A multi-source fault autonomous location and classification method

Abstract

The present application provides a multi-source fault autonomous positioning and classification method, comprising: locating the module related to the fault according to the code matching of the ship system fault signal, and determining the structural area where the fault is most likely to occur; calculating the probability of failure of all components in the area through the Bayesian network algorithm, and screening out the key components that are most likely to fail according to the calculation results; monitoring the real-time operation data according to the located key components containing the fault, judging whether abnormal fluctuations occur, and judging whether the fluctuations of the abnormal data are related to the abnormal external environment; drawing a probability interval distribution diagram of the fault spreading outward from the root component, and calculating the confidence that other faults may be hidden in different diffusion range intervals; constructing a component knowledge base and a system operation environment library according to the fault root component data of the ship system and the working environment data and task process data of the fault-related system, and identifying the key internal and external factors that affect the occurrence of the fault.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a multi-source fault autonomous positioning and classification method. Background Art

[0002] In modern ship engineering, the internal structure of the system is becoming increasingly complex, involving multiple modules and systems that are interdependent and interact with each other. These modules include but are not limited to power systems, navigation systems, communication systems, and various auxiliary control systems. Each system or module consists of dozens or even hundreds of small components that are highly specialized in function and closely connected to each other. In addition, the design and operation of ship systems must also take into account various marine environmental factors, such as waves, wind speed, currents, and the chemical and physical properties of seawater. The interaction of this complex internal structure and environmental factors has led to many challenges in the operation of ship systems. First, the complexity of the system makes it possible for even minor faults to spread rapidly and affect the stability and safety of the entire system. For example, a small sensor failure may cause errors in the navigation system, which in turn affects the direction of the entire ship. Second, ships operate in a complex and changeable marine environment. These external environmental factors are themselves extremely uncertain and random. For example, sudden changes in marine weather or abnormal hydrogeological conditions may have direct or indirect effects on ship systems. In this context, the diagnosis and location of ship system faults become extremely difficult. Due to the interdependence of internal systems and the uncertainty of external environmental factors, traditional fault diagnosis methods often have difficulty in quickly and accurately locating the source of the fault. For example, when a ship system shows abnormal fluctuations, it is difficult to determine whether it is due to the influence of the external environment or an internal mechanical failure. At the same time, even if it is determined to be an internal mechanical failure, it is challenging to further understand the specific location and formation process of the fault. In addition, due to the close connection between the various modules in the ship system, the failure of one module may trigger a chain reaction of other modules, resulting in performance degradation or complete failure of the entire system. In this case, the prediction and prevention of faults become particularly important, but under current technical conditions, how to effectively predict and prevent such chain failures is still an unresolved problem. Therefore, in the face of the high complexity of the ship system and the uncertainty of the external environment, how to quickly and accurately locate and diagnose system faults has become a technical problem that needs to be solved in the field of ship engineering. Summary of the invention

[0003] The present invention provides a multi-source fault autonomous location and classification method, which mainly includes:

[0004] According to the code matching of the ship system fault signal, locate the module related to the fault and determine the structural area where the fault is most likely to occur; calculate the probability of failure of all components in the area through the Bayesian network algorithm, and screen out the key components most likely to fail based on the calculation results; monitor the real-time operation data based on the located key components containing faults to determine whether abnormal fluctuations occur, and determine whether the fluctuations of abnormal data are related to abnormal external environment; build a component root fault knowledge base and root fault mode knowledge map based on the external environment data and historical fault reports during ship operation, and identify and evaluate the cause of the fault through correlation analysis between environmental factors and fault modes and virtual simulation verification; obtain the internal structure of key components based on the three-dimensional model data of key components of the ship The connection structure is calculated, the fault correlation between different structural nodes is calculated, and the associated components with high failure probability are marked. Combined with the spatial correlation evaluation of the structural units, the potential nodes and associated components for fault propagation are determined; based on the 3D scanning data and spatial grid analysis of key components, feature extraction and similarity calculation are performed on other mechanical components in the scanning area, the physical connection and load transfer are evaluated, and potential risk points of chain failures are identified; a probability interval distribution map of the fault spreading outward from the root component is drawn, and the confidence that other faults may be hidden in different diffusion range intervals is calculated; based on the fault root component data of the ship system and the working environment data and task process data of the fault-related system, a component knowledge base and a system operating environment library are constructed to identify the key internal and external factors that affect the occurrence of faults.

[0005] In one embodiment, the code matching of the ship system fault signal, locating the module related to the fault, and determining the structural area where the fault is most likely to occur includes:

[0006] When a fault signal appears in the ship system, the corresponding system module is matched according to the fault code contained in the fault signal, and the matched module is used as the fault-related module; if multiple modules are matched, the network hierarchy analysis method is used to calculate the correlation between the matched modules based on the communication frequency between the modules, functional dependencies, and historical fault data, and the modules with correlation higher than the preset first correlation threshold are retained as the main fault-related modules; the hull area involved in the fault is determined according to the location and function of the main fault-related modules as the structural area where the fault may occur; if the structural area where the fault may occur contains multiple components, based on the physical connection between the components, functional dependencies, or historical maintenance and fault records, Use the Dijkstra algorithm to calculate the correlation between components, and select components with a correlation greater than a preset second correlation threshold; determine whether the selected components are located in the same area. If they are located in the same area, determine the area as the area where failure is most likely to occur; based on the determined area where failure is most likely to occur, obtain detailed data on components in the area, including design parameters, operation history and maintenance records of the components, identify a preset number of components with the most complex connection relationships or the most critical functions in the area, and regard them as core components; if the core components are connected to the monitoring system, obtain real-time operation data of the core components; evaluate whether there are abnormal fluctuations in the real-time operation data. If so, it indicates that key components may fail.

[0007] In one embodiment, the Bayesian network algorithm is used to calculate the probability of failure of all components in the area, and the key components most likely to fail are screened out according to the calculation results, including:

[0008] Obtain all component data in the identified fault-related area and extract the feature vector of each component, including operation history, maintenance records and physical properties; construct a Bayesian network model based on the collected component data and historical fault data, and train the model to learn and predict the probability of each component having a specified type of fault; use the constructed Bayesian network model to predict the probability of component failure based on the latest component status data in the area; if the difference between the failure probabilities of multiple components is less than the preset difference threshold, sort them from high to low according to the correlation between the components; select components with a failure probability greater than the preset first fault threshold based on the prediction results of the Bayesian network model, check the correlation between the selected components, and determine the two components with the strongest correlation as key components; if the Bayesian network training data is insufficient, historical data of similar components are used for training to ensure network convergence; to avoid missing faulty components, components whose failure probability predicted by the Bayesian network model is greater than the preset second fault threshold are retained; check whether the components whose failure probability is greater than the preset second fault threshold belong to the same system. If so, retain the components whose failure probability is greater than the preset second fault threshold as key components; if not, retain all components whose failure probability is greater than the preset third fault threshold as key components; match the corresponding module through the fault code contained in the fault signal, confirm two components among the key components judged by the Bayesian network that are located adjacent to each other in physical position and have a correlation higher than the preset second correlation threshold, and determine these two components as core key components.

[0009] In one embodiment, the monitoring of real-time operation data based on the located key component containing the fault, determining whether abnormal fluctuations occur, and determining whether the fluctuations of the abnormal data are associated with an abnormal external environment include:

[0010] Obtain a list of located key components and determine whether all located key components are connected to the real-time monitoring system; for components that are not connected to the real-time monitoring system, search for replacement components whose functional or structural characteristics are similar to those of the components that are not connected to the monitoring system and whose similarity is greater than a preset similarity threshold, and obtain the real-time operating data of the replacement components; obtain the real-time operating data of all key components through data acquisition equipment, including current, voltage, temperature, and flow; set upper and lower limits of warning thresholds for each parameter to monitor whether the data exceeds the preset warning threshold range; if the monitoring data exceeds the preset warning threshold range, it is determined that the key component has abnormal fluctuations; if the monitoring data exceeds the threshold, it is determined that the key component has abnormal fluctuations, and evaluate whether the abnormal fluctuation amplitude meets the preset fault fluctuation amplitude requirement; obtain external route environment data that matches the monitoring time point, including sea conditions, temperature, and salinity; determine each of the external route environment data Check whether the environmental parameters are abnormal and match them with the data characteristics of component failure fluctuations, and use the method of calculating the Pearson correlation coefficient to determine the correlation between the two; if the external route environment data cannot be directly obtained, use the convolutional neural network algorithm to train the model based on the historical external monitoring screen marked with the external environmental conditions to determine the external environmental conditions; call the current external monitoring screen and determine the external environmental conditions based on the trained convolutional neural network model; based on historical data, evaluate the probability of failure of key components under specified environmental conditions, and match the abnormal parameter fluctuation characteristics of key components with the environmental abnormality characteristics; by comparing the similarity between the environmental abnormality data and the key component abnormality data, identify the possible correlation between the two; if the similarity between the abnormal fluctuation characteristics of the key component and the external environmental abnormality characteristics is greater than the preset similarity threshold, it is determined that the environmental abnormality induces or aggravates the failure of the key component.

[0011] In one embodiment, the component root fault knowledge base and root fault mode knowledge graph are constructed based on the external environment data and historical fault reports during ship operation, and the cause of the fault is identified and evaluated through correlation analysis between environmental factors and fault modes and virtual simulation verification, including:

[0012] Obtain various external environmental data during the operation of the ship, including temperature, humidity, salinity, sea conditions, and magnetic field; extract component types, fault parameters, and root fault modes in the reports by searching historical fault reports, and build a component root fault knowledge base, which includes the fault modes and related fault information of various components; according to the root fault mode and external environmental data, determine the potential relationship between the two through the Apriori algorithm, build a root fault mode knowledge graph, and display the influence network and association structure between environmental factors and root fault modes; for newly appeared faulty components, determine whether the component type already exists in the component root fault knowledge base; if not, build the root fault mode knowledge of this type of component; after completing the root fault mode determination of the faulty component, call the root fault mode knowledge graph, Automatically match the environmental influencing factor network corresponding to the fault mode and query the relevant environmental factor nodes; check whether the matched environmental factors have abnormal fluctuations or exceed the limit during the fault time period. If abnormal fluctuations or exceed the limit occur, it is judged that the environmental abnormality causes or aggravates the fault; if there is no relevant environmental impact match in the root fault mode knowledge graph, match the association between the corrosion mode and environmental factors corresponding to the component material or structure type to determine whether the fault is caused by environmental corrosion; for faults with uncertain judgment results, configure simulation parameters according to the actual operating characteristics and environmental interaction characteristics of the components, use virtual simulation software ANSYS to build a virtual simulation model, simulate the working status of the components under specified environmental conditions, and verify the influence of environmental factors. The simulation parameters include load conditions, thermal conductivity, and corrosion rate.

[0013] In one embodiment, the method of obtaining the internal connection structure of the key components based on the three-dimensional model data of the key components of the ship, calculating the fault correlation between different structural nodes, calibrating the high fault probability associated components, and combining the spatial correlation evaluation of the structural units to determine the potential nodes and associated components for fault propagation includes:

[0014] Obtain the 3D model data of the key components with locked faults, parse the connection structure network inside the components, and identify the connection nodes, including bolts, cables, and pipe interfaces; count the number of connections of each node, and determine the core nodes whose connection number is greater than the preset connection number threshold; define the distance measurement between nodes in the connection structure network, and the shorter the distance, the higher the correlation between the two nodes; use the Dijkstra algorithm to calculate the shortest path between all node pairs in the network; identify the node pairs whose fault correlation is greater than the preset fault correlation threshold by comparing the calculated shortest path lengths; based on the Bayesian network model, use historical fault statistics to determine the conditional probability between the structural nodes and the fault mode; calculate Calculate the posterior probability of the corresponding fault for the node pair whose fault correlation is greater than the preset fault correlation threshold, and determine the core node whose fault probability is greater than the preset fault probability threshold; obtain the front and rear structural components connected to the core node to form the influence propagation network of the key component fault; according to the network connection and fault correlation, judge the structural components with the correlation with the core node greater than the preset correlation threshold, and mark them as high fault probability associated components; obtain the information of all key components in the ship system, including the physical location, functional description, and connection status of key components with other components, and construct a fully connected topological network. In the connected topological network, each node represents a component and the edge represents the connection relationship between components; based on the existing Identify the faulty components and the physical, functional or data connection relationship between the faulty components and other components in the system, and build a fault impact network. The nodes in the fault impact network represent components, and the edges represent the interactions between components caused by fault propagation; combine the discrete sensor sampling signals to predict the fault hotspot propagation path and locate the high fault probability hidden danger components. The high fault probability hidden danger components include the key components with locked faults and the high fault probability associated components; obtain the structural components whose physical location distance is less than the preset distance threshold, and determine whether the structural unit belongs to the fault impact network; the physical location information of each structural component, the connection type between components and the historical fault data, the nearest neighbor algorithm is used for model training, and the fault impact network is constructed. The component spatial association model determines the spatial correlation of each structural component. The spatial correlation is used as a correction coefficient to correct the failure probability of the node and improve the model matching accuracy. The physical location information includes coordinates and distances, and the connection types include direct connection and indirect connection. Through the application of fully connected topological networks, Bayesian network models and component spatial association models, the failure probability of nodes and components is calculated in multiple dimensions. If the failure probability is greater than the preset failure probability threshold, it is determined to be a structural component that is prone to failure, and it is recommended to include it in the scope of linkage monitoring. It also includes: calculating the fault propagation impact index based on the analysis of the degree of dependence between components, operation history scores and the influence of environmental factors, and optimizing the maintenance and overhaul plan of the ship system.

[0015] The above-mentioned method calculates the fault propagation impact index based on the analysis of the dependency between components, the operation history score and the influence of environmental factors, and optimizes the maintenance and overhaul plan of the ship system, specifically including:

[0016] Use dependency matrix to represent direct and indirect connections between components, quantitatively analyze the degree of dependency between each component, and determine the functional dependency F. Obtain the usage frequency, maintenance record, and failure history of each component to determine the operation history score H. Determine the environmental factor influence E based on the impact of environmental parameters on each component; define the fault propagation impact index formula FPI = w 1 ×F+w 2 ×H+w 3 ×E,w 1 , w 2 , w 3 is the weight coefficient, which is used to adjust the influence of these three factors in FPI, w 1 , w 2 , w 3 The value of is obtained from historical data. Calculate the FPI value for each component, prioritize the components according to their FPI values, and optimize maintenance and overhaul plans. Establish a real-time monitoring system to continuously obtain parameters that affect FPI. Recalculate FPI values ​​regularly to reflect the latest operating and environmental conditions. Verify the accuracy of the FPI model through actual cases and historical failure data. Adjust the parameters and weights in the FPI formula based on actual operating data and failure occurrences.

[0017] In one embodiment, the three-dimensional scanning data and spatial grid analysis of key components are used to extract features and calculate similarities of other mechanical components in the scanning area, evaluate physical connections and load transfer, and identify potential cascading failure risk points, including:

[0018] Acquire the three-dimensional scanning data of the physical area where the determined high failure probability associated components are located; construct a local range spatial grid with each high failure probability associated component as the center, and set the area size to a preset multiple of the average size of the high failure probability associated component; extract features of other mechanical components in the scanning area, the features include structural features and functional features, the structural features include material, component size, interface form, and functional features include motion form and applied load; based on the extracted features, use the cosine similarity method to calculate the similarity between other components in the area and the high failure probability associated components in terms of structural features and functional features; set a similarity threshold to screen out faulty components whose structural or functional similarity is higher than the preset similarity threshold as similar components; determine whether there is a physical connection between the screened similar components and the fault-related components The physical connection includes fixed connection and load transmission; if there is a physical connection, it is determined that similar components may also fail; the workload and real-time status data of similar components are obtained, and the support vector machine algorithm is used for model training to predict the failure probability of the components, and the warning level is set for each component according to the failure probability; the cross-correlation function is used to evaluate the time relationship between failure events, and the Moran's I index is used to evaluate the spatial relationship between components; according to the calculated time and space correlation, the failure probability of each component is corrected; all components with structural and functional similarity higher than the preset similarity threshold and correlation with known faulty components higher than the preset correlation threshold are marked as chain failure risk points; when a key component fails, the status of all marked risk points is immediately checked to prevent the expansion of the chain consequences of the failure.

[0019] In one embodiment, the method of drawing a probability interval distribution diagram of the fault spreading outward from the root component and calculating the confidence that other faults may be hidden in different diffusion range intervals includes:

[0020] Based on the identified fault root component, a system connectivity network including the component is constructed. The system connectivity network represents the connection relationship and interaction between components. The Monte Carlo simulation method is used to calculate the probability path of the fault spreading outward from the root component. The range interval of fault diffusion is set, including the periphery of the core component, the local subsystem, and the overall system. Based on historical statistical data, the Bayesian network algorithm is used for model training to predict the potential fault probability distribution in each diffusion interval. A probability distribution diagram of the fault spreading outward from the core component is drawn to show the decreasing trend of the fault probability in different intervals. The key components or subsystems corresponding to each probability interval are marked on the probability distribution diagram as the key inspection objects. For those with a confidence level greater than the preset value, the key components or subsystems are marked as the key inspection objects. In the interval of the confidence threshold, the bypass sensor of the auxiliary detection equipment is started, and the bypass sensor signal is used to determine whether there are any faults that are not directly identified in the interval; the type and number of faults detected in the interval are counted; if the number of detected faults exceeds the preset number range or the severity exceeds the preset severity range, the fault diffusion speed or impact range is judged to be abnormal, and the detection range is expanded; based on the detection results and confidence level, the number distribution of hidden faults in each probability interval is determined, the risk level of each interval is evaluated, and the maintenance plan is optimized; it also includes: based on the structural analysis of the system connection network, the component interaction strength assessment and the environmental parameter impact assessment, the fault diffusion complexity index is calculated and integrated to monitor and update the failure risk of the ship system.

[0021] According to the structural analysis of the system interconnection network, the evaluation of the component interaction strength and the impact of environmental parameters, the fault diffusion complexity index is calculated and integrated, and the failure risk of the ship system is monitored and updated, which specifically includes:

[0022] Through the topological structure of the system's interconnected network, the connectivity distribution, average path length and network cluster coefficient of the network nodes are determined to quantify the complexity of the network connection N. Based on the energy conversion or information exchange frequency between components, the interaction intensity I between components is evaluated; environmental parameter data, including temperature changes and humidity fluctuations, is obtained to evaluate the degree of influence of environmental parameter data on system failure modes, and the influence of environmental parameters on fault propagation P is determined; the formula FDCI = a×N+b×I+c×P is defined to calculate the fault diffusion complexity index FDCI, where a, b, and c are weight coefficients determined based on historical fault data. The FDCI formula is integrated into the monitoring and maintenance platform of the ship system, and real-time data is used to calculate the FDCI to continuously monitor the fault diffusion risk of the system. The performance of the formula is regularly evaluated, and the parameters are updated based on the latest operating data and environmental changes.

[0023] In one embodiment, the component knowledge base and system operating environment base are constructed based on the fault root component data of the ship system and the working environment data and task process data of the fault-related system to identify the key internal and external factors that affect the occurrence of the fault, including:

[0024] Obtain the data of the identified root-cause components and build a component knowledge base containing material, structure, and load attributes; obtain the working environment data and task process data of the fault-related system and build a system operating environment library; use the Apriori algorithm to associate the root-cause component attributes with the variables in the environment library, calculate the support and confidence of each association rule, and screen out the internal and external factors that are most likely to affect the root-cause components; use the Bayesian network algorithm to train the model based on the identified internal and external influencing factor variables, build a component failure prediction model, and determine the root-cause component failure mode; for new fault cases, obtain the component data and environment data related to the case, use the component failure prediction model to obtain the component data and environment data The conditional probability of each variable associated with the fault is calculated; the internal and external root source variables that have the greatest impact on the fault are automatically determined based on the probability obtained from the component fault prediction model; if the probability difference of multiple variables is less than the preset probability difference threshold, the core root cause is determined based on the interaction relationship between the variables; numerical simulation and structural simulation are performed using the virtual simulation software ANSYS to verify the impact of the internal and external root sources output by the component fault prediction model; abnormal changes in the root source variables are simulated to confirm whether they will lead to the occurrence of the fault mode; based on all internal and external root source variables, and sorted according to the degree of relevant impact on the fault, a preset number of the most critical root source variables are marked to determine the most critical internal or external root sources of the system.

[0025] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0026] The present invention provides a method for autonomous positioning and classification of multi-source faults. Through in-depth analysis of each module and component inside the system, combined with dynamic monitoring of the external marine environment, the present invention can accurately locate the source of the fault and determine the root cause of the fault, whether it is caused by internal mechanical failure or external environmental factors. The core advantage is that it can comprehensively analyze the complexity of the internal structure of the system and the uncertainty of the external environment, and provide predictions of the fault propagation path by constructing a knowledge graph and real-time data analysis, so as to provide early warning of possible chain failures. This method improves the response speed to system failures, reduces the fault diagnosis time, and also improves the accuracy and efficiency of maintenance and prevention work. The present invention can identify and predict the fault diffusion trend inside the system, including determining the cause of abnormal fluctuations and locating the correlation between faulty components. This not only helps to maintain the stable operation of the ship system in a complex marine environment, but also significantly improves the safety and reliability of the ship, reduces the risk of accidents, and reduces maintenance costs. In general, the present invention provides an advanced fault management and prevention strategy for the field of ship engineering while ensuring the efficient operation of the ship system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention is a flowchart of a multi-source fault autonomous location and classification method.

[0028] Figure 2 A schematic diagram of a multi-source fault autonomous location and classification method of the present invention.

[0029] Figure 3 It is another schematic diagram of a multi-source fault autonomous location and classification method of the present invention. DETAILED DESCRIPTION

[0030] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0031] In this embodiment, a multi-source fault autonomous location and classification method may specifically include:

[0032] S101. According to the code matching of the ship system fault signal, locate the module related to the fault and determine the structural area where the fault is most likely to occur.

[0033] When a fault signal appears in the ship system, the corresponding system module is matched according to the fault code contained in the fault signal, and the matched module is used as the fault-related module. If multiple modules are matched, the network hierarchy analysis method is used to calculate the correlation between the matched modules based on the communication frequency, functional dependency, and historical fault data between the modules, and the modules with a correlation higher than the preset first correlation threshold are retained as the main fault-related modules. The hull area involved in the fault is determined by the location and function of the main fault-related modules as the structural area where the fault may occur. If the structural area where the fault may occur contains multiple components, the Dijkstra algorithm is used to calculate the correlation between the components based on the physical connection, functional dependency, or historical maintenance and fault records between the components, and the components with a correlation greater than the preset second correlation threshold are selected. It is determined whether the selected components are located in the same area. If they are located in the same area, the area is determined as the area where the fault is most likely to occur. According to the determined area where the fault is most likely to occur, the detailed data of the components in the area are obtained, including the design parameters, operation history and maintenance records of the components, and the preset number of components with the most complex connection relationship or the most critical function in the area are identified and regarded as core components. If the core components are connected to the monitoring system, the real-time operation data of the core components will be obtained. Evaluate whether the real-time operation data has abnormal fluctuations. If so, it indicates that the key components may have failed.

[0034] For example, when a fault signal appears in the ship system, the fault code is 1001, which can be matched to the corresponding system module, such as the power system module. In this way, the power system module becomes a fault-related module. If multiple modules are matched, such as the power system module and the steering gear system module, the network hierarchy analysis method can be used to calculate the correlation between the matched modules based on the communication frequency, functional dependency and historical fault data between the modules. If the correlation between the power system module and the steering gear system module is 8, which is higher than the preset first correlation threshold of 7, then both of them can be retained as the main fault-related modules. The hull area involved in the fault is determined by the location and function of the main fault-related modules, for example, it is determined that the fault may occur in the hull control area. If the structural area where the fault may occur contains multiple components, such as the ship power supply and the steering gear controller, the Dijkstra algorithm can be used to calculate the correlation between the components based on the physical connection, functional dependency or historical maintenance and fault records between the components. If the correlation between the ship power supply and the steering gear controller is 9, which is higher than the preset second correlation threshold of 7, then both of them can be selected as key components. Determine whether the selected components are located in the same area. If they are located in the same area, such as the hull control area, the area can be determined as the area where failure is most likely to occur. According to the area determined to be most likely to fail, obtain detailed data of the components in the area, including the design parameters, operation history and maintenance records of the components. A preset number of components with the most complex connection relationships or the most critical functions in the area can be identified, such as the power controller. If the power controller is connected to the monitoring system, its real-time operation data can be obtained. Evaluate whether there are abnormal fluctuations in the real-time operation data, such as detecting whether the current of the power controller suddenly increases or whether the temperature rises abnormally. If abnormal fluctuations occur, it can indicate that key components may fail, such as the power controller.

[0035] S102. Calculate the probability of failure of all components in the area through the Bayesian network algorithm, and select the key components most likely to fail based on the calculation results.

[0036] Obtain all component data for the identified fault-related area and extract the feature vector of each component, including operation history, maintenance records and physical properties. Based on the collected component data and historical fault data, construct a Bayesian network model and train the model to learn and predict the probability of each component having a specified type of fault. Based on the latest component status data in the area, use the constructed Bayesian network model to predict the probability of component failure. If the difference between the failure probabilities of multiple components is less than the preset difference threshold, sort them from high to low according to the correlation between the components. According to the prediction results of the Bayesian network model, select components with a failure probability greater than the preset first fault threshold, check the correlation between the selected components, and determine the two components with the strongest correlation as key components. If the Bayesian network training data is insufficient, use historical data of similar components for training to ensure network convergence. To avoid missing faulty components, retain components whose failure probability predicted by the Bayesian network model is greater than the preset second fault threshold. Check whether the components with a failure probability greater than the preset second failure threshold belong to the same system. If so, keep the components with a failure probability greater than the preset second failure threshold as key components; if not, keep all components with a failure probability greater than the preset third failure threshold as key components. Match the corresponding module with the fault code contained in the fault signal, confirm that two components in the key components determined by the Bayesian network are located adjacent to each other in physical position and have a correlation higher than the preset second correlation threshold, and determine these two components as core key components.

[0037] For example, a ship's navigation system contains multiple components, such as a GPS receiver, a compass, and a gyroscope. Data on these components, including operation history, maintenance records, and physical properties, were collected. The GPS receiver ran an average of 18 hours per day in the past year and had two failure records; the maintenance record of the gyroscope showed that it had undergone four routine maintenance and one emergency repair in the past three years. Using these data, a Bayesian network model was constructed to predict the probability of failure of each component of the navigation system. If the latest status data shows that some monitoring parameters of the GPS receiver, such as signal strength, have a slight abnormality recently, the model predicts that the probability of failure of the component is 0.4, and at the same time, the predicted probability of failure of the gyroscope is 0.2. Since the failure probability of the GPS receiver of 0.4 exceeds the preset first failure threshold of 0.3, and the failure probability of the gyroscope of 0.2 exceeds the preset second failure threshold of 0.15, these two components are identified as key components. Since the GPS receiver and the gyroscope are physically adjacent and functionally related in the navigation system, and both exceed their respective failure thresholds, they are identified as core key components.

[0038] S103. Based on the located key components with faults, monitor the real-time operation data to determine whether abnormal fluctuations occur, and determine whether the fluctuations in the abnormal data are related to abnormal external environment.

[0039] Obtain a list of key components that have been located, and determine whether all key components that have been located are connected to the real-time monitoring system. For components that are not connected to the real-time monitoring system, search for substitute components whose functional or structural characteristics are similar to those of components that are not connected to the monitoring system and whose similarity is greater than the preset similarity threshold, and obtain the real-time operation data of the substitute components. Obtain the real-time operation data of all key components through data acquisition equipment, including current, voltage, temperature, and flow. Set upper and lower limits of warning thresholds for each parameter, and monitor whether the data exceeds the preset warning threshold range. If the monitoring data exceeds the preset warning threshold range, it is determined that the key component has abnormal fluctuations. If the monitoring data exceeds the threshold, it is determined that the key component has abnormal fluctuations, and evaluate whether the abnormal fluctuation amplitude meets the preset fault fluctuation amplitude requirement. Obtain external route environmental data that matches the monitoring time point, including sea conditions, temperature, and salinity. Determine whether each environmental parameter of the external route environmental data is abnormal, and match it with the data characteristics of the component fault fluctuation, and use the method of calculating the Pearson correlation coefficient to determine the correlation between the two. If the external route environmental data cannot be directly obtained, the convolutional neural network algorithm is used to train the model based on the historical external monitoring screen marked with the external environmental conditions to determine the external environmental conditions. Call the current external monitoring screen and determine the external environmental conditions based on the trained convolutional neural network model; evaluate the probability of failure of key components under specified environmental conditions based on historical data, and match the abnormal parameter fluctuation characteristics of key components with the abnormal environmental characteristics. By comparing the similarity between the abnormal environmental data and the abnormal data of key components, identify the possible association between the two. If the similarity between the abnormal fluctuation characteristics of the key components and the abnormal characteristics of the external environment is greater than the preset similarity threshold, it is determined that the abnormal environment induces or aggravates the failure of the key component.

[0040] For example, a navigation system of a ship is being processed, in which the key components include a GPS receiver and a radar system. The GPS receiver is connected to the real-time monitoring system. The radar system is not connected to the real-time monitoring system, and another radar system is searched as a replacement component. Its functional and structural characteristics are similar to the original radar system at 0.9, and the preset similarity threshold is 0.8. The real-time data of the GPS receiver shows that the current is 1.2A, the normal range is 0.8A-1.5A, and the voltage is 12V, the normal range is 11V-13V. The data of the replacement radar system shows that the current is 2.5A, the normal range is 1.8A-2.3A, and the voltage is 11.5V, the normal range is 11V-13V, indicating that the current exceeds the warning threshold. The external route environment data at the monitoring time point shows that the sea condition fluctuates greatly, the temperature is 28℃, the normal range is 15℃-25℃, and the salinity is normal. Using the Pearson correlation coefficient method for analysis, it is found that the correlation coefficient between the abnormal current fluctuation of the radar system and the large fluctuation of the sea condition is 0.8, which is highly correlated. Since it is impossible to directly obtain sea condition data, a well-trained convolutional neural network model is used to analyze historical external monitoring images to confirm that the current sea condition is highly volatile. According to historical data, the probability of radar system failure increases to 0.7 under volatile sea conditions, compared to 0.2 under normal conditions. Since the abnormal current of the radar system is highly correlated with large fluctuations in sea conditions, it is judged that the abnormal fluctuation of the radar system this time is likely to be induced or aggravated by large fluctuations in sea conditions.

[0041] S104. Based on the external environment data and historical fault reports during ship operation, a knowledge base of component root faults and a knowledge graph of root fault modes are constructed, and the causes of faults are identified and evaluated through correlation analysis between environmental factors and fault modes and virtual simulation verification.

[0042] Obtain various external environmental data during the operation of the ship, including air temperature, humidity, salinity, sea conditions, and magnetic field; extract component types, fault parameters, and root fault modes in the reports by searching historical fault reports, and build a component root fault knowledge base, which contains the fault modes and related fault information of various components. According to the root fault mode and external environmental data, the potential relationship between the two is determined through the Apriori algorithm, and the root fault mode knowledge graph is constructed to display the influence network and association structure between environmental factors and root fault modes. For newly appeared faulty components, determine whether the component type already exists in the component root fault knowledge base. If not, build the root fault mode knowledge of this type of component. After the root fault mode of the faulty component is determined, call the root fault mode knowledge graph, automatically match the environmental influencing factor network corresponding to the fault mode, and query the relevant environmental factor nodes. Check whether the matched environmental factors have abnormal fluctuations or over-limit phenomena during the fault time period. If abnormal fluctuations or over-limit phenomena occur, it is determined that the environmental abnormality causes or aggravates the fault. If there is no relevant environmental impact match in the root fault mode knowledge graph, the association between the corrosion mode corresponding to the component material or structure type and the environmental factors is matched to determine whether the environmental corrosion causes the fault. For faults with uncertain judgment results, the simulation parameters are configured according to the actual operating characteristics and environmental interaction characteristics of the component, and the virtual simulation software ANSYS is used to build a virtual simulation model to simulate the working state of the component under specified environmental conditions and verify the influence of environmental factors. The simulation parameters include load conditions, thermal conductivity, and corrosion rate.

[0043] For example, a ship's cooling system failure is being handled. The system includes cooling pumps and pipelines. The collected ship operation data shows that the sea conditions are moderately fluctuating, the temperature is 28℃, the humidity is 60%, and the salinity is 3.5%. The failure modes of cooling pumps and pipelines are extracted from historical fault reports. The common failure of cooling pumps is bearing overheating, and the common failure of pipelines is corrosion perforation. The Apriori algorithm is used to analyze the potential relationship between failure modes and environmental factors, and it is found that high salinity environment has a significant correlation with pipeline corrosion. Recently, the cooling system failed, and the failure was manifested as pipeline leakage. This failure mode already exists in the knowledge base. It is confirmed that the recent salinity is 3.5%, which exceeds the safety threshold of 3% of the pipeline material. It is judged that high salinity may aggravate the corrosion of the pipeline. ANSYS is used to simulate the working state of the pipeline in a high salinity environment of 3.5%. The pipeline material is copper, the load condition is normal, the thermal conductivity is 400W / mK, and the corrosion rate is set according to historical data. The simulation results show that in a high salinity environment, the pipeline corrosion rate increases by 20%, resulting in the occurrence of perforation earlier.

[0044] S105. Based on the three-dimensional model data of the key components of the ship, the internal connection structure of the key components is obtained, the fault correlation between different structural nodes is calculated, the associated components with high fault probability are calibrated, and the potential nodes and associated components for fault propagation are determined in combination with the spatial correlation evaluation of the structural units.

[0045] Obtain the three-dimensional model data of the key components with locked faults, parse the connection structure network inside the components, and identify the connection nodes, including bolts, cables, and pipe interfaces. Count the number of connections of each node and determine the core nodes whose connection number is greater than the preset connection number threshold. Define the distance measurement between nodes in the connection structure network. The shorter the distance, the higher the correlation between the two nodes. Use the Dijkstra algorithm to calculate the shortest path between all node pairs in the network. By comparing the calculated shortest path lengths, identify the node pairs whose fault correlation is greater than the preset fault correlation threshold; based on the Bayesian network model, use historical fault statistics to determine the conditional probability between the structural nodes and the fault mode. Calculate the posterior probability of the corresponding fault for the node pairs whose fault correlation is greater than the preset fault correlation threshold, and determine the core nodes whose fault probability is greater than the preset fault probability threshold. Obtain the front and rear structural components connected to the core nodes to form the influence propagation network of the key component fault. According to the network connection and fault correlation, determine the structural components whose correlation with the core nodes is greater than the preset correlation threshold, and mark them as high fault probability associated components. Obtain information on all key components in the ship system, including the physical location, functional description, and connection status of key components with other components, and construct a fully connected topological network. In the connected topological network, each node represents a component, and the edge represents the connection relationship between components. Based on the identified faulty components and the physical, functional or data connection relationship between the faulty components and other components in the system, a fault impact network is constructed. The nodes in the fault impact network represent components, and the edges represent the interactions between components caused by fault propagation. Combined with discrete sensor sampling signals, the fault hotspot propagation path is predicted, and high fault probability hidden danger components are located. High fault probability hidden danger components include key components with locked faults and high fault probability associated components. Obtain structural components whose physical location distance is less than the preset distance threshold, and determine whether the structural unit belongs to the fault impact network; the physical location information of each structural component, the connection type between components and historical fault data, the nearest neighbor algorithm is used for model training, and the component spatial association model is constructed to determine the spatial correlation of each structural component. The spatial correlation is used as a correction coefficient to correct the node's fault probability and improve the model matching accuracy. The physical location information includes coordinates and distance, and the connection type includes direct connection and indirect connection. Through the application of fully connected topological networks, Bayesian network models and component spatial association models, the failure probability of nodes and components is calculated in multiple dimensions. If the failure probability is greater than the preset failure probability threshold, it is determined to be a structural component prone to failure, and it is recommended to include it in the scope of linkage monitoring.

[0046] For example, in a ship's power system, a main generator fails. The generator is connected to multiple components, including cables, switches, and control panels. The three-dimensional model data of the generator is obtained, its internal structure is parsed, and the connection nodes, such as the cable interface and the control connection point, are identified. The number of cable interface connections is 8, the number of control connection points connections is 5, and the preset connection number threshold is 3. The distance between nodes is defined in the model, and the Dijkstra algorithm is used to calculate the shortest path between nodes. The shortest path length from the cable interface to the control connection point is 2 units. The fault correlation threshold is set to 1.5 units. Since the path length of 2 units from the cable interface to the control connection point is greater than this threshold, the fault correlation between the two nodes is considered to be low. Since the number of connections of the cable interface is 8 and is higher than the preset connection number threshold of 3, it is determined to be a core node. The components directly connected to the generator cable interface and the switchboard connected to it are obtained to form a fault impact propagation network. The information of all key components in the ship system is obtained, including the physical location, functional description, and connection status of key components with other components, and a fully connected topological network is constructed. In the connected topological network, each node represents a component and the edge represents the connection relationship between components. Based on the faulty component and its physical, functional or data connection with other components in the system, a fault impact network is constructed, in which the cable interface and the switchboard are key nodes. Combined with discrete sensor sampling signals, such as current and voltage changes, the fault hotspot propagation path, such as current anomalies from the generator to the switchboard, is predicted. Other components that are physically close to the generator and switchboard, such as the backup power unit, are collected and model trained using the nearest neighbor algorithm to construct a component spatial association model and determine their spatial correlation. If the physical distance between the backup power unit and the generator is 1.2 units, which is less than the preset distance threshold of 2 units, it is included in the fault impact network. Based on the Bayesian network model and spatial correlation, the failure probability of each component is calculated. If the failure probability of the generator is 0.7, the switchboard is 0.5, and the backup power unit is 0.3. Since the failure probability of the generator and the switchboard exceeds the preset failure probability threshold of 0.4, they are judged as structural components prone to failure and are recommended to be included in the scope of linkage monitoring.

[0047] Based on the analysis of the degree of dependency between components, operation history scores and the influence of environmental factors, the fault propagation impact index is calculated to optimize the maintenance and overhaul plan of the ship system.

[0048] Use dependency matrix to represent direct and indirect connections between components, quantitatively analyze the degree of dependency between each component, and determine the functional dependency F. Obtain the usage frequency, maintenance record, and failure history of each component to determine the operation history score H. Determine the environmental factor influence E based on the impact of environmental parameters on each component; define the fault propagation impact index formula FPI = w1 ×F+w 2 ×H+w 3 ×E,w 1 , w 2 , w 3 is the weight coefficient, which is used to adjust the influence of these three factors in FPI, w 1 , w 2 , w 3 The value of is obtained from historical data. Calculate the FPI value for each component, prioritize the components according to their FPI values, and optimize maintenance and overhaul plans. Establish a real-time monitoring system to continuously obtain parameters that affect FPI. Recalculate FPI values ​​regularly to reflect the latest operating and environmental conditions. Verify the accuracy of the FPI model through actual cases and historical failure data. Adjust the parameters and weights in the FPI formula based on actual operating data and failure occurrences.

[0049] For example, there is a ship system that contains three key components, the engine, pump, and valve. Using dependency matrix analysis, it is found that the engine's dependence on the pump and valve is 0.8 and 0.6 respectively, while the pump's dependence on the valve is 0.7. Therefore, the F value of the engine is 0.8, the F value of the pump is 0.75, and the F value of the valve is 0.6. The engine is used frequently, has frequent maintenance records, and has a medium history of failures, so the H value is set to 0.7. The pump is used moderately, has few maintenance records and little history of failures, so the H value is set to 0.5. The valve is used infrequently, has few maintenance records and little history of failures, so the H value is set to 0.4. The engine operates in a high temperature environment, so the E value is 0.8. The pump and valve operate in a relatively stable environment, so the E values ​​are 0.5 and 0.4 respectively. If the weight coefficient is, w 1 =0.4,w 2 =0.3,w 3 =0.3. The FPI value of the engine is 0.4×0.8+0.3×0.7+0.3×0.8=0.77; the FPI value of the pump is 0.4×0.75+0.3×0.5+0.3×0.5=0.6. The FPI value of the valve is

[0050] 0.4×0.6+0.3×0.4+0.3×0.4=0.48. Based on the FPI value, the engine is identified as the weakest link in the system, followed by the pump and then the valve. A real-time monitoring system is established to regularly update the FPI values ​​of these components to reflect the latest operating and environmental conditions.

[0051] S106. Based on the 3D scanning data and spatial grid analysis of key components, perform feature extraction and similarity calculation on other mechanical components in the scanning area, evaluate physical connections and load transfer, and identify potential cascading failure risk points.

[0052] Obtain the three-dimensional scanning data of the physical area where the determined high failure probability associated components are located. With each high failure probability associated component as the center, construct a local range spatial grid, and set the area size to a preset multiple of the average size of the high failure probability associated component. Extract features of other mechanical components in the scanning area, including structural features and functional features. Structural features include material, component size, and interface form, and functional features include motion form and applied load. Based on the extracted features, the cosine similarity method is used to calculate the similarity of other components in the area with the high failure probability associated components in terms of structural features and functional features. Set a similarity threshold to filter out faulty components whose structural or functional similarity is higher than the preset similarity threshold as similar components. Determine whether there is a physical connection between the filtered similar components and the fault-related components, and the physical connection includes fixed connection and load transfer. If there is a physical connection, it is determined that the similar components may also fail. Obtain the workload and real-time status data of similar components, use the support vector machine algorithm to train the model, predict the failure probability of the components, and set the warning level for each component according to the failure probability. Use the cross-correlation function to evaluate the temporal relationship between failure events, and use the Moran's I index to evaluate the spatial relationship between components. Correct the failure probability of each component based on the calculated temporal and spatial correlations. Mark all components whose structural and functional similarities are higher than the preset similarity threshold and whose correlations with known faulty components are higher than the preset correlation threshold as cascading failure risk points. When a critical component fails, immediately check the status of all marked risk points to prevent the expansion of the cascading consequences of the failure.

[0053] For example, a critical electronic control unit ECU in the ship's engine control system is being analyzed for failure. The 3D scan data of the engine control unit ECU is obtained. The size of the ECU is 20cmx10cmx5cm. This size is set as the basic size of the local range spatial grid. In the grid around the ECU, other mechanical and electronic components such as sensors, connectors, and cables are scanned and identified. The features of these components, including material, size, and function, are extracted. The similarity between these components and the ECU is calculated using the cosine similarity method, and the similarity calculation result between a neighboring sensor and the ECU is 0.9. The similarity threshold is set to 0.85, so the sensor with a similarity of 0.9 is identified as a similar component. Check the physical connection between the sensor and the ECU, such as whether it is directly connected by a cable. Collect the sensor's workload and real-time status data, such as voltage and current values. Use the support vector machine algorithm to predict the sensor's failure probability based on these data. If the predicted failure probability is 0.7, it is higher than the preset warning threshold of 0.6. Using the cross-correlation function and Moran's I index, the temporal and spatial relationship between sensor failure and ECU failure is evaluated. The analysis results show that the sensor failure is closely related to the ECU failure time and the spatial location is close. This sensor is marked as a cascading failure risk point. When an ECU fails, the status of the sensor is checked immediately to prevent potential cascading failures.

[0054] S107. Draw a probability interval distribution diagram of the fault spreading outward from the root component, and calculate the confidence that other faults may be hidden in different diffusion range intervals.

[0055] Based on the identified fault root component, a system connectivity network including the component is constructed. The system connectivity network represents the connection relationship and interaction between components. The Monte Carlo simulation method is used to calculate the probability path of the fault spreading outward from the root component. The range interval of fault diffusion is set, including the periphery of the core component, the local subsystem, and the overall system. Based on historical statistical data, the Bayesian network algorithm is used for model training to predict the potential fault probability distribution in each diffusion interval. A probability distribution diagram of the fault spreading outward from the core component is drawn to show the decreasing trend of the fault probability in different intervals. The key components or subsystems corresponding to each probability interval are marked on the probability distribution diagram as the key inspection objects. For intervals with a confidence greater than the preset confidence threshold, the bypass sensor of the auxiliary detection equipment is started, and the bypass sensor signal is used to determine whether there are faults that are not directly identified in the interval. The type and number of faults detected in the interval are counted; if the number of faults detected exceeds the preset number range or the severity exceeds the preset severity range, it is judged that the fault diffusion speed or the impact range is abnormal, and the detection range is expanded. Based on the test results and confidence levels, the number and distribution of hidden faults in each probability interval are determined, the risk level of each interval is evaluated, and the maintenance plan is optimized.

[0056] For example, a system consists of core components A, B, and C. The fault source component is A. According to the system connectivity network, component A is connected to B and C. We want to calculate the probability path of the fault spreading outward from component A. First, a Monte Carlo simulation is performed. If 1,000 simulations are performed, a path starting from component A is randomly selected in each simulation, and the components passed through are recorded. Based on historical statistical data, the Bayesian network algorithm is used for model training to predict the potential fault probability distribution in each diffusion interval. If the range interval of the fault diffusion is the core component periphery, local subsystem, and overall system. The probability distribution is expressed as P(peripheral fault|A), P(local fault|A), and P(overall fault|A). Based on the simulation results and the prediction results of the Bayesian network algorithm, a probability distribution diagram of the fault spreading outward from the core component A can be drawn, and the decreasing trend of the fault probability in different intervals can be shown in the diagram. On the probability distribution diagram, the key components or subsystems corresponding to each probability interval can be marked as the key inspection objects. If the interval with a confidence greater than the preset confidence threshold is found to be the core component periphery and local subsystem. For these intervals with high confidence, auxiliary detection equipment can be started to use bypass sensor signals to determine whether there are faults in the interval that are not directly identified; in some cases, additional sensors may be deployed to supplement the monitoring of the main sensor, such as in areas that the main sensor cannot cover or to enhance the reliability of the system. These may be informally referred to as bypass sensors or auxiliary sensors, but this is not a standard industrial term. In actual applications, the selection and deployment of these sensors will vary according to the specific needs and design of the system. The types and number of faults detected in the statistical interval. If the number of detected faults exceeds the preset number range or the severity exceeds the preset severity range, it can be judged that the fault propagation speed or impact range is abnormal, so the detection range needs to be expanded. Based on the detection results and confidence levels, the distribution of the number of hidden faults in each probability interval can be determined, the risk level of each interval can be evaluated, and the maintenance plan can be optimized.

[0057] Based on the structural analysis of the system connectivity network, the component interaction strength assessment and the environmental parameter impact evaluation, the fault diffusion complexity index is calculated and integrated to monitor and update the failure risk of the ship system.

[0058] Through the topological structure of the system's interconnected network, the connectivity distribution, average path length and network cluster coefficient of the network nodes are determined to quantify the complexity of the network connection N. Based on the energy conversion or information exchange frequency between components, the interaction intensity I between components is evaluated; environmental parameter data, including temperature changes and humidity fluctuations, is obtained to evaluate the degree of influence of environmental parameter data on system failure modes, and the influence of environmental parameters on fault propagation P is determined; the formula FDCI = a×N+b×I+c×P is defined to calculate the fault diffusion complexity index FDCI, where a, b, and c are weight coefficients determined based on historical fault data. The FDCI formula is integrated into the monitoring and maintenance platform of the ship system, and real-time data is used to calculate the FDCI to continuously monitor the fault diffusion risk of the system. The performance of the formula is regularly evaluated, and the parameters are updated based on the latest operating data and environmental changes.

[0059] For example, the analysis of the system connectivity network shows that the average node degree is 5, the average path length is 3, and the network clustering coefficient is 0.6. The quantified network connection complexity N = (5 + 3 + 0.6) / 3 = 2.87. Through the analysis of energy conversion and information exchange frequency, the interaction intensity score is determined to be 7. After analyzing the environmental parameter data, such as temperature and humidity changes, the impact score on the failure mode is determined to be 6. Assume that based on historical data, the weight coefficients are determined as,

[0060] a=0.3, b=0.4, c=0.3a=0.3, b=0.4, c=0.3, then

[0061] FDCI = 0.3 × 2.87 + 0.4 × 7 + 0.3 × 6 = 0.861 + 2.8 + 1.8 = 5.461. The calculated FDCI value of 5.461 is integrated into the monitoring platform of the ship system. This value is used to continuously monitor the risk of fault propagation of the system and identify possible risk points. During subsequent operation, if there is a significant change in environmental parameters, such as increased temperature fluctuations, causing the P score to rise to 8, the FDCI is recalculated. The updated FDCI is FDCI = 0.3 × 2.87 + 0.4 × 7 + 0.3 × 8 = 6.061.

[0062] S108. Based on the fault root component data of the ship system and the working environment data and task process data of the fault-related system, a component knowledge base and a system operating environment library are constructed to identify key internal and external factors that affect the occurrence of faults.

[0063] Obtain the data of the identified fault root component and build a component knowledge base containing material, structure, and load attributes. Obtain the working environment data and task process data of the system related to the fault and build a system operation environment library. Use the Apriori algorithm to associate the root component attributes with the variables in the environment library, calculate the support and confidence of each association rule, and screen out the internal and external factors that are most likely to affect the root component. According to the identified internal and external influencing factor variables, use the Bayesian network algorithm to train the model, build a component failure prediction model, and determine the root component failure mode. For new fault cases, obtain the component data and environmental data related to the case, use the component failure prediction model to obtain the conditional probability of each variable in the component data and environmental data related to the fault. According to the probability obtained by the component failure prediction model, automatically determine the internal and external root source variables that have the greatest impact on this fault. If the probability difference of multiple variables is less than the preset probability difference threshold, the core root cause is determined based on the interaction relationship between the variables. Use the virtual simulation software ANSYS for numerical simulation and structural simulation to verify the influence of the internal and external root causes output by the component failure prediction model. Simulate abnormal changes in root variables to confirm whether they will lead to the occurrence of the fault mode. Based on all internal and external root variables, sort them according to the degree of their impact on the fault, mark a preset number of the most critical root variables, and determine the most critical internal or external root causes.

[0064] For example, to build a component knowledge base, which contains material, structure and load attributes. Data of multiple identified fault root components are collected, such as aluminum alloy material, support frame structure and heavy machinery load attribute. In addition, it is necessary to build a system operating environment library, which contains working environment data and task process data of fault-related systems. System operating environment data are collected, such as temperature of 40°C, humidity of 80%, and high-speed operation task process data. Next, the Apriori algorithm is used to associate the root component attributes with the variables in the environment library. By calculating the support and confidence of each association rule, the internal and external factors that are most likely to affect the root component can be screened out, and it is found that the failure probability of aluminum alloy components is higher in high temperature environments. Then, the Bayesian network algorithm is used for model training, a component failure prediction model is constructed, and the root component failure mode is determined. Through the identified internal and external influencing factor variables, the conditional probability of each variable in the component data and environmental data related to the fault can be obtained according to the model prediction. For a new fault case, the component data and environmental data related to the case are obtained, and the component fault prediction model is used to obtain the conditional probability of each variable related to the fault. According to the probability, the internal and external root source variables that have the greatest impact on this failure can be automatically determined. If the probability difference of multiple variables is less than the preset probability difference threshold, the core root cause can be determined based on the interaction between the variables. The virtual simulation software ANSYS is used for numerical simulation and structural simulation to verify the impact of the internal and external root causes output by the component failure prediction model, simulate the abnormal changes in the root source variables, and confirm whether the failure mode will be generated. Finally, based on all internal and external root source variables and sorted according to the degree of relevant impact on this failure, the most critical preset number of root source variables can be marked, and the most critical system internal or external root source can be determined. If the material and high temperature environment of the aluminum alloy component are both determined as key root source variables, it can be determined that the high temperature environment is the most critical external root source.

[0065] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A multi-source fault autonomous location and classification method, characterized in that: The method comprises: According to the code matching of the ship system fault signal, locate the module related to the fault and determine the structural area where the fault is most likely to occur; The probability of failure of all components in the area is calculated through the Bayesian network algorithm, and the key components most likely to fail are screened out based on the calculation results; Based on the located key components with faults, monitor real-time operation data to determine whether abnormal fluctuations occur, and determine whether the fluctuations in abnormal data are related to abnormal external environment; Based on the external environmental data and historical fault reports during ship operation, a knowledge base of component root faults and a knowledge graph of root fault modes are constructed. The causes of faults are identified and evaluated through correlation analysis between environmental factors and fault modes and virtual simulation verification. According to the 3D model data of key ship components, the internal connection structure of key components is obtained, the fault correlation between different structural nodes is calculated, and the high fault probability associated components are calibrated. Combined with the spatial correlation evaluation of structural units, the potential nodes and associated components for fault propagation are determined; Based on the 3D scanning data and spatial grid analysis of key components, feature extraction and similarity calculation are performed on other mechanical components in the scanning area to evaluate physical connections and load transfer and identify potential cascading failure risk points; Draw a probability interval distribution diagram of the fault spreading outward from the root component, and calculate the confidence that other faults may be hidden in different diffusion range intervals; According to the fault root component data of the ship system and the working environment data and task process data of the fault-related system, a component knowledge base and a system operating environment library are built to identify the key internal and external factors that affect the occurrence of faults; The determining of the structural area where the fault is most likely to occur comprises: determining the hull area involved in the fault as the structural area where the fault may occur through the location and function of the main fault-related modules; if the structural area where the fault may occur contains multiple components, then based on the physical connection, functional dependency, or historical maintenance and fault records between the components, using the Dijkstra algorithm, calculating the correlation between the components, and selecting components with a correlation greater than a preset second correlation threshold; judging whether the selected components are located in the same area, and if they are located in the same area, determining the area as the area where the fault is most likely to occur; according to the determined area where the fault is most likely to occur, obtaining detailed data of the components in the area, including the design parameters, operation history and maintenance records of the components, identifying a preset number of components with the most complex connection relationship or the most critical function in the area, and treating them as core components; if the core components are connected to the monitoring system, obtaining the real-time operation data of the core components; evaluating whether the real-time operation data has abnormal fluctuations, and if so, indicating that the key components may fail; The method of screening out the key components most likely to fail according to the calculation results includes: selecting components with a failure probability greater than a preset first failure threshold according to the prediction results of the Bayesian network model, checking the correlation between the selected components, and determining the two components with the strongest correlation as key components; if the Bayesian network training data is insufficient, using historical data of similar components for training to ensure network convergence; to avoid missing faulty components, retaining components with a failure probability greater than a preset second fault threshold predicted by the Bayesian network model; checking whether the components with a failure probability greater than the preset second fault threshold belong to the same system, if so, retaining the components with a failure probability greater than the preset second fault threshold as key components; if not, retaining all components with a failure probability greater than a preset third fault threshold as key components; matching the corresponding module with the fault code contained in the fault signal, confirming two components in the key components determined by the Bayesian network that are located adjacent in physical position and have a correlation higher than the preset second correlation threshold, and determining the two components as core key components; The method of judging whether an abnormal fluctuation occurs and judging whether the fluctuation of abnormal data is related to an abnormal external environment includes: setting upper and lower limits of warning thresholds for each parameter, and monitoring whether the data exceeds a preset warning threshold range; if the monitoring data exceeds a preset warning threshold range, it is judged that an abnormal fluctuation occurs in a key component; if the monitoring data exceeds the threshold, it is judged that an abnormal fluctuation occurs in a key component, and it is evaluated whether the amplitude of the abnormal fluctuation reaches a preset fault fluctuation amplitude requirement; obtaining external route environmental data matching the monitoring time point, including sea conditions, temperature and salinity; judging whether each environmental parameter of the external route environmental data is abnormal, and matching it with the data characteristics of the component fault fluctuation, and using the method of calculating the Pearson correlation coefficient to determine the correlation between the two. If the external route environment data cannot be directly obtained, the external environment conditions are determined by using the convolutional neural network algorithm to train the model based on the historical external monitoring screen marked with the external environment conditions; the current external monitoring screen is called to determine the external environment conditions based on the trained convolutional neural network model; based on the historical data, the probability of failure of key components under the specified environmental conditions is evaluated, and the abnormal parameter fluctuation characteristics of the key components are matched with the environmental abnormality characteristics; by comparing the similarity between the environmental abnormality data and the abnormal data of the key components, the possible association between the two is identified; if the similarity between the abnormal fluctuation characteristics of the key components and the external environment abnormality characteristics is greater than the preset similarity threshold, it is determined that the environmental abnormality induces or aggravates the failure of the key components; The identification and evaluation of the cause of the fault through the association analysis of environmental factors and fault modes and virtual simulation verification includes: for a newly appeared faulty component, determining whether the component type already exists in the component root fault knowledge base; if not, constructing the root fault mode knowledge of this type of component; after completing the root fault mode determination of the faulty component, calling the root fault mode knowledge graph, automatically matching the environmental influence factor network corresponding to the fault mode, and querying the relevant environmental factor nodes; checking whether the matched environmental factors have abnormal fluctuations or over-limit phenomena during the time period when the fault occurs, if abnormal fluctuations or over-limit phenomena occur, determining that the environmental abnormality causes or aggravates the fault; if there is no relevant environmental impact match in the root fault mode knowledge graph, matching the association between the corrosion mode corresponding to the component material or structure type and the environmental factor, and determining whether the environmental corrosion causes the fault; for faults with uncertain judgment results, configuring simulation parameters according to the actual operating characteristics and environmental interaction characteristics of the component, using the virtual simulation software ANSYS to build a virtual simulation model, simulating the working state of the component under specified environmental conditions, and verifying the influence of environmental factors, the simulation parameters include load conditions, thermal conductivity and corrosion rate; The spatial correlation evaluation of the combined structural unit determines the potential nodes and associated components for fault propagation, including: obtaining information on all key components in the ship system, including the physical location, functional description and connection status of the key components with other components, and constructing a fully connected topological network, in which each node represents a component and the edge represents the connection relationship between the components; constructing a fault impact network based on the identified faulty components and the physical and functional or data connection relationship between the faulty components and other components in the system, in which the nodes in the fault impact network represent the components and the edges represent the interactions between the components caused by the fault propagation; combining discrete sensor sampling signals, predicting the fault hotspot propagation path, and locating high fault probability hidden danger components, wherein the high fault probability hidden danger components include key components with locked faults; Components and high failure probability associated components; obtain structural components whose physical location distance is less than the preset distance threshold, and determine whether the structural unit belongs to the fault-affected network; the physical location information of each structural component, the connection type between components and historical fault data, the nearest neighbor algorithm for model training, build a component space association model, determine the spatial correlation of each structural component, and use the spatial correlation as a correction coefficient to correct the node's failure probability and improve the model matching accuracy. The physical location information includes coordinates and distances, and the connection types include direct connections and indirect connections; through the application of fully connected topological networks, Bayesian network models and component space association models, the failure probability of nodes and components is calculated in multiple dimensions; if the failure probability is greater than the preset failure probability threshold, it is determined to be a structural component prone to failure and is included in the scope of linkage monitoring; The method of evaluating physical connection and load transfer and identifying potential cascading failure risk points includes: setting a similarity threshold, screening out faulty components whose structural or functional similarity is higher than a preset similarity threshold as similar components; judging whether there is a physical connection between the screened similar components and the fault-related components, the physical connection includes fixed connection and load transfer; if there is a physical connection, judging that the similar components may also fail; obtaining the workload and real-time status data of the similar components, using a support vector machine algorithm to perform model training, predicting the failure probability of the components, and setting an early warning level for each component according to the failure probability; using a cross-correlation function to evaluate the time relationship between fault events, and using a Moran's I index to evaluate the spatial relationship between components; correcting the failure probability of each component according to the calculated time and space correlation; marking all components whose structural and functional similarity is higher than a preset similarity threshold and whose correlation with known faulty components is higher than a preset correlation threshold as cascading failure risk points; when a key component fails, immediately checking the status of all marked risk points to prevent the expansion of the cascading consequences of the failure.

2. The method according to claim 1, wherein: The method of locating the module related to the fault according to the code matching of the ship system fault signal comprises: When a fault signal appears in the ship system, the corresponding system module is matched according to the fault code contained in the fault signal, and the matched module is used as the fault-related module; If multiple modules are matched, the network hierarchy analysis method is used to calculate the correlation between the matched modules based on the communication frequency, functional dependency and historical fault data between the modules, and the modules with correlation higher than the preset first correlation threshold are retained as the main fault-related modules.

3. The method according to claim 1, wherein: The Bayesian network algorithm is used to calculate the probability of failure of all components in the area, and the key components most likely to fail are screened out according to the calculation results, including: Obtain all component data in the identified fault-related area and extract the feature vector of each component, including operation history, maintenance record and physical properties; construct a Bayesian network model based on the collected component data and historical fault data, and train the model to learn and predict the probability of each component having a specified type of fault; predict the probability of component failure based on the latest component status data in the area using the constructed Bayesian network model; if the difference between the failure probabilities of multiple components is less than the preset difference threshold, sort them from high to low according to the degree of correlation between the components.

4. The method according to claim 1, wherein: The monitoring of real-time operation data based on the located key component containing the fault includes: Obtain a list of located key components, and determine whether all the located key components are connected to the real-time monitoring system; for components that are not connected to the real-time monitoring system, search for substitute components whose functional or structural characteristics are similar to those of the components that are not connected to the monitoring system and whose similarity is greater than a preset similarity threshold, and obtain real-time operation data of the substitute components; Through data acquisition equipment, real-time operating data of all key components can be obtained, including current, voltage, temperature and flow.

5. The method according to claim 1, wherein: The method of constructing a component root fault knowledge base and a root fault mode knowledge graph based on external environment data and historical fault reports during ship operation includes: Obtain various external environmental data during the operation of the ship, including air temperature, humidity, salinity, sea conditions and magnetic field; extract component types, fault parameters and root failure modes in the reports by searching historical fault reports, and build a component root failure knowledge base, which includes the failure modes and related fault information of various components; according to the root failure mode and external environmental data, determine the potential relationship between the two through the Apriori algorithm, build a root failure mode knowledge graph, and display the influence network and association structure between environmental factors and root failure modes.

6. The method according to claim 1, wherein: The method of obtaining the internal connection structure of the key components based on the three-dimensional model data of the key components of the ship, calculating the fault correlation between different structural nodes, calibrating the high fault probability associated components, and combining the spatial correlation evaluation of the structural units to determine the potential nodes and associated components for fault propagation includes: Obtain the 3D model data of the key components with locked faults, parse the connection structure network inside the components, and identify the connection nodes, including bolts, cables, and pipe interfaces; count the number of connections for each node, and determine the core nodes whose connection number is greater than the preset connection number threshold; define the distance measurement between nodes in the connection structure network, and the shorter the distance, the higher the correlation between the two nodes; Using the Dijkstra algorithm, the shortest paths between all pairs of nodes in the network are calculated; by comparing the calculated shortest path lengths, node pairs whose fault correlation is greater than a preset fault correlation threshold are identified; Based on the Bayesian network model, the conditional probability between the structural nodes and the failure modes is determined using historical failure statistics. Calculate the posterior probability of the corresponding fault of the node pair whose fault correlation is greater than the preset fault correlation threshold, and determine the core node whose fault probability is greater than the preset fault probability threshold; obtain the front and rear stage structural components connected to the core node to form an impact propagation network of the key component fault; According to the network connection and fault correlation, the structural components whose correlation with the core node is greater than the preset correlation threshold are judged and marked as high fault probability associated components; it also includes: according to the analysis of the degree of dependence between components, operation history score and the influence of environmental factors, the fault propagation impact index is calculated to optimize the maintenance and repair plan of the ship system; The method calculates the fault propagation impact index based on the analysis of the dependency between components, the operation history score and the influence of environmental factors, and optimizes the maintenance and overhaul plan of the ship system, specifically including: using a dependency matrix to represent the direct and indirect connections between components, quantitatively analyzing the dependency between each component, and determining the functional dependency F; obtaining the use frequency, maintenance record and fault history of each component, and determining the operation history score H; determining the environmental factor influence E according to the influence of environmental parameters on each component; defining the fault propagation impact index formula FPI=w1×F+w2×H+w3×E, where w1, w2 and w3 are weight coefficients used to adjust the influence of these three factors in FPI, w1, w2 and w3 The value of is obtained through historical data; the FPI value of each component is calculated, and the components are prioritized according to their FPI values ​​to optimize maintenance and overhaul plans; a real-time monitoring system is established to continuously obtain parameters that affect FPI; the FPI value is recalculated regularly to reflect the latest operating and environmental conditions; the accuracy of the FPI model is verified through actual cases and historical failure data; the parameters and weights in the FPI formula are adjusted according to actual operating data and failure occurrence.

7. The method according to claim 1, wherein: The method of performing feature extraction and similarity calculation on other mechanical components in the scanning area based on the three-dimensional scanning data of the key components and the spatial grid analysis includes: Acquire three-dimensional scanning data of the physical area where the determined high failure probability associated components are located; construct a local range spatial grid with each high failure probability associated component as the center, and set the area size to a preset multiple of the average size of the high failure probability associated component; extract features of other mechanical components in the scanning area, the features include structural features and functional features, the structural features include material, component size and interface form, and the functional features include motion form and applied load; based on the extracted features, use the cosine similarity method to calculate the similarity between other components in the area and the high failure probability associated components in terms of structural features and functional features.

8. The method according to claim 1, wherein: The method of drawing a probability interval distribution diagram of the fault spreading outward from the root component and calculating the confidence that other faults may be hidden in different diffusion range intervals includes: Based on the identified fault root component, a system connectivity network including the component is constructed, wherein the system connectivity network represents the connection relationship and interaction between the components; the Monte Carlo simulation method is used to calculate the probability path of the fault spreading outward from the root component; the range interval of the fault spreading is set, including the periphery of the core component, the local subsystem and the overall system; based on historical statistical data, the Bayesian network algorithm is used for model training to predict the potential fault probability distribution in each diffusion interval; a probability distribution diagram of the fault spreading outward from the core component is drawn to show the decreasing trend of the fault probability in different intervals; the key components or subsystems corresponding to each probability interval are marked on the probability distribution diagram as the key inspection objects; for intervals with a confidence level greater than a preset confidence threshold, the bypass sensor of the auxiliary detection equipment is activated, and the bypass sensor signal is used to determine whether there is a fault that is not directly identified in the interval; The types and number of faults detected within the statistical interval; if the number of detected faults exceeds the preset number range or the severity exceeds the preset severity range, the fault diffusion speed or impact range is judged to be abnormal, and the detection range is expanded; based on the detection results and confidence level, the number distribution of hidden faults in each probability interval is determined, the risk level of each interval is evaluated, and the maintenance plan is optimized; it also includes: based on the structural analysis of the system connection network, the component interaction strength evaluation and the environmental parameter impact evaluation, the fault diffusion complexity index is calculated and integrated, and the failure risk of the ship system is monitored and updated; The method calculates and integrates the fault diffusion complexity index based on the structural analysis of the system connection network, the component interaction intensity assessment and the environmental parameter impact evaluation, and monitors and updates the failure risk of the ship system, specifically including: determining the connectivity distribution, average path length and network cluster coefficient of the network nodes through the topological structure of the system connection network, and quantifying the complexity N of the network connection; evaluating the interaction intensity I between components based on the energy conversion or information exchange frequency between components; obtaining environmental parameter data, including temperature changes and humidity fluctuations, evaluating the degree of influence of environmental parameter data on the system failure mode, and determining the influence P of environmental parameters on fault propagation; defining the formula FDCI=a×N+b×I+c×P, and calculating the fault diffusion complexity index FDCI, wherein a, b, and c are weight coefficients determined according to historical fault data; integrating the FDCI formula into the monitoring and maintenance platform of the ship system, using real-time data to calculate the FDCI, and continuously monitoring the fault diffusion risk of the system; regularly evaluating the performance of the formula, and updating the parameters according to the latest operating data and environmental changes.

9. The method according to claim 1, wherein: According to the fault root component data of the ship system and the working environment data and task process data of the fault-related system, a component knowledge base and a system operating environment library are constructed to identify the key internal and external factors that affect the occurrence of the fault, including: Obtain the data of the identified fault root component and build a component knowledge base containing material, structure and load attributes; obtain the working environment data and task process data of the fault-related system and build a system operation environment library; use the Apriori algorithm to associate the root component attributes with the variables in the environment library, calculate the support and confidence of each association rule, and screen out the internal and external factors that are most likely to affect the root component; use the Bayesian network algorithm to train the model based on the identified internal and external influencing factor variables, build a component failure prediction model, and determine the root component failure mode; for new fault cases, obtain the case-related component data and environmental data, use the component failure prediction model to obtain the conditional probability of each variable in the component data and environmental data being related to the fault; automatically determine the internal and external root source variables that have the greatest impact on this fault based on the probability obtained by the component failure prediction model; If the probability difference of multiple variables is less than the preset probability difference threshold, the core root cause is determined based on the interaction relationship between the variables; the virtual simulation software ANSYS is used to perform numerical simulation and structural simulation to verify the influence of the internal and external root causes output by the component failure prediction model; Simulate abnormal changes in root variables to confirm whether they will lead to the occurrence of this failure mode; based on all internal and external root variables, and sort them according to the degree of their relevant impact on this failure, mark a preset number of the most critical root variables to determine the most critical internal or external root causes of the system.

Images