial measurement units (IMUs) on the scanners themselves. Attitude compensation can only correct the vibration of the scanners themselves, and cannot correct the overall displacement of the mounting reference caused by hull deformation. However, step four introduces the instantaneous displacement field into the transformation matrix correction process, enabling the spatial pose of the hull-related scanners relative to the virtual global coordinate system to dynamically follow the hull deformation. This achieves the unification of references for multiple sensor sources. The correction does not involve modifying the scanner hardware, but rather an intelligent reconstruction of the data conversion process. It does not increase hardware costs, yet it fundamentally eliminates reference drift errors.

[0079] Taking the actual scenario in this embodiment as an example, during the unloading process, the midsection of the ship gradually rises, and the scanner on the hatch coaming moves up by 10 centimeters accordingly. Without correcting the transformation matrix, the point cloud output by the scanner will show the material pile position as 10 centimeters higher than the actual position in the virtual global coordinate system, resulting in a 10-centimeter vertical misalignment with the point cloud of the quay crane scanner. After this correction step, the translation component in the scanner's transformation matrix increases by 10 centimeters, pulling its point cloud back to its actual position and perfectly aligning it with the point cloud of the quay crane scanner. Similarly, if the ship undergoes torsional deformation, resulting in a 5-millimeter relative height difference between the left and right sides of the hatch coaming, the correction of the rotation component will cause the scanner coordinate system to rotate accordingly, ensuring that the point cloud attitude is consistent with the actual ship attitude.

[0080] Step five involves transforming the point cloud data collected by multiple 3D scanners into a virtual global coordinate system using a corrected transformation matrix. The transformed point clouds are then fused and registered to construct a 3D model of the cargo hold's internal load. The core of this step lies in converging the results of previous steps into the final 3D reconstruction, utilizing a dynamically corrected transformation matrix to eliminate reference deviations caused by hull deformation, thereby generating a digital model of the cargo hold's internal load that strictly conforms to the real physical world.

[0081] First, it's important to clarify that point cloud data refers to a set of discrete spatial points output by a 3D scanner. Each point contains 3D coordinate information, and some scanners also output reflection intensity or color information. In this embodiment, three 3D scanners were deployed: one installed in the middle of the seaside crossbeam of the quay crane to provide an overview of the entire ship's cabin; and two installed at the fore and aft ends of the hatch coaming to collect data from different angles into blind spots within the cabin. All scanners used phase-detection lidar with a scanning frequency of 200,000 points per second, a ranging accuracy of ±5 mm, and a field of view of 360 degrees horizontally multiplied by 120 degrees vertically. The scanners operated continuously during the operation, outputting one frame of point cloud data after each full-field scan, with a frame rate set to 1 frame per second. Each frame of point cloud data contains approximately 500,000 points, and the three scanners together generate 1.5 million point cloud data points per second.

[0082] After acquisition, the point cloud data is first stored locally and simultaneously transmitted in real time to the corresponding central processing unit of the system. Since the scanner and the central processing unit are usually connected via industrial Ethernet, and the point cloud data volume is large, this embodiment can adopt a data compression transmission method to losslessly compress the original point cloud data before transmission. The compression ratio is about 30%, which effectively reduces network bandwidth consumption. After receiving the data, the central processing unit decompresses it to restore the original point cloud data.

[0083] Next, coordinate transformation of the point cloud data is performed. For each frame of point cloud, the corrected transformation matrix corresponding to the scanner at the current moment is obtained based on its source scanner. This transformation matrix is ​​calculated in real time in step four and stored in shared memory. Its update timestamp strictly corresponds to the point cloud acquisition timestamp, ensuring that each point cloud frame is transformed using the transformation matrix closest to its acquisition time.

[0084] Because the scanner frame rate is 1 frame per second, while the instantaneous displacement field updates at 20 frames per second, there is a difference in temporal resolution. In this embodiment, a timestamp with millisecond precision is added to each scanner point cloud frame at the time of acquisition. Each time the transformation matrix correction module generates a new corrected transformation matrix, it simultaneously records the timestamp corresponding to that matrix, i.e., the moment the correction is triggered. When processing point cloud frames, the point cloud fusion module searches for the transformation matrix with a timestamp less than and closest to the point cloud frame's timestamp in shared memory for conversion. If the difference between the point cloud frame's timestamp and the latest transformation matrix's timestamp exceeds 100 milliseconds, a linear interpolation method is used to calculate the transformation matrix at the intermediate moment based on the two transformation matrices, ensuring that the time alignment error between the point cloud frame and the transformation matrix is ​​less than 5 milliseconds.

[0085] For each point in the point cloud, the transformation process multiplies the coordinates in the scanner coordinate system with the scanner's initial rotation matrix, and adds the translation vector to obtain the coordinates of the point in the virtual global coordinate system. This coordinate transformation is then performed on every point in every frame of the point cloud from all scanners to generate three sets of point cloud data unified in the virtual global coordinate system.

[0086] After coordinate transformation, although the three sets of point clouds are in the same coordinate system, there may still be misalignment or non-overlapping areas between them due to differences in scanning angle, scanner position, and scanning time. At this point, point cloud fusion and registration are required to precisely align and merge the three sets of point clouds into a complete point cloud of the cargo hold. This embodiment can employ a two-step registration strategy: coarse registration and fine registration.

[0087] In the coarse registration stage, the corrected transformation matrix provided in step four is used as the initial registration parameter. Since the corrected transformation matrix has eliminated the reference drift caused by hull deformation, the point clouds of the hatch coaxial scanner and the quay crane scanner are roughly aligned in the virtual global coordinate system. However, there may still be a small amount of residual error, such as minor deviations caused by scanner installation angle measurement errors or scanner clock drift. The purpose of coarse registration is to further reduce these residual errors. Specifically, the point cloud of the quay crane scanner is used as the reference point cloud, the point cloud of the hatch coaxial scanner is used as the point cloud to be registered, and the current transformation matrix of the point cloud to be registered is used as the initial value. A fast registration algorithm based on point cloud features, such as the sample consistency initial registration algorithm, is adopted. The registration algorithm extracts fast point feature histogram descriptors from the two point clouds, finds point pairs with similar features as corresponding points, and then solves for a more accurate transformation matrix based on these corresponding points. After coarse registration, the average distance error between the hatch coaxial point cloud and the quay crane point cloud is reduced from the corrected 5-8 mm to 2-3 mm.

[0088] In the fine registration stage, the Iterative Closest Point Algorithm (ILAB) is used to further optimize the point cloud after coarse registration. The basic principle of the ILAB is: for each point in the point cloud to be registered, find the nearest point in the reference point cloud as the corresponding point, and then solve for an optimal transformation matrix based on all corresponding point pairs to minimize the sum of squared distances between point pairs. This process is iteratively executed until convergence. In this embodiment, the convergence condition of the ILAB is set as follows: the translation change of the transformation matrix between two iterations is less than 0.1 mm and the rotation change is less than 0.001 degrees, or the number of iterations reaches 50. Since the coarse registration has already provided a good initial alignment, the ILAB usually converges within 10 iterations. After fine registration, the average distance error between the hatch cofferdam point cloud and the quay crane point cloud is reduced to less than 1 mm, reaching the limit of the scanner's own accuracy.

[0089] After pairwise registration, the three point clouds are merged into a single complete point cloud. During the merging process, there may be issues with excessively high point cloud density in overlapping areas, necessitating point cloud downsampling. In this embodiment, a voxel filter can be used for downsampling, dividing the virtual global coordinate system space into a 5mm cube grid. Only one point is retained within each grid, and the coordinates of this point are taken as the centroid coordinates of all points within that grid. After downsampling, the total number of point clouds is reduced from 1.5 million to approximately 300,000 points, preserving the main features of the material pile surface while significantly reducing the computational burden of subsequent 3D modeling.

[0090] The downsampled point cloud is the original 3D point cloud model of the material pile inside the hopper. However, the point cloud model is composed of discrete points, which is not convenient for direct volume calculation or visualization. Therefore, it is necessary to further construct a continuous 3D surface model of the material pile. This embodiment uses the Poisson surface reconstruction algorithm to convert the point cloud into a triangular mesh model. The basic principle of this algorithm is: taking the normal vector information of the point cloud as input, solving a Poisson equation, and obtaining an indicator function. The isosurface of this indicator function in space is the reconstructed surface. The normal vector information of the point cloud is estimated from the local neighborhood of the point cloud through principal component analysis. The depth parameter of the Poisson reconstruction is set to eight, corresponding to a mesh resolution of approximately five centimeters, meaning that the reconstructed triangular mesh model can distinguish surface concavity and convexity features at a scale of five centimeters. For areas with low point cloud density or voids, the Poisson algorithm can automatically fill the smooth surface according to the distribution of surrounding points, avoiding the appearance of holes.

[0091] In other embodiments, the point cloud fusion and registration method can be adjusted according to the actual hardware configuration. For example, if a higher-precision scanner is used, the convergence threshold of the iterative nearest-point algorithm can be reduced accordingly. If the number of scanners increases, a multi-path parallel registration strategy can be adopted, first registering pairs and then optimizing globally. The depth parameters of the Poisson reconstruction can be dynamically adjusted according to the point cloud density; higher resolution reconstruction is used for dense point cloud regions, while the resolution is reduced for sparse regions to avoid generating false surfaces. For applications with higher real-time requirements, the Poisson reconstruction step can be omitted, and the downsampled point cloud can be used directly for volume calculation, sacrificing some visualization effects for computational speed. In addition, the compression and transmission method of point cloud data can also be flexibly selected according to network conditions; compression can be turned off when network bandwidth is sufficient to reduce the burden on the central processing unit.

[0092] After surface reconstruction, a triangular mesh model of the cargo pile inside the hold is obtained, i.e., a three-dimensional cargo pile model. This model consists of hundreds of thousands of triangular facets, each containing the spatial coordinates of three vertices and the normal vector of the face. This model can be used for a variety of subsequent applications: calculating the cargo pile volume through integration, analyzing the distribution of the cargo pile at different depths through slicing, generating a three-dimensional visualization image for operator reference through rendering, or directly inputting it into an automated unloading system to guide grab bucket path planning.

[0093] Taking the actual unloading scenario in this embodiment as an example, when unloading reaches 50%, the material pile inside the hold presents a cone-shaped distribution, higher in the middle and lower around the edges. The quay crane scanner, viewed from above, can completely capture the cone's apex area, but there is a blind spot near the bottom of the cone close to the hold wall; the scanners at both ends of the hatch coaming supplement the point cloud data near the hold wall from the side. After correction and transformation matrix conversion, the three sets of point clouds naturally connect in the virtual global coordinate system, with the point clouds in the overlapping areas basically overlapping. After fine registration using the iterative nearest-point algorithm, the point cloud error in the overlapping areas is less than 1 mm. The triangular mesh model generated after Poisson reconstruction clearly presents the texture details of the material pile surface, including the pits left by the grab bucket and the slopes formed by the material sliding down. By performing difference calculations between this model and the model from the previous moment, the unloading volume of this material handling cycle can be accurately determined, with an error of less than 1.5% compared to the reading of the grab bucket weighing system.

[0094] To verify the model's accuracy, this embodiment can also include a verification process. For example, before the unloading operation begins, a 3D scanner is used to scan the empty hold and construct a baseline model of the empty hold. During the unloading process, the real-time constructed stockpile model is compared with the empty hold baseline model to calculate the volume of the unloaded area and compare it with the cumulative unloading volume of the unloader. The deviation between the two should be within 3%. If the deviation exceeds 3%, a system check is triggered. Possible causes include scanner calibration failure, strain sensor malfunction, or the need for finite element model calibration.

[0095] In summary, traditional methods are hampered by the fact that the ship's hull, as a large-scale flexible structure, continuously undergoes elastic deformation during unloading. This causes scanners installed at locations such as hatch coamings to move synchronously with the hull, while fixed scanners on the quay crane remain stationary. The spatial transformation relationship between the two, which should be fixed, is dynamically disrupted, and directly measuring this relative displacement faces the dual challenges of accuracy and occlusion. Our proposed method, however, introduces a finite element method (FEM) model as a physical knowledge engine. It pre-determines the deep correlation between strain and displacement, determined by structural mechanics laws, through simulation calculations. This transforms the continuous displacement field of the entire ship, which cannot be directly measured, into discrete strain values ​​at a few key points that are easily measured with high precision. When fiber optic strain sensors capture the dynamic strain at these key points in real time, the measured strain information is fed back into the pre-established physical model through distribution similarity matching and linear interpolation. The precise three-dimensional displacement of each position on the ship at the current moment is then deduced. This three-dimensional displacement is subsequently used to dynamically correct the transformation matrices of all scanners, ultimately ensuring that the point clouds from different scanners are precisely aligned in the virtual global coordinate system. This effectively eliminates the interference of hull deformation on 3D modeling and reduces the probability of accidents during ship unloading.

[0096] Example 2

[0097] Combination Figure 1 and Figure 2 This embodiment provides a shipborne three-dimensional material stack dynamic modeling system to implement the shipborne three-dimensional material stack dynamic modeling method disclosed in Embodiment 1. The system can be deployed at the bulk cargo terminal operation site and consists of hardware equipment installed on ships and quay cranes and software modules deployed in the central control room. The system includes a finite element modeling module, a fiber optic strain sensor array, a displacement field calculation module, a transformation matrix correction module, and a point cloud fusion module.

[0098] The finite element modeling module is installed on the server in the central control room and is used to build the finite element mechanical model of the ship. The input to the module is the ship's design drawings and material property parameters. Its output is the correspondence between the theoretical strain and theoretical displacement values ​​under different load conditions, as well as the theoretical strain distribution of the ship. This module completes the modeling work once before the system is put into operation, storing the calculation results in a database for other modules to use. The finite element modeling module can use commercial finite element analysis software as its calculation engine, and through secondary development, it can achieve batch calculations under various load conditions and automatic result extraction.

[0099] The fiber optic strain sensing array consists of multiple fiber optic strain sensors, which are installed on the ship's hatch coaming and deck girder, which are determined according to the theoretical strain distribution. The fiber optic strain sensing array is used to collect dynamic strain data of the ship's hull during actual operations caused by cargo unloading. Each sensor is connected to a fiber optic grating demodulator via fiber optic patch cord. The demodulator converts the optical signal into a digital signal and transmits it to the central control room via industrial Ethernet.

[0100] The displacement field calculation module is deployed in the real-time server in the central control room and connected to the fiber optic strain sensor array. The displacement field calculation module receives the dynamic strain data output by the demodulator, reads the correspondence between the theoretical strain value and the theoretical displacement value pre-stored in the finite element modeling module from the database, matches the dynamic strain data with the theoretical strain value, determines the actual three-dimensional displacement of each monitoring point at the current moment based on the matching result, and generates the instantaneous displacement field.

[0101] The transformation matrix correction module is connected to the displacement field calculation module and is also deployed on the real-time server. The transformation matrix correction module receives the instantaneous displacement field output by the displacement field calculation module and corrects the transformation matrices of multiple 3D scanners installed on the hatch coaming, quay crane, and ship unloader relative to a virtual global coordinate system based on this displacement field. The corrected transformation matrices are written to shared memory in real time for use by the point cloud fusion module. The transformation matrix correction module simultaneously monitors the displacement and strain data change rate in the instantaneous displacement field. When the 3D displacement of any monitoring point exceeds a preset displacement threshold or the change rate of the dynamic strain data exceeds a preset change rate threshold, a new round of correction calculation is automatically triggered.

[0102] The point cloud fusion module is connected to the transformation matrix correction module and multiple 3D scanners. The point cloud fusion module receives raw point cloud data from each 3D scanner in real time via industrial Ethernet, reads the corrected transformation matrix corresponding to each scanner from shared memory at the current moment, and transforms all point cloud data into a virtual global coordinate system according to the corrected transformation matrix. After the transformation is complete, the point cloud fusion module performs fusion registration on the transformed point cloud, including two stages: coarse registration and fine registration, ultimately constructing a 3D model of the material pile inside the hold. The completed 3D model is output in standard point cloud format or triangular mesh format to the unloader control system, inventory management system, and visualization monitoring platform.

[0103] The fiber optic strain sensor array can be arranged with optimized spacing. Specifically, the fiber optic strain sensor array is continuously arranged along the longitudinal direction of the hull. The spacing between adjacent fiber optic strain sensors is determined based on the strain distribution curvature output from the finite element mechanical model, and the spacing decreases as the strain distribution curvature increases. The strain distribution curvature reflects the degree of strain change along the longitudinal direction of the hull. Regions with high curvature indicate rapid strain changes, requiring a denser sensor arrangement to capture details; regions with low curvature indicate gradual strain changes, allowing for a more appropriate increase in sensor spacing to reduce costs.

[0104] Taking the 320-meter Cape-size bulk carrier in this embodiment as an example, the strain distribution curvature calculated by the finite element mechanical model reaches its maximum value of 0.35 microstrains per meter in the midship region, and the spacing between fiber optic strain sensors in this region is set to three meters. In the bow and stern regions, the strain distribution curvature is smaller, approximately 0.05 microstrains per meter, and the sensor spacing is correspondingly increased to 15 meters. This non-uniform arrangement method reduces the number of fiber optic strain sensors from eighty required for a uniform arrangement to forty-eight while ensuring the accuracy of deformation field reconstruction, thus reducing system cost and installation and maintenance workload.

[0105] The finite element modeling module also features segmented modeling. For ultra-large ships, a unified modeling of the entire ship could result in an excessively large number of computational meshes, impacting the efficiency of batch calculations under multiple operating conditions. Therefore, the finite element modeling module divides the finite element mechanical model into multiple segmented sub-models based on the compartment divisions and bulkhead locations in the ship's design drawings. Each segmented sub-model corresponds to a hull segment between two adjacent transverse bulkheads. The boundary conditions of the segmented sub-models are determined based on the interaction between adjacent segments, achieving mechanical coupling between segments through the transfer of forces and displacements. Each segmented sub-model independently calculates the correspondence between its theoretical strain and theoretical displacement values ​​under various load conditions. The calculation results of each segmented sub-model are then stitched together at the boundaries to form a complete strain-displacement correspondence for the entire ship.

[0106] Taking the 320-meter Cape-size bulk carrier in this embodiment as an example, the ship has seven cargo holds, thus it is divided into seven sub-models. The number of meshes in each sub-model is approximately one-fifth of the total ship model. The seven sub-models are computed in parallel, and the total computation time is only one-third of that of the serial computation of the entire ship model, thus improving modeling efficiency. At the same time, the sub-models facilitate subsequent detailed analysis of local deformations of individual compartments. For example, when the unloading speed of a certain compartment is abnormal, only the sub-model corresponding to that compartment can be locally calibrated without recalculating the entire ship.

[0107] The data interaction between the modules of this system adopts a standardized interface design. The output data of the finite element modeling module is stored in HDF5 format. The displacement field calculation module, the transformation matrix correction module and the point cloud fusion module achieve high-speed data exchange through shared memory and zero-copy technology to ensure the real-time performance of the entire system. The system is also equipped with a health management module to monitor the working status of each sensor and the computational load of each module in real time. When an abnormality is detected, redundant equipment is automatically switched and an alarm is issued.

[0108] Since this system uses a shipborne three-dimensional material stack dynamic modeling method from Example 1, it has the same effect, which will not be repeated here.

[0109] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the aforementioned scope.

[0110] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic modeling of a shipborne three-dimensional stockpile, characterized in that, include: A finite element mechanical model of the ship is established. The finite element mechanical model is constructed based on the ship design drawings and material properties. It is used to simulate the longitudinal bending and transverse torsional deformation of the ship under different load conditions. The theoretical strain value and theoretical displacement value of multiple preset monitoring points under different load conditions are calculated to obtain the theoretical strain distribution of the ship. The installation location of the fiber optic strain sensor is determined based on the theoretical strain distribution, and multiple fiber optic strain sensors are installed. The fiber optic strain sensors are used to collect dynamic strain data of the ship hull during actual operation caused by cargo unloading in real time. The dynamic strain data collected in real time by the fiber optic strain sensor is matched with the theoretical strain value. Based on the matching result, the actual three-dimensional displacement of each monitoring point at the current moment is determined from the correspondence between the theoretical strain value and the theoretical displacement value, and an instantaneous displacement field is generated. Based on the instantaneous displacement field, the transformation matrix of each of the multiple 3D scanners installed on the hatch coaming, quay crane and unloader relative to the virtual global coordinate system is corrected so that the spatial pose of each 3D scanner relative to the virtual global coordinate system is updated in real time with the deformation of the hull. Point cloud data collected by multiple 3D scanners are transformed into a virtual global coordinate system according to the corrected transformation matrix. The transformed point clouds are then fused and registered to construct a 3D model of the material stack inside the cabin.

2. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, When matching dynamic strain data with theoretical strain values, the distribution similarity between dynamic strain data and theoretical strain values ​​under each load condition is calculated. The load condition corresponding to the maximum distribution similarity is selected as the actual load condition of the current ship, and the theoretical displacement value under the corresponding load condition is used as the initial displacement estimate.

3. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, The trigger condition for correcting the transformation matrix is ​​that the actual three-dimensional displacement of any monitoring point in the instantaneous displacement field exceeds the preset displacement threshold, or the rate of change of the dynamic strain data collected by the fiber optic strain sensor exceeds the preset rate of change threshold.

4. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, The method further includes: After establishing the correspondence between theoretical strain values ​​and theoretical displacement values, when the deviation between the dynamic strain data collected N times and the theoretical strain values ​​all exceed the preset deviation threshold, the finite element mechanical model is calibrated according to the actual loading records of the ship, and the correspondence is recalculated and updated, where N is a preset positive integer.

5. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, When determining the installation location of the fiber optic strain sensor based on the theoretical strain distribution, the region in the theoretical strain distribution where the strain gradient exceeds a preset strain gradient threshold is selected as the installation location. The strain gradient threshold is determined based on the statistical characteristics of the theoretical strain distribution output by the finite element mechanical model.

6. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, The virtual global coordinate system is established with the fixed point of the dock as the origin and the direction of the quay crane track as the coordinate axis. The transformation matrix includes translation and rotation components. The correction amount of the translation component is equal to the three-dimensional displacement at the corresponding installation position of the 3D scanner in the instantaneous displacement field, and the correction amount of the rotation component is equal to the local torsion angle at the corresponding installation position of the 3D scanner in the instantaneous displacement field.

7. The method for dynamic modeling of a shipborne three-dimensional stockpile according to claim 1, characterized in that, When performing fusion registration on the transformed point clouds, the corrected transformation matrix is ​​used as the initial registration parameter, and a point cloud fine registration algorithm is used to register point clouds from different 3D scanners.

8. A shipborne three-dimensional material stack dynamic modeling system, used to implement the shipborne three-dimensional material stack dynamic modeling method according to any one of claims 1-7, characterized in that, The system includes: The finite element modeling module is used to establish a finite element mechanical model of a ship, simulate the longitudinal bending and transverse torsional deformation of the ship under different load conditions, calculate the correspondence between the theoretical strain values ​​and theoretical displacement values ​​of multiple preset monitoring points under different load conditions, and output the theoretical strain distribution of the ship. The fiber optic strain sensor array is installed at a location determined based on the theoretical strain distribution to collect dynamic strain data of the ship hull during actual operations caused by cargo unloading in real time. The displacement field calculation module is used to match dynamic strain data with theoretical strain values, and determine the actual three-dimensional displacement of each monitoring point at the current moment based on the correspondence between theoretical strain values ​​and theoretical displacement values ​​according to the matching results, thereby generating an instantaneous displacement field. The transformation matrix correction module is used to correct the transformation matrix of each of the multiple 3D scanners installed on the hatch coaming, quay crane and unloader relative to a virtual global coordinate system based on the instantaneous displacement field, so that the spatial pose of each 3D scanner relative to the virtual global coordinate system is updated in real time with the deformation of the hull. The point cloud fusion module is used to transform point cloud data collected by multiple 3D scanners into a virtual global coordinate system according to the corrected transformation matrix, and then perform fusion and registration on the transformed point cloud to construct a 3D model of the material stack inside the cabin.

9. A shipborne three-dimensional material stack dynamic modeling system according to claim 8, characterized in that, The fiber optic strain sensor array is arranged continuously along the longitudinal direction of the hull, and the spacing between adjacent fiber optic strain sensors is determined based on the strain distribution curvature output by the finite element mechanical model.

10. A shipborne three-dimensional material stack dynamic modeling system according to claim 8, characterized in that, The finite element modeling module is also used to divide the finite element mechanical model into multiple sub-models based on the compartment division and compartment location in the ship design drawings. Each sub-model corresponds to the hull section between two adjacent transverse bulkheads, and calculates the correspondence between the theoretical strain value and the theoretical displacement value of each sub-model under different load conditions.

Images