Self-service check-in equipment maintenance auxiliary processing method based on equipment anomaly knowledge graph

Abstract

This application discloses a maintenance assistance method for self-service check-in equipment based on an equipment anomaly knowledge graph, belonging to the field of equipment anomaly diagnosis technology. The method includes: constructing an ontology model of the self-service check-in equipment's anomaly knowledge graph and a directed weighted knowledge graph with causal chain tracing and operational context awareness; integrating operation logs, sensor time-series data, and user operation trajectories to map multi-source heterogeneous inputs into dynamic anomaly triples; using the dynamic anomaly triples as inference anchors, conducting multi-hop association inference in the anomaly knowledge graph based on graph neural networks, dynamically recalibrating the graph edge weights in conjunction with real-time equipment state vectors, aggregating candidate fault root causes along high-confidence paths, outputting a root cause priority ranking list, generating a context-aware maintenance assistance scheme, and dynamically arranging standard handling measures. This application deeply integrates the operating mechanism of self-service check-in equipment, user interaction behavior, and intelligence.

Description

Technical Field

[0001] This application relates to the field of equipment anomaly diagnosis technology, and in particular to a self-service check-in equipment maintenance auxiliary processing method based on equipment anomaly knowledge graph. Background Technology

[0002] With the development of intelligent transportation and unmanned services, self-service baggage drop-off equipment (such as self-service check-in baggage drop-off terminals) has been deployed on a large scale in transportation hubs such as airports and high-speed rail stations. These devices integrate multiple subsystems such as mechanical transmission, sensor control, human-computer interaction and network communication. The operating environment is complex, the usage frequency is high, and the user operation is non-standardized, which leads to frequent abnormal events with intertwined causes.

[0003] Currently, the operation and maintenance of self-service equipment mainly relies on two technical approaches: one is a fault code matching system based on rule engines, which pre-sets several fault codes and corresponding processing procedures, and directly pushes fixed maintenance steps when the equipment reports a specific error code; the other is a data mining method based on historical work orders, which generates recommended solutions by statistically analyzing high-frequency fault patterns.

[0004] However, the above methods have significant limitations: rule engines have difficulty covering combined, implicit, or novel anomalies and lack the ability to model complex scenarios such as user misoperation and multi-module coupling failure; while pure data-driven recommendation systems lack causal logic support and cannot explain "why the measure is applicable to the current anomaly", making it difficult for operations and maintenance personnel to judge the reliability of the solution.

[0005] In recent years, some studies have attempted to introduce knowledge graph technology to construct equipment fault knowledge bases, enabling the association query between fault phenomena and causes through entity relationship modeling. For example, existing solutions extract the "component-phenomenon-cause-measure" quadruple from maintenance manuals and historical records to construct equipment fault knowledge graphs, and combine semantic similarity for fault matching. However, such graphs are usually designed for general industrial equipment and do not consider the unique "human-machine-process" interaction characteristics of self-service check-in equipment. They lack the characterization of the causal chain between user operation events and equipment state changes. At the same time, their reasoning process is mostly static query or similarity calculation based on pre-trained embeddings, failing to integrate the real-time operating status of the equipment to dynamically adjust knowledge relationships, resulting in insufficient robustness in the face of instantaneous interference, sensor drift, or partial failure.

[0006] In summary, there is an urgent need for a knowledge-driven maintenance assistance method that can deeply integrate the operating mechanism of self-service check-in equipment, user interaction behavior, and intelligence. Summary of the Invention

[0007] In order to provide a knowledge-driven maintenance assistance method that can deeply integrate the operating mechanism of self-service check-in equipment, user interaction behavior and intelligence, this application provides a maintenance assistance method for self-service check-in equipment based on equipment anomaly knowledge graph.

[0008] Firstly, the objective of this invention is achieved through the following technical solution: A self-service baggage check-in device repair assistance method based on device anomaly knowledge graph includes: An anomaly knowledge graph ontology model for self-service check-in equipment is constructed. The anomaly knowledge graph ontology model uses equipment functional modules, user interaction events, anomaly representations, potential root causes and handling measures as nodes, and embeds domain-specific semantic relationships to construct a directed weighted knowledge graph with causal chain tracing and operational context awareness. By integrating the operation logs of self-service check-in equipment, sensor time-series data, and user operation trajectories, multi-source heterogeneous inputs are mapped into dynamic anomaly triples that adapt to the anomaly knowledge graph ontology model, and timestamps and confidence weights are assigned. Using the dynamic anomaly triples as inference anchors, multi-hop association inference is carried out in the anomaly knowledge graph based on graph neural networks. The graph edge weights are dynamically recalibrated in combination with real-time device state vectors. Candidate fault root causes are aggregated along high-confidence paths, and a root cause priority ranking list with attribution path explanations is output. Based on the root cause priority ranking list, a context-aware maintenance assistance plan is generated. Combined with equipment model, on-site spare parts, maintenance personnel skills and safety compliance constraints, standard handling measures are dynamically arranged and pushed to the maintenance terminal.

[0009] By adopting the above technical solutions, a domain-specific anomaly knowledge graph for self-service check-in equipment is constructed, and multi-source real-time operational data is integrated to drive dynamic reasoning and output maintenance assistance solutions. This application provides a knowledge-driven maintenance assistance method that deeply integrates the operating mechanism of self-service check-in equipment, user interaction behavior, and intelligence. Specifically, an anomaly knowledge graph ontology model and domain-specific semantic relationships of self-service check-in equipment are introduced, enabling the anomaly knowledge graph to depict the causal coupling mechanism between user behavior and equipment status. Combined with multi-hop reasoning of graph neural networks and edge weight recalibration based on real-time status awareness, it effectively distinguishes between apparent anomalies and deep-seated root causes, avoiding misjudgments caused by traditional rule matching or static similarity calculations, while outputting explanatory results with attribution paths. Unlike traditional fixed maintenance solutions, it can dynamically arrange standard handling measures based on on-site constraints such as the current equipment model, available spare parts inventory, maintenance personnel skill level, and safety compliance requirements, generating executable and implementable personalized maintenance guidelines, significantly improving maintenance efficiency and first-time repair rate. To adapt to real-time response efficiency in airport baggage handling scenarios, this application maps operational logs, sensor time-series data, and user operation trajectories in real time into dynamic anomaly events that can be processed by the knowledge graph, assigning timestamps and confidence weights. This enables the system to complete the entire process from anomaly detection to maintenance suggestion delivery within seconds, meeting the stringent availability requirements of high-traffic locations such as airports and high-speed rail stations. Finally, through real-time dynamic updates of the anomaly knowledge graph ontology model, the evolution from a "static knowledge base" to an "intelligent agent" is achieved.

[0010] In a preferred embodiment of this application, the domain-specific semantic relationships include operation triggering, state dependency, spatial adjacency, and disposal constraints; The multi-hop association reasoning includes: starting from the abnormal representation node in the dynamic abnormal triplet, associating it with the user interaction event that caused the abnormality and the corresponding device function module through the operation trigger relationship, extending it to the upstream or downstream associated function modules in the shipping process using the state dependency relationship, and identifying the potential fault propagation unit that is physically close by combining the spatial adjacency relationship; and dynamically scoring the candidate root cause by fusing the real-time device state vector and the historical fault statistical frequency to obtain the association reasoning result.

[0011] By adopting the above technical solutions, domain-specific semantic relationships, including operation triggers, state dependencies, spatial adjacency, and handling constraints, are defined, making the anomaly knowledge graph more closely aligned with the actual operating mechanism and physical structure of self-service check-in equipment, thus enhancing the graph's engineering practicality and reasoning rationality. Multi-hop association reasoning starts with anomaly representations, associating user interaction events with equipment functional modules through operation trigger relationships, enabling rapid location of the anomaly trigger source; expanding upstream and downstream related modules using state dependencies allows for identification of fault propagation paths; and combining spatial adjacency relationships to identify physically adjacent fault propagation units helps discover hidden related faults. Integrating real-time equipment state vectors and historical fault statistical frequencies for dynamic scoring of candidate root causes effectively improves the confidence of root cause judgments and reduces the false positive rate in complex anomaly scenarios. This enables intelligent association analysis from single-point anomalies to the entire link and multiple dimensions, making fault location more comprehensive and efficient.

[0012] In a preferred embodiment of this application, after pushing the maintenance assistance solution to the operation and maintenance terminal, the method further includes: Receive maintenance execution feedback signals from the terminal to determine whether the abnormal state has been eliminated; If the feedback indicates that the abnormal state has been eliminated, the path confidence between the potential root cause of this successful association and the adopted handling measures will be increased, and the edge weights of the corresponding operation triggers, state dependencies and handling constraints will be strengthened. If the abnormal feedback persists, based on the updated device sensor data and user interaction records, the candidate root causes in the original inference path are negatively marked, and local subgraph reconstruction is initiated in the abnormal knowledge graph ontology model. Multi-hop association inference is re-executed to generate supplementary maintenance guidance; the structured log of the entire maintenance process is used as the incremental training sample of the abnormal knowledge graph ontology model.

[0013] By adopting the above technical solutions, when anomalies are eliminated, the confidence of successful paths is increased and the weights of relevant relationships are strengthened, enabling the knowledge graph to be continuously optimized during continuous use, and the inference accuracy to gradually improve. When anomalies are not eliminated, candidate root causes of the original path are negatively marked, and local subgraph reconstruction and re-inference are initiated, which can correct deviations in a timely manner, generate supplementary maintenance guidelines, and avoid ineffective maintenance and repeated failures. The structured logs of the entire process are used as incremental training samples to continuously update the anomaly knowledge graph, enabling the model to self-learn and self-evolve, continuously enriching fault modes and handling experience, and adapting to dynamic operating conditions such as equipment aging, software upgrades, and scene changes.

[0014] In a preferred embodiment of this application, the construction of the anomaly knowledge graph ontology model for self-service check-in equipment includes: Collect historical maintenance work orders, equipment technical manuals, operation and maintenance expert experience, and user fault reports, and perform entity recognition and relationship extraction. The identified entities include equipment modules, sensors, actuators, abnormal phenomena, causes of failure, maintenance actions, and tools and consumables; the relationships include those that cause, belong to, require use, are repairable, and are often associated with. A multi-layer heterogeneous graph structure is constructed based on the entities and relationships, and initial confidence weights are assigned to each node and edge. By using graph neural networks to perform embedding learning on the multi-layer heterogeneous graph structure, vector representations that support semantic similarity matching are generated, forming a reasonable device anomaly knowledge graph; Based on the fault descriptions and replacement parts records in the historical maintenance work orders, as well as the module status jumps, actuator response delays and user interruption behaviors in the real-time operation logs, hidden fault modes or intermittent failure nodes not covered by the current anomaly detection are mined through preset association rules to obtain the anomaly types to be supplemented for diagnosis. The abnormal types to be diagnosed are added to the supplementary reasoning task, and the device abnormality knowledge graph is updated in combination with historical diagnostic results.

[0015] By adopting the above technical solutions, based on entity recognition and relation extraction, a comprehensive knowledge system covering equipment modules, abnormal phenomena, fault causes, and maintenance actions can be constructed by fully utilizing multi-source operation and maintenance knowledge, thereby improving the breadth and professionalism of knowledge coverage. A multi-layered heterogeneous graph structure is built based on entities and relations and assigned initial weights. Combined with graph neural network embedding learning to generate vector representations, quantitative matching of semantic similarity between abnormal nodes can be achieved, improving the efficiency of knowledge retrieval and reasoning. By mining hidden fault patterns and intermittent failure nodes, it is possible to supplement anomaly types that are difficult to cover by traditional detection methods, enhancing the system's diagnostic capabilities for complex and weak signal faults. The anomaly types to be supplemented for diagnosis are incorporated into the reasoning task, and the entire knowledge graph is continuously updated, ensuring that the model always aligns with the actual fault distribution and evolution patterns of the equipment.

[0016] In a preferred example, this application further includes, after constructing the anomaly knowledge graph ontology model for self-service check-in equipment: The abnormal keywords in the device's operating context features are subjected to synonym expansion and standardization processing to generate a standardized abnormal description vector. The standardized anomaly description vector is compared with the embedding vector of each anomaly node in the device anomaly knowledge graph ontology model to calculate the similarity, and candidate anomaly nodes with similarity higher than a preset similarity threshold are selected. By combining fault codes, equipment models, operating stages, and environmental constraints, the candidate abnormal nodes are filtered for context consistency, and abnormal nodes that meet the preset multi-dimensional constraints are retained as matching results.

[0017] By adopting the above technical solutions, standardized anomaly description vectors can eliminate matching errors caused by differences in natural language expressions, improving the alignment accuracy between anomaly information and knowledge graph nodes. Calculating the similarity between the standardized vectors and the embedded vectors of anomaly nodes allows for the rapid selection of highly relevant candidate anomaly nodes, improving matching efficiency. Contextual consistency filtering, incorporating multi-dimensional constraints such as fault codes, equipment models, operating stages, and environmental conditions, can eliminate interfering nodes with mismatched scenarios or inapplicable types, retaining reliable anomaly matching results and reducing the false matching rate.

[0018] In a preferred embodiment of this application, after obtaining the list of device anomalies, the method further includes: Based on the historical operation logs and real-time interactive behavior data of the target shipping equipment, a time sequence graph model of equipment status is constructed using a graph neural network, and behavioral feature vectors of functional module nodes are extracted. The behavioral feature vectors are input into a pre-trained abnormal state detection model to identify suspicious device nodes that deviate from the normal operating baseline and generate a high-risk fault candidate set. For the high-risk fault candidate set, the virtual maintenance simulation engine is invoked to dynamically generate a multi-step maintenance guidance sequence that matches the hardware interface and software protocol of the suspected device node, and an interactive maintenance simulation is performed. Based on the feedback from the maintenance simulation, and combined with the equipment anomaly knowledge graph and expert rule base, a multi-indicator dynamic evaluation is conducted on suspected faults. These multi-indicators include the actual impact, repair timeliness, user operation complexity, spare parts availability, and flight support priority. Based on the evaluation results, a maintenance confidence score is generated, and solutions with maintenance confidence scores higher than a preset threshold are marked as recommended maintenance solutions.

[0019] By adopting the above technical solutions, suspicious equipment nodes deviating from the normal baseline can be identified from a dynamic operational perspective, enabling early warning of faults and risk prediction, and improving the initiative in the operation and maintenance of self-service baggage check-in equipment. A virtual maintenance simulation engine generates multi-step maintenance guidance sequences that match hardware interfaces and software protocols, and executes interactive simulations. This allows for verification of the feasibility of the solution before actual maintenance, reducing on-site trial-and-error costs and operational risks. Combining knowledge graphs and expert rule bases, dynamic evaluation based on multiple indicators such as impact degree, repair timeliness, user complexity, spare parts availability, and flight support priority comprehensively measures the overall value of the maintenance solution, avoiding the one-sidedness of decisions based on a single indicator.

[0020] In a preferred embodiment of this application: the generation of a context-aware maintenance assistance plan based on the root cause priority ranking list, and the dynamic arrangement of standard handling measures in conjunction with equipment model, on-site spare parts, maintenance personnel skills, and safety compliance constraints, specifically includes: The equipment model is used to match the corresponding technical manual and special tool list; the on-site spare parts information includes the spare parts inventory quantity and expiration date; the maintenance personnel skill information includes certification level and historical repair success rate; the safety compliance constraints include civil aviation security inspection regulations and electrical safety operating procedures. Using each functional module of the equipment to be repaired as the basic unit, all related standard handling measures are retrieved from the preset maintenance knowledge base based on the high-confidence root causes in the root cause priority sorting list. For each standard handling measure retrieved, a feasibility score is calculated based on the equipment model, the on-site spare parts, the skills of the maintenance personnel, and the safety compliance constraints. Using the feasibility score and comprehensive index method, the various standard handling measures are ranked and combined to generate the optimal maintenance assistance plan.

[0021] By adopting the above technical solutions, the maintenance plans are not only technically feasible but also meet the multi-dimensional realistic conditions of resources, personnel, and standards. Using equipment functional modules as basic units, and based on high-confidence root cause retrieval and correlation standard handling measures, the solutions are guaranteed to be targeted and professional. Each measure is scored for feasibility, which quantitatively assesses the implementation difficulty, resource matching degree, and safety risks, eliminating steps that are not feasible. By using feasibility scores and a comprehensive index method for ranking and combination, the optimal maintenance assistance plan can be automatically generated, realizing intelligent orchestration and optimization of the maintenance process.

[0022] Secondly, the objective of this invention is achieved through the following technical solution: A self-service baggage check-in equipment repair assistance system based on equipment anomaly knowledge graph, the system comprising: One or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described above.

[0023] Thirdly, the objective of this invention is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described auxiliary processing method for self-service check-in equipment maintenance based on a device anomaly knowledge graph.

[0024] Fourthly, the objective of this invention is achieved through the following technical solution: A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the self-service check-in equipment maintenance assistance method based on a device anomaly knowledge graph as described above.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. Construct an ontology model of anomaly knowledge graph for self-service check-in equipment, with equipment functional modules, user interaction events, anomaly representations, potential root causes and handling measures as nodes, and embedding domain-specific semantic relationships to form a directed weighted knowledge graph with causal chain tracing and operational context awareness. This can systematically integrate the anomaly mechanism and operation and maintenance logic of self-service check-in equipment, and realize the structured expression of anomaly problems from phenomenon to root cause. 2. Real-time fusion of multi-source heterogeneous data such as operation logs, sensor time-series data and user operation trajectories, and mapping them into dynamic anomaly triples adapted to knowledge graphs can improve the standardization and real-time performance of anomaly information. 3. By conducting multi-hop association reasoning through graph neural networks and dynamically recalibrating graph edge weights based on real-time equipment status, candidate fault root causes can be accurately aggregated along high-confidence paths, outputting a priority list with attribution path explanations, significantly improving the accuracy and interpretability of fault location; based on root cause ranking, a context-aware maintenance assistance plan is generated, and combined with equipment model, spare parts, personnel skills, and safety constraints, dynamic arrangement of handling measures can be achieved, which can greatly reduce the reliance on the experience of maintenance personnel and improve maintenance efficiency and standardization. Attached Figure Description

[0026] Figure 1 This is a flowchart of a self-service baggage check-in equipment repair assistance method based on equipment anomaly knowledge graph in one embodiment of this application; Figure 2 This is another flowchart in a self-service baggage check-in equipment repair assistance method based on equipment anomaly knowledge graph in one embodiment of this application; Figure 3 This is a principle block diagram of a self-service check-in equipment maintenance assistance system based on an equipment anomaly knowledge graph, according to one embodiment of this application. Detailed Implementation

[0027] The present application will be further described in detail below with reference to the accompanying drawings.

[0028] In one embodiment, such as Figure 1 As shown, this application discloses a self-service baggage check-in equipment maintenance assistance method based on equipment anomaly knowledge graph, which specifically includes the following steps: S1: Construct an ontology model of anomaly knowledge graph for self-service check-in equipment. The ontology model of anomaly knowledge graph uses equipment functional modules, user interaction events, anomaly representations, potential root causes and handling measures as nodes, and embeds domain-specific semantic relationships to construct a directed weighted knowledge graph with causal chain tracing and operational context awareness.

[0029] In this embodiment, the device functional module refers to a hardware or software unit with independent functions in the self-service check-in device, such as a baggage weighing module, barcode scanner, printer, conveyor belt drive unit, payment terminal, etc.; user interaction events include the operations performed by passengers on the device, such as placing baggage, clicking to print a label, interrupting the process, etc.; abnormal manifestations are observable manifestations of abnormal device operation, such as weighing timeout, barcode scanning failure, screen unresponsiveness, etc.; potential root causes refer to the underlying reasons for the abnormality, such as dirty sensors, motor overheating, network interruption, etc.; and remedial measures are standard maintenance actions for specific root causes, such as cleaning sensors, restarting the control board, replacing the communication module, etc.

[0030] Domain-specific semantic relationships include operation triggers, state dependencies, spatial adjacency, and disposal constraints. Operation trigger relationships characterize how specific user interactions on self-service baggage check-in devices induce abnormal states in device functional modules. State dependencies reflect the temporal or logical dependencies between multiple functional modules in the baggage check-in process; for example, failure in one step will prevent the next step from starting. Spatial adjacency relationships describe the potential abnormal transmission effects between physically adjacent or tightly coupled hardware components in signal paths. Disposal constraints specify the preconditions for implementing a maintenance measure, such as whether specific tools, personnel qualifications, or the equipment being powered off are required. When constructing the knowledge graph, domain-specific semantic relationships are automatically labeled by analyzing causal descriptions in historical work orders.

[0031] Specifically, over 5,000 maintenance records related to self-service baggage check-in equipment from the past three years were extracted from the airport's historical work order database. Each record includes a description of the fault, the confirmed cause of the fault, the maintenance actions taken, and the equipment components involved. Named Entity Recognition (NER) was used to automatically extract the five types of entities mentioned above, and dependency parsing and rule template matching were employed to establish relationships between entities. For example, from the text "Passenger placed luggage but the weighing system did not respond; inspection revealed the weighing sensor was obstructed by a foreign object," the user interaction event "placed luggage," the anomaly "no response to weighing," the equipment functional module "weighing module," the potential root cause "sensor obstructed by a foreign object," and the remedial action "remove the foreign object and calibrate the sensor" were identified, establishing associations such as "placed luggage → no response to weighing" and "no response to weighing → sensor obstructed by a foreign object." Subsequently, all the extracted results are constructed into a directed graph structure, with each node labeled with its type and attributes, and each edge assigned an initial weight (such as normalized based on the frequency of the association in historical data), ultimately forming a directed weighted knowledge graph that supports causal tracing.

[0032] In one embodiment, in step S1, constructing the anomaly knowledge graph ontology model of the self-service check-in device includes: S101: Collect historical maintenance work orders, equipment technical manuals, operation and maintenance expert experience, and user fault reports, and perform entity recognition and relationship extraction.

[0033] In this embodiment, historical maintenance work orders refer to structured or semi-structured fault handling records stored in the airport or equipment manufacturer's operation and maintenance system, including multiple fields such as fault time, equipment number, phenomenon description, root cause conclusion, replaced parts, and maintenance personnel. Equipment technical manuals are PDF or XML documents provided by the manufacturer, covering module composition, interface protocols, fault code tables, and maintenance procedures; operation and maintenance expert experience exists in the form of interview records, internal wikis, or training materials, containing non-standardized but high-value fault diagnosis rules; user fault report text comes from natural language descriptions submitted by passengers through APP, service desk, or equipment interface. Entity recognition refers to automatically extracting noun units with independent semantics from the above multi-source text, while relation extraction identifies the semantic relationships between these entities.

[0034] S102: The identified entities include equipment modules, sensors, actuators, abnormal phenomena, causes of failure, maintenance actions, and tools and consumables. Relationships include causing, belonging to, needing to be used, repairable, and often accompanying.

[0035] In this embodiment, the equipment module refers to an independently maintainable functional unit in the self-service check-in terminal, such as a "baggage weighing platform," a "barcode scanning window," and a "thermal printer." Sensors include weight sensors, infrared photocells, temperature and humidity probes, and other sensing elements. Actuators refer to controlled output devices such as motors, solenoid valves, and indicator lights. Abnormal phenomena are abnormal states observable by the user or system, such as "screen frozen" or "conveyor belt not turning." Fault causes are deep-seated technical problems, such as "motor brush wear" or "communication protocol timeout." Repair actions are standard operating procedures, such as "tightening screws" or "updating firmware." Tools and consumables include "Phillips screwdrivers," "non-woven fabric," and "spare printheads." In the relationships described, "cause" indicates a causal relationship, "belongs to" indicates a relationship of attribution or composition, "requires" indicates the dependence of repair actions on tools, and "repairable" indicates a mapping between measures and causes, often reflecting the statistical co-occurrence of multiple anomalies or causes. All entities are assigned unique URI identifiers, and relationships are defined as RDF predicates, forming a preliminary ontology vocabulary.

[0036] S103: Construct a multi-layered heterogeneous graph structure based on entities and relationships, and assign initial confidence weights to each node and edge.

[0037] In this embodiment, the multi-layer heterogeneous graph structure refers to a complex network in which the knowledge graph contains multiple types of nodes (such as modules, causes, measures) and multiple types of edges (such as causing, belonging to, needing to use). Different layers correspond to different semantic dimensions (such as physical layer, logical layer, operational layer). The initial confidence weight is used to quantify the confidence level of each edge or each triple, and its value is between 0 and 1.

[0038] Specifically, the standardized triples output by S102 are imported into a graph database (such as Neo4j), and five types of node labels (:Module, :Sensor, :Anomaly, :RootCause, :Action) and five types of relation types (:CAUSES, :BELONGS_TO, :REQUIRES_TOOL, :RESOLVES, :CO_OCCURS_WITH) are constructed. For example, nodes (:Anomaly{name: "Weighing Timeout"}) and (:RootCause{name: "AD Converter Drift"}) are created, and connections are established through [:CAUSES{source: "Work Order #5821", freq: 37, weight: 0.84}]. The initial weight calculation formula is: in, To ensure source credibility, specific values ​​include: Technical Manual = 1.0, Work Order = 0.9, User Report = 0.6. To extract confidence levels; Let be the normalized frequency of the relation in the total sample, and and be the weight coefficient. .

[0039] S104: Utilize graph neural networks to perform embedding learning on multi-layer heterogeneous graph structures, generate vector representations that support semantic similarity matching, and form a reasonable device anomaly knowledge graph.

[0040] In this embodiment, a graph neural network (GNN) is used to map discrete graph nodes into continuous low-dimensional vectors. The embedding learning process aggregates neighbor information through a message passing mechanism, enabling each node's vector to encode its local graph structure features. The reasonable device anomaly knowledge graph not only supports precise queries but also enables fuzzy matching and analogical reasoning through vector similarity (e.g., "scanning failure" and "image blurry" are semantically similar).

[0041] Specifically, the system employs a heterogeneous graph neural network model (such as R-GCN or HAN), designing dedicated message passing functions for different node types and relation types. During training, positive sample triples in the graph (e.g., "Weighing timeout ← CAUSES—AD drift") are used as positive examples, and negative examples (e.g., "Weighing timeout ← CAUSES—Print head blockage") are generated through negative sampling. The optimization objective is to maximize the positive example score and minimize the negative example score (using the TransE or RotatE scoring function). After 100 rounds of iterative training, each node obtains a 128-dimensional embedding vector. For example, the vector for "lens dirt" has a cosine similarity of 0.91 with the vectors for nodes such as "window contamination" and "optical occlusion," while it only has a cosine similarity of 0.23 with the vector for "motor failure."

[0042] S105: Based on the fault descriptions and replacement part records in historical maintenance work orders, as well as the module status jumps, actuator response delays, and user interruption behaviors in the real-time operation logs, hidden fault modes or intermittent failure nodes not covered by the current anomaly detection are mined through preset association rules to obtain the anomaly types to be supplemented for diagnosis.

[0043] In this embodiment, latent fault modes refer to novel or rare fault combinations that are not explicitly modeled in the existing knowledge graph but can be discovered through data mining, such as "temperature rise + communication delay → motherboard solder joint failure"; intermittent failure nodes refer to fault points that occur only occasionally under specific conditions (such as peak hours or high-temperature environments); preset association rules include frequent itemset mining algorithms such as Apriori and FP-Growth. The supplementary reasoning task refers to temporarily constructing reasoning paths for newly discovered anomaly types and formally incorporating them into the knowledge graph after verification.

[0044] S106: Add the abnormal types to be diagnosed to the supplementary reasoning task, and update the equipment abnormality knowledge graph in combination with historical diagnostic results.

[0045] In this embodiment, the system periodically runs an association rule mining engine. Inputs include: 1) the "replace motherboard" record in the work order and the "CPU temperature > 80℃" and "CAN bus error rate > 5%" records in the concurrent logs; 2) the co-occurrence sequence of "exit after staying on the payment page for > 60 seconds" and "network latency > 1s" in the user interruption behavior logs. Using the FP-Growth algorithm, the rule "{temperature > 80℃, CAN error rate > 5%} ⇒ {motherboard cold solder joint}" (support 0.7%, confidence 92%) is discovered. Since the "motherboard cold solder joint" corresponding to this rule has not yet appeared as an independent root cause in the graph, the system temporarily stores it as a "fault root cause to be supplemented for diagnosis."

[0046] Subsequently, within the following week, if a new abnormal event matches this pattern (e.g., a device reports high temperature + communication anomaly), the system temporarily adds "motherboard solder joint cold" as a candidate root cause during inference and pushes a "check motherboard solder joints" guide. If the verification is successful more than three times, a "motherboard solder joint cold" node is officially created in the graph, connected to upstream nodes such as "temperature sensor" and "communication module", and an initial weight of 0.75 is set, completing the incremental expansion and update of the graph.

[0047] S2: Integrates the operation logs of self-service check-in equipment, sensor time-series data, and user operation trajectories, maps multi-source heterogeneous inputs into dynamic anomaly triples that adapt to the anomaly knowledge graph ontology model, and assigns timestamps and confidence weights.

[0048] In this embodiment, the operation log refers to the structured log stream output by the device's operating system or application layer, including module start / stop, error codes, and task status; sensor time-series data includes continuous sampling values ​​of multiple physical quantities such as weight sensor, infrared beam switch, motor current, and temperature; user operation trajectory is collected by embedded points at the device front end, recording the user's clicks, swipes, dwell time, and process interruption locations on the interface. Dynamic anomaly triples refer to structured events in the form of (subject entity, relation, object entity), the content of which must be compatible with the node types and relational semantics in the knowledge graph, and include a timestamp (accurate to milliseconds) and a confidence weight (between 0 and 1, reflecting the credibility of the event's authenticity).

[0049] Specifically, in a self-service baggage check-in terminal deployed at an airport, the system collects data from each sensor every 100 milliseconds and monitors log output and user interface events in real time. When it detects that "the weighing module has not returned a valid weight value for 3 consecutive seconds" and "the user has stayed in the weighing area for more than 15 seconds," the anomaly recognition engine is triggered. This engine first converts the raw signals into high-level semantic events: for example, combining abnormal current waveforms with motor stop logs, it determines "conveyor belt drive failure"; combining multiple failed barcode scans with camera image blur analysis, it determines "barcode scanner lens contamination." Subsequently, using a predefined mapping rule base (such as "weighing timeout + user placing luggage → anomaly representation: no response to weighing"), the above semantic events are converted into triples conforming to the knowledge graph ontology, such as (weighing module, occurrence, no response to weighing) and (user, execution, placing luggage). At the same time, based on the degree to which the sensor data deviates from the normal range (such as the standard deviation of the weight reading exceeding 3 times the threshold) and the log error level (such as ERROR level), the confidence weight of the triple (e.g., 0.87) is calculated and timestamped.

[0050] S3: Using dynamic anomaly triples as inference anchors, multi-hop association inference is carried out in the anomaly knowledge graph based on graph neural networks. The graph edge weights are dynamically recalibrated by combining real-time device state vectors. Candidate fault root causes are aggregated along high-confidence paths, and a root cause priority ranking list with attribution path explanations is output.

[0051] In this embodiment, the inference anchor point refers to the graph node or subgraph generated by S2 that represents the current abnormal state; the graph neural network is a deep learning model capable of message passing and feature aggregation on a graph structure, used to capture semantic associations on multi-hop paths; the real-time device state vector is a high-dimensional vector composed of all current sensor readings, module operating states, and environmental parameters (such as temperature and humidity). Attribution path explanation refers to the sequence of nodes and relational chains traversed from the abnormal representation back to the potential root cause.

[0052] Specifically, multi-hop association reasoning includes: starting with the anomaly representation node in the dynamic anomaly triplet, associating it with the user interaction event that triggered the anomaly and the corresponding device functional module through operation-triggered relationships; extending to upstream or downstream associated functional modules in the consignment process using state-dependent relationships; and identifying potential fault propagation units in physical proximity by combining spatial adjacency relationships. Furthermore, it integrates real-time device state vectors and historical fault statistical frequencies to dynamically score candidate root causes, obtaining the association reasoning results. The core of multi-hop association reasoning lies in using the current anomaly as the entry point and conducting directed exploration along different semantic relationship paths to construct a candidate root cause set covering multiple dimensions of "human operation—process logic—physical layout." The reasoning process does not simply traverse all neighbors, but assigns different exploration priorities based on relationship types and dynamically adjusts the path credibility based on the real-time device status.

[0053] For example, suppose a self-service baggage check-in terminal reports a dynamic anomaly triple (payment terminal, occurrence, payment interface frozen). The system first locates the "payment interface frozen" node in the knowledge graph as the starting point for inference. Step one, it searches backward along the "operation trigger" relationship and finds that this anomaly is often triggered by interactive events such as "user repeatedly and rapidly clicking the payment button" or "inserting an invalid bank card," with the corresponding functional modules being "touchscreen" or "card reader." Step two, it traces forward along the "state dependency" relationship and identifies that the "payment terminal" depends on the precondition that "baggage information has been successfully uploaded to the airline system"; if the current log shows upload failure, then "network communication module failure" becomes a high-priority candidate root cause. Step three, it expands outward along the "spatial adjacency" relationship, checking whether "printers" and "barcode scanners" sharing the same power module or installed in the same cabinet as the payment terminal also experience anomalies—if the printer is offline at the same time, then "power distribution unit failure" is included in the candidate set. Subsequently, the system obtains the real-time device state vector: the current network latency is 800ms (normal <100ms), the card reader has no card insertion signal, and the power supply voltage is stable. Based on the frequency of "network timeout leading to payment freeze" in historical data (accounting for 72% of similar anomalies), the system assigns high weight to the "network communication module failure" path and low weight to the "user misoperation" path.

[0054] Specifically, using the dynamic triple (weighing module, occurrence, no weighing response) as an anchor point, the system locates the "no weighing response" node in the knowledge graph and initiates multi-hop inference based on a graph attention network (GAT). This network retrieves all directly connected potential root cause nodes (such as "sensor occlusion," "AD converter failure," and "communication interruption") within a 1-hop range, and further extends to associated user events (such as "placing luggage") and functional modules (such as "main control board") within a 2-hop range. During this process, the system synchronously acquires the current device state vector—for example, if the infrared sensor shows that luggage is indeed present in the weighing area, but the weight reading is zero, the edge weight of the "sensor occlusion" path is increased; if the main control board CPU load is as high as 95%, the weight of the "communication delay" related path is increased. The graph neural network aggregates and scores each candidate root cause based on these dynamically adjusted edge weights, and finally outputs a sorted list: [(Sensor is blocked by foreign object, confidence level 0.92, path: no response to weighing ← caused by ← sensor blocking ← resulting in ← foreign object falling in when placing luggage), (AD conversion chip failure, confidence level 0.68, path: no response to weighing ← caused by ← AD chip failure)].

[0055] S4: Based on the root cause priority sorting list, generate context-aware maintenance assistance solutions, combine equipment model, on-site spare parts, maintenance personnel skills and safety compliance constraints, dynamically arrange standard handling measures, and push them to the maintenance terminal.

[0056] In this embodiment, context awareness means that the generation of a maintenance plan depends not only on the fault itself but also on the actual conditions of the current operational environment; the equipment model determines the available maintenance manual version and specialized tools; on-site spare parts information comes from the airport spare parts management system, including inventory quantity and expiration date; the skills of maintenance personnel are provided by the personnel authentication system, reflecting whether they are qualified to operate high-voltage components or upgrade software; safety and compliance constraints include the Civil Aviation Administration's safety procedures for self-service equipment maintenance, such as power-off operations and anti-static requirements. Dynamic orchestration refers to selecting the optimal combination from multiple feasible measures, rather than simply returning to a single solution.

[0057] For example, regarding the top-ranked root cause, "sensor obstructed by foreign object," the system retrieves the standard handling procedure from the maintenance knowledge base: "1. Disconnect the power to the weighing module; 2. Open the front maintenance panel; 3. Clean the sensor window with a non-woven cloth; 4. Power on again and perform a self-test." Subsequently, the system checks the current context: the device model is "SmartCheck-inV3.2," supporting quick-release panels; the spare parts inventory shows sufficient cleaning supplies; the currently logged-in maintenance personnel hold "Level 1 Electromechanical Maintenance" certification and have operating permissions; and there is no peak flight period at the moment, allowing for a 5-minute downtime. Therefore, the solution is deemed fully feasible. The system formats it into a step-by-step guide card, including text instructions, diagrams, and safety warnings (such as "Be sure to ensure the power indicator light is off before operation"), and pushes it to the personnel's handheld terminal via the airport maintenance APP. If there is a conflict in the context (such as no certified personnel present or missing spare parts), the system automatically downgrades to recommend a suboptimal solution (such as "attempt remote reset") or marks it as "requires senior technician intervention."

[0058] In one embodiment, after pushing the maintenance assistance solution to the operation and maintenance terminal, the self-service check-in equipment maintenance assistance processing method based on the equipment anomaly knowledge graph further includes: S401: Receive maintenance execution feedback signals from the terminal to determine whether to eliminate the abnormal state.

[0059] In this embodiment, the maintenance execution feedback signal refers to the structured confirmation information actively submitted by maintenance personnel through a handheld terminal, equipment maintenance interface, or voice interaction system after completing the maintenance operation, including status indicators such as the exception has been resolved, partially resolved, or not resolved.

[0060] S402: If the feedback indicates that the abnormal state has been eliminated, increase the path confidence between the potential root cause of this successful association and the adopted handling measures, and strengthen the edge weights of the corresponding operation triggers, state dependencies and handling constraints.

[0061] In this embodiment, path confidence enhancement refers to increasing the weights of all relevant edges in the reasoning link formed from the anomaly representation through the potential root cause to the treatment measures in the knowledge graph according to preset rules, so as to reflect the effectiveness of the path in the real scenario; the weight enhancement of the edge is not simply accumulated, but adopts exponential smoothing or Bayesian update strategy.

[0062] Specifically, taking the above successful case of "scanning failure → sensor contamination → lens cleaning" as an example, the system locates the path in the knowledge graph: (Barcode cannot be recognized) ← [caused by] — (lens contamination) ← [resulted by] — (passenger placed luggage with tape) (Lens contamination) — [Solved by] → (Clean the lens) Among them, "passenger placing luggage with tape" and "lens contamination" are "operation-triggered" relationships, "barcode unreadable" and "lens contamination" are causal relationships, and "cleaning the lens" and "lens contamination" implicitly involve "disposal constraints" (requiring the use of non-woven fabric, power-off operation, etc.). The system multiplies the original weights of the above three edges (e.g., 0.75, 0.82, 0.68) by an enhancement factor of 1.2 (maximum not exceeding 0.95), updating them to 0.90, 0.95, and 0.82.

[0063] S403: If the abnormal feedback status persists, based on the updated device sensor data and user interaction records, the candidate root causes in the original inference path are negatively marked, and local subgraph reconstruction is initiated in the abnormal knowledge graph ontology model. Multi-hop association inference is re-executed to generate supplementary maintenance guidance; the structured log of the entire maintenance process is used as the incremental training sample of the abnormal knowledge graph ontology model.

[0064] In this embodiment, the negation label refers to applying a negative weight or temporary masking label to the root cause-anomaly association that has been disproven in the anomaly knowledge graph; the local subgraph reconstruction refers to expanding the nodes and edges in the two-hop neighborhood of the original anomaly node as the center, and injecting new candidate root cause nodes (such as "unstable power supply" and "motherboard cold solder joint") in combination with newly collected sensor data (such as voltage fluctuations and temperature anomalies) to form an updated inference subgraph; the "incremental training sample" is to format the entire process data from the occurrence of the anomaly to the failure of the repair (including the original triplet, inference path, feedback result, and new sensor readings) into supervised learning samples.

[0065] Specifically, suppose another device is diagnosed as "sensor obstruction" due to "no weighing response," and a "remove foreign object" solution is pushed. However, maintenance personnel report that "there is still no weight reading after cleaning." The system immediately collects the latest data: the infrared sensor in the weighing area is normal (there is luggage), but the temperature of the AD conversion chip is as high as 85℃ (normal <60℃), and the motor drive current is abnormally low. Based on this, the system adds a "recently invalid" tag to the original "sensor obstruction" node, and adds candidate root cause nodes "AD converter thermal failure" and "insufficient power module output" in its neighborhood, and connects them to the weighing module through "spatial adjacency" relationship. Subsequently, the graph neural network inference is re-run on the updated subgraph, the new sensor data is fused, and a new ranking is output: [(AD chip overheating, confidence 0.85), (main control board communication failure, confidence 0.62)], and supplementary instructions are generated: "1. Check if the cooling fan is running; 2. Measure the AD chip power supply voltage." Meanwhile, the event (including the original error diagnosis, failure reasons, and new findings) is packaged into incremental samples in JSON format and uploaded to the central training server for the next round of graph neural network parameter fine-tuning.

[0066] In one embodiment, such as Figure 2 As shown, after constructing the ontology model of the anomaly knowledge graph of self-service check-in equipment, the auxiliary processing method for maintenance of self-service check-in equipment based on the equipment anomaly knowledge graph also includes: S100: Based on the historical operation logs and real-time interactive behavior data of the target shipping equipment, a time sequence diagram model of equipment status is constructed using a graph neural network, and behavioral feature vectors of functional module nodes are extracted.

[0067] In this embodiment, the device state sequence diagram model is a dynamic graph structure, where nodes represent functional modules of the device, edges represent dependencies or data flow relationships between modules, and the attributes of each node evolve over time, reflecting its operating status. Historical operation logs include the start / stop records of each module over the past 30 days, task completion time, and error code frequency. Real-time interactive behavior data refers to user click flows, process interruptions, and dwell times in the current session. The behavioral feature vector is a low-dimensional representation obtained by encoding the sequence diagram using a graph neural network, used to characterize whether a module is currently deviating from its normal behavior pattern.

[0068] For example, on a SmartCheck-in V3.2 device deployed at an airport, the system aggregates operational data every 5 minutes to construct a time-series graph containing 8 functional module nodes. Each node's attributes include: task success rate (e.g., successful scan / total attempts), response latency (ms), error rate (%), and user interruption frequency. The system uses a Temporal Graph Convolutional Network (T-GCN) to model this graph sequence: first, GCN aggregates the states of neighboring modules in the spatial dimension (e.g., the payment terminal state is affected by the weighing module), and then GRU captures the state evolution trend in the temporal dimension. After training (using data from fault-free devices over the past 6 months as positive samples), the device state time-series graph model outputs a 128-dimensional behavioral feature vector for each module.

[0069] S200: Input the behavioral feature vector into the pre-trained abnormal state detection model to identify suspicious device nodes that deviate from the normal operating baseline and generate a high-risk fault candidate set.

[0070] In this embodiment, the pre-trained abnormal state detection model is a binary classifier trained on a large amount of normal and abnormal equipment operation data, used to determine whether the current behavior features belong to the normal distribution; the normal operation baseline refers to the average performance index of the equipment model under similar airports and similar traffic conditions; the high-risk fault candidate set is a list of modules that are judged as abnormal by the model and have a high probability of failure.

[0071] Specifically, the behavioral feature vectors of each module output by S100 are input into an anomaly detection model based on a variational autoencoder (VAE). During the training phase, this model learns the behavioral reconstruction error distribution of normal devices and sets the 99th percentile as the anomaly threshold. When the feature vector reconstruction error of a device's "conveyor belt drive module" reaches 0.32 (threshold 0.25), and exceeds the limit for two consecutive cycles, the system adds it to the high-risk candidate set. Simultaneously, business rules are used for filtering: if no passengers are currently using the device, no alarm is triggered; if it is during peak flight hours (>200 passengers per hour), the priority is increased. The final candidate set is generated as: [conveyor belt drive module (anomaly score 0.88), main control board (anomaly score 0.76)].

[0072] S300: For high-risk fault candidate sets, the virtual maintenance simulation engine is invoked to dynamically generate a multi-step maintenance guidance sequence that matches the hardware interface and software protocol of the suspected device node, and to perform interactive maintenance simulation.

[0073] In this embodiment, the virtual maintenance simulation engine is a simulation system that integrates a digital twin model of the equipment, with built-in hardware interface definitions, software API protocols, and mechanical disassembly and assembly logic for each module. The multi-step maintenance guidance sequence is a structured maintenance process generated according to safety and operating procedures; interactive maintenance simulation refers to simulating the system response after maintenance personnel perform each step in a virtual environment to verify the feasibility of the solution.

[0074] For example, in response to a "conveyor belt drive module" anomaly, the system loads the 3D model and interface documentation of model V3.2 from the device's digital twin library. Based on the potential root causes associated with the knowledge graph (such as "motor carbon brush wear" or "loose belt"), the virtual engine automatically generates two guiding sequences: Sequence A (for carbon brush wear): 1. Disconnect the main power supply; 2. Remove the front maintenance cover; 3. Use a multimeter to measure the motor resistance; 4. If the resistance is >10Ω, replace the carbon brush assembly.

[0075] Sequence B (for loose belts): 1. Enter maintenance mode; 2. Perform belt tension self-check; 3. If tension < 5N, adjust the tension wheel.

[0076] Subsequently, the system "executes" sequence A in the simulation environment: after simulating a power outage, it checks whether the motor signal disappears; after simulating removing the cover, it verifies whether there is enough operating space; and it simulates measuring resistance. If the return value is 8Ω (normal), the path is determined to be invalid, and the system automatically switches to sequence B.

[0077] S400: Based on the feedback from maintenance simulations, combined with the equipment anomaly knowledge graph and expert rule base, a multi-indicator dynamic evaluation is conducted on suspected faults. The multi-indicators include the actual impact, repair timeliness, user operation complexity, spare parts availability, and flight support priority.

[0078] In this embodiment, the maintenance simulation feedback results include the feasibility of the solution, the estimated time, and the required tools; the expert rule base stores the operation and maintenance strategies. Multi-indicator dynamic evaluation scores each candidate maintenance solution across five dimensions and then weights the scores. The definitions of each indicator are as follows: Actual impact level: The impact on passenger throughput if the fault is not addressed (1-5 points). Repair timeliness: Estimated repair time (shorter score, higher score); User operation complexity: Required skill level and number of steps (the simpler the operation, the higher the score); Spare parts availability: Is there stock in the local warehouse (Yes = 1, No = 0); Flight priority: Are there any international flights about to depart in the area where the current device is located (yes = high priority).

[0079] For example, the weighting coefficients for the five indicators are 0.3, 0.25, 0.2, 0.15, and 0.1, respectively.

[0080] S500: Generates a maintenance confidence score based on the evaluation results, and marks solutions with maintenance confidence scores higher than a preset threshold as recommended maintenance solutions.

[0081] In this embodiment, the maintenance confidence score is a final confidence score (0–1) calculated by comprehensively considering the success rate of simulation, the multi-indicator evaluation score, and the repair rate of similar historical cases; the preset threshold can be dynamically adjusted according to the operation and maintenance strategy (e.g., 0.7 on a daily basis and 0.6 during peak hours).

[0082] For example, the system calculates the confidence level of sequence B as follows: Confidence level = 0.5 × projection success rate + 0.3 × (multi-index score / 5) + 0.2 × historical repair rate.

[0083] In one embodiment, the self-service baggage check-in equipment maintenance assistance method based on equipment anomaly knowledge graph in step S4 further includes: S41: Using each functional module of the equipment to be repaired as the basic unit, retrieve all related standard handling measures from the preset maintenance knowledge base based on the high-confidence root causes in the root cause priority sorting list.

[0084] In this embodiment, the equipment model is used to match the corresponding technical manual and special tool list; the on-site spare parts information includes the quantity and expiration date of spare parts inventory, which is synchronized in real time by the airport's central spare parts management system, including the current inventory, batch number, and expiration date of each consumable or module; the maintenance personnel skill information includes certification level and historical repair success rate; safety compliance constraints include civil aviation security regulations and electrical safety operating procedures, among which civil aviation security regulations, such as the "Rules for Security and Protection of Civil Aviation Transport Airports," require that operations involving passenger data modules must be supervised by two people; electrical safety operating procedures, such as the "Electrical Installation Specifications," stipulate that the voltage for live-line work shall not exceed 36V.

[0085] Equipment functional modules serve as the basic granularity for maintenance tasks. The root cause priority ranking list is output by step S3 and typically includes the top-3 high-confidence root causes. The preset maintenance knowledge base is a structured database that stores all standard handling measures and their prerequisites for each root cause, derived from manufacturer manuals, internal SOPs, and historical success cases.

[0086] For example, in response to a "payment failure" exception, the system receives a root cause sorting list from S3: [(Airline interface timeout, confidence 0.89), (Reader firmware error, confidence 0.76), (PCIe communication interruption, confidence 0.65)]. Taking the "payment terminal" module as a unit, the system retrieves information from the maintenance knowledge base: For "Airline interface timeout": Measures A1 = "Restart payment service", A2 = "Switch to backup communication link"; For the "card reader firmware error": Solution B1 = "Flash the card reader firmware (requires a dedicated USB flash drive)"; For "PCIe communication interruption": Measures C1 = "Re-insert / remove the PCIe card", C2 = "Replace the motherboard".

[0087] Five standard handling measures were retrieved, each accompanied by the required tools, estimated time, and safety requirements.

[0088] S42: For each standard handling measure retrieved, a feasibility score is given based on the equipment model, on-site spare parts, maintenance personnel skills, and safety compliance constraints.

[0089] In this embodiment, the feasibility score is a value between 0 and 1. The score uses a weighted linear combination model, where each constraint dimension is scored independently and then summed with weights. The specific rules are as follows: Equipment model matching degree: 1.0 if the measure supports the model, otherwise 0; Spare parts availability: 1.0 if inventory > 0 and not expired; 0.6 if inventory = 0 but can be redistributed within 2 hours; otherwise 0. Personnel skill matching: 1.0 for having the required certification and a historical success rate >90%; 0.7 for having certification but a success rate <90%; 0 for having no certification. Safety compliance satisfaction: 1.0 for full compliance, 0.5 for partial compliance (if additional approval is required), and 0 for violation of mandatory regulations.

[0090] For example, each of the above five measures will be scored: A1 (Restart Payment Service): Model Support (1.0), No Spare Parts Required (1.0), Personnel Have Permissions (1.0), No Security Conflicts (1.0), then Feasibility = 1.0; A2 (Switching to backup communication link): Same as above, feasibility = 1.0; B1 (Refresh Card Reader Firmware): Model Support (1.0), 2 Dedicated USB Flash Drives in Stock (1.0), Personnel are Certified and Success Rate is 94% (1.0), but the back cover of the device needs to be opened (which is "operation with electricity"), and the current environment does not have an insulating pad, which violates electrical safety regulations, so the safety score is 0 and the feasibility is 0. C1 (Reinserting PCIe Card): Power off operation is required, but it is currently peak flight time and downtime of more than 2 minutes is not allowed. The safety compliance score is calculated to be 0.5. All other items are full marks, so the feasibility score is (1+1+1+0.5) / 4=0.875. C2 (Replace Motherboard): The motherboard inventory is 0 and it needs to arrive in 4 hours, so the spare parts score is 0 and the feasibility is 0.

[0091] Ultimately, three feasible measures, A1, A2, and C1, were retained.

[0092] S43: Using feasibility scoring and comprehensive index methods, sort and combine various standard treatment measures to generate the optimal maintenance assistance plan.

[0093] In this embodiment, the comprehensive index method refers to calculating a comprehensive score by further integrating multiple business indicators such as repair efficiency, resource consumption, and risk level on the basis of feasibility scoring. The ranking is based on descending order of scores. Combination refers to arranging multiple measures into a multi-step solution in a logical order when a single measure is insufficient to solve the problem.

[0094] For example, the system calculates a composite index for the three retained measures: The overall score is calculated as follows: 0.6 × Feasibility + 0.25 × (1 - Estimated Time / T_max) + 0.15 × (1 - Risk Level), where T_max is the maximum allowable time for a single maintenance operation in minutes, typically set to 10 minutes during peak hours. The risk level is defined by the expert rule base (A1 / A2 = Low Risk = 0.1, C1 = Medium Risk = 0.4). Substituting these values, we get: A1: 0.6×1.0+0.25×(1−1 / 10)+0.15×(1−0.1)=0.6+0.225+0.135=0.96; A2: 0.6×1.0+0.25×(1−2 / 10)+0.15×(1−0.1)=0.6+0.20+0.135=0.935; C1: 0.6×0.875+0.25×(1−5 / 10)+0.15×(1−0.4)=0.525+0.125+0.09=0.74; The system selects A1, which has the highest score, as the preferred solution and generates a combined strategy: "1. Execute A1 (restart payment service); 2. If the anomaly is not resolved within 5 minutes, automatically upgrade to execute A2."

[0095] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0096] In one embodiment, a self-service baggage check-in equipment repair assistance system based on a device anomaly knowledge graph is provided, which corresponds to the self-service baggage check-in equipment repair assistance method based on a device anomaly knowledge graph in the above embodiments.

[0097] The system in the embodiments of this invention is described from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of a self-service check-in equipment maintenance auxiliary processing system based on equipment anomaly knowledge graph provided in this application embodiment.

[0098] It should be noted that, Figure 3 The structure of the system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0099] like Figure 3 As shown, the system includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in ROM 302 or a program loaded from storage section 308 into RAM 303, such as executing the methods described in the above embodiments. RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0100] The following components are connected to I / O interface 305: input section 306 including a camera, infrared sensor, etc.; output section 307 including a liquid crystal display (LCD) and speakers, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card and a modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0101] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the self-service check-in equipment maintenance assistance method based on a device anomaly knowledge graph.

[0102] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0103] In one embodiment, particularly according to embodiments of the present invention, the processes described above with reference to the flowchart can be implemented as a computer software program. For example, embodiments of the present invention include a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the self-service check-in equipment repair assistance method based on a device anomaly knowledge graph. In such embodiments, the computer program can be downloaded and installed from a network via a communication module, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the various functions defined in the present invention.

[0104] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0105] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for auxiliary maintenance of self-service baggage check-in equipment based on equipment anomaly knowledge graph, characterized in that: include: An anomaly knowledge graph ontology model for self-service check-in equipment is constructed. The anomaly knowledge graph ontology model uses equipment functional modules, user interaction events, anomaly representations, potential root causes and handling measures as nodes, and embeds domain-specific semantic relationships to construct a directed weighted knowledge graph with causal chain tracing and operational context awareness. By integrating the operation logs of self-service check-in equipment, sensor time-series data, and user operation trajectories, multi-source heterogeneous inputs are mapped into dynamic anomaly triples that adapt to the anomaly knowledge graph ontology model, and timestamps and confidence weights are assigned. Using the dynamic anomaly triples as inference anchors, multi-hop association inference is carried out in the anomaly knowledge graph based on graph neural networks. The graph edge weights are dynamically recalibrated in combination with real-time device state vectors. Candidate fault root causes are aggregated along high-confidence paths, and a root cause priority ranking list with attribution path explanations is output. Based on the root cause priority ranking list, a context-aware maintenance assistance plan is generated. Combined with equipment model, on-site spare parts, maintenance personnel skills and safety compliance constraints, standard handling measures are dynamically arranged and pushed to the maintenance terminal.

2. The method according to claim 1, characterized in that, The domain-specific semantic relationships include operation triggering, state dependency, spatial adjacency, and disposal constraints. The multi-hop association reasoning includes: starting from the abnormal representation node in the dynamic abnormal triplet, associating it with the user interaction event that caused the abnormality and the corresponding device function module through the operation trigger relationship, extending it to the upstream or downstream associated function modules in the shipping process using the state dependency relationship, and identifying the potential fault propagation unit that is physically close by combining the spatial adjacency relationship; and dynamically scoring the candidate root cause by fusing the real-time device state vector and the historical fault statistical frequency to obtain the association reasoning result.

3. The method according to claim 1, characterized in that, After pushing the maintenance assistance solution to the operation and maintenance terminal, the method further includes: Receive maintenance execution feedback signals from the terminal to determine whether the abnormal state has been eliminated; If the feedback indicates that the abnormal state has been eliminated, the path confidence between the potential root cause of this successful association and the adopted handling measures will be increased, and the edge weights of the corresponding operation triggers, state dependencies and handling constraints will be strengthened. If the abnormal feedback persists, based on the updated device sensor data and user interaction records, the candidate root causes in the original inference path are negatively marked, and local subgraph reconstruction is initiated in the abnormal knowledge graph ontology model. Multi-hop association inference is re-executed to generate supplementary maintenance guidance; the structured log of the entire maintenance process is used as the incremental training sample of the abnormal knowledge graph ontology model.

4. The method according to claim 1, characterized in that, The construction of the anomaly knowledge graph ontology model for self-service baggage check-in equipment includes: Collect historical maintenance work orders, equipment technical manuals, operation and maintenance expert experience, and user fault reports, and perform entity recognition and relationship extraction. The identified entities include equipment modules, sensors, actuators, abnormal phenomena, causes of failure, maintenance actions, and tools and consumables; the relationships include those that cause, belong to, require use, are repairable, and are often associated with. A multi-layer heterogeneous graph structure is constructed based on the entities and relationships, and initial confidence weights are assigned to each node and edge. By using graph neural networks to perform embedding learning on the multi-layer heterogeneous graph structure, vector representations that support semantic similarity matching are generated, forming a reasonable device anomaly knowledge graph; Based on the fault descriptions and replacement parts records in the historical maintenance work orders, as well as the module status jumps, actuator response delays and user interruption behaviors in the real-time operation logs, hidden fault modes or intermittent failure nodes not covered by the current anomaly detection are mined through preset association rules to obtain the anomaly types to be supplemented for diagnosis. The abnormal types to be diagnosed are added to the supplementary reasoning task, and the device abnormality knowledge graph is updated in combination with historical diagnostic results.

5. The method according to claim 4, characterized in that, After constructing the anomaly knowledge graph ontology model for self-service baggage check-in equipment, the following is also included: The abnormal keywords in the device's operating context features are subjected to synonym expansion and standardization processing to generate a standardized abnormal description vector. The standardized anomaly description vector is compared with the embedding vector of each anomaly node in the device anomaly knowledge graph ontology model to calculate the similarity, and candidate anomaly nodes with similarity higher than a preset similarity threshold are selected. By combining fault codes, equipment models, operating stages, and environmental constraints, the candidate abnormal nodes are filtered for context consistency, and abnormal nodes that meet the preset multi-dimensional constraints are retained as matching results.

6. The method according to claim 4, characterized in that, After obtaining the list of device anomalies, the method further includes: Based on the historical operation logs and real-time interactive behavior data of the target shipping equipment, a time sequence graph model of equipment status is constructed using a graph neural network, and behavioral feature vectors of functional module nodes are extracted. The behavioral feature vectors are input into a pre-trained abnormal state detection model to identify suspicious device nodes that deviate from the normal operating baseline and generate a high-risk fault candidate set. For the high-risk fault candidate set, the virtual maintenance simulation engine is invoked to dynamically generate a multi-step maintenance guidance sequence that matches the hardware interface and software protocol of the suspected device node, and an interactive maintenance simulation is performed. Based on the feedback from the maintenance simulation, and combined with the equipment anomaly knowledge graph and expert rule base, a multi-indicator dynamic evaluation is conducted on suspected faults. These multi-indicators include the actual impact, repair timeliness, user operation complexity, spare parts availability, and flight support priority. Based on the evaluation results, a maintenance confidence score is generated, and solutions with maintenance confidence scores higher than a preset threshold are marked as recommended maintenance solutions.

7. The method according to claim 4, characterized in that, The process involves generating a context-aware maintenance assistance plan based on the root cause priority ranking list. This plan dynamically arranges standard handling measures, taking into account equipment model, on-site spare parts, maintenance personnel skills, and safety and compliance constraints. Specifically, this includes: The equipment model is used to match the corresponding technical manual and special tool list; the on-site spare parts information includes the spare parts inventory quantity and expiration date; the maintenance personnel skill information includes certification level and historical repair success rate; the safety compliance constraints include civil aviation security inspection regulations and electrical safety operating procedures. Using each functional module of the equipment to be repaired as the basic unit, all related standard handling measures are retrieved from the preset maintenance knowledge base based on the high-confidence root causes in the root cause priority sorting list. For each standard handling measure retrieved, a feasibility score is calculated based on the equipment model, the on-site spare parts, the skills of the maintenance personnel, and the safety compliance constraints. Using the feasibility score and comprehensive index method, the various standard handling measures are ranked and combined to generate the optimal maintenance assistance plan.

8. A self-service baggage check-in equipment maintenance assistance system based on equipment anomaly knowledge graph, characterized in that, The system includes: One or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the self-service check-in equipment maintenance assistance method based on the equipment anomaly knowledge graph as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the self-service check-in equipment maintenance assistance method based on the equipment anomaly knowledge graph as described in any one of claims 1 to 7.

Images