A new energy ship power equipment operation and maintenance analysis method

Abstract

The application discloses a new energy ship power equipment operation and maintenance analysis method, through collecting multi-physical field operation, microstructure and historical failure data and correcting and standardizing processing time sequence deviation, a high-quality coupled data sample set is generated. By constructing a physical constraint layer embedded with a physical field coupling equation and a fault mechanism knowledge graph and injecting a model training loss function, the obtained multi-physical condition calculation model can accurately depict the multi-field coupling characteristics and fault evolution law, and improve the complex working condition adaptation ability. By extracting multi-dimensional features and weighted fusion dimension reduction, a core fault feature vector is generated, and the fault essence is accurately captured. Through the meta-learning algorithm, the pre-training model is fine-tuned to obtain a fault analysis large model, which improves the small sample scene analysis accuracy and generalization ability, and can clearly trace the fault evolution path, define the coupling critical condition and establish the micro-macro mapping relationship. Through the feedback of the real ship, the model and the knowledge graph are updated in a closed loop, and the operation and maintenance accuracy is ensured to be stable for a long time.

Description

Technical Field

[0001] This application relates to the field of marine power technology, and more specifically, to a method for operation and maintenance analysis of new energy marine power equipment. Background Technology

[0002] New energy ships using batteries, fuel cells, and other technologies have become the mainstream of development in the shipping industry. Lithium batteries, fuel cell stacks, and propulsion motors are the core power equipment of new energy ships, and their operational safety directly determines the safety of ship navigation and the efficiency of operation and maintenance.

[0003] Failures in power equipment are all caused by the coupling of multiple physical fields including "electricity, heat, force, and chemistry," exhibiting typical characteristics such as complex mechanisms, scarce small samples, and high requirements for generalization across operating conditions. Current fault diagnosis technologies for power equipment in new energy ships suffer from three major pain points, and existing operation and maintenance analysis methods are insufficient to meet actual needs: First, physical models are limited. Traditional multiphysics coupling simulations rely on precise mathematical equations and component parameters, making it difficult to adapt to complex operating conditions such as extreme sea states and dynamic loads. The models have poor adaptability to differentiated navigation scenarios and cannot characterize the evolution of irregular multiphysics coupling faults. Second, data-driven adaptation is insufficient. Traditional deep learning models rely on large-scale labeled fault data, but data on rare faults in new energy ships (such as lithium battery thermal runaway and fuel cell membrane rupture) are difficult to obtain, resulting in data hunger and a significant drop in diagnostic accuracy in small sample scenarios. Third, interpretability is lacking. Traditional machine learning models are mostly "black box" structures, unable to establish quantitative correlations between microscopic failures such as SEI membrane decomposition and membrane electrode porosity changes and macroscopic failures such as capacity decay and power interruption. This makes it difficult to reveal the physical nature of the faults, and fault tracing and prevention lack clear basis.

[0004] In addition, existing technologies suffer from problems such as non-standard data processing, incomplete feature extraction, and lack of dynamic optimization mechanisms, which further restrict the accuracy and reliability of operation and maintenance analysis and cannot meet the actual needs of efficient and precise operation and maintenance of new energy ship power equipment. There is an urgent need for an operation and maintenance analysis technology that can adapt to complex working conditions, solve the small sample dilemma, and has strong interpretability. Summary of the Invention

[0005] This application provides a method for the operation and maintenance analysis of power equipment for new energy ships, which enables accurate diagnosis of power equipment failures, clear analysis of mechanisms, and efficient operation and maintenance, effectively ensuring ship navigation safety, improving operational efficiency, and reducing operation and maintenance costs.

[0006] A method for analyzing the operation and maintenance of power equipment for new energy ships, including:

[0007] Collect multi-physics field operation data, microstructure data and historical fault data of new energy ship power equipment, and generate a multi-physics field coupled data sample set through time series deviation correction and data standardization processing;

[0008] Based on the data sample set, a physical constraint layer is constructed by embedding physical field coupling equations and fault mechanism knowledge graphs, and the physical constraint layer is injected into the training loss function of the data-driven model to train a multi-physics condition calculation model.

[0009] The time-domain statistical features, frequency-domain spectral features, microstructure features, and multi-field coupling features of the multi-physics field coupled data sample set are extracted using the multi-physics condition calculation model. Based on the physical contribution of each feature to the fault occurrence mechanism, weighted fusion and dimensionality reduction compression are performed to generate a core fault feature vector.

[0010] A guided prompt template is constructed based on the core fault feature vector and the node attributes of the fault mechanism knowledge graph. The initial analytical model, which has been pre-trained with fault data in the power equipment field, is fine-tuned with small sample scenario parameters through a meta-learning algorithm to obtain a large fault analysis model.

[0011] Using the aforementioned fault analysis model and the fault mechanism knowledge graph as the link, by tracing the fault evolution path, defining the critical conditions of multi-field coupling, and establishing the micro-macro parameter mapping relationship, the fault mechanism analysis results containing fault causes, propagation links, and quantification thresholds are generated.

[0012] Based on the fault mechanism analysis results, an operation and maintenance solution is generated, and an incremental closed-loop update is performed on the multi-physical condition calculation model and the fault mechanism knowledge graph based on the actual ship operation feedback data.

[0013] Optionally, the timing deviation correction is performed using a multiphysics timing data alignment model, based on the time difference correction amount determined by the time asynchronous deviation between sensor timing data and microscopic test synchronization data.

[0014] The expression for the multiphysics temporal data alignment model is:

[0015] ;

[0016] In the formula, For the aligned timestamp; This is the original data timestamp; This is the time difference correction amount; This is sensor timing data; Synchronize data for microscopic testing; It is an L2 norm.

[0017] Optionally, the process of constructing the fault mechanism knowledge graph includes:

[0018] Constructing a graph structure The fault mechanism knowledge graph represents the following:

[0019] Node set ; Nodes representing fault types such as thermal runaway, internal short circuit, and capacity decay; Microscopic failure nodes characterizing SEI film decomposition, lithium dendrite growth, and membrane electrode porosity changes; A node representing a multi-field coupling condition for temperature threshold, stress threshold, and voltage descent threshold;

[0020] Edge set This represents the causal, propagational, and consequential relationships between nodes, with each edge representing these relationships. Configured with weights ;

[0021] Attribute Matrix Store the physical parameter threshold vector corresponding to each node;

[0022] The entity linking model maps the feature parameters of the multiphysics coupled data sample set to the corresponding graph nodes of the fault mechanism knowledge graph. The expression of the entity linking model is as follows:

[0023] ;

[0024] In the formula, These are the standardized feature parameters; For graph nodes eigenvectors; The similarity is cosine similarity; a link is considered valid when the similarity is not less than 0.75.

[0025] Optionally, the training loss function is a composite loss function that includes data loss and physical constraint loss, and its expression is:

[0026] ;

[0027] ;

[0028] In the formula, The data loss is determined based on the cross-entropy loss between the model's predicted values ​​and the actual fault labels; The deviation norm between the theoretical physical field distribution value calculated based on the multi-physics coupling equation of the dynamic equipment and the predicted physical field distribution value output by the model is determined as the physical constraint loss. Basic coefficient; To ensure data reliability, when the input data is actual ship operation data, The value ranges from [0.8, 1.0]. When the input data is laboratory simulation data, The value is in the range [0.5, 0.8].

[0029] Optionally, the physical constraint loss is constructed based on a set of thermo-electro-mechanical coupled partial differential equations for the lithium battery in the power equipment, wherein the set of equations is:

[0030] ;

[0031] In the formula, For battery thermal conductivity; For temperature; Current density; For internal resistance, It is the heat of a chemical reaction; Battery density; Specific heat capacity at constant pressure; Electrical conductivity; It is the electric potential; The lithium-ion diffusion coefficient; Lithium ion concentration; In response to the situation; It is the extrusion pressure; is the coefficient of thermal expansion.

[0032] Optionally, a core fault feature vector is generated by weighted fusion and dimensionality reduction based on the physical contribution of each feature to the fault occurrence mechanism, including:

[0033] Based on the physical mechanism of the failure of the target power equipment, the influence of the time-domain statistical characteristics, frequency-domain spectral characteristics, microstructure characteristics and multi-field coupling characteristics on the failure mechanism is determined, and weight coefficients are assigned to each characteristic based on the degree of influence.

[0034] The time-domain statistical feature vector, frequency-domain spectral feature vector, microstructure feature vector, and multi-field coupling feature vector are multiplied by their respective weighting coefficients and then summed element by element to obtain the fused feature vector.

[0035] Principal component analysis is performed on the fused feature vector to extract the main feature components, and the dimensionality-reduced vector formed by the main feature components is used as the core fault feature vector.

[0036] Optionally, tracing the fault evolution path includes:

[0037] Using the initial cause node in the fault mechanism knowledge graph as the search starting point and the target fault node as the search ending point, a depth-first search algorithm is used to traverse the fault mechanism knowledge graph. During the search process, only strongly related edges with edge weights greater than a first threshold are retained, and the fault propagation chain is output:

[0038] ;

[0039] Calculate the path confidence of the fault propagation chain. When the path confidence is greater than a second threshold, the fault propagation chain is determined to be a valid fault evolution path.

[0040] The formula for calculating the path confidence is as follows:

[0041] ;

[0042] In the formula, This is a fault propagation chain; This is the initial trigger node; The faulty node; For path confidence; The edge weight is denoted as .

[0043] Optionally, an exponential nonlinear fitting function can be used to establish the micro-macro parameter mapping relationship, the expression of which is:

[0044] ;

[0045] In the formula, These are parameters characterizing macroscopic faults, including capacity decay rate or voltage fluctuation amplitude. These are microstructure parameters, including SEI film thickness or electrode porosity; , , These are the fitting coefficients calibrated using the least squares method based on multi-field stress accelerated aging experimental data.

[0046] Optionally, the update strategy for fine-tuning parameters in small-sample scenarios using the meta-learning algorithm is as follows:

[0047] ;

[0048] In the formula, These are the initial analytical model parameters after pre-training with fault data from the power equipment field. These are the parameters after fine-tuning; Fine-tuning the step size for meta-learning; A task loss function constructed based on the accuracy of fault type parsing; It is a small sample support set containing a preset number of fault samples.

[0049] Optionally, an operation and maintenance solution is generated based on the fault mechanism analysis results, and an incremental closed-loop update is performed on the multi-physical condition calculation model and the fault mechanism knowledge graph based on actual ship operation feedback data, including:

[0050] By using a rule engine with built-in operation and maintenance knowledge rules, the fault type, multi-field coupling critical conditions and fault propagation path information in the fault mechanism analysis results are matched by rules to generate an operation and maintenance processing solution that includes standardized operation instructions and a visual map.

[0051] Collect no less than a first preset number of experimental data and a second preset number of actual ship operation feedback data according to a preset cycle. Based on the newly collected data, perform incremental training on the multi-physical condition calculation model and fault mechanism knowledge graph to obtain an updated parameter vector.

[0052] An exponentially weighted moving average parameter fusion strategy is adopted to weight and fuse the current parameter vector with the updated parameter vector according to a preset historical parameter retention coefficient, so as to obtain the updated model parameters and map parameters.

[0053] As can be seen from the above technical solutions, the new energy ship power equipment operation and maintenance analysis method provided in this application collects multi-physics field operation, microstructure and historical fault data and performs standardized processing to construct a data-driven model with a physical constraint layer, extracts multi-dimensional features and weighted fusion, obtains a large fault analysis model based on meta-learning fine-tuning, analyzes the fault mechanism and generates operation and maintenance solutions, and realizes incremental closed-loop updates of the model and knowledge graph.

[0054] This application effectively eliminates the heterogeneity and temporal bias of multi-source data through time-series deviation correction and data standardization, generating a high-quality multi-physics coupled data sample set. This provides a reliable data foundation for subsequent model training and feature extraction, significantly improving the accuracy of data support for operation and maintenance analysis. By constructing a physical constraint layer that embeds physical field coupling equations and a fault mechanism knowledge graph, and injecting it into the training loss function of the data-driven model, a multi-physics condition calculation model is trained. This model can accurately characterize the complex operational characteristics of "electric-thermal-mechanical-chemical" multi-physics coupling in new energy ship power equipment, clearly capture the evolution law of irregular multi-field coupling faults, and greatly improve the model's adaptability to complex working conditions such as extreme sea states and dynamic loads, as well as differentiated navigation scenarios. By extracting time-domain statistical features, frequency-domain spectral features, microstructural features, and multi-field coupling features through the multi-physics condition calculation model, and combining the weighted fusion and dimensionality reduction of each feature to the physical contribution of the fault occurrence mechanism, a core fault feature vector is generated. This vector can accurately capture the core essence of the fault, providing comprehensive and accurate feature support for fault analysis. By fine-tuning the parameters of an initial analytical model pre-trained with fault data in the field of power equipment using meta-learning algorithms for small-sample scenarios, a large-scale fault analysis model is obtained. This model can effectively adapt to scenarios where rare fault data is difficult to obtain, significantly improving the accuracy and generalization ability of fault analysis in small-sample scenarios. Through this large-scale fault analysis model, using a fault mechanism knowledge graph as a link, the model traces the fault evolution path, defines multi-field coupling critical conditions, and establishes a micro-to-macro parameter mapping relationship. This clearly reveals the physical essence of the fault, achieving accurate analysis of fault causes, propagation paths, and quantification thresholds, providing a clear and reliable theoretical basis for operation and maintenance control. Incremental closed-loop updates are performed on the multi-physical condition calculation model and the fault mechanism knowledge graph using real-ship operation feedback data, continuously optimizing the adaptability and accuracy of the model and knowledge graph, ensuring the long-term stability of operation and maintenance analysis results.

[0055] In summary, this application, through the aforementioned series of technical means, achieves accurate diagnosis, clear mechanism analysis, and efficient operation and maintenance of power equipment failures in new energy ships, effectively ensuring ship navigation safety, improving operational efficiency, and reducing operation and maintenance costs. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0057] Figure 1 This is a flowchart of an operation and maintenance analysis method for new energy ship power equipment disclosed in an embodiment of this application. Detailed Implementation

[0058] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] This application can be used in a wide variety of general-purpose or special-purpose computing device environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor devices, distributed computing environments including any of the above devices, etc.

[0060] The following section introduces the solution proposed in this application. The technical solution is as follows, and details are provided below.

[0061] Figure 1 This is a flowchart of an operation and maintenance analysis method for new energy ship power equipment disclosed in an embodiment of this application.

[0062] like Figure 1 As shown, the method may include:

[0063] Step S1: Collect multi-physics field operation data, microstructure data and historical fault data of new energy ship power equipment, and generate a multi-physics field coupled data sample set through time series deviation correction and data standardization processing.

[0064] Specifically, the first step is to comprehensively and multidimensionally collect multi-physics field operation data, microstructure data, and historical fault data of the power equipment of new energy ships throughout their entire life cycle. Multi-physics field operation data includes real-time monitoring data of electro-thermal-mechanical-chemical coupled fields such as voltage, current, temperature, stress, vibration, and humidity. Microstructure data is obtained through microscopic characterization methods, covering microscopic evolution information such as electrode micromorphology, film thickness, porosity, and interface structure. Historical fault data integrates real-ship fault cases, accelerated aging test data, classification society standard data, and typical fault sample information, forming a comprehensive and diverse set of raw data. To address the temporal asynchrony problem of multi-source data, a multi-physics field temporal data alignment model is used to correct temporal deviations. Using the temporal operation data collected by sensors and the synchronous data of microscopic tests as a benchmark, the original timestamps are optimized and adjusted by calculating the time difference correction, ensuring precise matching of multi-physics field data and microscopic test data in the time dimension and eliminating analytical biases caused by temporal misalignment. Based on this, the corrected data undergoes standardization processing, including noise filtering, outlier removal, missing value completion, and dimensional unification and numerical normalization, effectively reducing the impact of data heterogeneity, environmental interference, and equipment differences. Through dual processing of time-series deviation correction and data standardization, a high-quality, highly consistent, and highly coupled multiphysics coupling data sample set is finally generated, providing stable, reliable, and directly usable data support for subsequent multiphysics computational model training, fault feature extraction, and fault mechanism analysis.

[0065] After multi-source data acquisition, this application addresses the time asynchrony deviation between sensor time-series data and synchronous microscopic test data by specifically designing a time-series deviation correction step. This is achieved through a multi-physics time-series data alignment model. This model uses the original data timestamp as a benchmark and matches the sensor-acquired voltage, temperature, vibration, and other time-series data with the synchronous microscopic structure test data. By calculating and optimizing the time difference correction, it minimizes the deviation distance between the two sets of data, unifying data from different sources and sampling times to the same time benchmark, resulting in aligned timestamps. This completely resolves the coupling analysis distortion problem caused by asynchronous sampling, transmission delays, and test timing differences in multi-physics monitoring data and microscopic characterization data, ensuring strict temporal correspondence and physical consistency of multi-field data. Furthermore, the time-aligned multi-source data undergoes standardization processing, including noise reduction, normalization, outlier removal, and missing value completion, eliminating the effects of dimensional differences, environmental interference, and equipment errors.

[0066] Among them, the timing deviation correction is performed by using a multi-physics timing data alignment model to correct the time difference based on the time asynchronous deviation between the sensor timing data and the microscopic test synchronous data.

[0067] The expression for the multiphysics temporal data alignment model is:

[0068] ;

[0069] In the formula, For the aligned timestamp; This is the original data timestamp; This is the time difference correction amount; This is sensor timing data; Synchronize data for microscopic testing; It is an L2 norm.

[0070] Step S2: Based on the data sample set, a physical constraint layer is constructed by embedding the physical field coupling equation and the fault mechanism knowledge graph, and the physical constraint layer is injected into the training loss function of the data-driven model to train a multi-physics condition calculation model.

[0071] Specifically, the process of constructing a fault mechanism knowledge graph includes:

[0072] Constructing a graph structure The fault mechanism knowledge graph represents the following:

[0073] Node set ; Nodes representing fault types such as thermal runaway, internal short circuit, and capacity decay; Microscopic failure nodes characterizing SEI film decomposition, lithium dendrite growth, and membrane electrode porosity changes; A node representing a multi-field coupling condition for temperature threshold, stress threshold, and voltage descent threshold;

[0074] Edge set This represents the causal, propagational, and consequential relationships between nodes, with each edge representing these relationships. Configured with weights ;

[0075] Attribute Matrix Store the physical parameter threshold vector corresponding to each node;

[0076] The entity linking model maps the feature parameters of the multiphysics coupled data sample set to the corresponding graph nodes of the fault mechanism knowledge graph. The expression of the entity linking model is as follows:

[0077] ;

[0078] In the formula, These are the standardized feature parameters; For graph nodes eigenvectors; The similarity is cosine similarity; a link is considered valid when the similarity is not less than 0.75.

[0079] Specifically, based on the aforementioned multi-physics coupled data sample set, a multi-physics condition calculation model that deeply integrates the physical mechanisms of equipment is constructed. First, a physical constraint layer is built, embedding physical field coupling equations and a fault mechanism knowledge graph. The fault mechanism knowledge graph is represented using a graph structure for standardization, abstracting fault-related information of new energy ship power equipment into three parts: nodes, edges, and an attribute matrix. The node set is divided into fault type nodes, micro-failure nodes, and multi-field coupling condition nodes, corresponding to macro-fault modes, micro-deterioration mechanisms, and boundary conditions triggering faults, respectively. The edge set represents the relationships between causes, transmission, and consequences, and the strength of these relationships is distinguished by weights. The attribute matrix stores the threshold values ​​of key physical parameters such as temperature, stress, and voltage for each node. Simultaneously, an entity linking model is used to match and map the standardized multi-physics coupled data feature parameters with knowledge graph nodes. Valid links are determined based on similarity, achieving precise binding of data information with domain knowledge. This enables the physical constraint layer to possess a structured physical logic that is computable, inferable, and associative.

[0080] Furthermore, the training loss function is a composite loss function that includes data loss and physical constraint loss, and its expression is:

[0081] ;

[0082] ;

[0083] In the formula, The data loss is determined based on the cross-entropy loss between the model's predicted values ​​and the actual fault labels; The deviation norm between the theoretical physical field distribution value calculated based on the multi-physics coupling equation of the dynamic equipment and the predicted physical field distribution value output by the model is determined as the physical constraint loss. Basic coefficient; To ensure data reliability, when the input data is actual ship operation data, The value ranges from [0.8, 1.0]. When the input data is laboratory simulation data, The value is in the range [0.5, 0.8].

[0084] Building upon this foundation, a composite training loss function integrating data fitting and physical mechanisms is designed. This loss function comprises two parts: data loss and physical constraint loss. The data loss measures the deviation between the model's predicted results and the actual fault labels, ensuring the model's data fitting capability. The physical constraint loss is constructed based on the thermo-electric-mechanical coupling partial differential equations of core equipment such as lithium batteries, constraining the model's output results to maintain consistency with the objective evolution laws of multiphysics fields, avoiding prediction deviations that violate physical mechanisms. Simultaneously, a dynamically adjustable balance coefficient is introduced, determined jointly by the base coefficients and data reliability. Different weights are assigned to actual ship operation data and laboratory simulation data, achieving an adaptive balance between data-driven and physical constraints. This physical constraint layer is injected into the training process of the data-driven model in the form of a composite loss function, enabling the model to strictly adhere to physical mechanisms while learning data patterns. After iterative optimization, a multiphysics-condition computational model is obtained. This model can accurately characterize the coupling laws of electro-thermal-mechanical-chemical fields, adapting to complex navigation conditions such as extreme sea states and dynamic loads, effectively improving the physical consistency and generalization ability of fault characterization. The balance coefficient in the composite loss function is jointly determined by the base coefficient and the data reliability. The base coefficient is a preset base weight coefficient with a value of 0.1, used to establish the basic balance ratio between physical constraints and data fitting. The data reliability index is dynamically assigned based on the source and reliability of the input data. When the input data is actual ship operation data, its value ranges from [0.8, 1.0], giving higher weight to physical constraints to ensure operational accuracy. When the input data is laboratory simulation data, its value ranges from [0.5, 0.8], appropriately reducing the constraint weight to adapt to the noise characteristics of the simulation data. Through this dynamic adjustment mechanism, an adaptive balance between data-driven and physical constraints is achieved under different data scenarios, ensuring the robustness of the model while avoiding the risk of underfitting due to over-constraints.

[0085] The aforementioned physical constraint layer is injected into the training process of the data-driven model in the form of a composite loss function, enabling the model to strictly adhere to the physical laws of thermo-electric-mechanical coupling while learning data patterns. The multi-physics computational model trained after iterative optimization can accurately characterize the laws of multi-field coupling of electro-thermal-mechanical-chemical fields, adapt to complex navigation conditions such as extreme sea states and dynamic loads, and effectively improve the physical consistency and cross-condition generalization ability of fault characterization.

[0086] Among them, the physical constraint loss is constructed based on the thermo-electro-mechanical coupling partial differential equations of the lithium battery in the power equipment, and the equations are as follows:

[0087] ;

[0088] In the formula, For battery thermal conductivity; For temperature; Current density; For internal resistance, It is the heat of a chemical reaction; Battery density; Specific heat capacity at constant pressure; Electrical conductivity; It is the electric potential; The lithium-ion diffusion coefficient; Lithium ion concentration; In response to the situation; It is the extrusion pressure; is the coefficient of thermal expansion.

[0089] The calculation of physical constraint loss is based on the three core physical equations of lithium batteries under complex operating conditions, and the coupling relationship is constructed from three dimensions: heat transfer, electrochemical reaction and mechanical deformation.

[0090] Heat transfer law: This law describes the dynamic change process of the battery's internal temperature, comprehensively considering the conduction and diffusion of heat inside the battery, the Joule heat generated by current flowing through the internal resistance, and the heat generated by electrochemical reactions during charging and discharging, ultimately corresponding to the overall change in the battery's internal energy. This law ensures that the model strictly adheres to energy conservation when predicting temperature evolution, providing accurate thermal field analysis basis for early warning of heat-related faults such as thermal runaway.

[0091] Electrochemical reaction law: This law establishes the correlation between the electric field and the ion concentration field, describes the formation and distribution mechanism of the current inside the battery, and considers the ohmic current driven by the potential gradient and the diffusion current driven by the lithium ion concentration gradient. It truly reflects the microscopic process of the electrochemical reaction inside the battery and ensures that the model's prediction of the battery's electrical state conforms to the basic principles of electrochemistry.

[0092] Mechanical Deformation Law: This law describes the internal stress and deformation distribution of a battery caused by factors such as thermal expansion and contraction and external loads. It achieves strong coupling between the thermal field and the force field, and can accurately analyze the changes in mechanical stress caused by temperature rise and compression, providing a reliable mechanical basis for the analysis of mechanical damage and structural failure.

[0093] Step S3: Using the multi-physics condition calculation model, extract the time-domain statistical features, frequency-domain spectral features, microstructure features, and multi-field coupling features of the multi-physics field coupled data sample set. Based on the physical contribution of each feature to the fault occurrence mechanism, perform weighted fusion and dimensionality reduction compression to generate the core fault feature vector.

[0094] Specifically, based on the multi-physics condition calculation model trained above, in-depth and multi-dimensional feature mining and analysis are performed on the preprocessed multi-physics field coupling data sample set. In the specific execution process, the strong nonlinear fitting and physical representation capabilities of the model are first utilized to simultaneously extract four core features covering the full state characterization of the equipment: First, time-domain statistical features, capturing the statistical laws of the mean, variance, kurtosis, skewness, etc. of the operating parameters of the power equipment from the time dimension, reflecting the steady-state and dynamic fluctuation characteristics of the equipment operation; Second, frequency-domain spectral features, converting the time-domain signal to the frequency domain through spectrum analysis, extracting the amplitude, energy, and spectral concentration of each frequency component, revealing the frequency drift and energy distribution changes during the equipment degradation process; Third, microstructure features, extracting microstructure evolution information such as electrode morphology, interface state, and porosity based on microstructure characterization data, directly related to the material-level failure state of the equipment; Fourth, multi-field coupling features, combining the multi-physics field coupling equations injected by the physical constraint layer, deeply characterizing the interaction and coupling strength between multiple fields such as electro-thermal-mechanical-chemical fields, and accurately capturing the fault characterization caused by the synergy of multiple fields.

[0095] After completing multi-dimensional feature extraction, this scheme introduces a weighted fusion mechanism based on physical contribution to address the technical problems of high information redundancy and uneven fault identification across different feature dimensions. In specific implementation, firstly, based on the physical mechanism of the target power equipment's faults, the influence of time-domain statistical features, frequency-domain spectral features, microstructural features, and multi-field coupling features on the fault occurrence, development, and propagation mechanism is quantified through mechanism analysis and historical fault data statistics. Based on this influence level, corresponding weight coefficients are assigned to each feature, giving higher weights to features directly and strongly correlated with the fault cause and weakening the influence of irrelevant or interfering features. Subsequently, the feature vectors of each dimension are multiplied by their corresponding weight coefficients, and then element-wise summation is performed to obtain a fused feature vector that integrates multi-dimensional information and possesses physical orientation.

[0096] To further reduce feature dimensionality, eliminate redundant information, and improve the efficiency and accuracy of subsequent model inference, this step performs principal component analysis (PCA) dimensionality reduction on the fused feature vector. PCA extracts the most informative and variance-contributing key feature components from the fused feature vector, discarding low-contribution redundant components. The retained key feature components form a low-dimensional, highly discriminative core fault feature vector. This core fault feature vector not only fully preserves the key information reflecting the essence of the fault but also significantly reduces the computational overhead caused by feature dimensionality. Furthermore, weighted assignment strengthens feature attributes strongly correlated with the fault mechanism, providing high-quality, highly discriminative feature input support for subsequent small-sample fault analysis and fault mechanism interpretation.

[0097] Based on the weighted fusion and dimensionality reduction of each feature's physical contribution to the fault occurrence mechanism, a core fault feature vector is generated, including:

[0098] ① Based on the physical mechanism of the failure of the target power equipment, determine the degree of influence of the time-domain statistical characteristics, frequency-domain spectral characteristics, microstructure characteristics and multi-field coupling characteristics on the failure mechanism, and assign weight coefficients to each characteristic based on the degree of influence;

[0099] ② Multiply the time-domain statistical feature vector, frequency-domain spectral feature vector, microstructure feature vector, and multi-field coupling feature vector by their respective weighting coefficients, and then sum them element by element to obtain the fused feature vector;

[0100] ③ Perform principal component analysis on the fused feature vector to extract the main feature components of the fused feature vector, and use the dimensionality-reduced vector formed by the main feature components as the core fault feature vector.

[0101] Step S4: Based on the core fault feature vector and the node attributes of the fault mechanism knowledge graph, a guided prompt template is constructed. The initial analytical model, which has been pre-trained with fault data in the power equipment field, is then fine-tuned using a meta-learning algorithm to obtain a large fault analysis model.

[0102] Specifically, based on the core fault feature vectors generated in the preceding steps and combined with the physical attributes corresponding to each node in the fault mechanism knowledge graph, a guided prompt template specifically for faults in new energy ship power equipment is constructed. This provides the subsequent model reasoning process with clear physical logic guidance and structured analytical constraints, enabling accurate reasoning according to the fault mechanism path. In the model construction stage, an analytical initial model that has been pre-trained using fault data in the power equipment field is used as the foundation. This model has fully learned the common fault rules and feature patterns in the field and has good initial reasoning capabilities. To enable the model to quickly adapt to small-sample scenarios with scarce rare fault samples and diverse fault types, this application introduces a meta-learning algorithm to fine-tune the parameters of the analytical initial model, achieving rapid adaptation of the model to the target fault scenario with only a small number of samples.

[0103] The meta-learning algorithm's parameter update strategy starts with the pre-trained model parameters and iterates along the gradient direction of the task loss function according to a set fine-tuning step size. The fine-tuning step size controls the magnitude of parameter updates, ensuring model convergence stability without destroying learned domain knowledge. The task loss function is constructed based on fault type analysis accuracy and measures the model's fault discrimination and mechanism analysis accuracy on a small sample set. The small sample support set consists of a preset number of typical fault samples, providing target scene supervision information for the fine-tuning process. Through this meta-learning update strategy, adaptive optimization of model parameters can be completed under conditions with very few labeled samples, enabling the model to quickly acquire specific analytical capabilities for multi-physics coupled faults in new energy ship power equipment. Ultimately, a large-scale fault analysis model with strong adaptability and high inference accuracy is obtained. This model effectively solves the problems of traditional deep learning's excessive reliance on large-scale labeled data and weak generalization ability for rare fault scenarios, maintaining high stability and high analytical accuracy under small sample conditions.

[0104] The update strategy for fine-tuning parameters in few-sample scenarios using meta-learning algorithms is as follows:

[0105] ;

[0106] In the formula, These are the initial analytical model parameters after pre-training with fault data from the power equipment field. These are the parameters after fine-tuning; Fine-tuning the step size for meta-learning; A task loss function constructed based on the accuracy of fault type parsing; It is a small sample support set containing a preset number of fault samples.

[0107] The meta-learning parameter update strategy employed in this step primarily utilizes gradient descent optimization to achieve rapid model adaptation in small-sample scenarios. Starting with analytical initial model parameters pre-trained on fault data from the power equipment domain, these parameters have fully learned the general patterns of faults in new energy ship power equipment, providing a solid domain knowledge foundation for the model and avoiding underfitting issues caused by the model learning from scratch during small-sample fine-tuning. Using a task loss function constructed based on fault type analysis accuracy as the optimization objective, the gradient of the loss function relative to the model parameters is calculated. This gradient clearly indicates the direction of parameter optimization; adjusting the parameters can reduce the fault analysis error in small-sample scenarios and improve the accuracy of fault identification and mechanism analysis. The meta-learning fine-tuning step size controls the magnitude of each parameter update, ensuring rapid model convergence while avoiding over-updating and destroying the domain-general knowledge already learned by the model. Finally, the initial parameters are adjusted along the gradient direction to obtain the fine-tuned parameters adapted to the target small-sample scenario, completing the small-sample adaptation optimization of the model.

[0108] This update strategy combines domain-wide knowledge with small-sample scenario adaptation, which can quickly improve the model's fault analysis capability in the target scenario with only a small number of labeled fault samples, and adapt to application scenarios where rare fault data of new energy ship power equipment is scarce and fault types are diverse.

[0109] Step S5: Using the fault analysis big model and the fault mechanism knowledge graph as the link, by tracing the fault evolution path, defining the critical conditions of multi-field coupling, and establishing the micro-macro parameter mapping relationship, the fault mechanism analysis results containing the fault cause, propagation link and quantification threshold are generated.

[0110] Specifically, based on the trained fault analysis model, and using the fault mechanism knowledge graph constructed in the previous step as the core link, the entire chain mechanism analysis of faults in new energy ship power equipment is carried out from three dimensions: fault evolution tracing, multi-field coupling critical condition definition, and micro-macro parameter mapping. Finally, a complete fault mechanism analysis result containing fault causes, propagation links, and quantification thresholds is generated.

[0111] In the fault evolution path tracing stage, a fault mechanism knowledge graph is used as the reasoning carrier. The initial trigger node in the graph is the search starting point, and the target fault node is the search endpoint. A depth-first search algorithm traverses the nodes and related edges of the knowledge graph. During the search, only strongly related edges with weights greater than a preset first threshold are retained, while weakly related and interfering node links are filtered out, thus outputting a complete fault propagation chain from the initial trigger to the target fault. Based on this, the path confidence of the generated fault propagation chain is calculated. By multiplying the weights of each strongly related edge in the propagation chain, the physical rationality and credibility of the propagation chain are quantitatively evaluated. When the path confidence exceeds a preset second threshold, the fault propagation chain is determined to be a valid fault evolution path, clarifying the complete transmission process of the fault from a microscopic trigger to a macroscopic fault, and accurately locating the root cause and propagation law of the fault.

[0112] In the multi-field coupling critical condition definition stage, the fault analysis big model relies on the multi-field coupling characteristics output by the multi-physics condition calculation model, combined with the physical properties of the multi-field coupling condition nodes in the fault mechanism knowledge graph, to conduct in-depth analysis of the multi-physics field interaction of power equipment under complex working conditions, accurately identify and quantify the multi-field coupling critical conditions for fault occurrence, clarify the cooperative state of different physical field parameters that will trigger the corresponding fault, and provide a physical basis for setting the fault warning threshold.

[0113] In the process of establishing the micro-macro parameter mapping relationship, an exponential nonlinear fitting function is used to construct a quantitative mapping relationship between microstructural parameters and macroscopic fault characterization parameters. This fitting function takes microstructural parameters as input and macroscopic fault characterization parameters as output. The fitting coefficients are calibrated using multi-field stress-accelerated aging experimental data, achieving a precise mapping from microstructural degradation to macroscopic fault performance. This allows for direct prediction of macroscopic operational faults based on microstructural conditions, providing microscopic-level support for early fault warning and condition assessment.

[0114] Through the above multi-dimensional analysis, the final fault mechanism analysis results, which include the cause of the fault, the complete propagation link and the quantitative threshold, are generated. The results clearly reveal the physical nature of the faults in the power equipment of new energy ships, and serve as the data basis for the formulation of operation and maintenance solutions.

[0115] Tracing the failure evolution path includes:

[0116] Using the initial cause node in the fault mechanism knowledge graph as the search starting point and the target fault node as the search ending point, a depth-first search algorithm is used to traverse the fault mechanism knowledge graph. During the search process, only strongly related edges with edge weights greater than a first threshold are retained, and the fault propagation chain is output:

[0117] ;

[0118] Calculate the path confidence of the fault propagation chain. When the path confidence is greater than a second threshold, the fault propagation chain is determined to be a valid fault evolution path.

[0119] The formula for calculating the path confidence is as follows:

[0120] ;

[0121] In the formula, This is a fault propagation chain; This is the initial trigger node; The faulty node; For path confidence; The edge weight is denoted as .

[0122] When tracing the failure evolution path, a failure mechanism knowledge graph is used as the core reasoning vehicle. Specifically, the initial trigger node in the graph serves as the search starting point, and the target failure node as the search endpoint. A depth-first search algorithm is employed to traverse the failure mechanism knowledge graph. During the traversal, a strong-association edge filtering strategy is implemented, retaining only edges with weights greater than a first threshold, while filtering out edges and nodes with weak or insufficient physical association. This eliminates interfering links and ensures that the output failure propagation chain possesses rigorous physical logic. Through this algorithm, an ordered failure propagation chain is output, from the initial trigger node through intermediate associated nodes to the target failure node, clearly presenting the complete transmission process of the failure from its initial initiation to its final manifestation.

[0123] Based on this, the path confidence of the fault propagation chain is calculated. The path confidence is obtained by multiplying the weights of all strongly correlated edges in the fault propagation chain. This calculation method comprehensively reflects the cumulative effect of the correlation strength between nodes in the entire propagation chain. The calculated path confidence is compared with a second threshold. When the path confidence is greater than the second threshold, the fault propagation chain is determined to be a valid fault evolution path, confirming that the chain truly reflects the actual evolution logic and transmission mechanism of the fault, providing clear structured path support for fault cause localization and propagation mechanism analysis.

[0124] Meanwhile, an exponential nonlinear fitting function can be used to establish the micro-macro parameter mapping relationship, the expression of which is:

[0125] ;

[0126] In the formula, These are parameters characterizing macroscopic faults, including capacity decay rate or voltage fluctuation amplitude. These are microstructure parameters, including SEI film thickness or electrode porosity; , , These are the fitting coefficients calibrated using the least squares method based on multi-field stress accelerated aging experimental data.

[0127] To establish a quantitative correlation between microstructural degradation and macroscopic failure performance, this step employs an exponential nonlinear fitting function to construct a micro-macro parameter mapping relationship. This function uses microstructural parameters as input variables and macroscopic failure characterization parameters as output variables, characterizing the impact of microstructural evolution on macroscopic performance through an exponential nonlinear relationship. In practice, based on a large amount of sample data obtained from multi-field accelerated aging experiments, the coefficients in the fitting function are calibrated using the least squares method to ensure that the mapping relationship has high fitting accuracy and high prediction precision.

[0128] Once this mapping relationship is established, a precise quantitative conversion from microstructural parameters to macroscopic fault characterization parameters is achieved. By monitoring changes in microstructural parameters, the degree of degradation of the macroscopic operating state can be directly predicted, and conversely, the degree of damage to the microstructure can be inferred from macroscopic fault manifestations. This mechanism overcomes the limitation of the separation between microscopic state and macroscopic manifestation in traditional analysis, providing crucial quantitative data for early fault warning, lifetime prediction, and joint micro-macroscopic diagnosis.

[0129] Step S6: Generate an operation and maintenance solution based on the fault mechanism analysis results, and perform incremental closed-loop updates on the multi-physical condition calculation model and the fault mechanism knowledge graph based on actual ship operation feedback data.

[0130] Specifically, this step, based on the complete fault mechanism analysis results output by the aforementioned fault analysis big model, carries out intelligent generation of operation and maintenance solutions, and relies on real ship operation feedback data to realize incremental closed-loop updates of multi-physical condition calculation models and fault mechanism knowledge graphs, forming a full-process closed-loop iterative mechanism of data collection, model calculation, fault analysis, operation and maintenance feedback, and model optimization, continuously ensuring the long-term stability of operation and maintenance analysis accuracy and scenario adaptability.

[0131] In the operation and maintenance (O&M) solution generation stage, a rule engine with built-in O&M knowledge rules performs structured matching and reasoning on the core information in the fault mechanism analysis results. Specifically, the rule engine takes key information such as fault type, multi-field coupling critical conditions, and fault propagation path as input, and performs precise matching with a preset O&M knowledge rule base. Combining the O&M specifications, industry standards, and historical handling experience of new energy ship power equipment, it automatically generates an O&M solution that includes standardized operating instructions, handling priorities, operating sequences, and safety precautions. At the same time, the solution uses a fault mechanism knowledge graph as a visualization carrier to present the fault causes, propagation links, and quantification thresholds in an intuitive graph form, providing O&M personnel with clear fault logic and handling basis, realizing intelligent, standardized, and interpretable O&M decision-making, and significantly improving the efficiency and accuracy of O&M handling.

[0132] In the incremental closed-loop update phase, sufficient experimental data and real-ship operation feedback data are collected according to a preset cycle. The experimental data is used to supplement the sample support under extreme operating conditions and rare fault scenarios, while the real-ship operation feedback data truly reflects the actual operating status and fault handling effects of the ship during navigation. Based on the newly collected incremental data, incremental training is performed on the multi-physics condition calculation model and fault mechanism knowledge graph. Through small-batch iterative optimization, without destroying the core knowledge of the original model and graph, the model learns the feature patterns and fault mechanisms in the new data, generates corresponding update parameter vectors, and realizes the rapid adaptation of the model and graph to new operating conditions and new fault types.

[0133] To ensure the stability and continuity of model parameter updates and avoid parameter oscillations caused by fluctuations in single incremental data, this step employs an exponentially weighted moving average parameter fusion strategy. This strategy uses a preset historical parameter retention coefficient as weight to weight and fused the current model parameters, graph parameters, and the updated parameter vector obtained from incremental training. While preserving historical training results and domain knowledge, it incorporates optimization information from new data, ultimately yielding updated model and graph parameters. This fusion strategy ensures continuous iterative optimization of the model and graph while maintaining the smoothness and robustness of parameter updates, achieving dynamic closed-loop optimization throughout the entire lifecycle. This ensures long-term stability of the accuracy of operation and maintenance analysis for new energy ship power equipment, continuously adapting to dynamic changes in ship operation scenarios, and providing full-cycle technical support for ship navigation safety and efficient operation and maintenance.

[0134] Based on the fault mechanism analysis results, an operation and maintenance solution is generated, and incremental closed-loop updates are performed on the multi-physical condition calculation model and the fault mechanism knowledge graph based on actual ship operation feedback data, including:

[0135] ① By using a rule engine with built-in operation and maintenance knowledge rules, the fault type, multi-field coupling critical conditions and fault propagation path information in the fault mechanism analysis results are matched by rules to generate an operation and maintenance processing solution that includes standardized operation instructions and a visual map.

[0136] ② Collect no less than a first preset number of experimental data and a second preset number of actual ship operation feedback data according to a preset cycle. Based on the newly collected data, perform incremental training on the multi-physical condition calculation model and fault mechanism knowledge graph to obtain an updated parameter vector.

[0137] ③ An exponentially weighted moving average parameter fusion strategy is adopted to weight and fuse the current parameter vector with the updated parameter vector according to a preset historical parameter retention coefficient to obtain the updated model parameters and map parameters.

[0138] As can be seen from the above technical solutions, the new energy ship power equipment operation and maintenance analysis method provided in this application collects multi-physics field operation, microstructure and historical fault data and performs standardized processing to construct a data-driven model with a physical constraint layer, extracts multi-dimensional features and weighted fusion, obtains a large fault analysis model based on meta-learning fine-tuning, analyzes the fault mechanism and generates operation and maintenance solutions, and realizes incremental closed-loop updates of the model and knowledge graph.

[0139] This application effectively eliminates the heterogeneity and temporal bias of multi-source data through time-series deviation correction and data standardization, generating a high-quality multi-physics coupled data sample set. This provides a reliable data foundation for subsequent model training and feature extraction, significantly improving the accuracy of data support for operation and maintenance analysis. By constructing a physical constraint layer that embeds physical field coupling equations and a fault mechanism knowledge graph, and injecting it into the training loss function of the data-driven model, a multi-physics condition calculation model is trained. This model can accurately characterize the complex operational characteristics of "electric-thermal-mechanical-chemical" multi-physics coupling in new energy ship power equipment, clearly capture the evolution law of irregular multi-field coupling faults, and greatly improve the model's adaptability to complex working conditions such as extreme sea states and dynamic loads, as well as differentiated navigation scenarios. By extracting time-domain statistical features, frequency-domain spectral features, microstructural features, and multi-field coupling features through the multi-physics condition calculation model, and combining the weighted fusion and dimensionality reduction of each feature to the physical contribution of the fault occurrence mechanism, a core fault feature vector is generated. This vector can accurately capture the core essence of the fault, providing comprehensive and accurate feature support for fault analysis. By fine-tuning the parameters of an initial analytical model pre-trained with fault data in the field of power equipment using meta-learning algorithms for small-sample scenarios, a large-scale fault analysis model is obtained. This model can effectively adapt to scenarios where rare fault data is difficult to obtain, significantly improving the accuracy and generalization ability of fault analysis in small-sample scenarios. Through this large-scale fault analysis model, using a fault mechanism knowledge graph as a link, the model traces the fault evolution path, defines multi-field coupling critical conditions, and establishes a micro-to-macro parameter mapping relationship. This clearly reveals the physical essence of the fault, achieving accurate analysis of fault causes, propagation paths, and quantification thresholds, providing a clear and reliable theoretical basis for operation and maintenance control. Incremental closed-loop updates are performed on the multi-physical condition calculation model and the fault mechanism knowledge graph using real-ship operation feedback data, continuously optimizing the adaptability and accuracy of the model and knowledge graph, ensuring the long-term stability of operation and maintenance analysis results.

[0140] In summary, this application, through the aforementioned series of technical means, achieves accurate diagnosis, clear mechanism analysis, and efficient operation and maintenance of power equipment failures in new energy ships, effectively ensuring ship navigation safety, improving operational efficiency, and reducing operation and maintenance costs.

[0141] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0143] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for analyzing the operation and maintenance of power equipment for new energy ships, characterized in that, include: Collect multi-physics field operation data, microstructure data and historical fault data of new energy ship power equipment, and generate a multi-physics field coupled data sample set through time series deviation correction and data standardization processing; Based on the data sample set, a physical constraint layer is constructed by embedding physical field coupling equations and fault mechanism knowledge graphs, and the physical constraint layer is injected into the training loss function of the data-driven model to train a multi-physics condition calculation model. The time-domain statistical features, frequency-domain spectral features, microstructure features, and multi-field coupling features of the multi-physics field coupled data sample set are extracted using the multi-physics condition calculation model. Based on the physical contribution of each feature to the fault occurrence mechanism, weighted fusion and dimensionality reduction compression are performed to generate a core fault feature vector. A guided prompt template is constructed based on the core fault feature vector and the node attributes of the fault mechanism knowledge graph. The initial analytical model, which has been pre-trained with fault data in the power equipment field, is fine-tuned with small sample scenario parameters through a meta-learning algorithm to obtain a large fault analysis model. Using the aforementioned fault analysis model and the fault mechanism knowledge graph as the link, by tracing the fault evolution path, defining the critical conditions of multi-field coupling, and establishing the micro-macro parameter mapping relationship, the fault mechanism analysis results containing fault causes, propagation links, and quantification thresholds are generated. Based on the fault mechanism analysis results, an operation and maintenance solution is generated, and an incremental closed-loop update is performed on the multi-physical condition calculation model and the fault mechanism knowledge graph based on the actual ship operation feedback data. The process of constructing the fault mechanism knowledge graph includes: Constructing a graph structure The fault mechanism knowledge graph represents the following: Node set ; Nodes that characterize fault types such as thermal runaway, internal short circuit, and capacity decay; Microscopic failure nodes characterizing SEI film decomposition, lithium dendrite growth, and membrane electrode porosity changes; A node representing a multi-field coupling condition for temperature threshold, stress threshold, and voltage descent threshold; Edge set This represents the causal, propagational, and consequential relationships between nodes, with each edge representing these relationships. Configured with weights ; Attribute Matrix Store the physical parameter threshold vectors corresponding to each node; The entity linking model maps the feature parameters of the multiphysics coupled data sample set to the corresponding graph nodes of the fault mechanism knowledge graph. The expression of the entity linking model is as follows: ； In the formula, These are the standardized feature parameters; For graph nodes eigenvectors; The similarity is cosine similarity; a link is considered valid when the similarity is not less than 0.

75.

2. The method according to claim 1, characterized in that, The timing deviation correction is performed by using a multi-physics timing data alignment model, based on the time difference correction amount determined by the time asynchronous deviation between sensor timing data and microscopic test synchronous data. The expression for the multiphysics temporal data alignment model is: ； In the formula, For the aligned timestamp; This is the original data timestamp; This is the time difference correction amount; This is sensor timing data; Synchronize data for microscopic testing; It is an L2 norm.

3. The method according to claim 1, characterized in that, The training loss function is a composite loss function that includes data loss and physical constraint loss, and its expression is: ； ； In the formula, The data loss is determined based on the cross-entropy loss between the model's predicted values ​​and the actual fault labels; The deviation norm between the theoretical physical field distribution value calculated based on the multi-physics coupling equation of the dynamic equipment and the predicted physical field distribution value output by the model is determined as the physical constraint loss. Basic coefficient; To ensure data reliability, when the input data is actual ship operation data, The value ranges from [0.8, 1.0]. When the input data is laboratory simulation data, The value is located in the range [0.5, 0.8]. This is the balance coefficient.

4. The method according to claim 1, characterized in that, Based on the weighted fusion and dimensionality reduction of each feature's physical contribution to the fault occurrence mechanism, a core fault feature vector is generated, including: Based on the physical mechanism of the failure of the target power equipment, the influence of the time-domain statistical characteristics, frequency-domain spectral characteristics, microstructure characteristics and multi-field coupling characteristics on the failure mechanism is determined, and weight coefficients are assigned to each characteristic based on the degree of influence. The time-domain statistical feature vector, frequency-domain spectral feature vector, microstructure feature vector, and multi-field coupling feature vector are multiplied by their respective weighting coefficients and then summed element by element to obtain the fused feature vector. Principal component analysis is performed on the fused feature vector to extract the main feature components, and the dimensionality-reduced vector formed by the main feature components is used as the core fault feature vector.

5. The method according to claim 1, characterized in that, The tracing of the fault evolution path includes: Using the initial cause node in the fault mechanism knowledge graph as the search starting point and the target fault node as the search ending point, a depth-first search algorithm is used to traverse the fault mechanism knowledge graph. During the search process, only strongly related edges with edge weights greater than a first threshold are retained, and the fault propagation chain is output: ； Calculate the path confidence of the fault propagation chain. When the path confidence is greater than a second threshold, the fault propagation chain is determined to be a valid fault evolution path. The formula for calculating the path confidence is as follows: ； In the formula, This is a fault propagation chain; This is the initial trigger node; The faulty node; For path confidence; The edge weight is denoted as .

6. The method according to claim 1, characterized in that, An exponential nonlinear fitting function is used to establish the micro-macro parameter mapping relationship, and its expression is as follows: ； In the formula, These are parameters characterizing macroscopic faults, including capacity decay rate or voltage fluctuation amplitude. These are microstructure parameters, including SEI film thickness or electrode porosity; , , These are the fitting coefficients calibrated using the least squares method based on multi-field stress accelerated aging experimental data.

7. The method according to claim 1, characterized in that, The update strategy for fine-tuning parameters in few-sample scenarios using a meta-learning algorithm is as follows: ； In the formula, These are the initial analytical model parameters after pre-training with fault data from the power equipment field. These are the parameters after fine-tuning; Fine-tuning the step size for meta-learning; A task loss function constructed based on the accuracy of fault type parsing; It is a small sample support set containing a preset number of fault samples.

8. The method according to claim 1, characterized in that, Based on the fault mechanism analysis results, an operation and maintenance solution is generated, and incremental closed-loop updates are performed on the multi-physical condition calculation model and the fault mechanism knowledge graph based on actual ship operation feedback data, including: By using a rule engine with built-in operation and maintenance knowledge rules, the fault type, multi-field coupling critical conditions and fault propagation path information in the fault mechanism analysis results are matched by rules to generate an operation and maintenance processing solution that includes standardized operation instructions and a visual map. Collect no less than a first preset number of experimental data and a second preset number of actual ship operation feedback data according to a preset cycle. Based on the newly collected data, perform incremental training on the multi-physical condition calculation model and fault mechanism knowledge graph to obtain an updated parameter vector. An exponentially weighted moving average parameter fusion strategy is adopted to weight and fuse the current parameter vector with the updated parameter vector according to a preset historical parameter retention coefficient, so as to obtain the updated model parameters and map parameters.

Images