A fault assessment method and system based on ship maintenance

Abstract

The present invention relates to the field of fault assessment technology and discloses a ship maintenance-based fault assessment method and system, comprising: utilizing a knowledge graph to perform node fault assessment, and constructing a health risk distribution based on the results of the fault assessment; migrating the health risk distribution between nodes through analysis of ship maintenance records to obtain a risk distribution model adapted to the ship; actively generating risk probabilities of fault combinations in the node space and risk probabilities of single node failures based on the risk distribution model; generating fault hypotheses based on the risk probabilities of fault combinations in the node space and risk probabilities of single node failures, and evaluating the coverage of each fault hypothesis with respect to the risk distribution model, ultimately outputting a fault assessment result. This method effectively overcomes the problem of traditional methods' lack of adaptability to single ship types and specific operating conditions, and enhances the personalization and generalization capabilities of the risk model.

Description

Technical Field

[0001] The present invention relates to the technical field of fault assessment, and in particular to a fault assessment method and system based on ship maintenance. Background Art

[0002] With the continuous development of the global shipping industry, modern ships are constantly improving in terms of functional integration, structural complexity, and degree of automation. As large-scale engineering equipment, ships involve numerous subsystems such as power systems, propulsion systems, navigation, electronic and electrical systems, hydraulic systems, piping, and outfitting. These components operate under long-term high loads and complex marine environments, making them extremely susceptible to various physical and chemical factors, leading to problems such as wear, corrosion, loosening, aging, and failure. Equipment failures can not only lead to voyage delays and increased operating costs, but also pose serious risks to personal safety and environmental pollution.

[0003] At present, ship maintenance management is mostly based on regular inspections, planned maintenance and experience-based judgment; traditional maintenance models rely on manual inspections, periodic disassembly or simple threshold alarms; these methods often have delayed responses and find it difficult to detect complex faults caused by hidden, gradual or multi-point coupling, and cannot meet the high requirements of modern ships for safety and economy; in recent years, with the development of sensor technology, ship automation and intelligent monitoring systems, the ability to collect ship operation data has been greatly improved, providing more basis for systematic health management and scientific decision-making; however, the effective integration of massive data, the accurate extraction of abnormal features, and the fault propagation and causal correlation analysis of multi-node complex systems still face huge challenges; how to make full use of the new generation of information technology and data analysis tools to improve the scientificity and intelligence level of early identification of ship faults and risk warnings has become an important technical development direction in the field of ship engineering and intelligent operation and maintenance. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a fault assessment method based on ship maintenance, comprising:

[0006] During the normal operation of the ship, multimodal data of each monitoring part is collected;

[0007] Using the knowledge graph to perform node fault assessment, and constructing a health risk distribution based on the results of the fault assessment;

[0008] By analyzing ship maintenance records, the distribution of health risks between nodes is migrated to obtain a risk distribution model that adapts to ships;

[0009] Actively generate risk probabilities of failure combinations and single node failures in terms of node space based on the risk distribution model;

[0010] According to the risk probability of the combination of faults in the node space and the risk probability of a single node failure, a fault hypothesis is generated, and the coverage of each fault hypothesis for the risk distribution model is evaluated, and finally a fault assessment result is output.

[0011] As a preferred embodiment of the ship maintenance-based fault assessment method of the present invention, the monitoring portion includes mechanical components and subsystems in the ship structure that can be independently identified and are suitable for installing sensors;

[0012] The multimodal data includes data used to describe the operation characteristics of the monitored part according to the type of sensor in each monitored part.

[0013] As a preferred embodiment of the ship maintenance-based fault assessment method of the present invention, the knowledge graph includes: treating each component in the ship as a node, and constructing unidirectional or bidirectional causal relationships based on actual connections, functional dependencies, and relationships between entity nodes extracted from historical maintenance data of the sample; wherein the causal relationships between nodes are learned through a Bayesian network;

[0014] If the same monitoring part contains multiple nodes, the multimodal data of the nodes in the monitoring part are generalized so that each node in the monitoring part synchronizes data according to the data distribution law;

[0015] The data distribution rule includes, for each monitoring part, mapping each node in the monitoring part to a corresponding monitoring data value through a pre-trained mapping relationship and a measurement value of a sensor.

[0016] As a preferred embodiment of the ship maintenance-based fault assessment method of the present invention, the distribution of health risks includes, based on the monitoring data of each node, taking the health status of each node as the target variable, and combining the causal relationship obtained based on the Bayesian network in the knowledge graph;

[0017] Output the probability distribution of each node under different fault conditions to form the health risk distribution D.

[0018] As a preferred embodiment of the ship maintenance-based fault assessment method of the present invention, the migration of the health risk distribution between nodes includes generating an adjusted health risk distribution using each maintenance record and the maintenance records of ships of the same type; and comprehensively generating a health risk distribution migration result by combining all maintenance records of the current ship;

[0019] The specific steps for generating the adjustment distribution are:

[0020] Step 1: Generate the distribution D2 of health risks of ships of the same type based on the maintenance records of ships of the same type;

[0021] Step 2: Generate weights and fault distribution based on each maintenance record of the current ship; D 1,i It represents the probability distribution of failure when the current ship is maintained for the i-th time; the weight of the current ship at the i-th maintenance time is defined as: w 1,i =α×F(f i ), α represents the preset adjustment coefficient; f i It represents the time interval between the i-th maintenance and the previous maintenance; F() represents the mapping function, which is to transform f i The corresponding value is a reference value obtained through a preset function relationship; when the current ship is maintained for the i-th time, the weight of the health risk distribution for the same type of ship is defined as: w 2,i =1-w 1,i ;

[0022] Step 3: Assume that the current ship has undergone I maintenance, and combine the I maintenance records:

[0023]

[0024] Among them, D ’ represents the probability distribution after combination; G() represents the normalization function; D 2,i Always equal to D2;

[0025] The risk distribution model includes generating an attention weight for each node based on the adjusted distribution; performing a weighted adjustment on the probability distribution of the health risk distribution D according to the attention weight at each node and then normalizing the weighted adjustment to achieve distribution migration of the health risk.

[0026] As a preferred embodiment of the ship maintenance-based fault assessment method of the present invention, the risk probability of a single node failure comprises: performing probability calculation on each node i in the risk distribution model to obtain the failure probability of node i in the risk distribution model; obtaining a risk probability sequence of a single node failure;

[0027] The risk probability of the fault combination includes calculating the probability of a fault occurring for each group of multiple nodes with an associated relationship using the risk distribution model; and obtaining a risk probability sequence of the fault combination using the average value of the failure probabilities of the nodes with an associated relationship;

[0028] The fault hypothesis includes randomly combining the risk probability sequence of the single node failure and the risk probability sequence of the fault combination;

[0029] Evaluating the coverage of each portfolio with respect to the risk distribution model;

[0030] The coverage includes re-assigning a value to each node in the fault hypothesis and recalculating the value distribution of the node space according to the assignment result; calculating the similarity between the value distribution and the risk distribution model, and using the similarity as the coverage of each fault hypothesis;

[0031] If the node is a node failure in the fault combination in the fault hypothesis, the node is assigned a value according to the risk probability of the corresponding fault combination; if the node is a single node failure in the fault hypothesis, the node is directly assigned a value according to the risk probability of the corresponding single node failure.

[0032] As a preferred solution of the ship maintenance-based fault assessment method of the present invention, the fault hypotheses are output in descending order of coverage as fault assessment results.

[0033] A fault assessment system based on ship maintenance using the method of the present invention, wherein:

[0034] The acquisition unit collects multimodal data of each monitoring part during the normal operation of the ship;

[0035] A distribution unit, which uses the knowledge graph to perform fault assessment of nodes and constructs a distribution of health risks based on the results of the fault assessment;

[0036] The migration unit migrates the distribution of health risks between nodes by analyzing the ship maintenance records to obtain a risk distribution model adapted to the ship; based on the risk distribution model, it actively generates the risk probability of failure combinations in the node space and the risk probability of single node failure;

[0037] The output unit generates a fault hypothesis based on the risk probability of the fault combination in the node space and the risk probability of a single node failure, evaluates the coverage of each fault hypothesis for the risk distribution model, and finally outputs a fault assessment result.

[0038] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: when the processor executes the computer program, the steps of any one of the methods of the present invention are implemented.

[0039] A computer-readable storage medium stores a computer program, wherein: when the computer program is executed by a processor, the steps of any one of the methods of the present invention are implemented.

[0040] Beneficial effects of the present invention: The fault assessment method based on ship maintenance provided by the present invention realizes accurate assessment of the health status of key components of the ship and dynamic risk distribution modeling by integrating multimodal monitoring data, knowledge graph modeling and Bayesian causal reasoning, and significantly improves the early identification capability of systemic faults. Combining maintenance history with empirical data of similar ships, the adaptive migration and attention weighting mechanism is adopted to effectively overcome the problem of insufficient adaptability of traditional methods to single ship types and specific working conditions, and enhance the personalization and generalization capabilities of risk models. The present invention can actively infer the probability of single-point and combined failures, and introduce a probability distribution coverage evaluation mechanism to systematically output highly representative fault hypotheses and evaluation results, thereby providing a scientific basis for intelligent operation and maintenance, precise maintenance and safety decision-making of ships, and greatly improving the safety and economy of ship operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is an overall flow chart of a fault assessment method based on ship maintenance provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0043] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0044] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a fault assessment method based on ship maintenance, comprising:

[0045] S1: During the normal operation of the ship, multimodal data of each monitoring part is collected.

[0046] The monitoring part includes mechanical components and subsystems in the ship structure that can be independently identified and suitable for installing sensors; including but not limited to: the main cylinders, bearings, connecting rods, cylinder heads, and exhaust pipe sections of the main engine (diesel engine / main propulsion unit). Bearings, couplings, sealing chambers, tail shaft sleeves, etc. of the propulsion shaft system. Rotating or power units such as gearboxes, reducers, pumps, fans, hydraulic cylinders, and oil valve groups. Electrical components such as electrical control cabinets, generators, and motors. Critical compartments (such as engine rooms, pump rooms, cooling water tanks) or special structural nodes. And local functional units of the above components (such as lubricating oil pump seats, cooling water pipe ports, local pressure nodes, etc.).

[0047] Furthermore, the multimodal data includes data used to describe the operating characteristics of the monitored part according to the type of sensor in each monitored part.

[0048] By deploying multiple types of sensors in key mechanical components and subsystems within the ship's structure, it is possible to continuously collect and acquire multimodal operating parameters including vibration, temperature, pressure, current, flow, and sound. These multimodal data can fully reflect the true operating status and dynamic change characteristics of each monitored part under different operating conditions. This not only lays a solid data foundation for subsequent intelligent analysis steps such as fault identification, health assessment, and risk modeling, but also significantly enhances the system's ability to perceive early hidden dangers, complex coupled faults, and atypical anomalies. This helps realize the transformation of ship operation and maintenance from experience-driven to data-driven, and from passive response to active prediction, thereby ensuring the safety and efficiency of ship operations.

[0049] S2: Use the knowledge graph to perform node fault assessment, and construct a health risk distribution based on the results of the fault assessment.

[0050] Furthermore, the knowledge graph includes treating each component of a ship as a node and constructing unidirectional or bidirectional causal relationships based on actual connections, functional dependencies, and relationships between entity nodes extracted from historical maintenance data of samples (collected from multiple ships). The causal relationships between nodes are learned using a Bayesian network. This is essentially transfer learning of relationships between nodes. By learning from multiple ship samples, the resulting knowledge graph can reflect the associations between nodes within the broad category of "ship." Then, through the "Using the current ship's maintenance records for migration" described later, migration is performed based on the unique characteristics of the current ship.

[0051] By abstracting the various components of a ship as nodes in a knowledge graph and systematically extracting and modeling multi-level, multi-type causal relationships between nodes based on structural connections, functional dependencies, and large-sample historical maintenance data, we can fully reflect the linkage and influence mechanisms between the various elements in a complex ship system. By learning and expressing causal relationships with the help of Bayesian networks, the knowledge graph is equipped with dynamic reasoning and uncertainty modeling capabilities. Furthermore, through transfer learning of sample data from multiple ship types, the industry-wide "ship system" universal laws can be summarized and refined. Furthermore, by combining the current ship's own maintenance history, migration adaptation based on individual differences can be achieved, enhancing the accuracy and interpretability of the model for different ships and different operating scenarios.

[0052] If the same monitoring section contains multiple nodes, the multimodal data of the nodes in the monitoring section is generalized, and each node in the monitoring section is synchronized according to the data distribution rules. The data distribution rules include, for each monitoring section, mapping each node in the monitoring section to the corresponding monitoring data value through pre-trained mapping relationships and sensor measurements. Generalizing and synchronizing the data distribution of multiple nodes within the monitoring section ensures the consistency of feature expression and the standardization of model inputs, providing reliable data support and knowledge guarantee for subsequent health risk inference and fault identification.

[0053] Based on the monitoring data of each node, the health status of each node is used as the target variable, and combined with the causal relationship obtained based on the Bayesian network in the knowledge graph, the probability distribution of each node under different fault states is output to form the health risk distribution D.

[0054] It's important to note that a Bayesian network is a directed acyclic graph (DAG) structure that effectively represents and calculates conditional dependencies between variables in a system. In the context of ship health management, a Bayesian network treats each monitoring node as a random variable node within the network, embedding the structural, functional, and maintenance history dependencies between components into the network structure as edges. By inputting node monitoring data, the Bayesian network utilizes conditional probabilistic inference techniques to calculate the posterior probability distribution for each node under different health / fault states. This inference method not only handles multi-source, incomplete, and uncertain data, but also reflects the complex causal linkages and anomaly propagation mechanisms between nodes. By combining knowledge graphs with Bayesian networks, the global representation of ship system structure and operation and maintenance knowledge is ensured, while also enhancing the ability to characterize industry-wide common laws and ontological semantics during the inference process. As a result, the health risk distribution D output by the model not only possesses strict probabilistic meaning but also exhibits good physical interpretability and industry adaptability.

[0055] S3: By analyzing the ship maintenance records, the distribution of health risks between nodes is migrated to obtain a risk distribution model that adapts to the ship.

[0056] The migration of the distribution of health risks between nodes includes using each maintenance record and the maintenance records of ships of the same type to generate an adjusted distribution of health risks; combining all maintenance records of the current ship to comprehensively generate the distribution migration results of health risks.

[0057] The adjusted distribution is derived from the health risk distribution obtained by introducing maintenance data of similar ships. This serves as an industry baseline or group knowledge, providing a strong empirical reference for risk modeling of current ships. At the same time, by combining each maintenance record of the current ship and assigning dynamic weights to each maintenance interval (frequency), it can reflect the unique fault evolution and maintenance response patterns of the target ship in actual operation. The specific generation steps are:

[0058] Step 1: Generate the distribution D2 of health risks of ships of the same type based on the maintenance records of ships of the same type.

[0059] Step 2: Generate weights and fault distribution based on each maintenance record of the current ship; D 1,i It represents the probability distribution of failure when the current ship is maintained for the i-th time; the weight of the current ship at the i-th maintenance time is defined as: w 1,i =α×F(f i ), α represents the preset adjustment coefficient; f i represents the time interval between the i-th maintenance and the previous maintenance (this is actually the frequency factor, which is the concentration of the current maintenance in time in the entire maintenance history); F() represents the mapping function, which is to transform f i The corresponding value is a reference value obtained by a preset function relationship (which can be nonlinear or linear, but the result value must be negatively correlated with the input value); when the current ship is maintained for the i-th time, the weight of the health risk distribution for the same type of ship is defined as: w 2,i =1-w 1,i .

[0060] Step 3: Assume that the current ship has undergone I maintenance, and combine the I maintenance records:

[0061]

[0062] Among them, D ’ represents the probability distribution after combination; G() represents the normalization function; D 2,i Always equal to D2.

[0063] The risk distribution model includes generating an attention weight for each node based on the adjusted distribution; performing a weighted adjustment on the probability distribution of the health risk distribution D according to the attention weight at each node and then normalizing the weighted adjustment to achieve distribution migration of health risks (personalized configuration for subtle features of the ship, that is, this step is to learn and migrate the current fault through the maintenance records of similar and current ships).

[0064] D ’ →θ

[0065] D + =θ·D

[0066] Wherein, θ represents the attention weight of each node. → represents the mapping relationship. In this embodiment, the preset linear transformation is not used. In other feasible embodiments, the output can be obtained through a pre-trained neural network. + Represents the risk distribution model.

[0067] First, by introducing maintenance data from similar vessels, an industry baseline for health risk distribution is formed, enabling the model to inherit and leverage group experience and provide a good foundation for generalization. Second, by combining all maintenance records of the current target vessel itself, dynamic weights are assigned based on the actual frequency of each maintenance and the manifestation of the failure, reflecting the unique risk evolution characteristics of the target vessel during actual operation, management, and maintenance. This weighting design based on maintenance intervals (frequency) and failure distribution can highlight the impact of recent, frequent, or abnormal maintenance behaviors on risk distribution, fully embodying the core concept of "personalized migration."

[0068] Through multiple fusions of these weights and distributions, combined with normalization, we ultimately achieve a risk distribution migration result that accurately describes the target ship's health status and adaptively reflects its evolution. This adjustment is then used to generate attention weights for each node, and the risk distribution model in the node space is probabilistically weighted and normalized. This allows the model to dynamically learn and efficiently adapt to the ship's structure, operating conditions, and historical behavior, providing a solid data and knowledge foundation for intelligent risk assessment and proactive operation and maintenance decision-making.

[0069] S4: Actively generate risk probabilities of failure combinations in the node space and risk probabilities of single node failures based on the risk distribution model.

[0070] Specifically, the risk probability of a single node failure includes performing probability calculation on each node i in the risk distribution model to obtain the failure probability of node i in the risk distribution model; and obtaining a risk probability sequence of a single node failure.

[0071] The risk probability of the fault combination includes calculating the probability of failure for each group of multiple nodes with an associated relationship using the risk distribution model, and obtaining the risk probability sequence of the fault combination using the average value of the failure probabilities of the nodes with an associated relationship.

[0072] S5: Generate a fault hypothesis based on the risk probability of the fault combination in the node space and the risk probability of a single node failure, evaluate the coverage of each fault hypothesis for the risk distribution model, and finally output a fault assessment result.

[0073] Furthermore, the risk probability sequence for single-node failures and the risk probability sequence for combinations of failures are randomly combined. By randomly combining the risk probability sequence for single-node failures with the risk probability sequence for combinations of failures, a large number of fault hypotheses of varying levels and structures can be proactively generated. This approach can simulate various single-point failures, multi-point linkage failures, and rare high-risk scenarios that ships may encounter in actual operation, avoiding the limitation of assessment results to a limited set of a priori risk scenarios.

[0074] This randomized hypothesis generation mechanism not only expands the coverage of risk analysis and enhances early warning capabilities for anomalies and boundary events, but also provides a diverse input basis for subsequent distribution coverage calculations and priority sorting, thereby supporting scientific screening of failure modes and intelligent decision-making. Ultimately, this helps improve the systematicity, completeness, and robustness of fault assessments in practical applications.

[0075] Evaluate each combination's coverage of the risk distribution model. This coverage involves reassigning values ​​to each node in the fault hypothesis, recalculating the value distribution in the node space based on the assigned values, and calculating the similarity between this value distribution and the risk distribution model, using the similarity as the coverage of each fault hypothesis. By reassigning values ​​to each fault hypothesis, constructing its corresponding node space probability distribution, and calculating the similarity between this distribution and the risk distribution model, the representativeness of the hypothesis within the global risk landscape can be effectively reflected. Coverage, as an evaluation criterion, not only quantitatively measures the degree of fit of a fault hypothesis with the overall risk distribution but also allows comparison of the impact and importance of different fault modes within the global risk landscape. This effectively identifies highly representative or typical fault modes, allowing risk assessment results to focus more closely on risk scenarios with significant actual threats, providing a more targeted reference for intelligent operations and maintenance, resource allocation, and emergency decision-making. It also avoids excessive focus on low-relevance or marginal hypotheses, improving the efficiency and scientific nature of risk management.

[0076] If the node belongs to a node failure in the fault combination in the fault hypothesis, the node is assigned the risk probability of the corresponding fault combination; if the node belongs to a single node failure in the fault hypothesis, it is directly assigned according to the risk probability of the corresponding single node failure. By assigning the nodes belonging to the fault combination with the risk probability of the corresponding combined failure, it is possible to accurately express the linkage, amplification or special impact risks faced by the system when multiple nodes fail at the same time, reflecting the systemic complexity brought about by the combined failure; while directly assigning single-point risk probability to the nodes belonging to single node failures highlights the true role and weight of independent failures in the overall risk pattern. This type and situation-based assignment strategy not only avoids the repeated superposition of risk probabilities, but also prevents the dilution of risk information in space, making the node spatial distribution generated by each hypothesis more consistent with actual engineering semantics and operation and maintenance requirements.

[0077] Ultimately, this assignment method can provide a scientific probabilistic basis for subsequent distribution similarity calculation and coverage evaluation, improving the physical interpretability of fault hypothesis assessment and the accuracy of intelligent decision-making.

[0078] The fault hypotheses are output in descending order of coverage as fault assessment results.

[0079] Example 2: This embodiment further provides a fault assessment system based on ship maintenance, which includes:

[0080] The acquisition unit collects multimodal data of each monitoring part during the normal operation of the ship.

[0081] The distribution unit uses the knowledge graph to perform fault assessment of the node and constructs the distribution of health risks according to the results of the fault assessment.

[0082] The migration unit migrates the distribution of health risks between nodes by analyzing the ship maintenance records, and obtains a risk distribution model adapted to the ship; based on the risk distribution model, it actively generates the risk probability of failure combinations in the node space and the risk probability of single node failure.

[0083] The output unit generates a fault hypothesis based on the risk probability of the fault combination in the node space and the risk probability of a single node failure, evaluates the coverage of each fault hypothesis for the risk distribution model, and finally outputs a fault assessment result.

[0084] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0085] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0086] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0087] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or combination of the following technologies known in the art can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A fault assessment method based on ship maintenance, characterized in that: include: During the normal operation of the ship, multimodal data of each monitoring part is collected; Using the knowledge graph to perform node fault assessment, and constructing a health risk distribution based on the results of the fault assessment; By analyzing ship maintenance records, the distribution of health risks between nodes is migrated to obtain a risk distribution model that adapts to ships; Actively generate risk probabilities of failure combinations and single node failures in terms of node space based on the risk distribution model; Generate fault hypotheses based on the risk probability of fault combinations in the node space and the risk probability of single node failures, evaluate the coverage of each fault hypothesis for the risk distribution model, and ultimately output a fault assessment result; The coverage includes re-assigning a value to each node in the fault hypothesis and recalculating the value distribution of the node space according to the assignment result; Calculating similarity between the value distribution and the risk distribution model, and using the similarity as the coverage of each fault hypothesis; The migration of the health risk distribution between nodes includes generating an adjusted distribution of health risks using each maintenance record and the maintenance records of ships of the same type; and generating a comprehensive migration result of the health risk distribution by combining all maintenance records of the current ship; The specific steps for generating the adjustment distribution are: Step 1: Generate the distribution of health risks of ships of the same type based on their maintenance records ; Step 2: Generate weights and fault distribution based on each maintenance record of the current ship; represents the probability distribution of failure when the current ship is maintained for the i-th time; the weight of the current ship at the i-th maintenance time is defined as: , Indicates the preset adjustment coefficient; It represents the time interval between the i-th maintenance and the previous maintenance; Represents the mapping function, which is The corresponding value is a reference value obtained through a preset function relationship. When the current ship is undergoing maintenance for the i-th time, the weight of the health risk distribution for the same type of ship is defined as: ; Step 3: Assume that the current ship has undergone I maintenance, and combine the I maintenance records: ； in, represents the probability distribution after combination; represents the normalization function; Identical to ; The risk distribution model includes generating an attention weight for each node according to the adjusted distribution; performing a weighted adjustment on the probability distribution of the health risk distribution D according to the attention weight at each node and then normalizing the weighted adjustment to achieve distribution migration of the health risk; The knowledge graph includes treating each component of a ship as a node and constructing unidirectional or bidirectional causal relationships based on actual connections, functional dependencies, and relationships between entity nodes extracted from historical maintenance data of the sample. The causal relationships between nodes are learned through a Bayesian network. If the same monitoring part contains multiple nodes, the multimodal data of the nodes in the monitoring part are generalized so that each node in the monitoring part synchronizes data according to the data distribution law; The data distribution rule includes, for each monitoring part, mapping each node in the monitoring part to a corresponding monitoring data value through a pre-trained mapping relationship and a measurement value of a sensor.

2. The ship maintenance-based fault assessment method according to claim 1, wherein: The monitoring part includes mechanical components and subsystems in the ship structure that can be independently identified and are suitable for installing sensors; The multimodal data includes data used to describe the operation characteristics of the monitored part according to the type of sensor in each monitored part.

3. The ship maintenance-based fault assessment method according to claim 2, wherein: The distribution of health risks includes, based on the monitoring data of each node, taking the health status of each node as the target variable, and combining the causal relationship obtained based on the Bayesian network in the knowledge graph; Output the probability distribution of each node under different fault conditions to form the health risk distribution D.

4. The ship maintenance-based fault assessment method according to claim 3, wherein: The risk probability of a single node failure includes performing probability calculation on each node i in the risk distribution model to obtain the failure probability of node i in the risk distribution model; The risk probability of the fault combination includes calculating the probability of a fault occurring for each group of multiple nodes with an associated relationship using the risk distribution model; and obtaining a risk probability sequence of the fault combination using the average value of the failure probabilities of the nodes with an associated relationship; The fault hypothesis includes randomly combining the risk probability sequence of the single node failure and the risk probability sequence of the fault combination; Evaluating the coverage of each portfolio with respect to the risk distribution model; If the node belongs to the node failure in the fault combination in the fault hypothesis, the node is assigned the risk probability of the corresponding fault combination; If the node belongs to a single node failure in the fault hypothesis, the risk probability of the corresponding single node failure is directly assigned.

5. The ship maintenance-based fault assessment method according to claim 4, wherein: The fault hypotheses are output in descending order of coverage as fault assessment results.

6. A ship maintenance-based fault assessment system using the method according to any one of claims 1 to 5, characterized in that: The acquisition unit collects multimodal data of each monitoring part during the normal operation of the ship; A distribution unit, which uses the knowledge graph to perform fault assessment of nodes and constructs a distribution of health risks based on the results of the fault assessment; The migration unit migrates the distribution of health risks between nodes by analyzing the ship maintenance records, and obtains a risk distribution model that adapts to the ship; Actively generate risk probabilities of failure combinations and single node failures in terms of node space based on the risk distribution model; The output unit generates a fault hypothesis based on the risk probability of the fault combination in the node space and the risk probability of a single node failure, evaluates the coverage of each fault hypothesis for the risk distribution model, and finally outputs a fault assessment result.

7. A computer device comprising: memory and processor; The memory stores a computer program, wherein the processor implements the steps of any one of the methods according to claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images