Distributed fault prediction system for military ship control units

Abstract

The application provides a distributed fault prediction system of a military ship control unit, relates to the technical field of fault prediction, and comprises: a framework building module, which is used for building a distributed fault prediction framework; a feature extraction module, which is used for collecting and acquiring a working signal set of a sub-control unit, and performing fault feature extraction on the working signal set of the sub-control unit; a path analysis module, which is used for mining and acquiring a fault correlation rule base, and activating a regional convergence node to perform fault path analysis on an edge node fault feature set based on the fault correlation rule base; and a fault operation and maintenance module, which is used for calling a military ship fault knowledge base, performing integrated prediction on a distributed regional fault path based on the military ship fault knowledge base, obtaining a fault integrated prediction result, and performing fault early warning operation and maintenance of the military ship control unit based on the fault integrated prediction result. Through the application, the technical problem of insufficient fault identification precision in the prior art can be solved, and the technical effect of improving fault early warning accuracy is achieved.

Description

Technical Field

[0001] This application relates to the field of fault prediction technology, and in particular to a distributed fault prediction system for military ship control units. Background Technology

[0002] As the complexity of modern military ship systems continues to increase, their control units are becoming increasingly complex in terms of structural hierarchy, functional distribution, and information interaction. Ensuring their efficient and reliable operation has become the key to the intelligent development of military equipment.

[0003] Currently, in the fault prediction and diagnosis of shipborne systems, traditional methods mostly rely on centralized data acquisition and rule-matching analysis strategies, that is, unified processing and analysis of all monitoring signals through a single control center. This approach has obvious limitations in dealing with massive information growth, increasing real-time requirements, and the fusion of multi-source heterogeneous data.

[0004] In summary, existing technologies suffer from technical problems such as reliance on centralized data processing architecture for shipborne system fault prediction, lack of regional collaborative fault modeling and dynamic knowledge updating capabilities, resulting in response delays and insufficient fault identification accuracy in multi-source heterogeneous data environments, which further affect the real-time fault warning efficiency and intelligent operation and maintenance capabilities of the entire ship's control system. Summary of the Invention

[0005] The purpose of this application is to provide a distributed fault prediction system for the control unit of military ships, in order to solve the technical problems in the prior art that shipborne system fault prediction relies on a centralized data processing architecture and lacks the ability to model regional collaborative faults and update dynamic knowledge, resulting in response delays and insufficient fault identification accuracy in a multi-source heterogeneous data environment, which further affects the real-time fault warning efficiency and intelligent operation and maintenance capabilities of the entire ship control system.

[0006] In view of the above problems, this application provides a distributed fault prediction system for military ship control units, comprising: an architecture building module, used to build a distributed fault prediction architecture based on the military ship control unit, the distributed fault prediction architecture including edge computing nodes, regional aggregation nodes and a central decision node; a feature extraction module, used to collect and obtain the working signal set of the sub-control unit through the edge computing nodes, and extract fault features from the working signal set of the sub-control unit to obtain the edge node fault feature set; a path analysis module, used to mine and obtain a fault association rule base, activate the regional aggregation node to perform fault path analysis on the edge node fault feature set based on the fault association rule base, and determine the distributed regional fault path; and a fault operation and maintenance module, used to call the military ship fault knowledge base through the central decision node, perform integrated prediction on the distributed regional fault path based on the military ship fault knowledge base, obtain the fault integrated prediction result, and the military ship control unit performs fault early warning operation and maintenance based on the fault integrated prediction result.

[0007] Preferably, the distributed fault prediction system for the military vessel control unit further includes: a sub-unit decomposition unit, used to decompose the military vessel control unit into sub-units to obtain a set of sub-control units; an edge computing node acquisition unit, used to sequentially deploy sensors and configure computing resources based on the monitoring requirements information of the sub-control unit set to obtain edge computing nodes; a region division unit, used to collect and acquire historical fault datasets of military vessels, use the historical fault datasets of military vessels to divide the edge computing nodes into regions, and determine the region aggregation nodes; and a global federated learning unit, used to perform global federated learning based on the region aggregation nodes to obtain a central decision node, and build the distributed fault prediction architecture based on the edge computing nodes, the region aggregation nodes, and the central decision node.

[0008] Preferably, the distributed fault prediction system for the military vessel control unit further includes: a fault label dataset acquisition unit, used to normalize and add fault labels to the historical fault dataset of the military vessel to obtain a military vessel fault label dataset; a fault association feature set acquisition unit, used to extract association features based on the military vessel fault label dataset to obtain a military vessel fault association feature set; a clustering result acquisition unit, used to perform K-means spatial clustering on the military vessel fault association feature set to obtain a sub-control unit clustering result; and a computing resource allocation unit, used to allocate computing resources based on the amount of association data in the sub-control unit clustering result to determine the regional aggregation node.

[0009] Preferably, the distributed fault prediction system of the military vessel control unit further includes: an association and distribution unit, used to associate and distribute the historical fault dataset of the military vessel based on the regional aggregation node to obtain a regional node historical fault dataset; a fault identification training unit, used to train the fault identification on the regional node historical fault dataset using a deep neural network to generate a regional node fault prediction network set; a fault prediction twin acquisition unit, used to extract the model parameters of the regional node fault prediction network set for global federated learning to obtain a global node fault prediction twin; and a verification, tuning, and storage unit, used to verify, tune, and store the global node fault prediction twin to obtain the central decision node.

[0010] Preferably, the distributed fault prediction system for the military ship control unit further includes: a program acquisition unit, used to acquire a node preprocessing calculation program and a node feature extraction calculation program based on the edge computing nodes; a standardization preprocessing unit, used to perform standardization preprocessing on the sub-control unit working signal set using the node preprocessing calculation program to acquire a standard sub-control unit working signal set; a sub-unit associated working feature set acquisition unit, used to extract associated features from the standard sub-control unit working signal set based on the node feature extraction calculation program to acquire a sub-unit associated working feature set; and an edge node fault feature set acquisition unit, used to extract fault features from the sub-unit associated working feature set according to the sub-unit normal operation threshold to acquire the edge node fault feature set.

[0011] Preferably, the distributed fault prediction system of the military vessel control unit further includes: a fault acquisition unit, used to identify fault types and extract fault features from the historical fault dataset of the military vessel to obtain a fault type dataset and a fault feature dataset; a timestamp alignment unit, used to perform timestamp alignment and association analysis on the fault type dataset and the fault feature dataset to construct a fault type-feature association table; and an association rule mining unit, used to perform association rule mining based on the fault type-feature association table to obtain a fault association rule library.

[0012] Preferably, the distributed fault prediction system of the military ship control unit further includes: a format conversion unit, used to perform feature discretization and format conversion on the fault type-feature association table to obtain a fault association feature transaction dataset; and a frequent itemset mining unit, used to set a confidence benchmark threshold, and use the confidence benchmark threshold to perform frequent itemset mining and association rule generation on the fault association feature transaction dataset to obtain the fault association rule library.

[0013] Preferably, the distributed fault prediction system of the military ship control unit further includes: an associated region node set acquisition unit, used to activate the region convergence node according to the edge node fault feature set to obtain an associated region node set; a region matching fault rule set acquisition unit, used to call the fault association rule library through the associated region node set to perform fault diagnosis on the associated region node set to obtain a region matching fault rule set; and a distributed region fault path determination unit, used to perform propagation path deduction based on the region matching fault rule set to determine the distributed region fault path.

[0014] Preferably, the distributed fault prediction system of the military ship control unit further includes: a fusion processing unit, used to fuse the distributed regional fault paths to construct a global fault path graph; an associated path description unit, used to extract fault modes and describe associated paths from the military ship fault knowledge base to generate a fault path pattern prediction network; and an integrated prediction unit, used to perform integrated prediction on the global fault path graph based on the fault path pattern prediction network to obtain the fault integrated prediction result.

[0015] Preferably, the distributed fault prediction system of the military ship control unit further includes: a simulation verification unit, used to simulate and verify the fault integrated prediction result using the global node fault prediction twin to obtain fault prediction accuracy parameters; and an adaptive learning optimization unit, used to adaptively learn and optimize the fault path pattern prediction network based on the fault prediction accuracy parameters.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goal of constructing a multi-level collaborative distributed fault prediction architecture and achieving the technical goal of edge-region-central node linkage perception and intelligent judgment, the technical effects of improving fault detection response speed, enhancing model generalization ability and real-time early warning accuracy are achieved.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the distributed fault prediction system for the military ship control unit of this application.

[0020] Figure 2 This is a schematic diagram of the architecture of the distributed fault prediction system for the military ship control unit of this application.

[0021] Figure labeling: Architecture building module 1, feature extraction module 2, path analysis module 3, fault operation and maintenance module 4, sub-unit decomposition unit 11, edge computing node acquisition unit 12, region division unit 13, global federated learning unit 14. Detailed Implementation

[0022] This application provides a distributed fault prediction system for military ship control units, addressing the technical problems in existing technologies where shipborne system fault prediction relies on a centralized data processing architecture and lacks the ability to model regional collaborative faults and dynamically update knowledge. This results in response delays and insufficient fault identification accuracy in multi-source heterogeneous data environments, further impacting the real-time fault warning efficiency and intelligent operation and maintenance capabilities of the entire ship's control system. The system achieves the technical goals of constructing a multi-level collaborative distributed fault prediction architecture and realizing edge-region-central node linkage perception and intelligent judgment, thereby improving fault detection response speed, enhancing model generalization ability, and increasing real-time warning accuracy.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Please see the appendix Figure 1 This application provides a distributed fault prediction system for the control unit of military ships, specifically including:

[0025] Architecture building module 1 is used to build a distributed fault prediction architecture based on the military ship control unit. The distributed fault prediction architecture includes edge computing nodes, regional aggregation nodes and central decision nodes.

[0026] Specifically, the control unit of a military vessel refers to the control module used to manage and coordinate the operation of various mission systems within the ship, such as the power system, navigation system, and weapon control system, each composed of different sub-control units. Due to their complex operating environment and high workload, these sub-control units are highly susceptible to potential malfunctions. To achieve real-time monitoring and early fault prediction of these sub-control units, a systematic, hierarchical prediction framework needs to be built to obtain a distributed fault prediction architecture, thereby improving response speed and system robustness. The distributed fault prediction architecture is deployed physically close to the specific sub-control unit, responsible for collecting sensor data and performing preprocessing and preliminary feature extraction, such as determining whether current or voltage is abnormal and identifying local vibration patterns.

[0027] Next, multiple edge computing nodes upload the extracted fault feature data to the regional aggregation node. The regional aggregation node is responsible for integrating and analyzing the fault features within the same region, further establishing regional-level fault correlations and propagation paths. The central decision-making node is responsible for integrating the fault path information uploaded by multiple regional aggregation nodes, ultimately generating a unified fault prediction result for the entire ship.

[0028] Feature extraction module 2 is used to acquire the working signal set of the sub-control unit through the edge computing node, extract fault features from the working signal set of the sub-control unit, and obtain the edge node fault feature set.

[0029] Specifically, the system acquires the working signal sets of the sub-control units through edge computing nodes. Then, using computing devices deployed near the control unit of the military vessel, the working signals generated by each sub-control unit are acquired and processed locally in real time. A sub-control unit refers to the smallest independent functional module within the military vessel control unit, such as an independent module for controlling propulsion, power distribution, or communication. The working signal set consists of the physical or logical output data of the sub-control unit under normal or abnormal conditions, such as voltage, current, frequency, and operating status codes.

[0030] Fault features are extracted from the working signal set of the sub-control unit. Feature analysis is performed on the acquired raw signals to mine feature values ​​or indicators closely related to potential fault states, including signal amplitude abrupt changes, continuous oscillations, zero drift, and spectral distortion that exceed the normal fluctuation range. Extraction methods can be adopted, such as time-domain statistical analysis, frequency-domain filtering, principal component analysis, or machine learning models. This transforms the content in the raw signals that does not have direct identifiability into indicators that can be used to judge the equipment status, resulting in a fault feature set for edge nodes, which is used for subsequent fault diagnosis, predictive analysis, and federated model training.

[0031] Path analysis module 3 is used to mine and obtain a fault association rule base, activate the regional aggregation node to perform fault path analysis on the fault feature set of the edge node based on the fault association rule base, and determine the distributed regional fault path.

[0032] Specifically, mining and acquiring a fault association rule base refers to conducting in-depth analysis of the statistical relationships between fault types and related features in a large amount of historical fault data from military vessels, extracting a set of rules that can represent the causal relationships and frequent associations between different fault modes. Activating regional aggregation nodes means that when the fault features reported by edge computing nodes meet the triggering conditions of certain rules in the fault association rule base, the regional computing center is triggered to jointly analyze the status of multiple edge nodes within the relevant region. Regional aggregation nodes are a type of mid-level node that integrates and processes data across multiple edge nodes, used to collaboratively handle complex events and fault trends across nodes, improving the comprehensiveness and accuracy of judgments.

[0033] Based on the fault association rule base, fault path analysis is performed on the fault feature set of edge nodes. The rules in the fault association rule base are used to match and deduce the fault features uploaded by the current edge nodes, thereby identifying the possible propagation path or chain reaction structure of a certain fault. Determining the distributed regional fault path means integrating the above path analysis results regionally to identify the fault evolution line based on multiple sub-control units and regional nodes in the entire military ship system.

[0034] The fault operation and maintenance module 4 is used to call the military ship fault knowledge base through the central decision node, perform integrated prediction of the fault path of the distributed area based on the military ship fault knowledge base, and obtain the fault integrated prediction result. The military ship control unit performs fault early warning operation and maintenance based on the fault integrated prediction result.

[0035] Specifically, the central decision-making node accesses and calls upon the military ship fault knowledge base, which stores a large amount of historical fault information, rules, and patterns. This knowledge base contains information on various fault types, occurrence conditions, corresponding characteristics, propagation paths, and remedial measures. Based on this knowledge base, distributed regional fault paths are integrated and predicted. Utilizing the logical relationships between historical fault patterns and features in the fault association rule base, a comprehensive evaluation and trend judgment are made on the distributed regional fault paths currently uploaded from multiple regions. The integrated prediction process incorporates the commonalities, differences, and potential coupling effects among multiple paths, thereby outputting intelligent predictions of future system states. Modeling typically employs methods such as Bayesian networks, graph neural networks, and logistic regression ensembles to obtain integrated fault prediction results. The output includes a set of prediction results containing information such as fault occurrence probability, potential impact range, and propagation trend paths, reflecting the severity and urgency of risks faced by different control units.

[0036] Military ship control units perform fault early warning and maintenance based on fault integration prediction results, and take response measures in advance, such as switching to backup systems, reducing operating load, and issuing maintenance warnings, to prevent or mitigate potential faults, thereby achieving proactive prevention, reducing losses, and improving system stability.

[0037] Furthermore, this application also includes: a sub-unit disassembly unit 11, used to disassemble the military ship control unit into sub-units to obtain a set of sub-control units; an edge computing node acquisition unit 12, used to sequentially deploy sensors and configure computing resources based on the monitoring requirement information of the set of sub-control units to obtain edge computing nodes; a region division unit 13, used to collect and acquire historical fault datasets of military ships, use the historical fault datasets of military ships to divide the edge computing nodes into regions, and determine the region aggregation nodes; and a global federated learning unit 14, used to perform global federated learning based on the region aggregation nodes to obtain a central decision node, and build the distributed fault prediction architecture based on the edge computing nodes, the region aggregation nodes, and the central decision node.

[0038] Specifically, the control unit of military ships is disassembled into sub-units. The automatic control system of the entire ship is subdivided and decomposed according to its functional structure to form several relatively independent module units with specific control responsibilities, such as propulsion control sub-unit, cabin environment regulation sub-unit, weapon launching sub-unit, etc., which constitute a set of sub-control units. This allows subsequent monitoring, analysis and operation and maintenance work to be carried out on specific functional blocks, improving the accuracy of system management and response efficiency.

[0039] Next, based on the monitoring requirements of each sub-control unit, sensors of the appropriate type and quantity are deployed one by one, and localized computing resources are allocated to construct edge computing nodes. The monitoring requirements include sampling frequency, data dimensions, and real-time requirements.

[0040] Subsequently, historical fault datasets generated during the operation of military vessels were collected and organized, recording the fault types, characteristics, and evolution processes of different sub-control units under specific conditions. Using these historical fault datasets, edge computing nodes were divided into regions. Edge nodes with similar functions, high data correlation, or similar historical fault types were grouped into regions. Within each region, nodes with stronger computing and communication capabilities were selected as regional aggregation nodes.

[0041] Then, by leveraging the aggregation nodes in each region, the local model parameters are uploaded to participate in global federated learning, thereby establishing a unified central decision-making node. Federated learning is a distributed machine learning method that enables collaborative model training without centralizing raw data, protecting data privacy while achieving knowledge fusion. The central decision-making node aggregates the model knowledge from each region to form a global predictive model, and ultimately coordinates and manages the fault prediction task of the entire military ship control unit.

[0042] Furthermore, this application also includes: a fault label dataset acquisition unit, used to normalize and add fault labels to the historical fault dataset of the military vessel to obtain a fault label dataset of the military vessel; a fault association feature set acquisition unit, used to extract association features based on the fault label dataset of the military vessel to obtain a fault association feature set of the military vessel; a clustering result acquisition unit, used to perform K-means spatial clustering on the fault association feature set of the military vessel to obtain the clustering result of the sub-control unit; and a computing power resource allocation unit, used to allocate computing power resources based on the amount of association data of the clustering result of the sub-control unit and determine the regional aggregation node.

[0043] Specifically, the historical fault dataset of military vessels undergoes normalization and fault labeling. The raw fault data, collected from different dimensions and units, is mathematically standardized, for example, by converting values ​​such as temperature, voltage, and vibration amplitude into dimensionless values ​​between 0 and 1, ensuring comparability and a unified scale for different features. Simultaneously, clear fault type labels, such as short circuit, overheating, and jamming, are added to each data record in the historical fault dataset, resulting in a fault-labeled dataset for military vessels. This allows subsequent machine learning algorithms to conduct supervised learning based on these labels.

[0044] Next, using the military ship fault label dataset, key features closely related to the occurrence of faults can be extracted through statistical analysis, information entropy filtering, or model training, forming a military ship fault association feature set. Association features refer to sensor signals or system state values ​​that have repeatedly co-occurred with specific fault labels in history, such as a significant correspondence between a specific pressure increase and a valve jamming fault.

[0045] Subsequently, K-means spatial clustering is performed on the extracted fault association feature set. K-means is an unsupervised clustering algorithm that can divide high-dimensional data into several closely related subgroups in the feature space. Through K-means spatial clustering, sub-control units with similar fault association features of military vessels can be automatically grouped together to form sub-control unit clustering results.

[0046] Finally, computing resources are allocated based on the amount of associated data contained in each subclass of the subcontrol unit clustering results. The data pressure and computing requirements of each subclass are comprehensively evaluated based on indicators such as the number of nodes, data transmission frequency, and historical failure frequency. Then, computing resources are dynamically allocated to each cluster to determine the regional aggregation nodes that undertake aggregation analysis tasks.

[0047] Furthermore, this application also includes: an association and routing unit, used to perform association and routing on the historical fault dataset of the military vessels based on the regional aggregation node to obtain a regional node historical fault dataset; a fault identification training unit, used to train fault identification on the regional node historical fault dataset using a deep neural network to generate a regional node fault prediction network set; a fault prediction twin acquisition unit, used to extract the model parameters of the regional node fault prediction network set for global federated learning to obtain a global node fault prediction twin; and a verification, tuning, and storage unit, used to verify, tune, and store the global node fault prediction twin to obtain the central decision node.

[0048] Specifically, the historical fault dataset of military vessels is correlated and distributed based on regional aggregation nodes. It is categorized according to its functional region and assigned to various aggregation nodes to obtain regional node historical fault datasets. Correlation and distribution utilize factors such as time synchronization, physical location, signal similarity, or functional coupling between fault data to establish mapping relationships between data. This divides the historical fault dataset of military vessels into logically consistent regional nodes, making the data managed by each node more concentrated and possessing specific domain characteristics, thus facilitating subsequent independent training.

[0049] Next, a deep neural network, a multilayer perceptron structure, possesses the ability to express nonlinearities and automatically extract features. A deep neural network is used to train fault identification on a historical fault dataset of regional nodes to identify potential fault feature patterns in the data. During training, the historical fault dataset of regional nodes is first divided into training, validation, and test sets according to time or proportion, such as a ratio of 7:2:1. Based on the temporal and spatial correlation of the historical fault dataset, a deep neural network structure containing an input layer, multiple hidden layers, and an output layer is designed. The input layer receives the training set, the hidden layers use activation functions such as ReLU to construct nonlinear mappings, and the output layer uses the Softmax activation function to output the probability distribution of each fault. Forward propagation is performed using the training set to calculate the output, using cross-entropy loss or mean squared error loss. Each iteration during training is called an epoch, such as setting 50 to 200 epochs, and a preset number of iterations is set to prevent overfitting. After each training epoch, performance metrics such as accuracy are evaluated on the validation set. After completing the basic model training of the deep neural network, a regional node fault prediction network set is obtained.

[0050] Subsequently, model parameters are extracted from the regional node fault prediction network set, and a global federated learning operation is performed to finally construct a global node fault prediction twin. Federated learning is a distributed collaborative modeling method in which each regional node completes model training locally, uploading only the model parameters and not the original data, which can effectively protect sensitive information. In the global federated learning process, weighted averaging, difference constraints, and global consistency optimization methods are used to aggregate all model parameters to construct a predictive model that can represent the overall operating state and fault trend, which is the global node fault prediction twin, and can realistically reflect the response mode of the control system under different states.

[0051] Finally, the global node fault prediction twin was validated, optimized, and stored to verify its prediction accuracy and generalization ability, ensuring its applicability to real-world decision-making scenarios. The validation and optimization process included test set evaluation, error backpropagation correction, and parameter refinement. After passing both stability and accuracy standards, the global node fault prediction twin was officially deployed as the central decision-making node. The central decision-making node possesses a global perspective, coordinating prediction results from nodes in various regions to achieve multi-regional collaborative judgment and ship-wide fault early warning decision-making.

[0052] Furthermore, this application also includes: a program obtaining unit, used to obtain a node preprocessing calculation program and a node feature extraction calculation program based on the edge computing node; a standardization preprocessing unit, used to perform standardization preprocessing on the sub-control unit working signal set using the node preprocessing calculation program to obtain a standard sub-control unit working signal set; a sub-unit associated working feature set obtaining unit, used to perform associated feature extraction on the standard sub-control unit working signal set based on the node feature extraction calculation program to obtain a sub-unit associated working feature set; and an edge node fault feature set obtaining unit, used to perform fault feature extraction on the sub-unit associated working feature set according to the sub-unit normal operation threshold to obtain the edge node fault feature set.

[0053] Specifically, edge computing nodes are miniature intelligent computing modules deployed in local control areas of military vessels. They possess certain data processing and model calculation capabilities and can independently execute some tasks without relying on a central server. Based on the task type and number of sensors of the sub-control units connected to each edge node, two types of computing programs can be assigned: one is a node preprocessing computing program, used for basic processing such as cleaning, alignment, and interpolation of the raw data; the other is a node feature extraction computing program, used to extract key features related to the health status of the equipment from the preprocessed data, such as fluctuation frequency, rate of rise, and mean change.

[0054] Next, the main objective of the node preprocessing computation procedure is to perform normalization preprocessing on the working signal set of the sub-control unit. The working signal set of the sub-control unit refers to the raw data collected by the edge computing nodes, such as temperature, voltage, and pressure signals. Normalization can eliminate differences in scale, units, and sampling frequency among different signals, so that subsequent models can fairly process various data features.

[0055] Then, the node feature extraction calculation program performs feature extraction on the standardized signal set to extract correlated features that reflect the operating status of the military ship control unit. Correlated features refer to the coupling behavior between multiple signals, such as whether current changes are always accompanied by temperature increases, or whether a certain pressure anomaly always occurs after a voltage drop. Extracting correlated features can reveal the potential cooperative change patterns of the military ship control unit, which helps to build a more interpretable health model.

[0056] Subsequently, based on the preset normal operating threshold of the sub-unit, fault features are extracted from the associated features. The normal operating threshold of the sub-unit is a stable operating range defined in experience or historical data. Once a feature exceeds this range, it can be marked as an edge node fault feature set.

[0057] Furthermore, this application also includes: a fault acquisition unit, used to identify fault types and extract fault features from the historical fault dataset of the military vessels to obtain a fault type dataset and a fault feature dataset; a timestamp alignment unit, used to perform timestamp alignment and association analysis on the fault type dataset and the fault feature dataset to construct a fault type-feature association table; and an association rule mining unit, used to perform association rule mining based on the fault type-feature association table to obtain a fault association rule library.

[0058] Specifically, the historical fault dataset for military vessels refers to all relevant data recorded during the operation of military vessels when fault events occurred, including sensor signals, alarm records, maintenance logs, and other information before and after the fault occurred. The historical fault dataset is then categorized into fault types, classifying different fault phenomena into several distinct fault categories, such as "main motor overload," "hydraulic system leakage," and "control module failure." Fault feature extraction is performed on the dataset, identifying data features representative of fault behavior from the raw signals, such as sudden temperature increases, voltage drops, or abnormal vibration frequencies, thereby constructing a feature set describing the behavioral characteristics of each type of fault. Finally, a fault type dataset and a fault feature dataset are obtained.

[0059] Next, the fault type dataset and fault feature dataset are time-stamp aligned and correlation analyzed. Time-stamp alignment unifies the time information recorded in the two datasets, ensuring that a fault event occurring at a specific moment corresponds to the extracted feature changes at that time. For example, recording "pressure sensor anomaly" at a certain moment allows us to trace the feature changes at that time. Correlation analysis involves data mining on the aligned records to identify features whose change patterns consistently appear before specific fault types, thus forming a data foundation of "strongly correlated faults and features." Finally, a fault type-feature correlation table is constructed. Table 1 shows a portion of the records from the most recently obtained fault type-feature correlation table.

[0060] Table 1: Partial records of the most recently obtained fault type-feature association table

[0061]

[0062] Furthermore, association rule mining is conducted based on the fault type-feature association table to generate a fault association rule base. Association rule mining is a data mining technique that aims to discover potential causal or co-occurrence relationships between multiple variables. For example, it can be used to infer that "if features A and B are detected simultaneously, then fault type C is highly likely to occur." Complex data can be transformed into a set of high-probability rules, i.e., the fault association rule base.

[0063] Furthermore, this application also includes: a format conversion unit, used to perform feature discretization and format conversion on the fault type-feature association table to obtain a fault association feature transaction dataset; and a frequent itemset mining unit, used to set a confidence benchmark threshold, and use the confidence benchmark threshold to perform frequent itemset mining and association rule generation on the fault association feature transaction dataset to obtain the fault association rule library.

[0064] Specifically, the fault type-feature association table undergoes feature discretization, transforming the original continuous numerical features (such as temperature, voltage, vibration frequency, etc.) into several discrete intervals or classification labels. For example, temperature can be divided into three discrete intervals: "below 50 degrees Celsius," "50 to 80 degrees Celsius," and "above 80 degrees Celsius." The purpose of feature discretization is to enable subsequent data mining processes to handle discrete values, thereby improving computational efficiency and facilitating rule extraction. Next, a format conversion is performed, organizing the discretized features according to the input format of the association rule mining algorithm. Each data point is represented as a transaction (i.e., a group of simultaneously occurring feature items), thus constructing a fault association feature transaction dataset.

[0065] Next, setting a confidence benchmark threshold refers to a minimum confidence standard customized by those skilled in the art based on actual conditions, used to filter unreliable rules generated during data mining. Confidence is an indicator of the reliability of a rule, representing the probability of a target fault occurring given the given preconditions. For example, if a rule "large current fluctuations → motor overheating" holds true in 80% of the samples, then the confidence of that rule is 80%. Setting a confidence benchmark threshold, such as 70%, means that only rules with a confidence of at least 70% will be retained. Frequent itemset mining is performed on the fault association feature transaction dataset to identify frequently occurring feature combinations, such as "high temperature + high pressure" or "large vibration + abnormal voltage." Subsequently, specific association rules are derived from the frequent itemsets, such as "if high temperature and large vibration occur, the probability of control module failure is 85%." Finally, a fault association rule base is formed, which is a set of predictive rules that meet the confidence requirements.

[0066] Furthermore, this application also includes: an associated region node set acquisition unit, used to activate the region aggregation node according to the edge node fault feature set to obtain an associated region node set; a region matching fault rule set acquisition unit, used to call the fault association rule library through the associated region node set to perform fault diagnosis on the associated region node set to obtain a region matching fault rule set; and a distributed region fault path determination unit, used to perform propagation path deduction based on the region matching fault rule set to determine the distributed region fault path.

[0067] Specifically, regional aggregation nodes are activated based on the fault feature set of edge nodes. When an edge node detects representative fault features, it triggers its associated regional aggregation nodes to participate in analysis and processing, thereby obtaining a set of associated regional nodes. Regional aggregation nodes are intermediate layer nodes that centrally manage, process, and judge data uploaded from multiple edge nodes.

[0068] Next, fault diagnosis is performed by calling the fault association rule base through the associated regional node set. This involves querying the pre-established fault association rule base in the system to match the observed feature patterns. The associated regional node set refers to a collection of geographically or functionally related nodes that may collectively influence or propagate a certain type of fault. The fault diagnosis process involves determining whether the current observation matches a certain fault mode through rule matching, thereby outputting a regional matching fault rule set.

[0069] Subsequently, based on the regional matching fault rule set, propagation path simulation is performed to analyze the possible spread of the fault, forming a path diagram from a certain starting point to multiple affected points. The propagation path simulation relies on the connection relationships, characteristic influence relationships, and temporal sequence between control units. For example, a fault in a cooling system may first affect the adjacent temperature control module and then spread to the energy supply module. The distributed regional fault path is the final determined fault propagation path, composed of multiple nodes and possessing a causal chain structure, used to guide emergency response or precise maintenance.

[0070] Furthermore, this application also includes: a fusion processing unit, used to fuse the distributed regional fault paths to construct a global fault path graph; an associated path description unit, used to extract fault modes and describe associated paths in the military ship fault knowledge base to generate a fault path pattern prediction network; and an integrated prediction unit, used to perform integrated prediction on the global fault path graph based on the fault path pattern prediction network to obtain the fault integrated prediction result.

[0071] Specifically, the distributed regional fault paths are fused to construct a global fault path graph, unifying and integrating the fault propagation paths derived from the aggregation nodes of various regions. The fusion process involves analyzing the overlapping nodes, correlation logic, and propagation order of causal relationships between different fault nodes in multiple local regions, piecing together fragmented information into a complete global graph, thereby obtaining the global fault path graph.

[0072] Next, fault mode extraction and path description were performed on the military ship fault knowledge base. Utilizing a large amount of accumulated historical fault data, typical and representative fault evolution patterns were identified, and their propagation paths within the system structure were labeled. The military ship fault knowledge base is a structured database containing fault records, feature descriptions, and processing results collected during past operations. Fault mode extraction uses statistical analysis or machine learning methods to identify regular combinations. Path description supplements the spatial, functional, or temporal propagation relationships between nodes, ultimately constructing a fault path pattern prediction network.

[0073] This paper utilizes a fault path pattern prediction network to perform integrated predictions on the global fault path map. The extracted typical patterns are compared with the current global path map to assess potential future fault evolution trends. Integrated prediction refers to fusing historical experience and current observation data to obtain more accurate and robust prediction results. The fault integrated prediction result is a comprehensive judgment reflecting the severity, scope of impact, and potential risks of the current fault state.

[0074] Furthermore, this application also includes: a simulation verification unit, used to simulate and verify the fault ensemble prediction result using the global node fault prediction twin, and obtain fault prediction accuracy parameters; and an adaptive learning optimization unit, used to perform adaptive learning optimization on the fault path pattern prediction network based on the fault prediction accuracy parameters.

[0075] Specifically, a global node fault prediction twin is used to simulate and verify the integrated fault prediction results in order to evaluate their accuracy. The simulation and verification process includes setting fault triggering conditions, reproducing sensor behavior, simulating control response, etc. By comparing with real historical data or simulation target results, the deviation of fault prediction is calculated, and finally, fault prediction accuracy parameters such as mean absolute error and prediction hit rate are output.

[0076] Then, the fault path pattern prediction network is adaptively learned and optimized based on the fault prediction accuracy parameter, and targeted structural adjustments or parameter updates are made so that the fault path pattern prediction network can be continuously fine-tuned according to its own prediction results. For example, the backpropagation algorithm is used to reduce the loss value, update the associated weights, and enhance the weak feature signal processing capability, thereby improving its performance in future prediction tasks.

[0077] In summary, the distributed fault prediction system for military ship control units provided in this application has the following technical effects: by realizing the technical goal of constructing a multi-level collaborative distributed fault prediction architecture and achieving the linkage perception and intelligent judgment of edge-region-central nodes, it achieves the technical effects of improving fault detection response speed, enhancing model generalization ability and real-time early warning accuracy.

[0078] 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.

[0079] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A distributed fault prediction system for a military ship control unit, characterized in that, The system comprises: An architecture building module, configured to build a distributed fault prediction architecture according to a military ship control unit, the distributed fault prediction architecture comprising an edge computing node, a regional aggregation node and a central decision node, wherein the edge computing node is a miniature intelligent computing module deployed in a local control area of the military ship, has data processing and model calculation capabilities, and independently performs tasks without relying on a central server, the regional aggregation node is a middle layer node that integrates and processes data and performs calculation on multiple edge nodes, and is used for cooperatively processing complex events and fault trends across nodes, the central decision node aggregates model knowledge of each region to form a global prediction model, and plans and manages fault prediction tasks of the entire military ship control unit; A feature extraction module, configured to collect and acquire a working signal set of a sub-control unit through the edge computing node, perform fault feature extraction on the working signal set of the sub-control unit, and obtain an edge node fault feature set; A path analysis module, configured to mine and acquire a fault association rule base, activate the regional aggregation node to perform fault path analysis on the edge node fault feature set based on the fault association rule base, and determine a distributed regional fault path; A fault operation and maintenance module, configured to call a military ship fault knowledge base through the central decision node, perform integrated prediction on the distributed regional fault path based on the military ship fault knowledge base, obtain a fault integrated prediction result, and perform fault early warning operation and maintenance based on the fault integrated prediction result.

2. The distributed fault prediction system for a naval vessel control unit of claim 1, wherein, The architecture building module comprises: A sub-unit disassembling unit, configured to disassemble the military ship control unit into a set of sub-control units; An edge computing node obtaining unit, configured to sequentially perform sensor deployment and computing resource configuration based on monitoring requirement information of the set of sub-control units, and obtain an edge computing node; A regional division unit, configured to collect and acquire a historical fault data set of the military ship, divide the edge computing node into regional aggregation nodes using the historical fault data set of the military ship, and determine the regional aggregation nodes; A global federated learning unit, configured to perform global federated learning based on the regional aggregation nodes, obtain a central decision node, and build the distributed fault prediction architecture according to the edge computing node, the regional aggregation nodes and the central decision node.

3. The distributed fault prediction system for a naval ship control unit of claim 2, wherein, The regional division unit comprises: A fault label data set obtaining unit, configured to perform normalization processing and fault label addition on the historical fault data set of the military ship, and obtain a military ship fault label data set; A fault association feature set obtaining unit, configured to extract association features based on the military ship fault label data set, and obtain a military ship fault association feature set; A clustering result obtaining unit, configured to perform K-means spatial clustering on the military ship fault association feature set, and obtain a sub-control unit clustering result; An algorithm resource allocation unit, configured to allocate algorithm resources based on an association data volume of the sub-control unit clustering result, and determine the regional aggregation nodes.

4. The distributed fault prediction system for a naval ship control unit of claim 2, wherein, The global federated learning unit comprises: The association shunt unit is configured to perform association shunting on the military ship historical fault data set based on the regional convergence node, and obtain a regional node historical fault data set; The fault identification training unit is configured to perform fault identification training on the regional node historical fault data set by using a deep neural network, and generate a regional node fault prediction network set; The fault prediction twin obtaining unit is configured to extract model parameters of the regional node fault prediction network set for global federated learning, and obtain a global node fault prediction twin; The verification tuning storage unit is configured to perform verification tuning storage on the global node fault prediction twin, and obtain the central decision node.

5. The distributed prognostics system for a military ship control unit of claim 1, wherein, The feature extraction module comprises: The program obtaining unit is configured to obtain a node preprocessing calculation program and a node feature extraction calculation program according to the edge computing node; The standardization preprocessing unit is configured to perform standardization preprocessing on the sub-control unit working signal set by using the node preprocessing calculation program, and obtain a standard sub-control unit working signal set; The sub-unit association working feature set obtaining unit is configured to perform association feature extraction on the standard sub-control unit working signal set based on the node feature extraction calculation program, and obtain a sub-unit association working feature set; The edge node fault feature set obtaining unit is configured to perform fault feature extraction on the sub-unit association working feature set according to a sub-unit normal working threshold, and obtain the edge node fault feature set.

6. The distributed fault prediction system for a naval ship control unit of claim 2, wherein, The path analysis module comprises: The fault obtaining unit is configured to perform fault type identification and fault feature extraction on the military ship historical fault data set, and obtain a fault type data set and a fault feature data set; The timestamp alignment unit is configured to perform timestamp alignment and association analysis on the fault type data set and the fault feature data set, and construct a fault type-feature association table; The association rule mining unit is configured to perform association rule mining based on the fault type-feature association table, and obtain a fault association rule library.

7. The distributed fault prediction system for a naval vessel control unit of claim 6, wherein, The association rule mining unit comprises: The format conversion unit is configured to perform feature discretization and format conversion on the fault type-feature association table, and obtain a fault association feature transaction data set; The frequent item set mining unit is configured to set a confidence benchmark threshold, perform frequent item set mining and association rule generation on the fault association feature transaction data set by using the confidence benchmark threshold, and obtain the fault association rule library.

8. The distributed prognostics system for a military ship control unit of claim 1, wherein, The path analysis module further comprises: The associated regional node set obtaining unit is configured to activate the regional convergence node according to the edge node fault feature set, and obtain an associated regional node set; The regional matching fault rule set obtaining unit is configured to perform fault diagnosis on the associated regional node set by calling the fault association rule library through the associated regional node set, and obtain a regional matching fault rule set; The distributed regional fault path determination unit is configured to perform propagation path deduction based on the regional matching fault rule set, and determine the distributed regional fault path.

9. The distributed fault prediction system for a naval ship control unit of claim 4, wherein, The fault operation and maintenance module comprises: The fusion processing unit is configured to perform fusion processing on the distributed regional fault path, and construct a global fault path graph; The association path description unit is configured to perform fault mode extraction and association path description on the military ship fault knowledge base, and generate a fault path mode prediction network. The integrated prediction unit is configured to perform integrated prediction on the global fault path graph based on the fault path mode prediction network, and obtain the fault integrated prediction result.

10. The distributed prognostics system for a naval vessel control unit of claim 9, wherein, The integrated prediction unit comprises: The simulation verification unit is configured to perform simulation verification on the fault integrated prediction result by using the global node fault prediction twin, and obtain a fault prediction accuracy parameter. The adaptive learning optimization unit is configured to perform adaptive learning optimization on the fault path mode prediction network based on the fault prediction accuracy parameter.

Images