FPSO hull structure fatigue life real-time prediction method and system

Abstract

The application provides a kind of FPSO hull structure fatigue life real-time prediction method and system, it is related to ship structure health monitoring technical field, the method includes: obtaining the real-time six degrees of freedom pose data and sparse sensor measured stress data of FPSO hull;Pose data is input into deep residual regression network, and the predicted stress characteristic coefficient is obtained;According to the measured stress data, the predicted characteristic coefficient is fused and corrected;Inverse principal component analysis transformation is carried out to the corrected characteristic coefficient, and the real-time full-field stress data of the key node of the whole ship is reconstructed;According to full-field stress data, fatigue damage accumulation calculation is carried out, and the cumulative damage degree of each key node is obtained and the remaining life is predicted.The application combines off-line simulation learning and on-line real-time deduction, realizes millisecond-level full-ship stress field reconstruction and fatigue life prediction, and achieves real-time, accurate and full-coverage technical effects.

Description

Technical Field

[0001] This invention relates to the field of ship structural health monitoring technology, and in particular to a method and system for real-time prediction of fatigue life of FPSO hull structure. Background Technology

[0002] FPSOs (Floating Production Storage and Offloading Units) are core facilities for offshore oil and gas development, operating long-term in harsh sea conditions. Their hull structures are subjected to alternating environmental loads from wind, waves, and currents, making structural fatigue damage a key factor affecting operational safety and remaining service life. Due to the complex and variable marine environment, hull structures are highly susceptible to fatigue cracks at stress concentration points such as welds and joints, leading to decreased structural strength and even catastrophic fracture accidents. Accurately predicting the fatigue life of hull structures is crucial for ensuring operational safety and guiding maintenance decisions.

[0003] Currently, fatigue analysis of FPSO hull structures primarily relies on the finite element method (FEM). This method establishes a high-precision geometric model, performs fluid-structure interaction time-domain analysis under typical sea conditions, calculates the stress response of key nodes throughout the ship, and then assesses fatigue life based on rainflow counting and Miner's linear cumulative damage rule. However, FEM simulation involves enormous computational costs, making real-time analysis difficult and unable to respond immediately to sudden changes in sea conditions or structural conditions. On the other hand, deploying strain sensors at key locations on the hull can acquire real-time stress data for localized areas, but limitations in sensor deployment costs and the complexity of the hull structure restrict the monitoring coverage, making it impossible to directly obtain the stress state of numerous fatigue hotspots.

[0004] Therefore, how to overcome the limitations of simulation calculation speed and sensor coverage to achieve real-time and accurate prediction of the fatigue life of the entire FPSO structure is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address at least one of the technical problems mentioned above, this invention provides a method and system for real-time prediction of fatigue life of FPSO hull structures. By constructing a digital twin framework that integrates offline simulation learning, online real-time extrapolation, and virtual-real deviation correction, it achieves millisecond-level real-time extrapolation of the entire ship's stress field and accurate mapping from local monitoring points to the entire ship's stress field.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for real-time prediction of the fatigue life of an FPSO hull structure, comprising:

[0007] Acquire real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data acquired in real time based on sparse sensors;

[0008] The real-time six-degree-of-freedom pose data is input into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients.

[0009] Based on the measured stress data, the predicted stress characteristic coefficients are corrected by virtual-real fusion to obtain the corrected stress characteristic coefficients;

[0010] The modified stress characteristic coefficients are subjected to inverse principal component analysis transformation to reconstruct real-time full-field stress data of key nodes of the entire ship.

[0011] Fatigue damage accumulation is calculated based on the real-time full-field stress data to obtain the cumulative damage degree of each key node.

[0012] The remaining lifespan of each critical node is predicted based on the cumulative damage level and design service life of each critical node.

[0013] Furthermore, the predicted stress characteristic coefficients are corrected by a virtual-real fusion process, including:

[0014] Extract the submatrix corresponding to the sensor location from the projection matrix of principal component analysis;

[0015] Calculate the virtual anchor stress value at the sensor location based on the predicted stress characteristic coefficients and the submatrix;

[0016] Calculate the residual between the measured stress data and the virtual anchor point stress value;

[0017] The correction amount of the residual in the feature space is solved by the least squares method;

[0018] The predicted stress characteristic coefficients are corrected using the correction amount.

[0019] Furthermore, the training process of the deep residual regression network includes:

[0020] An offline simulation database is constructed, which contains six-degree-of-freedom hull pose data and corresponding full-field key node stress data under multiple time samples.

[0021] The stress data of the key nodes in the entire field are centered and a covariance matrix is ​​constructed.

[0022] Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, and select the first k eigenvectors to form the projection matrix based on the cumulative variance contribution rate;

[0023] The projection matrix is ​​used to map high-dimensional full-field key node stress data into low-dimensional stress characteristic coefficients.

[0024] Using the six-DOF pose data of the hull as input and the corresponding stress characteristic coefficients as output, a depth residual regression network is trained to obtain the pre-trained depth residual regression network.

[0025] Furthermore, inverse principal component analysis is performed on the corrected stress characteristic coefficients to reconstruct real-time full-field stress data for key nodes throughout the ship, including:

[0026] Obtain the average stress vector and projection matrix calculated during principal component analysis;

[0027] Multiplying the corrected stress characteristic coefficients by the projection matrix yields the deviation stress;

[0028] The deviation stress is added to the average stress vector to obtain the real-time full-field stress data of the key nodes of the entire ship.

[0029] Furthermore, fatigue damage accumulation calculations are performed based on the real-time full-field stress data to obtain the cumulative damage degree of each key node, including:

[0030] Rainflow counting is performed on the stress time history of each key node in the real-time full-field stress data to extract the amplitude and mean of stress cycles;

[0031] Select the corresponding SN curve parameters according to the type of welded joint at the node;

[0032] The stress amplitude is corrected for average stress based on the mean value;

[0033] Based on Miner's linear cumulative damage rule, the cumulative damage degree of each critical node is calculated according to the corrected stress amplitude and the corresponding SN curve parameters.

[0034] Furthermore, it also includes operation and maintenance strategies, which include:

[0035] Based on the preset damage threshold, each key node is divided into normal state, warning state, or dangerous state.

[0036] For the normal state, maintain routine inspections;

[0037] For the aforementioned warning status, a list of key areas of concern is generated and it is recommended to increase the frequency of visual inspections.

[0038] In the event of the aforementioned dangerous situation, an audible and visual alarm is triggered, and a non-destructive testing task order is generated.

[0039] The second aspect of the present invention provides a real-time fatigue life prediction system for FPSO hull structure, comprising: a data acquisition module for acquiring real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data acquired in real time based on sparse sensors;

[0040] The online simulation module is used to input the real-time six-degree-of-freedom pose data into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients.

[0041] The virtual-real fusion correction module is used to perform virtual-real fusion correction on the predicted stress characteristic coefficients based on the measured stress data, so as to obtain the corrected stress characteristic coefficients.

[0042] The full-field stress reconstruction module is used to perform inverse principal component analysis transformation on the corrected stress characteristic coefficients to reconstruct real-time full-field stress data of key nodes of the entire ship.

[0043] The fatigue damage calculation module is used to perform fatigue damage accumulation calculation based on the real-time full-field stress data to obtain the cumulative damage degree of each key node.

[0044] The life prediction module is used to predict the remaining life of each critical node based on the cumulative damage of each critical node and its designed service life.

[0045] A third aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps in the real-time prediction method for fatigue life of FPSO hull structures as described in the first aspect of the present invention.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the real-time prediction method for fatigue life of FPSO hull structures as described in the first aspect of the present invention.

[0047] A fifth aspect of the present invention provides a computer program product comprising software code, wherein the program in the software code performs the steps of the real-time prediction method for fatigue life of FPSO hull structure as described in the first aspect of the present invention.

[0048] Compared with existing technologies, the real-time prediction method and system for fatigue life of FPSO hull structure provided by this invention has the following advantages:

[0049] (1) This invention performs feature dimensionality reduction on high-dimensional finite element full-field stress data through principal component analysis, and establishes a mapping relationship between pose and stress features through deep neural network. The data-driven proxy model replaces the time-consuming traditional finite element solution process. It not only retains the high-fidelity mechanical information of the entire ship structure, but also realizes millisecond-level real-time calculation, solving the problem of difficulty in balancing computational efficiency and model accuracy in digital twin systems.

[0050] (2) This invention introduces a virtual-real fusion correction mechanism driven by sparse sensors, and uses a small amount of measured data from the actual FPSO ship to perform online calibration of the surrogate model based on simulation training; it can effectively eliminate the objective deviation between the simulation model and the physical entity caused by factors such as construction tolerance, corrosion loss, and actual load changes, and significantly improve the credibility of the life prediction results in the real engineering environment.

[0051] (3) This invention achieves full-field stress reconstruction through inverse principal component analysis, overcoming the limitations of the limited number of physical sensors and the inability to cover all potential fatigue hotspots; it can realize synchronous monitoring and life evolution analysis of hundreds or thousands of key nodes of FPSO hull, providing comprehensive and reliable data support for maintenance personnel to formulate scientific hull structure integrity management strategies. Attached Figure Description

[0052] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0053] Figure 1 A flowchart of the real-time fatigue life prediction method for FPSO hull structure provided in Embodiment 1 of the present invention;

[0054] Figure 2 The overall flowchart of the real-time prediction method for fatigue life of FPSO hull structure provided in Embodiment 1 of the present invention is shown below.

[0055] Figure 3 This is a schematic diagram of the proxy model structure provided in Embodiment 1 of the present invention;

[0056] Figure 4 This is a schematic diagram of the online correction principle of virtual-real fusion based on sparse sensors provided in Embodiment 1 of the present invention;

[0057] Figure 5 This is an architecture diagram of the real-time fatigue life prediction system for FPSO hull structure provided in Embodiment 2 of the present invention. Detailed Implementation

[0058] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0059] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0060] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0061] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0062] Example 1

[0063] like Figure 1 This embodiment provides a method for real-time prediction of fatigue life of FPSO hull structure, including:

[0064] S1. Acquire real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data collected in real time based on sparse sensors;

[0065] S2. Input the real-time six-degree-of-freedom pose data into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients;

[0066] S3. Based on the measured stress data, the predicted stress characteristic coefficients are corrected by virtual-real fusion to obtain the corrected stress characteristic coefficients;

[0067] S4. Perform inverse principal component analysis transformation on the corrected stress characteristic coefficients to reconstruct the real-time full-field stress data of key nodes of the entire ship.

[0068] S5. Based on the real-time full-field stress data, perform fatigue damage accumulation calculation to obtain the cumulative damage degree of each key node;

[0069] S6. Predict the remaining lifespan of each critical node based on the cumulative damage level and design service life of each critical node.

[0070] As shown in S1 to S6, the method provided in this embodiment is as follows: First, six-degree-of-freedom pose data and measured stress data at sparse locations of the hull are collected using fiber optic gyroscopes and fiber optic strain sensors deployed on the actual ship. These data serve as inputs and correction benchmarks for online simulation. Then, the pose data is input into a pre-trained deep residual regression network. This network has learned the nonlinear mapping relationship between pose data and stress characteristic coefficients in the offline simulation database, thus enabling it to quickly output predicted stress characteristic coefficients. Next, the predicted characteristic coefficients are corrected using virtual-real fusion correction based on measured data from sparse sensors. This step eliminates the deviation between the simulation model and the real physical world, yielding corrected characteristic coefficients. Then, through inverse principal component analysis, the low-dimensional corrected characteristic coefficients are reconstructed into real-time full-field stress data for key nodes throughout the ship. This overcomes the limitation of sensor coverage. Finally, fatigue damage accumulation is calculated based on the reconstructed full-field stress data to obtain the cumulative damage degree of each node, and the remaining life is predicted in conjunction with the design service life.

[0071] The aforementioned prediction method achieves millisecond-level reconstruction of the entire ship's stress field through an architecture combining offline simulation learning and online real-time extrapolation. The accuracy of the prediction results is ensured by a sparse sensor correction mechanism, and a complete closed loop from stress monitoring to life prediction is realized through online fatigue damage calculation.

[0072] In one specific embodiment, such as Figure 2 As shown, the overall process of the real-time fatigue life prediction method for FPSO hull structures provided in this embodiment includes an offline modeling and training stage (offline stage) and an online real-time prediction stage (online stage). In the offline stage, a database is built through high-fidelity simulation, and PCA dimensionality reduction and deep residual regression network training are performed. In the online stage, pose and sparse stress data are collected in real time, and the full-field stress is reconstructed through virtual-real fusion correction and inverse PCA, thereby calculating fatigue damage and predicting the remaining life. The following section combines... Figure 2 This method will be explained in detail.

[0073] (I) Offline Modeling and Training Phase

[0074] S101, establish a high-fidelity physical model and offline simulation environment.

[0075] A detailed geometric model of the FPSO hull structure is created using 3D modeling software. The model must include key components such as the hull shell, internal compartments, longitudinal and transverse frames, and superstructure support piers. Meshing is performed in finite element analysis software. Considering the accuracy requirements of fatigue assessment, the mesh is refined for key hotspot areas such as cable guide holes, mooring support roots, and deck connections to ensure that the mesh size in these hotspot areas meets fatigue analysis standards. Typically, the mesh size is set to t × t of the plate thickness.

[0076] Set up simulation conditions. Based on long-term hydrological and meteorological data (wave scattering diagram) of the FPSO operating area, select conditions including different wave heights ( ), spectral peak period ( Combination of wind speed and wave direction Typical sea conditions were studied. Hydrodynamic analysis software was used to perform fluid-structure interaction time-domain analysis to calculate the dynamic response of the ship under wave loads.

[0077] S102, Vectorized definition and acquisition of data.

[0078] To facilitate processing by deep learning models, physical quantities are given standardized mathematical definitions.

[0079] Define the six-DOF pose vector of the ship at any time t as the input variable. :

[0080]

[0081] in, These represent the linear displacements of the FPSO hull in the longitudinal, transverse, and helical directions, respectively. These represent the angles of roll, pitch, and yaw, respectively. This data is output by the hydrodynamic software in the simulation and collected by the inertial measurement unit on the actual ship.

[0082] Define the stress vector of all critical nodes at any time t as the output variable. :

[0083]

[0084] in, This represents the total number of finite element nodes selected within the critical hotspot areas of the entire ship. For the first Von-Mises equivalent stress at each node.

[0085] Through the Long-term series simulations of various sea states were performed, data were extracted at a fixed sampling frequency, and a model containing... The offline training database for each sample:

[0086] Input matrix ;

[0087] Stress snapshot matrix .

[0088] S103, PCA feature dimensionality reduction of full-field stress data.

[0089] Since the number of critical nodes M is typically enormous (thousands to tens of thousands), directly training a neural network for end-to-end prediction would lead to the "curse of dimensionality." This step uses principal component analysis to analyze the stress snapshot matrix. Perform orthogonal decomposition and dimensionality reduction.

[0090] First, the stress data is centered, and the average stress vector is calculated. :

[0091]

[0092] Obtain a centralized matrix .

[0093] Constructing the covariance matrix And solve its eigenvalue problem:

[0094]

[0095]

[0096] in, Eigenvalues, sorted in descending order ; These are the corresponding eigenvectors, i.e., principal component modes.

[0097] The number of principal components to be retained is determined based on the cumulative variance contribution rate. Set a threshold :

[0098]

[0099] Before selection The projection matrix is ​​composed of eigenvectors. .

[0100] Using the projection matrix to transform the high-dimensional stress vector Mapped to a low-dimensional stress characteristic coefficient vector :

[0101]

[0102] Thus, the prediction target of the neural network has shifted from... The physical stress field of dimension simplifies to eigenvalues ​​of dimension (usually) ).

[0103] S104, Deep Residual Regression Network Design.

[0104] like Figure 3 As shown, a deep regression network based on residual connections is designed to establish a high-precision nonlinear mapping from "six-degree-of-freedom pose of the hull" to "stress characteristic coefficients".

[0105] Network architecture design:

[0106] Input layer: 6 neurons, corresponding to a six-degree-of-freedom pose vector. .

[0107] Feature extraction layer: Maps the input to a high-dimensional feature space (256 dimensions), and includes linear layers, batch normalization (BN) layers, and the ReLU activation function.

[0108] Residual coding module: composed of indivual( The residual blocks are connected in series with identical structures. Each residual block contains two fully connected layers and introduces skip connections between the input and output, allowing the network to learn the residuals of the identity mapping.

[0109] Output layer: Number of neurons is The corresponding stress characteristic coefficient vector The output layer does not use an activation function; it directly outputs the regression values.

[0110] The formula for calculating the residual block is:

[0111]

[0112] The final output of the network is:

[0113]

[0114] in, , For network weights and bias parameters, It is the ReLU activation function. For batch normalization operations, The input features for the residual block are... For the first The output feature vectors of each residual block This is the weight matrix of the first fully connected layer within the residual block. This is the weight matrix of the second fully connected layer within the residual block. To output the final weight matrix of the regression layer, This refers to the final high-dimensional feature vector extracted by the residual module of the network. This is the bias vector of the first fully connected layer within the residual block. This is the bias vector for the second fully connected layer within the residual block. This is the bias vector for the final output regression layer.

[0115] S105, Model Training and Loss Function Optimization.

[0116] Mean squared error (MSE) is used as the loss function to measure the predicted feature coefficients. With true feature coefficients Differences between them:

[0117]

[0118] in, This is the batch size during training.

[0119] The Adam optimizer is used for backpropagation parameter updates, with an initial learning rate of 0.001, and a cosine annealing learning rate decay strategy is applied. Training stops when the validation set loss no longer decreases, the optimal model parameters are saved, and a pre-trained deep residual regression network is obtained.

[0120] S103 to S105 define the training process of the deep residual regression network, which is the core step of the offline simulation learning stage. First, the dynamic response of the FPSO hull under various typical sea conditions is analyzed in the time domain using finite element simulation software. Data is extracted at a fixed sampling frequency to construct an offline simulation database containing multiple time samples. This database includes the hull's six-DOF pose data and its corresponding full-field key node stress data. Then, the stress data is centered to construct a covariance matrix, and its eigenvalues ​​and eigenvectors are solved. The top k principal components are selected according to a preset cumulative variance contribution rate threshold to form a projection matrix. This projection matrix achieves orthogonal decomposition and feature extraction of the high-dimensional stress field. Next, the projection matrix is ​​used to map the high-dimensional stress data into low-dimensional stress characteristic coefficients. At this point, the prediction target of the neural network is simplified from an M-dimensional physical stress field to k-dimensional characteristic coefficients, effectively avoiding the "curse of dimensionality." Finally, using the hull's six-DOF pose data as input and the corresponding stress characteristic coefficients as output, the deep residual regression network is trained to learn the nonlinear mapping relationship between pose and stress characteristics.

[0121] The above steps address the technical problems of "curse of dimensionality," low computational efficiency, and difficulty in convergence caused by directly using end-to-end neural networks to predict high-dimensional stress fields in existing technologies. Principal component analysis is used to reduce the dimensionality of high-dimensional stress data, extracting the main feature modes and reducing the prediction target from thousands of dimensions to tens of dimensions. This significantly reduces the learning difficulty and computational complexity of the neural network, improving model training efficiency and prediction speed.

[0122] (II) Online Real-time Prediction Stage

[0123] S201, Sparse Sensor Deployment and Online Data Acquisition.

[0124] Hardware deployment is carried out on the actual FPSO vessel.

[0125] Pose and position sensing: A high-precision fiber optic gyroscope and accelerometer are installed at the ship's center of gravity to collect six-degree-of-freedom motion data in real time. This refers to the real-time six-degree-of-freedom pose data of the FPSO hull.

[0126] Stress sensing: A small number of fiber optic strain sensors are installed at typical locations on the ship's hull (such as the midship deck edge and the bow mooring area) as real anchor points to collect measured stress data in real time. The locations of these sensors correspond to the set of node indices in the finite element model. .

[0127] Data synchronization: The inertial measurement unit data and strain data are timestamped and aligned through an industrial switch, and then transmitted to a shore-based or shipborne edge computing server via 4G / 5G or satellite link.

[0128] S202, Virtual-Real Fusion Correction Based on Residuals.

[0129] Because the simulation model contains idealized assumptions (not considering construction tolerances, corrosion, and actual load variations), directly using the surrogate model to predict results may lead to systematic deviations from the true values. This step utilizes sparse measured data to perform online correction of the model output.

[0130] Figure 4 This diagram illustrates the principle of online correction based on sparse sensor fusion. It visually demonstrates the overall process, data transfer relationships, and core computational logic for online calibration of stress characteristic coefficients output by a deep neural network surrogate model using measured data from sparsely arranged strain sensors on an actual FPSO vessel. This correction process is a crucial step in the online monitoring phase of this invention, eliminating systematic deviations between the simulation model and the physical FPSO entity caused by factors such as construction tolerances, corrosion losses, and actual load variations. This provides a realistic and reliable low-dimensional characteristic coefficient foundation for accurate reconstruction of the full-field stress.

[0131] Specifically, the virtual-real fusion correction of the predicted stress characteristic coefficients includes:

[0132] Projection matrix from principal component analysis Extract the sub-matrix corresponding to the sensor position. Where P is the number of sensors. The dimension is the feature coefficient.

[0133] The real-time six-degree-of-freedom pose data Inputting a pre-trained deep residual regression network yields the predicted stress characteristic coefficients. ;

[0134] Based on the predicted stress characteristic coefficients and the sub-matrix, the virtual anchor stress value at the sensor location is calculated. :

[0135]

[0136] in, The full-field average stress vector calculated offline. The subset at the corresponding node index of the sensor;

[0137] Obtain the measured sensor stress value Calculate the residual between the measured stress data and the virtual anchor point stress value. :

[0138]

[0139] The correction amount of the residual in the characteristic space is solved by the least squares method. :

[0140]

[0141] in, This is a regularization term to prevent over-modification.

[0142] The predicted stress characteristic coefficients are corrected using the correction amount to obtain the corrected stress characteristic coefficients. :

[0143]

[0144] in, This is a smoothing factor used to control the smoothness of the correction process.

[0145] This completes the entire process of online correction for virtual-real fusion. This process adjusts the stress characteristic coefficients initially predicted by the deep neural network surrogate model. With correction amount Combined, the final output is the calibrated feature coefficients. These coefficients are the characteristic coefficients after eliminating the virtual-to-real bias, and can be directly used in the subsequent inverse principal component analysis full-field stress reconstruction step.

[0146] S202 further specifies the concrete implementation method of virtual-real fusion correction: First, the projection matrix obtained during the principal component analysis process... In the process, rows corresponding to the actual shipboard sensor deployment locations are extracted to form sub-matrices. This submatrix establishes the correlation between the feature space and the physical location of the sensor. Then, the submatrix is ​​used... and predictive characteristic coefficients The virtual anchor stress value at the sensor location is calculated. This virtual value represents the stress estimate at the sensor location based on the current model prediction. Next, the measured stress value is calculated. With virtual anchor stress value residual The residual quantifies the instantaneous deviation between the simulation model's predictions and the real physical world. Then, the correction amount of the residual in the feature space is solved using the least squares method. The regularization term This is used to prevent overcorrection. Finally, the predicted feature coefficients are corrected using this correction amount to obtain... Smoothing factor Used to control the smoothness of corrections.

[0147] The virtual-real fusion correction method addresses the technical problem in existing technologies where simulation models contain idealized assumptions (such as not considering construction tolerances, corrosion, and actual load differences), leading to systematic deviations between predicted results and true values. By introducing a residual correction mechanism based on sparse sensors, model bias is dynamically compensated, achieving virtual-real fusion and online calibration between the digital twin model and the physical entity. This further improves the accuracy and robustness of the prediction results.

[0148] S203, Real-time Reconstruction of Full-Field Stress (Inverse PCA).

[0149] The modified stress characteristic coefficients are subjected to inverse principal component analysis transformation to reconstruct real-time full-field stress data for key nodes throughout the ship, including:

[0150] Obtain the average stress vector calculated during principal component analysis. and projection matrix ;

[0151] The corrected stress characteristic coefficient With the projection matrix Multiplying them together yields the deviation stress;

[0152] Adding the deviation stress to the average stress vector yields the whole ship. Real-time full-field stress data at key nodes :

[0153]

[0154] This step enables the estimation of stress states in hundreds or thousands of blind spot areas across the entire ship using only a small amount of sensor data.

[0155] S204, cumulative fatigue damage calculation.

[0156] Fatigue damage accumulation is calculated based on the real-time full-field stress data to obtain the cumulative damage degree of each key node, including:

[0157] Stress time history data for each key node obtained from the reconstruction The data is processed by performing rainflow counting on the stress time history of each key node in the real-time full-field stress data and extracting the amplitude of stress cycles. and mean;

[0158] Select the corresponding SN curve parameters according to the type of welded joint at the node. ;

[0159] The stress amplitude is corrected for average stress based on the mean value (e.g., using Goodman correction).

[0160] Based on Miner's linear cumulative damage rule, the cumulative damage degree of each critical node is calculated according to the corrected stress amplitude and the corresponding SN curve parameters. :

[0161]

[0162] in, Stress amplitude Number of times it appears The fatigue life is given by this amplitude, where m represents the fatigue strength index of the material (the slope parameter of the SN curve). Category index representing stress cycle amplitude ( ), This represents the fatigue strength coefficient of the material (SN curve intercept parameter), which is determined by referring to a table based on the type of welded joint at the node and the corresponding classification society specifications. Indicates the first Stress amplitude of a stress cycle. Indicates the first Key nodes ( Total cumulative damage.

[0163] This step, through rainflow counting, SN curves, and Miner's linear cumulative damage rule, ensures the consistency of fatigue assessment results with industry standards, achieving a scientific and standardized quantitative assessment of fatigue damage.

[0164] S205, Lifetime Prediction and Operation and Maintenance Strategy Generation.

[0165] Based on the cumulative damage of each key node With design service life Predict the remaining lifetime of each critical node. :

[0166]

[0167] Overall ship damage Mapping onto a 3D digital twin model generates a red-green color cloud map.

[0168] Based on the preset damage threshold, each key node is divided into normal state, warning state, or dangerous state:

[0169] Normal state (green): Maintain routine inspections;

[0170] Warning status (yellow): Generate a list of key areas of concern and recommend increasing the frequency of visual inspections;

[0171] Dangerous state (red): This triggers an audible and visual alarm and generates a non-destructive testing task sheet, which includes the specific node number, suggested testing method, and location coordinates.

[0172] This step correlates the damage calculation results with preset thresholds and automatically generates specific, executable operation and maintenance strategies, realizing intelligent closed-loop management from condition monitoring to risk warning and maintenance guidance, thereby improving operation and maintenance efficiency and ensuring the safety of the ship's structure.

[0173] Example 2

[0174] like Figure 5 As shown, this embodiment provides a real-time fatigue life prediction system for FPSO hull structures, including: a data acquisition module, used to acquire real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data acquired in real time based on sparse sensors;

[0175] The online simulation module is used to input the real-time six-degree-of-freedom pose data into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients.

[0176] The virtual-real fusion correction module is used to perform virtual-real fusion correction on the predicted stress characteristic coefficients based on the measured stress data, so as to obtain the corrected stress characteristic coefficients.

[0177] The full-field stress reconstruction module is used to perform inverse principal component analysis transformation on the corrected stress characteristic coefficients to reconstruct real-time full-field stress data of key nodes of the entire ship.

[0178] The fatigue damage calculation module is used to perform fatigue damage accumulation calculation based on the real-time full-field stress data to obtain the cumulative damage degree of each key node.

[0179] The life prediction module is used to predict the remaining life of each critical node based on the cumulative damage of each critical node and its designed service life.

[0180] Example 3

[0181] Embodiment 3 of the present invention provides an electronic device.

[0182] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. The processor includes, but is not limited to, at least one of a central processing unit (CPU), a graphics processing unit (GPU), a neural network processor (NPU), a tensor processor (TPU), or an artificial intelligence acceleration chip. The program is used to execute the steps in the real-time fatigue life prediction method for FPSO hull structures as described in Embodiment 1 of the present invention.

[0183] The detailed steps are the same as those of the real-time fatigue life prediction method for FPSO hull structure provided in Example 1, and will not be repeated here.

[0184] Example 4

[0185] Embodiment 4 of the present invention provides a computer-readable storage medium.

[0186] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the real-time prediction method for fatigue life of FPSO hull structure as described in Embodiment 1 of the present invention.

[0187] The detailed steps are the same as those of the real-time fatigue life prediction method for FPSO hull structure provided in Example 1, and will not be repeated here.

[0188] Example 5

[0189] Embodiment 5 of the present invention provides a computer program product.

[0190] A computer program product includes software code, wherein the program in the software code performs the steps of the real-time fatigue life prediction method for FPSO hull structure as described in Embodiment 1 of the present invention.

[0191] The detailed steps are the same as those of the real-time fatigue life prediction method for FPSO hull structure provided in Example 1, and will not be repeated here.

[0192] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages. For example, in one implementation, the methods and systems can be developed based on deep learning frameworks (such as TensorFlow, PyTorch, etc.) and using the Python language. Those skilled in the art will understand that other suitable programming languages ​​or tools can also be used for implementation without departing from the core ideas of the present invention.

[0193] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0194] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0195] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0196] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for real-time prediction of fatigue life of FPSO hull structure, characterized in that, include: Acquire real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data acquired in real time based on sparse sensors; The real-time six-degree-of-freedom pose data is input into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients. Based on the measured stress data, the predicted stress characteristic coefficients are corrected by virtual-real fusion to obtain the corrected stress characteristic coefficients; The predicted stress characteristic coefficients are corrected by a virtual-real fusion process, including: Extract the submatrix corresponding to the sensor location from the projection matrix of principal component analysis; Calculate the virtual anchor stress value at the sensor location based on the predicted stress characteristic coefficients and the submatrix; Calculate the residual between the measured stress data and the virtual anchor point stress value; The correction amount of the residual in the feature space is solved by the least squares method; The predicted stress characteristic coefficients are corrected using the correction amount; The modified stress characteristic coefficients are subjected to inverse principal component analysis transformation to reconstruct real-time full-field stress data of key nodes of the entire ship. The training process of the deep residual regression network includes: An offline simulation database is constructed, which contains six-degree-of-freedom hull pose data and corresponding full-field key node stress data under multiple time samples. The stress data of the key nodes in the entire field are centered and a covariance matrix is ​​constructed. Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, and select the first k eigenvectors to form the projection matrix based on the cumulative variance contribution rate; The projection matrix is ​​used to map high-dimensional full-field key node stress data into low-dimensional stress characteristic coefficients. Using the six-DOF pose data of the hull as input and the corresponding stress characteristic coefficients as output, a depth residual regression network is trained to obtain the pre-trained depth residual regression network. Fatigue damage accumulation is calculated based on the real-time full-field stress data to obtain the cumulative damage degree of each key node. The remaining lifespan of each critical node is predicted based on the cumulative damage level and design service life of each critical node.

2. The method as described in claim 1, characterized in that, The modified stress characteristic coefficients are subjected to inverse principal component analysis transformation to reconstruct real-time full-field stress data for key nodes throughout the ship, including: Obtain the average stress vector and projection matrix calculated during principal component analysis; Multiplying the corrected stress characteristic coefficients by the projection matrix yields the deviation stress; The deviation stress is added to the average stress vector to obtain the real-time full-field stress data of the key nodes of the entire ship.

3. The method as described in claim 1, characterized in that, Fatigue damage accumulation is calculated based on the real-time full-field stress data to obtain the cumulative damage degree of each key node, including: Rainflow counting is performed on the stress time history of each key node in the real-time full-field stress data to extract the amplitude and mean of stress cycles; Select the corresponding SN curve parameters according to the type of welded joint at the node; The stress amplitude is corrected for average stress based on the mean value; Based on Miner's linear cumulative damage rule, the cumulative damage degree of each critical node is calculated according to the corrected stress amplitude and the corresponding SN curve parameters.

4. The method as described in claim 1, characterized in that, It also includes operation and maintenance strategies, which include: Based on the preset damage threshold, each key node is divided into normal state, warning state, or dangerous state. For the normal state, maintain routine inspections; For the aforementioned warning status, a list of key areas of concern is generated and it is recommended to increase the frequency of visual inspections. In the event of the aforementioned dangerous situation, an audible and visual alarm is triggered, and a non-destructive testing task order is generated.

5. A real-time fatigue life prediction system for FPSO hull structures, characterized in that, include: The data acquisition module is used to acquire real-time six-degree-of-freedom pose data of the FPSO hull and measured stress data acquired in real-time based on sparse sensors. The online simulation module is used to input the real-time six-degree-of-freedom pose data into a pre-trained deep residual regression network to obtain the predicted stress characteristic coefficients. The virtual-real fusion correction module is used to perform virtual-real fusion correction on the predicted stress characteristic coefficients based on the measured stress data, so as to obtain the corrected stress characteristic coefficients. The predicted stress characteristic coefficients are corrected by a virtual-real fusion process, including: Extract the submatrix corresponding to the sensor location from the projection matrix of principal component analysis; Calculate the virtual anchor stress value at the sensor location based on the predicted stress characteristic coefficients and the submatrix; Calculate the residual between the measured stress data and the virtual anchor point stress value; The correction amount of the residual in the feature space is solved by the least squares method; The predicted stress characteristic coefficients are corrected using the correction amount; The full-field stress reconstruction module is used to perform inverse principal component analysis transformation on the corrected stress characteristic coefficients to reconstruct real-time full-field stress data of key nodes of the entire ship. The training process of the deep residual regression network includes: An offline simulation database is constructed, which contains six-degree-of-freedom hull pose data and corresponding full-field key node stress data under multiple time samples. The stress data of the key nodes in the entire field are centered and a covariance matrix is ​​constructed. Solve for the eigenvalues ​​and eigenvectors of the covariance matrix, and select the first k eigenvectors to form the projection matrix based on the cumulative variance contribution rate; The projection matrix is ​​used to map high-dimensional full-field key node stress data into low-dimensional stress characteristic coefficients. Using the six-DOF pose data of the hull as input and the corresponding stress characteristic coefficients as output, a depth residual regression network is trained to obtain the pre-trained depth residual regression network. The fatigue damage calculation module is used to perform fatigue damage accumulation calculation based on the real-time full-field stress data to obtain the cumulative damage degree of each key node. The life prediction module is used to predict the remaining life of each critical node based on the cumulative damage of each critical node and its designed service life.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of the real-time prediction method for fatigue life of FPSO hull structure as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the real-time prediction method for the fatigue life of FPSO hull structures as described in any one of claims 1 to 4.

8. A computer program product, comprising software code, characterized in that, The program in the software code executes the steps of the real-time prediction method for the fatigue life of FPSO hull structure as described in any one of claims 1 to 4.

Images