Dynamic fmea analysis method and system across structure hierarchy

Abstract

The present application relates to the technical field of computer aided engineering and reliability analysis, and discloses a method and system for supporting dynamic FMEA analysis across structure levels, wherein the method performs bidirectional graph traversal to render a failure network view by extracting a focus node; in response to a cross-layer connection instruction between nodes, three-level legality verification is performed based on directed graph connectivity and physical level parameters to prevent loop, sub-level and parent level misplacement, and the relationship is solidified; cross-layer extreme value pooling calculation of severity parameters is performed along a reverse sequence to output action priority; a local connected subgraph is extracted based on focus shift to reconstruct an activity memory area; and the system is constructed from bottom to top according to processing depth into data layer, network construction layer, interaction layer and application layer modules to execute the above method. The present application guarantees the accuracy of cross-layer modeling logic, realizes objective derivation of risk priority, and reduces the memory overhead of large-scale views.

Description

Technical Field

[0001] This invention relates to the field of computer-aided engineering and reliability analysis technology, specifically to a dynamic FMEA analysis method and system that supports cross-structural levels. Background Technology

[0002] Failure Mode and Effects Analysis (FMEA) is an important technical means in the field of systems engineering to assess product reliability and identify potential risks. Existing FMEA auxiliary analysis tools usually use tabular or strict direct parent-child tree structure to organize system component and functional data. However, when facing highly integrated complex electromechanical equipment, the interaction between internal components often spans multiple physical structure and logical function levels. The failure relationship is no longer a unidirectional linear structure, but presents a complex network topology.

[0003] When dealing with such complex cross-structure hierarchy relationships, traditional techniques often rely on human experience to establish upper and lower level connections between failure modes. They lack automated verification of the underlying data topology. When establishing complex cross-layer links, this approach is prone to logical dead loops in the underlying related data due to human judgment errors, or to the demotion and inversion of causes and effects at the physical structure level, thereby compromising the accuracy of the underlying data model of the entire system.

[0004] Because it is difficult to guarantee absolute continuity and compliance of the underlying topology links, existing tools cannot automatically transfer parameters based on network paths when assessing risk indicators. In particular, the assessment of severity parameters is usually limited to subjective judgment and manual input within a single structural level. When the top-level system experiences a serious failure, such risk parameters cannot be automatically transmitted back to the root cause node at the bottom layer along cross-layer links and extreme value aggregation can occur. This makes it easy for the priority rating of actions for underlying design changes to deviate from the objective reality of the global system perspective.

[0005] Furthermore, in the front-end graphical interaction and underlying computing processing of large-scale complex systems, the existing network topology rendering mechanism usually loads all node data and relation edges directly into the device's active memory. This coarse global data scheduling method not only consumes a large amount of computing device memory resources when dealing with cross-layer failure networks, but also causes serious computing delays in the graphical interface when users perform view focus switching or drag-and-drop editing, making it difficult to meet the dynamic interaction needs under massive engineering data. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a dynamic FMEA analysis method and system that supports cross-structural levels. It solves the problems of existing FMEA-assisted analysis technologies when dealing with cross-level failure associations in complex systems, such as the lack of automated topology verification leading to easily disordered underlying data logic, the inability to objectively derive risk assessment parameters along network links, and the excessive memory resource consumption of computing devices caused by globally loading large-scale node primitives.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a dynamic FMEA analysis method supporting cross-structural levels. The method is configured to execute in a computing device, characterized in that the computing device includes an underlying physical storage medium, a processor, and a graphics interaction component. The processor establishes data access with the underlying physical storage medium through a memory environment and provides a graphics computing and interactive processing environment through the graphics interaction component. The method includes the following steps:

[0009] S1, lock the focus node of the current system view, and extract the corresponding failed node data from the underlying physical storage medium;

[0010] S2, using the focal node as a reference, perform a bidirectional graph traversal operation to generate multi-level topological entities, calculate coordinate parameters, and render a failed network view;

[0011] S3, extract the failure node data corresponding to the heterogeneous level across structural levels as the target node, and render the floating free node primitive in the failure network view;

[0012] S4. Respond to the drag-and-drop connection command from the target node to the focus node to construct cross-layer candidate edges. Based on the extracted absolute physical level parameters of the target node and the focus node, execute the three-level legality verification mechanism, which includes anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules. After compliance, generate invalid relationship edge data.

[0013] S5, based on the global directed graph model composed of the multi-level topological entities and the failure relationship edge data, reverse traversal is performed to execute the cross-layer unidirectional transmission and extreme value pooling calculation logic of severity parameters, and the actual aggregated severity parameters of the root failure cause nodes at the physical bottom layer are output.

[0014] S6, extract the occurrence frequency parameter and detectability parameter associated with the root cause node, combine the actual aggregate severity parameter, the occurrence frequency parameter and the detectability parameter to perform risk order number product operation, and map and output action priority rating result;

[0015] S7 captures the focus shift command, triggering viewport geometric center drift and dynamic memory reconstruction, and extracts local connected subgraph data to reconstruct the active memory region using the underlying data entity node associated with the target active node primitive as the origin.

[0016] By adopting the above technical solution, through the synchronous processing of graphical interaction and underlying physical data, cross-layer associations are established in response to drag-and-drop interaction, and a three-level legality verification mechanism is introduced to block connections that violate logical rules. Furthermore, based on the global directed graph model, cross-level reverse propagation of execution parameters and extreme value pooling operations are performed to realize dynamic aggregation of the severity of underlying root cause nodes. In addition, memory dynamic reconstruction is triggered according to the focus transfer instruction, and active memory regions are managed according to the local connected subgraph, reducing the consumption of hardware resources for large-scale network view rendering.

[0017] Preferably, before extracting the corresponding failed node data, the method further includes:

[0018] Establish a structure-independent unified data abstraction model for failure nodes;

[0019] For failed node data to be persistently written, the underlying mapping validity check function is executed: if the first foreign key pointing to the product structure tree entity or the second foreign key pointing to the system function in the node to be written belongs to the set of valid global identifiers, then the write is allowed; if both foreign keys cannot be addressed, the write is rejected to block the orphaned node.

[0020] By adopting the above technical solution, and by performing bidirectional foreign key detection, write operations on isolated nodes that lack basic associations are intercepted, ensuring the structural integrity and traceability of data within the underlying physical storage medium.

[0021] Furthermore, the logic for generating multi-level topological entities by performing bidirectional graph traversal operations includes:

[0022] Based on the breadth-first search algorithm, heterogeneous node identifiers are extracted layer by layer from the failure cause side and the failure impact side, respectively, along the deterministic association direction recorded in the failure relationship edge data.

[0023] During the expansion process, potential related edges and deterministic failure relationship edges generated dynamically by the training module based on the preset topology model are integrated to participate in the traversal, generating a global directed graph model containing explicit and implicit paths.

[0024] By adopting the above technical solution, the range of topological entity generation is expanded by using breadth-first search and fusion traversal of potential related edges; and by combining explicit and implicit paths, a directed graph basic model is provided for the subsequent calculation of risk parameters.

[0025] Preferably, the logic for calculating coordinate parameters and rendering the failed network view in the interactive interface includes:

[0026] A two-dimensional Cartesian coordinate system is established using a hierarchical layout algorithm based on tree branching tiling. The absolute horizontal coordinate of a node primitive on the rendering canvas is obtained by multiplying the logical level depth of the node primitive by the standard horizontal spacing constant between levels. The absolute vertical coordinate of a node primitive on the rendering canvas is obtained by subtracting half the total number of nodes in the same level from the vertical index number of the node primitive in the same level node sequence, and then multiplying it by the standard vertical spacing constant between nodes in the same level.

[0027] By adopting the above technical solution, the primitives are automatically laid out using horizontal and vertical coordinate calculation rules, avoiding overlap or occlusion of primitives in the rendering canvas, generating a hierarchical network view, and improving the readability of the interactive interface.

[0028] Furthermore, the logic for extracting the failed node data corresponding to the heterogeneous levels across structural levels as the target node includes:

[0029] By bypassing the hierarchical access restrictions of the direct parent-child physical hierarchy, a global index scan is performed on the full database based on the feature text carried by the structure retrieval command to filter out the set of nodes that meet the fuzzy matching conditions of the user input characters;

[0030] By using the difference operation, historical nodes with established directed connections and hierarchical nodes that are determined to be illegal are removed from the set of nodes that meet the fuzzy matching conditions of the user input characters, and a clean target candidate set is output.

[0031] By adopting the above technical solution, using global index scanning combined with difference operation filtering, unrelated legal target nodes are extracted, historical related data and illegal data are excluded, and the invalid rendering of detached nodes in the graphical interface is reduced.

[0032] Preferably, the three-tiered validity verification mechanism, which includes anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules, comprises:

[0033] Anti-loop rule: Detect whether there is a directed connected path from the target node to the focus node in the global directed graph model. If it exists, trigger an interception.

[0034] Anti-sub-level misalignment rule: For the connection in the direction of failure, if the structural level depth parameter of the target node is greater than the sum of the structural level depth parameter of the focus node and the preset cross-level tolerance constant, it is judged as a downgrade inversion and interception is triggered.

[0035] Anti-parent misalignment rule: For the connection in the direction of the failure cause, if the structural hierarchy depth parameter of the target node is less than the difference between the structural hierarchy depth parameter of the focus node and the preset cross-level tolerance constant, it is judged as a cross-level misalignment and interception is triggered.

[0036] By adopting the above technical solution, and by executing interception commands with rules that include prevention of loops, prevention of child-level misalignment, and prevention of parent-level misalignment, the system prevents the generation of dead loop paths or the inversion of logical levels when connecting across layers, thus maintaining the unidirectional flow of directed graph relational data at the bottom layer of the system.

[0037] Furthermore, the cross-layer unidirectional transfer and extreme value pooling calculation logic for the severity parameter includes:

[0038] Extract the inherent initial severity scalar of the currently traversed node, as well as the propagated severity scalars of all directly affected nodes of the currently traversed node.

[0039] The inherent initial severity scalar and the transmitted severity scalar are placed in the same computation pool for numerical comparison, and the maximum value is selected to update the actual aggregate severity parameter of the currently traversed node.

[0040] By adopting the above technical solution, and through extreme value pooling by comparing the initial parameters with the subsequent passed parameters and taking the maximum value, the updated aggregate severity parameter covers all upper-layer risk impacts caused by this node.

[0041] Preferably, the logic for performing the risk sequence number product operation and mapping the output action priority rating result includes:

[0042] The actual aggregate severity parameter, the occurrence frequency parameter, and the detectability parameter of the root cause node are multiplied by a scalar to output the risk order number;

[0043] The discrete numerical combinations of the actual aggregate severity parameter, the occurrence frequency parameter, and the detectivity parameter are extracted and matched in a preset action priority multidimensional decision matrix to map and output action priority rating results corresponding to high, medium, and low processing urgency.

[0044] By adopting the above technical solution, based on the quantification of risk sequence number, address matching is performed through a multi-dimensional decision matrix to output targeted action priority rating results, providing standardized risk management decision reference.

[0045] Furthermore, the logic for triggering viewport geometric center drift and dynamic memory reconstruction to extract locally connected subgraph data by the capture focus shift command includes:

[0046] Capture the focus shift command, extract the initial absolute physical coordinates and target absolute physical coordinates of the source focus node primitive and the target active node primitive in the screen coordinate system, construct the viewport geometric center drift vector, and use the nonlinear easing function to perform interpolation calculation to generate a smooth viewport transition animation.

[0047] Using the underlying data entity node associated with the target active node primitive as the addressing root node, candidate associated nodes with a shortest undirected edge hop count less than or equal to the allowed logical hop count limit constant are extracted to form a locally connected subgraph data and loaded into the active memory region. Simultaneously, historical redundant node data outside the active memory range are cleaned up.

[0048] By adopting the above technical solution, smooth viewport transitions are achieved using drift vectors and interpolation calculations. The size of the local connected subgraph is limited by the number of undirected edge jumps. The memory data region is dynamically reconstructed, and processor memory resources are concentrated on the local network associated with the currently active node, thus avoiding memory overflow caused by global view loading.

[0049] Secondly, the present invention also provides a dynamic FMEA analysis system supporting cross-structural levels. The system is deployed in a computing device, which includes an underlying physical storage medium, a processor, and a graphical user interface. The system is constructed from bottom to top according to processing depth into four logical levels that map to the hardware environment of the computing device, specifically including:

[0050] The data layer module, configured on the underlying physical storage medium, is used to persistently store uniformly defined failure node data and failure relationship edge data, and to perform a global index scan of heterogeneous data.

[0051] The network construction layer module, configured in the processor's memory environment, is used to call the failed relationship edge data to perform bidirectional graph traversal operations to generate multi-level topological entities, and to perform cross-layer unidirectional transfer calculation of severity parameters based on extreme value pooling calculation logic.

[0052] The interaction layer module is configured in the graphics computing and interaction processing environment of the graphics interaction component, and is used to provide a graphical user interface with an independent free editing area, capture drag and drop connection instructions and calculate and render a failed network view.

[0053] The application layer module, configured at the top of the overall system control, is used to perform legality interception and verification, including anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules, before cross-layer connection writing. It also performs risk order number product operation and action priority multi-dimensional decision matrix mapping by comprehensively aggregating actual severity parameters, occurrence frequency parameters, and detection parameters.

[0054] By adopting the above technical solution, the system is constructed into four logically mapped levels, reducing the coupling between functional modules and enabling data access, view rendering, and rule verification tasks in the cross-level FMEA analysis process to be processed independently in their respective hardware environments.

[0055] This invention provides a dynamic FMEA analysis method and system that supports cross-structural levels. It offers the following advantages:

[0056] 1. This invention introduces a three-level legality verification mechanism, including anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules, during the cross-layer candidate edge construction stage. Based on the directed graph connectivity parameters and absolute physical level parameters, the drag-and-drop connection command is pre-verified. This feature can intercept illogical failure relationship writing requests, solve the problem of dead loops in failure chains and inverted hierarchical structures caused by human cross-layer operation errors during the FMEA modeling of complex systems, and ensure the accuracy of the underlying network topology data.

[0057] 2. This invention proposes a method for cross-layer unidirectional transfer and extreme value pooling calculation of severity parameters along the reverse hierarchical traversal sequence. Based on a global directed graph model containing failure relationship edge data, this method compares and updates the aggregated severity parameters returned by lower-level influencing nodes with the maximum value, automatically deriving the actual severity of the root cause node. This calculation mechanism based on deterministic topology paths replaces the traditional purely manual subjective rating method, making the risk parameter product calculation and the final output action priority rating result more objective and traceable.

[0058] 3. This invention adopts a mechanism based on reconstructing the active memory region using locally connected subgraphs in front-end graphical interaction and underlying resource scheduling. After monitoring the focus activation operation, the system smoothly switches the view by drifting the viewport center and only extracts data within a specified range of undirected edge hops with the target active node as the origin and loads it into memory. At the same time, redundant historical data outside the range is cleaned up. This data processing logic of on-demand extraction and dynamic release reduces the device memory overhead when rendering large-scale complex failed network views and ensures the stability of the interactive interface response. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0060] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0061] Figure 3 This is a schematic diagram of the unified data model and focus node locking principle of the present invention;

[0062] Figure 4 This is a schematic diagram illustrating the principle of bidirectional graph traversal and multi-level failure network topology generation of the present invention;

[0063] Figure 5 This is a schematic diagram illustrating the heterogeneous hierarchical feature extraction and target node free loading principle of the present invention;

[0064] Figure 6 This is a schematic diagram illustrating the principle of cross-layer failure association establishment and three-dimensional legality interception calculation in this invention;

[0065] Figure 7 This is a schematic diagram illustrating the principle of cross-layer unidirectional transfer and extreme value pooling of severity parameters in this invention.

[0066] Figure 8 This is a schematic diagram illustrating the principle of risk parameter product operation and action priority qualitative mapping in this invention;

[0067] Figure 9 This is a schematic diagram illustrating the interaction principle of viewport geometric center drift and dynamic memory reconstruction in this invention;

[0068] Figure 10 This is a comparison chart of high-level indicators of the present invention;

[0069] Figure 11 This is a comparison chart of percentages and low-level indicators of the present invention.

[0070] The module consists of: 10. Data layer module; 20. Network construction layer module; 30. Interaction layer module; 301. Dynamic network visualization unit; 302. Node center switching unit; 303. Failure association editing unit; 40. Application layer module; 401. Failure chain rule verification unit; 50. Underlying physical storage medium; 60. Processor; and 70. Graphical interaction component. Detailed Implementation

[0071] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings.

[0072] See attached document Figure 1 , Figure 1 This is a schematic diagram of a system architecture according to an embodiment of the present invention. In order to realize the dynamic FMEA analysis method supporting cross-structural levels provided by the present invention, the present invention provides a computing device hardware environment to support the operation of the method. The computing device may include: a bottom-level physical storage medium 50, a processor 60, and a graphics interaction component 70. Based on the above hardware environment, the present invention constructs a system logical architecture, which may include, from bottom to top, the following processing depth: a data layer module 10, a network construction layer module 20, an interaction layer module 30, and an application layer module 40.

[0073] The data layer module 10 is configured in the underlying physical storage medium 50 and is used to persistently store various basic entity data and associated mapping data that constitute the failure network. The data layer module 10 has a structured data area and a relational data area. The structured data area is configured with product structure tree data, functional data and failure node data. The failure node data is configured with a unified data structure model, including node unique identifier attributes and severity score attributes, and forms underlying data dependencies with product structure tree data and functional data through foreign key mapping, thereby performing homogeneous storage for physical hardware nodes and software logical nodes. The relational data area is configured with failure relationship edge data. Each failure relationship edge data is configured with source node ID, target node ID and business logic description fields, which is used to bypass the physical hierarchy restrictions of product structure tree data and directly record the unidirectional failure propagation path across levels.

[0074] The network construction layer module 20 is configured in the memory environment of the processor 60. It is used to receive the underlying read instructions and perform multi-level graph topology entity logic assembly. The network construction layer module 20 may include a failure association engine, a failure network graph generator, and an S-value transfer calculator. The failure association engine is used to read the business logic description fields inside the failure relationship edge data and establish data connections and addressing paths between heterogeneous physical layer nodes. The failure network graph generator is used to synchronously extract the failure node data and its connected failure relationship edge data, and complete the mesh topology construction operation based on the graph theory traversal algorithm to generate memory-level entities. The S-value transfer calculator is used to read the node evaluation parameters under the multi-dimensional intersection path in the memory-level entity, and perform severity inheritance calculation and cross-layer extreme value determination on the affected nodes according to the directed path of data transmission.

[0075] The interaction layer module 30 is configured in the graphics computing and interaction processing environment of the graphics interaction component 70. It is used to provide operation interface and receive visual rendering instructions returned from the underlying layer. The interaction layer module 30 may include a dynamic network visualization unit 301, a node center switching unit 302, and a failure association editing unit 303. The dynamic network visualization unit 301 is used to receive the mesh topology data generated by the network construction layer module 20 and calculate the layout coordinates in the graphical user interface to render and expand the multi-level failure association view.

[0076] The node center switching unit 302 is used to listen for double-click focus actions input by the operation terminal, extract the unique identifier of the corresponding entity and set it as the geometric center origin of the viewport, thereby triggering the reloading of the associated node data sequence and dynamic reconstruction of the interface. The invalid association editing unit 303 is used to capture drag-and-drop creation actions inside the view, parse the logical hierarchy parameters corresponding to the start coordinates and end coordinates, and construct cross-level association connection request data accordingly.

[0077] Application layer module 40 is configured at the top of the overall system control and is used to perform risk quantification analysis index deduction and engineering entity connection rule verification and control. Application layer module 40 may include failure chain rule verification unit 401, RPN / AP calculation engine and risk analysis report generator. Failure chain rule verification unit 401 is used to perform global relationship detection before cross-level association connection request data is sent to data layer module 10. Based on the physical absolute level parameter comparison results and closed-loop path detection algorithm, it identifies and intercepts illegal connection requests that cause directed graph circular dependency or physical level misalignment.

[0078] The RPN / AP calculation engine receives the severity update value output by the S-value propagation calculator, combines it with independently configured failure occurrence frequency and detectability parameters, performs numerical multiplication of risk priority numbers and comparison mapping of action priority correlation matrices. The risk analysis report generator extracts the indicator values ​​generated by the RPN / AP calculation engine and the topology link trajectory sorted out by the network construction layer module 20, and encapsulates and outputs a quantitative evaluation document format stream that displays complete failure propagation network data.

[0079] See attached document Figure 2 , Figure 2 This is a schematic diagram of a method flow according to an embodiment of the present invention. The present invention provides a dynamic FMEA analysis method that supports cross-structural layers. The method relies on the collaborative operation of functional modules such as data layer module 10, network construction layer module 20, interaction layer module 30, and application layer module 40, and may include the following steps:

[0080] S1, the user selects a failure mode as the focus failure node, the interaction layer module 30 receives the external operation command, extracts the product structure tree data and the corresponding failure node data from the data layer module 10, and the node center switching unit 302 extracts the globally unique identifier of the node of the selected failure mode and configures it as the focus origin of the system view.

[0081] S2, the failure network of the focus node is automatically expanded. The network construction layer module 20 calls the failure association engine with the focus node as the starting point and performs bidirectional traversal along the business logic of the data record along the failure relationship edge. The failure network graph generator gathers the upstream and downstream node relationships to generate a multi-level topology entity. The dynamic network visualization unit 301 receives the topology entity and calculates the coordinate parameters, and outputs the horizontally expanded failure network view on the interactive interface.

[0082] S3, the user selects the failure mode that needs to be connected across levels as the target node. The interaction layer module 30 responds to the structure retrieval command, reads the product structure tree data of the level that is not the current focus node, extracts the failure node data corresponding to the heterogeneous level as the target node, and loads the target node into the free editing area of ​​the interaction interface.

[0083] S4, establish the failure association between the focus node and the target node. The failure association editing unit 303 parses the displacement coordinates and connection direction of the target node dragged to the focus node, and generates cross-level association request parameters. The failure chain rule verification unit 401 imports the absolute level attribute parameters and performs graph theory deadlock detection and level inversion comparison. After determining that the parameters are compliant, the instruction data layer module 10 generates the failure relationship edge data that records the logic of the source node and the target node.

[0084] S5, S-value transfer calculation: The S-value transfer calculator extracts the data of failed nodes that establish cross-level connections, performs severity inheritance calculation based on the directed path from higher level to lower level and the set attenuation coefficient. When there are multiple connection paths intersecting nodes, the S-value transfer calculator extracts the parameter values ​​of each input path, performs extreme value solving, and outputs the effective severity value of the target node.

[0085] S6, Risk Analysis and Automatic RPN / AP Calculation: The RPN / AP calculation engine obtains the effective severity value, calls the pre-set independent parameters of occurrence frequency and detectability in the data layer module 10, performs system-level parameter product operation to generate risk priority number, sets rules to map and output qualitative assessment indicators according to action priority, and performs data persistence operation.

[0086] S7, Dynamic Network Visualization and Interaction: The node center switching unit 302 monitors the double-click trigger signal of the node entity in the failed network view, extracts the unique identifier of the touched entity and sends it to the network construction layer module 20. The failed network graph generator performs memory reclamation to clean up the historical topology structure, and calls the node relationship traversal instruction again with the new identifier as the data origin and performs dynamic view reconstruction.

[0087] See attached document Figure 3 , Figure 3 This is a schematic diagram of a unified data model and a focused node locking principle according to an embodiment of the present invention. The node center switching unit 302 coordinates the data layer module 10 and the interaction layer module 30 to perform homogenization processing of the underlying heterogeneous data and the establishment of the initial interaction focus of the system. In this embodiment, the homogenization processing of the underlying heterogeneous data and the establishment of the initial interaction focus of the system specifically include the following execution steps.

[0088] In this embodiment, the data layer module 10 establishes a structure-independent unified failure node data abstraction model in the underlying physical storage medium 50 to homogenize the hardware entity components at the physical layer and the software function code at the logical layer. In order to realize the intelligent structural transformation of massive unstructured engineering data, the data layer module 10 has a built-in pre-training engine to build a natural language parsing model to automatically instantiate the above-mentioned structure-independent unified failure node data abstraction model.

[0089] The training dataset for the structure-independent unified failure node data abstraction model comes from historical failure mode and impact analysis documents from within and outside the enterprise, equipment maintenance logs across physical layers, and software defect work order records. In the training preparation phase, the system performs lexicalization and text cleaning on the training dataset to generate feature vectors. In the training phase, the system uses a named entity recognition network based on the Transformer architecture to supervise the learning of the feature vectors. The backpropagation algorithm is used to continuously adjust the network weights to minimize the cross-entropy loss function between predicted classification and manually labeled classification, and finally outputs a natural language parsing model with natural language parsing capabilities.

[0090] In the specific parsing and mapping process, the natural language parsing model uses entity slot extraction technology to map the physical component codes identified in the text to the first foreign key pointing to the product structure tree entity, to map the software function identifier to the second foreign key pointing to the system function, and to convert the parsed risk level quantifiers into initial severity score attributes.

[0091] Based on the trained and mapped natural language parsing model, the system fills the parsed structured attributes into the data abstraction model. Thus, the data layer module 10 generates isomorphic failure node data. The failure node data includes a globally unique identifier for the node, a name text description field, a node classification attribute, and an initial severity score attribute. The initial severity score attribute is usually set to an integer range of one to ten according to the standard failure mode and impact analysis specifications to quantify the severity of the failure consequences. At the same time, the failure node data is stripped of the mandatory hierarchical relationship attached to the product structure tree in FMEA analysis at the database level. It is defined uniformly using an independent data table and associated with the structured data area by configuring external constraint keys. This allows the failure node data to reserve a first foreign key pointing to the product structure tree data and a second foreign key pointing to the functional data.

[0092] The construction of the underlying foreign key constraints and cross-table join query engine for relational databases is carried out in a conventional manner based on the existing standard Structured Query Language interface.

[0093] As a preferred implementation, in cross-level heterogeneous networking scenarios, it is usually necessary to ensure that the failed nodes written to the underlying layer not only have independent business attributes, but also need to establish a mapping with entities or logic to prevent the generation of free and dirty data. Therefore, the system defines a structure-independent unified failed node data abstract model as a five-tuple model. The data layer module 10 executes the underlying mapping legality detection function for the persistently written failed node data to verify the compliance status of the failed node data, thereby blocking the risk of storing isolated nodes at the source of data writing.

[0094] The aforementioned structure-independent unified failure node data abstraction model is represented by a quintuple formula, which is as follows:

[0095]

[0096] In the formula, A five-tuple data model for failed node data; It is a globally unique identifier for the node; Categorize node attributes; This is the initial severity rating attribute; It is the first foreign key pointing to the product structure tree entity; This is a second foreign key that points to a system function.

[0097] The aforementioned underlying mapping validity check function is expressed as a piecewise logic judgment form. Its general principle is to check whether any node to be written has at least one valid upper-level attribution mapping through Boolean algebra operations. If any valid foreign key reference exists, the judgment is passed; if all foreign keys cannot be addressed in the existing global identifier set, the judgment is rejected. The specific formula is as follows:

[0098]

[0099] In the formula, This is a function for validating a specific node instance. For a specific node instance; The first foreign key in a specific node instance that points to the product structure tree entity; A second foreign key containing a pointer to a system function within a specific node instance; This is the set of global identifiers for the valid physical product structure tree data within the current system. A set of global identifiers for valid logical function data.

[0100] Based on the design logic of the underlying mapping validity detection function, the recorded heterogeneous failure node entities are associated with at least one of the underlying physical hardware set or the upper-level logical function set. This enables electronic component nodes and software function abnormal nodes to be stored as equivalent entities in a unified dimensional space based on the five-tuple model, thus laying the technical foundation for subsequent cross-structure hierarchical heterogeneous addressing.

[0101] Furthermore, the interaction layer module 30 sends a loading command to the underlying data interface to extract product structure tree data and render it as a multi-level nested product structure tree view in the graphical user interface. Subsequently, the interaction layer module 30 continuously listens for touch and selection events inside the front-end product structure tree view to capture the target structure node identifier parameters carried by the operation command, so that the data layer module 10 receives the target structure node identifier parameters and routes them to the query analyzer.

[0102] Based on the established first or second foreign key index relationship, precise filtering and matching are performed. The data layer module 10 selects all failure node data belonging to the target structure node corresponding to the target structure node identifier parameter from the full data table corresponding to the unified failure node data abstract model, and assembles all failure node data into a serialized data stream and returns it to the interaction layer module 30 to generate a failure mode candidate operation list in the corresponding area of ​​the graphical user interface. For the asynchronous on-demand loading logic of the tree data structure in the multi-level structure view and the expansion and collapse rendering mechanism of the front-end component node, the existing document object model operation interface and asynchronous data retrieval framework can be configured in a conventional manner.

[0103] With the successful rendering of the failure mode candidate operation list, the node center switching unit 302 receives the operation instruction, selects a single service record in the failure mode candidate operation list, and extracts the globally unique identifier of the node corresponding to the single service record. Subsequently, the node center switching unit 302 loads the globally unique identifier of the node into the system's viewport status register and configures and locks the globally unique identifier of the node as the focus node of the current system view. At the same time, the node center switching unit 302 synchronously extracts the complete attribute parameter set of the focus node and serializes the complete attribute parameter set of the focus node into a memory-level entity object to reside in the high-speed cache of the processor 60. Finally, the memory-level entity object, as the logical root node of the graph data structure, opens read access to the network construction layer module 20, allowing subsequent related modules to initiate multi-level graph traversal operations across physical layers with the memory-level entity object as the spatial origin.

[0104] See attached document Figure 4 , Figure 4 This is a schematic diagram of the principle of bidirectional graph traversal and multi-level failure network topology generation according to an embodiment of the present invention. The network construction layer module 20, together with the data layer module 10 and the interaction layer module 30, performs multi-level graph topology entity logic assembly and graphics rendering output actions. In this embodiment, the multi-level graph topology entity logic assembly and graphics rendering output actions specifically include the following execution steps.

[0105] In this embodiment, the failure association engine inside the network construction layer module 20 extracts the globally unique identifiers of the focused nodes residing in the cache of the processor 60 and uses the globally unique identifiers of the nodes as the addressing reference points. In order to fully trace the failure propagation link and prevent memory overflow caused by the dead loop of the underlying relationship, the failure association engine performs a bidirectional graph traversal operation with the maximum level depth constraint based on the breadth-first search algorithm mechanism. While extracting heterogeneous node identifiers layer by layer along the deterministic association direction recorded in the failure relationship edge data to the failure cause side and the failure influence side respectively, it integrates the potential association edges dynamically predicted by the following topology model training module. The two participate in the traversal to construct a hybrid association dataset containing explicit and implicit paths, and finally generate a global directed graph model.

[0106] To improve the accuracy of the global directed graph model in extracting complex failure propagation paths, the network construction layer module 20 incorporates a graph neural network-based topology model training module to construct a graph neural network link prediction model and performs dynamic training and updates on it. This topology model training module uses real cascading alarm sequence stream data generated in the system's history and cross-domain maintenance traceability work order graphs as training datasets; it uses the confirmed real fault propagation frequency and its propagation path between nodes as positive sample labels, and uses logical paths that have not experienced real cascading as negative sample labels.

[0107] Furthermore, the initial severity score attribute and classification attribute in the aforementioned unified failure node data abstract model are extracted and one-hot encoded, and then concatenated with the text semantic vector output by the natural language parsing model. This is used as the initial input feature vector of the node to input into the network. The distribution of the node's neighbor features is aggregated through the graph convolutional network algorithm, and the gradient descent optimizer is used to minimize the binary cross-entropy path prediction loss function. The weight parameters of each associated edge in the set of implicit directed relation edges generated by the prediction are dynamically adjusted, thereby completing the deep learning training of the graph neural network link prediction model. This enables the graph neural network link prediction model to have the ability to dynamically and adaptively traverse dead zones based on historical data.

[0108] In actual use, the failure association engine reads the aforementioned weight parameters and introduces a weight threshold filtering mechanism during the queue expansion phase of the breadth-first search. This mechanism prunes (removes) low-probability failure propagation edges with weights lower than the system's preset confidence level, thereby improving the extraction accuracy of core links while controlling memory consumption. It should be noted that this pruning mechanism only applies to recommended association edges predicted by the graph neural network link prediction model. For deterministic failure relationship edges that have been confirmed and persistently stored by engineers in the database, the failure association engine forcibly retains their connectivity to ensure lossless source tracing of subsequent severity parameters under extremely low probabilities.

[0109] In conjunction with the aforementioned pruning mechanism, the layer-by-layer expansion logic of bidirectional graph traversal operations can be expressed as a piecewise set derivation. The specific formula for the layer-by-layer expansion logic of bidirectional graph traversal operations is as follows:

[0110]

[0111] In the formula, To trace back to the fundamental cause in the left direction Layer cause node set; This is the cause node that has been traversed so far; This is the cause node at the next higher level; For the first Layer cause node set; It is the set of directed relation edges in a global directed graph model; This represents the current search level depth.

[0112]

[0113] In the formula, To trace back to the top level and the direction of ultimate influence Layers affect the set of nodes; The currently traversed nodes that are affected; It is an influential node at the next higher level; For the first Layers affect the set of nodes; It is the set of directed relation edges in a global directed graph model; This represents the current search level depth.

[0114] Furthermore, the failure network graph generator receives the set of failure cause-side nodes and failure influence-side nodes, along with the corresponding failure relationship edge data, output by the failure association engine. Based on the graph theory adjacency list structure, it aggregates upstream and downstream node relationships in memory space and generates multi-level topological entities, transforming discrete data records into a network data structure with connected states. Simultaneously, the failure network graph generator performs isolated node removal logic on the multi-level topological entities to filter out invalid data objects that fail to form complete links due to missing underlying foreign key data during traversal. To prevent accidental deletion of core origin nodes, the system skips the isolated node removal logic only when the focus node itself does not have any input or output relationship edges, retaining the focus node as a single entity to provide an initial visual mounting point in the interactive interface, thereby ensuring the data continuity of subsequent graphics rendering.

[0115] Based on this, the dynamic network visualization unit 301 receives multi-level topology entities and calls the layout algorithm engine to calculate and allocate the front-end view coordinate parameters. As a preferred implementation, in order to make the failure propagation logic have an intuitive causal relationship at the visual level, the dynamic network visualization unit 301 adopts a hierarchical layout algorithm based on tree branch tiling to establish a two-dimensional Cartesian coordinate system and anchors the focus node at the origin of the two-dimensional Cartesian coordinate system. The horizontal dimension spacing parameter and the vertical dimension spacing parameter are calculated based on the front-end preset primitive bounding box size and the safety margin distance to ensure that no primitive physical overlap occurs in the network at any scaling level.

[0116] The calculation and allocation logic of the front-end view coordinate parameters is expressed as a coordinate mapping form through formulas. The general principle of the calculation and allocation logic of the front-end view coordinate parameters is to determine the horizontal axis coordinate by the network topology depth of the nodes to reflect the transmission order, and to determine the vertical axis coordinate by the arrangement index number distribution of nodes at the same level to avoid overlapping and occlusion of front-end visual primitives. The specific formula for the calculation and allocation logic of the front-end view coordinate parameters is as follows:

[0117]

[0118] In the formula, The absolute horizontal coordinate of the currently processed node object on the rendering canvas; The currently processed node object; This represents the logical hierarchy depth of the currently processed node object. This is the standard horizontal spacing constant between levels preset by the system.

[0119]

[0120] In the formula, The absolute vertical coordinate of the currently processed node object on the rendering canvas; The currently processed node object; The vertical index number of the currently processed node object in the same level node sequence; This represents the total number of nodes at the level of the currently processed node object; This is the standard longitudinal spacing constant between nodes on the same layer that is preset for the system.

[0121] Finally, the dynamic network visualization unit 301 drives the underlying graphics application programming interface based on the set of coordinate parameters calculated and output. In the interface canvas component of the interaction layer module 30, it draws the rectangular primitive objects corresponding to each node and renders the Bézier curve path connecting the rectangular primitive objects based on the directed edge entity data. Thus, a horizontally unfolded failure network view containing multi-level causal transmission logic is stably output in the interactive interface. The underlying Bézier curve path smoothing drawing algorithm of the canvas component and the view anti-shaking control mechanism bound to the primitive drag response event can be implemented based on existing vector graphics rendering engine technology.

[0122] See attached document Figure 5 , Figure 5 This is a schematic diagram of the heterogeneous hierarchical feature extraction and target node free loading principle according to an embodiment of the present invention. The interaction layer module 30, in conjunction with the data layer module 10, performs cross-level target data retrieval and interactive interface preloading actions that break through the traditional hierarchical tree constraints. In this embodiment, the cross-level target data retrieval and interactive interface preloading actions that break through the traditional hierarchical tree constraints specifically include the following execution steps.

[0123] In this embodiment, the interaction layer module 30 responds to the structure retrieval command sent by the front end and reads the product structure tree data that is not directly at the current focus node level. In order to break the rigid constraint that traditional failure mode and impact analysis can only be mounted layer by layer along the physical structure tree, the query engine inside the data layer module 10 bypasses the hierarchical access restriction of the direct parent-child physical level and directly performs a global index scan in the full database based on the feature text carried by the structure retrieval command. In this way, it extracts the failure node data that matches the retrieval conditions across heterogeneous structure levels and uses it as the target node to be connected. In order to ensure retrieval efficiency in the scenario of concurrent querying of massive data and avoid the return of a large number of useless results and memory overflow caused by global unlimited search, the data layer module 10 generates a dynamic query filter view by calculating the absolute level parameters of the focus node before performing the retrieval. The system top-level nodes or cross-domain bottom-level nodes that are determined by engineering logic to be generally not feasible for business association are pre-isolated to reduce the retrieval space and avoid the performance dead zone of the search algorithm.

[0124] The filtering and extraction logic for heterogeneous hierarchical target nodes can be expressed as a set intersection and difference operation. The general principle of this logic lies in eliminating nodes already existing in the current network topology and hierarchical nodes deemed invalid according to engineering rules through global matching results, thereby outputting a clean set of target candidates. The specific formula for the filtering and extraction logic of heterogeneous hierarchical target nodes is as follows:

[0125]

[0126] In the formula, This is the final set of valid candidate nodes output to the front-end interactive interface; This refers to the current complete set of failed node entities in the system database. A set of nodes that satisfy the fuzzy matching conditions of the user's input characters; This is the set of historically associated nodes that have established directed topological connections with the current focus node; This is the set of nodes that are deemed illegal based on cross-domain association rules.

[0127] After obtaining a clean set of target candidates, the data layer module 10 serializes each entity object in the heterogeneous hierarchical target node set into a standardized data message containing node identifiers and initial attribute parameters, and pushes the standardized data message to the interaction layer module 30 through the internal communication bus. This allows the interaction layer module 30 to parse the heterogeneous hierarchical target node set and extract the core view fields used for front-end rendering. In this process, to prevent communication timeouts and instantaneous peaks in front-end memory caused by the centralized transmission of large-volume messages, the system adopts a cursor-based data block transmission mechanism to paginate the heterogeneous hierarchical target node set. The threshold for the size of the paginated data block is dynamically and adaptively allocated based on the available memory capacity of the front-end device and the network bandwidth conditions to ensure that the front-end interactive components maintain a high frame rate response of the graphical interface when receiving continuous data streams, thereby avoiding main thread rendering blockage.

[0128] As the paginated data of the detached nodes arrives, the interaction layer module 30 opens an independent detached editing area on the surface of the current network topology view. Then, the dynamic network visualization unit 301 performs floating loading rendering of the parsed heterogeneous hierarchical target nodes in the detached editing area to generate detached node primitives independent of the currently failed network. As a preferred implementation, the detached node primitives are visually configured with a semi-transparent overlay outline highlight style that is different from the regular network entity primitives to represent that the detached node primitives are in an independent state where no failure association has been established. At the same time, the detached editing area serves as a floating panel component that independently responds to the drift of the positioning coordinates in order to receive cross-level network connection interaction actions initiated by the user for the detached node primitives. The independent layer overlay rendering control and floating panel drag-and-drop event listening mechanism of the detached editing area can be implemented based on the existing front-end view state management and cascading style sheet absolute positioning technology.

[0129] See attached document Figure 6 , Figure 6 This is a schematic diagram of the cross-layer failure association establishment and three-dimensional legality interception calculation principle according to an embodiment of the present invention. The network construction layer module 20, in collaboration with the interaction layer module 30 and the data layer module 10, executes a free connection mechanism that ignores the physical layer and a three-dimensional low-level interception action to prevent engineering logic errors. In this embodiment, the cross-layer failure association establishment and three-dimensional legality interception calculation action specifically includes the following execution steps.

[0130] In this embodiment, the interactive editing module inside the interaction layer module 30 responds to the drag-and-drop connection command triggered by the user in the free editing area and parses the planar orientation features of the connection landing point relative to the focus node. It breaks through the traditional hierarchical limitation and directly assigns heterogeneous hierarchical target nodes as the failure causes or failure effects of the current focus node based on business logic definition, thereby constructing cross-layer candidate edges. In this process, the system extracts the source node identifier and target node identifier based on the directed graph data structure to generate candidate relation object serialization entities carrying relation direction enumeration values, thereby providing a basic data inference carrier for the subsequent graph algorithm interception mechanism based on engineering business specifications. The above-mentioned cross-layer candidate edge construction mechanism completely decouples the strong binding relationship between the physical product bill of materials and the failure logic tree, giving flexibility to cross-domain analysis of complex system engineering.

[0131] To prevent the system from crashing into infinite derivation in a multi-level interconnected network, along with the generation of cross-layer candidate edges, the failure chain rule verification unit 401 inside the application layer module 40 extracts graph theory path features to execute the anti-loop rule of the three-level legality verification mechanism. The failure chain rule verification unit 401 calls the directed graph depth-first search algorithm or breadth-first search algorithm to extract the globally connected subgraph from the target node to the source node to perform reachability calculation, thereby intercepting illegal connection requests at the source that attempt to reverse the downstream failure consequences back to the upstream, causing a closed-loop logic deadlock, and triggering operation rollback.

[0132] The underlying interception logic of the three-level legality verification mechanism's anti-loop rule is expressed as a segmented judgment form through a formula. The general principle of the underlying interception logic of the anti-loop rule lies in determining whether adding a new edge will form a directed loop by detecting whether there is a reverse connected path in the network topology. The specific formula for the underlying interception logic of the anti-loop rule is as follows:

[0133]

[0134] In the formula, This is a circular dependency detection function; This is the source and focus node of the current network structure; For target candidate nodes; This represents the directed connected path from the target candidate node to the source focus node. This represents the complete set of directed connected paths in the current global directed graph model.

[0135] Under the premise that the anti-loop rule verification is passed, in order to constrain the disordered connection of excessive cross-levels and maintain the rationality of the engineering analysis, the failure chain rule verification unit 401 initiates the anti-sub-level misalignment rule for cross-level candidate edges defined as failure effects. The failure chain rule verification unit 401 extracts the absolute physical level parameters of the source node and the target node in the product physical structure tree and performs numerical comparison calculations to intercept the engineering level inversion instruction that is defined as a failure effect despite the absolute physical level of the target node being extremely lower than that of the focus node. It should be noted that in the structural level depth parameter system of the present invention, the root node (top level of the system) of the physical structure tree has the smallest depth parameter value. As the physical level is subdivided down to the bottom component, the depth parameter value of this level increases.

[0136] The interception calculation logic of the three-level legality verification mechanism for preventing child-level misalignment rules is expressed as a logical composite condition through a formula. The general principle of the interception calculation logic for preventing child-level misalignment rules lies in combining business direction constraints and hierarchical depth tolerance to determine the legality of upward propagation of influence. The specific formula for the interception calculation logic of preventing child-level misalignment rules is as follows:

[0137]

[0138] In the formula, This is an interception function that affects the directional structure degradation. The current focus node; For target candidate nodes; This refers to the direction attribute of the connection between the focus node and the target candidate node. These are logical enumeration values ​​representing the direction of the failure's impact; The structural hierarchy depth parameter of the target candidate node; The structural hierarchy depth parameter for the focus node; This is a preset cross-level tolerance constant for the system.

[0139] Following the interception and judgment of the anti-child misalignment rule, the failure chain rule verification unit 401 uses equivalent mirror logic to execute the anti-parent misalignment rule for the cross-layer candidate edge defined as the direct cause. The failure chain rule verification unit 401 blocks the reverse misalignment operation that is defined as the direct cause but is defined as a direct cause by the target node whose absolute physical structure level is much higher than the focus node (i.e., the structural level depth parameter value of the target node is extremely smaller than that of the focus node) through reverse hierarchical verification comparison. When all three interception calculations pass, the data layer module 10 officially solidifies the cross-layer candidate relationship edge as a legal failure relationship edge data into the underlying storage medium and drives the front end to trigger the local topology redrawing mechanism. The dynamic response logic of the front end primitive connection drag state and mouse pointer snapping effect can be performed based on the existing hypertext markup language canvas event listening technology.

[0140] The interception calculation logic of the three-level legality verification mechanism's anti-parent misalignment rule is expressed as a logical compound condition through a formula. The general principle of the anti-parent misalignment rule interception calculation logic lies in combining the characteristics of business cause tracing and downward propagation tolerance to prevent high-level system anomalies from being attributed to the failure of minor components at the lower level. The specific formula for the anti-parent misalignment rule interception calculation logic is as follows:

[0141]

[0142] In the formula, The function is a bypass interception function for the cause-direction structure; For target candidate nodes; The current focus node; The direction attribute of the connection between the target candidate node and the focus node; These are logical enumeration values ​​representing the direction of the failure cause; The structural hierarchy depth parameter of the target candidate node; The structural hierarchy depth parameter for the focus node; This is a preset cross-level tolerance constant for the system.

[0143] See attached document Figure 7 , Figure 7 This is a schematic diagram of the principle of cross-layer unidirectional transmission and extreme value pooling of severity parameters according to an embodiment of the present invention. The network construction layer module 20, in collaboration with the data layer module 10, executes a system-level dynamic severity evaluation mechanism that breaks the traditional isolated assignment mode. In this embodiment, the cross-layer unidirectional transmission and extreme value pooling of severity parameters specifically includes the following execution steps.

[0144] In this embodiment, with the successful establishment of cross-layer failure association edges, the S-value transfer calculator inside the network construction layer module 20 is triggered and extracts the global directed graph model in the current failure network view. In order to support the reverse propagation of severity parameters along the topology link from the final failure impact node at the top layer of the system to the root failure cause node at the bottom layer of the physical layer, the S-value transfer calculator executes the topology sorting algorithm to generate a reverse hierarchical traversal sequence based on the global directed graph model generated by the failure network graph generator in the previous step and presented as a directed acyclic structure after passing the anti-loop check. During this process, the S-value transfer calculator identifies all nodes with zero out-degree and uses all nodes with zero out-degree as the source reference point for severity transfer. Then, based on the reverse hierarchical traversal sequence, a top-down cross-layer unidirectional data flow channel is constructed. The cross-layer unidirectional data flow channel breaks through the limitation of relying solely on manual experience for local static assignment in traditional failure mode and impact analysis, and realizes the automatic inference and quantitative tracing of the hazard of the bottom layer failure in complex cross-domain networks.

[0145] Based on the computational order formed by the completed cross-layer unidirectional data flow channel, the S-value propagation calculator performs extreme value pooling calculation logic for each intermediate node in the reverse hierarchical traversal sequence. The S-value propagation calculator extracts the inherent initial severity scalar of the intermediate node itself and the propagated severity scalar returned by all direct downstream adjacent nodes of the intermediate node along the reverse hierarchical traversal sequence. The inherent initial severity scalar of the intermediate node itself and all the propagated severity scalars are put into the same computational pool for numerical comparison, and the maximum value in the computational pool is selected as the actual aggregated severity parameter of the intermediate node under the current network topology. When the intermediate node has no direct downstream adjacent nodes, the actual aggregated severity parameter of the intermediate node is equal to the inherent initial severity scalar of the intermediate node itself. In this way, any minor failure at the bottom layer, as long as it has a propagation path that can cause catastrophic consequences at the top layer, will have its actual aggregated severity parameter increased to the same business alert level as the catastrophic consequences at the top layer according to business rules, thereby reducing the dead zone of misjudgment of severity degradation caused by the limited line of sight of manual assessment.

[0146] The extreme value pooling calculation logic for severity parameters is expressed as a recursive search for the maximum value. The general principle of this logic is to ensure that the worst-case scenario at the system level can be directly mapped to the underlying cause node without loss of intermediate layers. The specific formula for the extreme value pooling calculation logic of severity parameters is as follows:

[0147]

[0148] In the formula, This is the actual aggregation severity parameter for the currently traversed node; The current network node being traversed; This is the initial severity score attribute for the currently traversed node; This is the set of all directly successor nodes that affect the currently traversed node. For nodes that directly affect subsequent nodes; This is the actual aggregation severity parameter that directly affects subsequent nodes.

[0149] Once the actual aggregate severity parameters of all network nodes have been calculated, as a preferred approach, the data layer module 10 receives the update message output by the S-value delivery calculator and overwrites and persistently stores the actual aggregate severity parameters in the data structure of the corresponding failed node data. Simultaneously, the dynamic network visualization unit 301 inside the interaction layer module 30 responds to the numerical change event of the data layer module 10, triggering the view redrawing and rendering action of the corresponding node primitives. The dynamic network visualization unit 301 maps the preset warning color gradient value according to the numerical range of the actual aggregate severity parameters and updates the background fill attribute of the corresponding node primitives, thereby providing engineers with an intuitive and real-time linked severity heat map distribution view on the interactive interface. For the color interpolation calculation and layer incremental update processing mechanism in the front-end graphics rendering, it can be done based on the existing two-dimensional graphics drawing interface encapsulation technology.

[0150] See attached document Figure 8 , Figure 8 This is a schematic diagram of the principle of risk parameter product operation and action priority qualitative mapping according to an embodiment of the present invention. The network construction layer module 20, in collaboration with the data layer module 10, performs the conversion calculation action from single-dimensional hazard assessment to multi-dimensional comprehensive risk quantification. In this embodiment, the risk parameter product operation and action priority qualitative mapping action specifically includes the following execution steps.

[0151] In this embodiment, as the cross-layer update of the actual aggregated severity parameters of all network nodes is completed, the RPN / AP calculation engine inside the application layer module 40 initiates a full risk dimension data retrieval for each root cause node at the physical bottom layer in the current failed network topology (i.e., a node identified as having an in-degree of zero in the global directed graph model). Since the five-tuple of the bottom-layer node only carries the basic structure and severity attributes, the RPN / AP calculation engine uses the globally unique identifier of the node as the primary key for the joint query to extract the occurrence frequency parameters and detectivity parameters associated with the corresponding root cause node from the risk parameter extended data table independently configured in the underlying storage medium.

[0152] To prevent engineers from omitting input during the early analysis phase, which could disrupt the risk assessment process or result in false low-risk calculations, the RPN / AP calculation engine incorporates a null value safety check mechanism. When the frequency or detectability parameter is detected to be unassigned, the RPN / AP calculation engine will forcibly assign the unassigned frequency or detectability parameter to the maximum penalty threshold set by the system with a value of ten (according to the FMEA assessment standard, a value of ten represents the worst business scenario with extremely high occurrence frequency or absolute undetectability). This constructs a complete three-dimensional risk assessment data structure and avoids the risk omission dead zone caused by missing data in the data layer module 10.

[0153] Based on the extracted and verified occurrence frequency parameters and detectability parameters, the RPN / AP calculation engine performs a risk order number product operation in conjunction with the actual aggregate severity parameters corresponding to the root cause node. The RPN / AP calculation engine inputs the actual aggregate severity parameters, occurrence frequency parameters, and detectability parameters into its internal risk product operation logic for scalar multiplication, thereby outputting a risk order number to measure the overall risk scale. As a basic quantitative reference indicator, the risk order number can provide the engineering team with a preliminary basis for risk ranking. Given that different combinations of risk parameters may produce the same risk order number but correspond to completely different actual physical risk scenarios, the system introduces the risk order number as an auxiliary reference dimension into the subsequent decision matrix according to business standards.

[0154] The calculation logic of the risk parameter product operation is expressed as a series of products. The general principle of the risk parameter product operation is to combine the severity of the failure consequences, the probability of failure, and the effectiveness of existing control measures to output a single risk metric value. The specific formula for the calculation logic of the risk parameter product operation is as follows:

[0155]

[0156] In the formula, This is the risk sequence number calculated for the current root cause of failure node; This is the instance of the node that is currently being traversed and is the root cause of the failure. The actual aggregate severity parameter inherited by the node instance that is the root cause of the failure; This is the frequency parameter for the occurrence of the node instance that is the root cause of the failure. This refers to the probeness parameter of the node instance that is the root cause of the failure.

[0157] Following the generation of risk sequence numbers, the RPN / AP calculation engine performs coordinate addressing and matching in a pre-set multi-dimensional decision matrix for action priorities based on discrete numerical combinations of actual aggregate severity parameters, occurrence frequency parameters, and detectability parameters. The multi-dimensional decision matrix for action priorities divides the three-dimensional parameter space into non-linear mapping regions corresponding to high, medium, and low processing urgency levels. The RPN / AP calculation engine outputs action priority rating results that reflect the actual intervention needs through a lookup table positioning mechanism. This alleviates the resource misallocation problem caused by relying solely on risk sequence numbers for sorting, and helps ensure that fatal failures with high actual aggregate severity parameters but low occurrence frequency parameters can be prioritized by the system and marked as a red alert state that requires priority corrective action.

[0158] The lookup logic for the qualitative mapping of action priorities is expressed as a multivariate piecewise mapping. The general principle of this lookup logic lies in using industry-standard nonlinear weight allocation rules to map the three-dimensional input to a defined discrete set of action levels. The specific formula for the lookup logic of the qualitative mapping of action priorities is as follows:

[0159]

[0160] In the formula, This is the action priority rating result output based on the node mapping of the current root cause of failure; This is the instance of the node that is currently being traversed and is the root cause of the failure. A multidimensional decision matrix function for action priorities that is persistently stored within the system; The actual aggregate severity parameter inherited by the node instance that is the root cause of the failure; This is the frequency parameter for the occurrence of the node instance that is the root cause of the failure. This refers to the probeness parameter of the node instance that is the root cause of the failure.

[0161] As the calculation and mapping of risk sequence number and action priority rating results are completed, as a preferred approach, the data layer module 10 receives the result dataset issued by the RPN / AP calculation engine and triggers a batch update operation of the relational database to solidify the risk assessment results. At the same time, the dynamic network visualization unit 301 inside the interaction layer module 30 responds to the data structure state flip event of the data layer module 10. The dynamic network visualization unit 301 applies visual enhancement rendering to the view primitives corresponding to the root cause node based on the action priority rating results. The interactive interface assigns a high-saturation red outline style to the node primitives with high priority action priority rating results to guide users to directly focus on the weakest and most resource-intensive underlying links in the current engineering design. For the polymorphic style switching mechanism of specific primitives in the front-end interface, it can be done based on the existing cascading style sheet attribute binding technology.

[0162] See attached document Figure 9 , Figure 9 This is a schematic diagram of the interaction principle of viewport geometric center drift and memory dynamic reconstruction according to an embodiment of the present invention. The interaction layer module 30, in conjunction with the data layer module 10, performs on-demand loading and smooth viewpoint switching actions to cope with complex network topologies. In this embodiment, the interaction action of viewport geometric center drift and memory dynamic reconstruction specifically includes the following execution steps.

[0163] In this embodiment, as the engineer initiates a focus activation operation on a non-focused node primitive in a detached state in the interactive interface, the dynamic network visualization unit 301 inside the interaction layer module 30 captures the focus transfer instruction and extracts the initial absolute physical coordinates of the source focused node primitive in the current screen coordinate system and the target absolute physical coordinates of the target activated node primitive in the screen coordinate system. The dynamic network visualization unit 301 constructs a viewport geometric center drift vector that spans the viewport plane based on the initial absolute physical coordinates and the target absolute physical coordinates. To alleviate the loss of spatial orientation perception caused by the instantaneous change of the viewport, the dynamic network visualization unit 301 discretizes the viewport geometric center drift vector based on a nonlinear easing interpolation algorithm to generate continuous viewport transition animation frame parameters. The viewport geometric center drift and memory dynamic reconstruction interaction mechanism improves the coherence of visual tracking during engineering analysis at the micro level.

[0164] The interpolation calculation logic for viewport geometric center drift is expressed as a time-driven position analysis form. The general principle of the interpolation calculation logic for viewport geometric center drift lies in using a time normalization factor to smoothly calculate the spatial distance between the starting and ending points. The specific formula for the interpolation calculation logic of viewport geometric center drift is as follows:

[0165]

[0166] In the formula, These are the real-time center coordinates of the viewport camera at the current rendering time. The initial absolute physical coordinates of the source focus node primitive in the screen coordinate system; The target's absolute physical coordinates in the screen coordinate system are used to activate the node primitives of the target. A nonlinear easing function preset for the system; This is the absolute time parameter accumulated since the focus transfer instruction was triggered; The total duration for smooth viewport transitions set for the system.

[0167] During the viewport coordinate translation process, to reduce the probability of front-end memory overflow and rendering stuttering caused by loading the infinitely extended system-level failure association graph all at once, the network construction layer module 20 synchronously responds to the focus transfer command and activates the failure network graph generator. The failure network graph generator uses the underlying data entity nodes associated with the target active node as the addressing root node. Based on the graph search algorithm, the failure network graph generator expands outward to extract locally connected subgraph data that meets the preset logical hop count limit. When the target active node has no adjacent connected path in the global directed graph model, the locally connected subgraph data degenerates into a single set of elements containing only the underlying data entity node itself. The failure network graph generator loads the locally connected subgraph data into the active memory area. At the same time, the failure network graph generator identifies and cleans up historical redundant node data that is outside the locally connected subgraph data in the active memory area. The on-demand loading memory reconstruction strategy helps maintain the dynamic balance of client system resource consumption.

[0168] The data extraction logic for locally connected subgraphs in dynamic memory reconstruction is expressed as a set-condition filtering formula. The general principle of this logic lies in extracting topologically related network segments within a fixed step size, centered on the new focal node. The specific formula for this logic is as follows:

[0169]

[0170] In the formula, This is a set of locally connected subgraph data generated for reconstructing the target activated node primitives; These are candidate associated node entities obtained through traversal and search in the global directed graph model; Activate the underlying data entity node associated with the target node element; This represents the shortest undirected edge hop count between the underlying data entity node and the candidate associated node entity in the directed graph structure. This is a preset constant that limits the number of logical jumps allowed by the system.

[0171] As the local connected subgraph data set is reconstructed in the network construction layer module 20, as a preferred approach, the dynamic network visualization unit 301 extracts the updated topology information within the active memory region and performs incremental rendering comparison. The dynamic network visualization unit 301 pushes the primitive instances that newly enter the viewport range into the rendering pipeline and performs destruction and transparency decay animations on the primitive instances that fade out of the logical hop count limit boundary. This creates a dynamic interactive experience at the visual level where the focus shifts and the surrounding related branches naturally emerge. The specific coordinate calculation mapping and overlay context management mechanism for nodes and connections in the graphical interface can be performed based on the existing Hypertext Markup Language canvas interface call technology.

[0172] Application Examples:

[0173] Taking the failure mode and effects analysis of the electronic and electrical system of the automotive intelligent driving domain controller as the implementation scenario, the physical structure of this product includes four levels: system-level intelligent driving domain controller, assembly-level power management assembly, component-level power management board, and component-level filter capacitor.

[0174] In the traditional analysis mode, the failure of components needs to be propagated to the system level by establishing a correlation step by step through the intermediate structure. In this embodiment, the data layer module 10 performs physical absolute level stripping and stores the node data of the four levels in a unified five-tuple data model. During the system initialization phase, the node center switching unit 302 extracts the globally unique identifier of the system-level intelligent driving function loss failure node and loads it into the viewport status register. The network construction layer module 20 generates local topology entities and anchors the local topology entities to the origin of the two-dimensional Cartesian coordinate system in the interaction layer module 30.

[0175] The operator issues a cross-level structure retrieval command, the query analyzer matches the target parameters in the bottom set, extracts the capacitor short-circuit failure node at the fourth physical level and converts it into an independent data stream and sends it to the free editing area of ​​the graphical interface. The failure association editing unit 303 captures the displacement coordinate features of the target node in the component layer to the source node in the system layer and constructs cross-level connection request data.

[0176] At this time, the application layer module 40 calls the failure chain rule verification unit 401 to perform connectivity verification calculation. The failure chain rule verification unit 401 uses breadth-first search to detect whether there is a loop path in the global directed graph and performs anti-loop judgment. Then, it extracts the absolute level parameters of the product structure tree associated with the target node and the source node for numerical comparison to check whether the failure impact transmission direction follows the unidirectional engineering rule of flowing from high numerical level to low numerical level.

[0177] After the parameter comparison is successful, the data layer module 10 generates the corresponding failure relationship edge data. The network construction layer module 20 triggers the S-value transfer calculator, extracts the newly created directed path based on the topology sorting algorithm, and the severity assessment scalar configured for the system-level failure node is propagated backward along the directed edge to the component-level failure node. The S-value transfer calculator reads the parameter values ​​of multiple input paths of the component-level node, performs extreme value solving, and outputs the aggregated severity value. Finally, the RPN / AP calculation engine extracts the independently configured occurrence frequency parameter and detectability parameter, performs system-level parameter multiplication with the aggregated severity value, outputs the risk order number, and performs coordinate positioning mapping in the preset three-dimensional decision matrix to output the action priority parameter of the component node, completing the multi-dimensional risk quantification data extrapolation across physical layers.

[0178] Table 1: Performance Comparison of Traditional Relational Hierarchical Analysis Systems and Dynamic Cross-Level FMEA Systems

[0179]

[0180] According to Table 1 and Figure 10 and Figure 11 As can be seen, the dynamic cross-level FMEA system recorded an average time of 1.03 seconds for constructing cross-four-level associations, which is lower than the 12.47 seconds of the traditional relational hierarchical analysis system. The difference comes from the unified failure node data abstraction model configured in the data layer module 10. The system does not scan the intermediate relationship tables level by level along the physical bill of materials. The failure association engine relies on the second foreign key mapping relationship to cross the physical level limit of the product structure tree and directly call the failure relationship edge data in the relationship data area, reducing the hierarchical depth required for data addressing.

[0181] The time taken for updating and extrapolating the severity parameters of all nodes was recorded as 114 milliseconds, while the control group's value was 4861 milliseconds. The test data reflects the execution efficiency of the S-value cross-layer transfer extreme value pooling calculation logic. The network construction layer module 20 pre-generates topology entity objects in memory. The S-value transfer calculator relies on the directed graph topology sorting sequence and uses memory pointers to directly extract the transfer severity scalar of all successor nodes. The set is then substituted into the calculation formula for recursively finding the maximum value to output the result.

[0182] The peak memory usage for concurrent 5000-node rendering decreased from 1024.8MB to 315.6MB. The recalculation latency for the viewport geometric drift center was recorded as 18 milliseconds. After receiving the operation terminal command, the interaction layer module 30 parses the absolute physical coordinates of the source node and the target node to calculate the viewport geometric center drift vector. The dynamic network visualization unit 301 outputs the transition animation frame parameters based on the interpolation algorithm. At the same time, the network construction layer module 20 expands outward based on the preset hop count limit constant to extract local connected subgraph data, loads it into the cache, and executes the historical object cleanup mechanism.

[0183] Traditional relational hierarchical analysis systems output 0 values ​​for cross-domain logical loop deadlock interception rate and hierarchical misalignment and inverted connection interception rate. This is because traditional systems are form-driven architectures, and the data entry layer does not have graph topology connectivity verification rules and hierarchical polarity judgment logic configured. The process of constructing failed associations is done by manually reading a two-dimensional list and performing checkmarks. The system itself cannot perform machine calculations on the physical hierarchical flow and closed-loop deadlock characteristics during the data submission stage, resulting in the failure of automated interception at the system level.

[0184] The dynamic cross-level FMEA system outputs 100% results for both cross-domain logical loop deadlock interception rate and hierarchical misalignment and inverted connection interception rate. Simultaneously, it outputs 100% accuracy for severity cross-level attenuation transmission values. The failure chain rule verification unit 401, configured in the application layer module 40, enters the workflow during the data request stage of the association connection construction in the interaction layer module 30. The failure chain rule verification unit 401 imports the circular dependency detection function and the structural over-level interception function to calculate the state of the relationship edges to be written. The three-dimensional interception calculation excludes directed circular connected paths and illegal requests that violate the physical depth level difference, ensuring the integrity of the directed acyclic graph data structure in the storage medium. The RPN / AP calculation engine receives the aggregated valid values ​​output by the S-value transmission calculator, calls the preset data set to perform risk quantification multi-dimensional decision matrix coordinate addressing calculations, calculates the single risk sequence number and maps it to the corresponding action priority discrete level, generating quantitative indicators for engineering entity analysis and verification.

[0185] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic FMEA analysis method supporting cross-structural levels, wherein the method is configured to be executed on a computing device, characterized in that, The computing device includes an underlying physical storage medium, a processor, and a graphics interaction component. The processor establishes data access with the underlying physical storage medium through a memory environment and provides a graphics computing and interactive processing environment through the graphics interaction component. The method includes the following steps: S1, lock the focus node of the current system view, and extract the corresponding failed node data from the underlying physical storage medium; S2, using the focal node as a reference, perform a bidirectional graph traversal operation to generate multi-level topological entities, calculate coordinate parameters, and render a failed network view; S3, extract the failure node data corresponding to the heterogeneous level across structural levels as the target node, and render the floating free node primitive in the failure network view; S4. Respond to the drag-and-drop connection command from the target node to the focus node to construct cross-layer candidate edges. Based on the extracted absolute physical level parameters of the target node and the focus node, execute the three-level legality verification mechanism, which includes anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules. After compliance, generate invalid relationship edge data. S5, based on the global directed graph model composed of the multi-level topological entities and the failure relationship edge data, reverse traversal is performed to execute the cross-layer unidirectional transmission and extreme value pooling calculation logic of severity parameters, and the actual aggregated severity parameters of the root failure cause nodes at the physical bottom layer are output. S6, extract the occurrence frequency parameter and detectability parameter associated with the root cause node, combine the actual aggregate severity parameter, the occurrence frequency parameter and the detectability parameter to perform risk order number product operation, and map and output action priority rating result; S7 captures the focus shift command, triggering viewport geometric center drift and dynamic memory reconstruction, and extracts local connected subgraph data to reconstruct the active memory region using the underlying data entity node associated with the target active node primitive as the origin.

2. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, Before extracting the corresponding failed node data, the method further includes: Establish a structure-independent unified data abstraction model for failure nodes; For failed node data to be persistently written, the underlying mapping validity check function is executed: if the first foreign key pointing to the product structure tree entity or the second foreign key pointing to the system function in the node to be written belongs to the set of valid global identifiers, then the write is allowed; if both foreign keys cannot be addressed, the write is rejected to block the orphaned node.

3. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The logic for generating multi-level topological entities by performing bidirectional graph traversal operations includes: Based on the breadth-first search algorithm, heterogeneous node identifiers are extracted layer by layer from the failure cause side and the failure impact side, respectively, along the deterministic association direction recorded in the failure relationship edge data. During the expansion process, potential related edges and deterministic failure relationship edges generated dynamically by the training module based on the preset topology model are integrated to participate in the traversal, generating a global directed graph model containing explicit and implicit paths.

4. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The logic for calculating coordinate parameters and rendering the failed network view in the interactive interface includes: A two-dimensional Cartesian coordinate system is established using a hierarchical layout algorithm based on tree branching tiling. The absolute horizontal coordinate of a node primitive on the rendering canvas is obtained by multiplying the logical level depth of the node primitive by the standard horizontal spacing constant between levels. The absolute vertical coordinate of a node primitive on the rendering canvas is obtained by subtracting half the total number of nodes in the same level from the vertical index number of the node primitive in the same level node sequence, and then multiplying it by the standard vertical spacing constant between nodes in the same level.

5. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The logic for extracting failed node data corresponding to heterogeneous levels across structural levels as target nodes includes: By bypassing the hierarchical access restrictions of the direct parent-child physical hierarchy, a global index scan is performed on the full database based on the feature text carried by the structure retrieval command to filter out the set of nodes that meet the fuzzy matching conditions of the user input characters; By using the difference operation, historical nodes with established directed connections and hierarchical nodes that are determined to be illegal are removed from the set of nodes that meet the fuzzy matching conditions of the user input characters, and a clean target candidate set is output.

6. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The three-tiered validity verification mechanism, which includes anti-loop rules, anti-child-level misalignment rules, and anti-parent-level misalignment rules, includes: Anti-loop rule: Detect whether there is a directed connected path from the target node to the focus node in the global directed graph model. If it exists, trigger an interception. Anti-sub-level misalignment rule: For the connection in the direction of failure, if the structural level depth parameter of the target node is greater than the sum of the structural level depth parameter of the focus node and the preset cross-level tolerance constant, it is judged as a downgrade inversion and interception is triggered. Anti-parent misalignment rule: For the connection in the direction of the failure cause, if the structural hierarchy depth parameter of the target node is less than the difference between the structural hierarchy depth parameter of the focus node and the preset cross-level tolerance constant, it is judged as a cross-level misalignment and interception is triggered.

7. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The cross-layer unidirectional transfer and extreme value pooling calculation logic for the severity parameter includes: Extract the inherent initial severity scalar of the currently traversed node, as well as the propagated severity scalars of all directly affected nodes of the currently traversed node. The inherent initial severity scalar and the transmitted severity scalar are placed in the same computation pool for numerical comparison, and the maximum value is selected to update the actual aggregate severity parameter of the currently traversed node.

8. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The logic for performing the risk sequence number product operation and mapping to output the action priority rating result includes: The actual aggregate severity parameter, the occurrence frequency parameter, and the detectability parameter of the root cause node are multiplied by a scalar to output the risk order number; The discrete numerical combinations of the actual aggregate severity parameter, the occurrence frequency parameter, and the detectivity parameter are extracted and matched in a preset action priority multidimensional decision matrix to map and output action priority rating results corresponding to high, medium, and low processing urgency.

9. The dynamic FMEA analysis method supporting cross-structural levels according to claim 1, characterized in that, The logic for triggering viewport geometric center drift and dynamic memory reconstruction to extract locally connected subgraph data by the capture focus shift command includes: Capture the focus shift command, extract the initial absolute physical coordinates and target absolute physical coordinates of the source focus node primitive and the target active node primitive in the screen coordinate system, construct the viewport geometric center drift vector, and use the nonlinear easing function to perform interpolation calculation to generate a smooth viewport transition animation. Using the underlying data entity node associated with the target active node primitive as the addressing root node, candidate associated nodes with a shortest undirected edge hop count less than or equal to the allowed logical hop count limit constant are extracted to form a locally connected subgraph data and loaded into the active memory region. Simultaneously, historical redundant node data outside the active memory range are cleaned up.

10. A dynamic FMEA analysis system supporting cross-structural levels, characterized in that, The system is applied to execute the dynamic FMEA analysis method supporting cross-structural levels as described in any one of claims 1 to 9. The system is deployed in a computing device, which includes an underlying physical storage medium, a processor, and a graphics interaction component. The system is constructed from bottom to top according to processing depth into four logical levels that map to the hardware environment of the computing device, specifically including: The data layer module, configured on the underlying physical storage medium, is used to persistently store uniformly defined failure node data and failure relationship edge data, and to perform a global index scan of heterogeneous data. The network construction layer module, configured in the processor's memory environment, is used to call the failed relationship edge data to perform bidirectional graph traversal operations to generate multi-level topological entities, and to perform cross-layer unidirectional transfer calculation of severity parameters based on extreme value pooling calculation logic. The interaction layer module is configured in the graphics computing and interaction processing environment of the graphics interaction component, and is used to provide a graphical user interface with an independent free editing area, capture drag and drop connection instructions and calculate and render a failed network view. The application layer module, configured at the top of the overall system control, is used to perform legality interception and verification, including anti-loop rules, anti-child misalignment rules, and anti-parent misalignment rules, before cross-layer connection writing. It also performs risk order number product operation and action priority multi-dimensional decision matrix mapping by comprehensively aggregating actual severity parameters, occurrence frequency parameters, and detection parameters.

Images