Equipment operation knowledge graph construction and health state evaluation method based on Internet of Things

Abstract

The invention belongs to the technical field of industrial Internet of Things, and particularly relates to an equipment operation knowledge graph construction and health state evaluation method based on Internet of Things, which aims at solving the problems of knowledge graph staticization and diagnosis isolation in the existing industrial equipment fault diagnosis, acquires and preprocesses equipment operation data through a distributed sensor, and improves the fault diagnosis accuracy. A knowledge graph is dynamically constructed in combination with historical fault records, and multi-source information fusion is achieved; health indexes are calculated by using a weighted time decay model, so that the evaluation accuracy is improved; performing path reasoning based on a knowledge graph in an abnormal state to realize fault mode intelligent matching and natural language explanation generation; and on the basis of an edge-cloud collaboration mechanism, the real-time performance of the system and continuous optimization of knowledge are guaranteed. According to the method, the comprehensiveness of equipment state sensing and the intelligence and interpretability of fault diagnosis are enhanced, and the efficiency and reliability of industrial equipment health management are improved.

Description

Technical Field

[0001] This invention belongs to the field of industrial Internet of Things (IoT) technology, specifically relating to a method for constructing an IoT-based knowledge graph of equipment operation and assessing its health status. Background Technology

[0002] In the operation of industrial equipment, timely detection of potential faults and effective maintenance are crucial for ensuring production continuity and equipment reliability. Traditional equipment health status assessment methods mainly rely on single data sources from sensors and determine whether the equipment is in an abnormal state by setting thresholds or statistical analysis. While these methods are simple to implement, they have significant limitations when dealing with complex equipment systems: on the one hand, it is difficult to effectively integrate and semantically represent multidimensional data; on the other hand, there is a lack of effective utilization of historical failure modes, resulting in low accuracy in fault diagnosis and an inability to provide clear fault explanations.

[0003] In recent years, with the development of IoT and big data technologies, some systems have attempted to introduce knowledge graphs for equipment status modeling to enhance the intelligence level of fault identification. However, in existing technologies, knowledge graphs are often used as static information storage structures, failing to fully integrate real-time operational data for dynamic updates, and also failing to effectively support path reasoning and fault explanation generation based on semantic relationships. Therefore, how to construct a knowledge graph in the industrial IoT environment that can integrate real-time sensor data and historical knowledge, and possess dynamic update capabilities, and on this basis achieve intelligent assessment of equipment health status and interpretable fault matching, has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing an equipment operation knowledge graph and assessing its health status based on the Internet of Things. This method effectively solves the problems of static knowledge graphs and isolated fault diagnosis in the prior art, and significantly improves the intelligence level and practicality of health status assessment of industrial equipment.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for constructing an equipment operation knowledge graph and assessing its health status based on the Internet of Things, comprising the following steps:

[0006] Real-time operating data of the equipment is collected by distributed sensors, and the real-time operating data is filtered and smoothed by sliding window to generate normalized sensor values. Based on the sensor values ​​and historical fault records, a knowledge graph containing equipment entities, component entities, fault mode entities and their relationships is constructed. According to the entity relationships in the knowledge graph and the current sensor data, the health index of the equipment is calculated. When the health index is lower than the warning threshold, path reasoning is performed using the knowledge graph to match the current equipment status with historical fault modes and generate fault explanation text. Through the collaboration mechanism between edge computing nodes and the cloud, the health index and fault explanation are uploaded to the cloud, and the knowledge graph weights are optimized based on the cloud knowledge graph update mechanism.

[0007] Preferably, the formula for the sliding window filtering is: ;in, Here is the original data at time t, and n is the window size. This is the filtered output value.

[0008] Preferably, the formula for calculating the health index is: ;in, The result is a weighted summation of the sensor values, where λ is the time decay coefficient and t is the time interval since the last maintenance.

[0009] Preferably, the time decay coefficient λ ranges from 0.01 to 0.1.

[0010] Preferably, the construction of the knowledge graph includes: extracting fault mode entities from maintenance logs and establishing causal relationship edges between fault mode entities and component entities.

[0011] Preferably, the generation of the fault explanation text includes: extracting nodes in the fault link based on the path reasoning results, and generating a natural language description containing maintenance suggestions by combining historical cases.

[0012] Preferably, the edge computing node and cloud collaboration mechanism includes: transmitting normalized sensor values ​​and health indices via the MQTT protocol, and employing a differential upload strategy to reduce data transmission volume.

[0013] Preferably, the fault mode matching uses the cosine similarity calculation formula: ; For the current feature vector, Let be the feature vector of the k-th fault mode.

[0014] Preferably, the knowledge graph update mechanism includes: when the similarity S of a new fault mode is greater than or equal to τ, where τ is a set threshold, the weight of the existing node is updated; otherwise, a new node is added to the knowledge graph.

[0015] Preferably, the path reasoning includes: using breadth-first search to find the shortest path from the current node to the high-probability failure mode, where the weight of each edge is determined by the historical co-occurrence frequency.

[0016] Technical effects and advantages of the present invention: The method for constructing equipment operation knowledge graphs and assessing health status based on the Internet of Things proposed in this invention has the following advantages compared with the prior art:

[0017] This invention collects and preprocesses equipment operation data through distributed sensors, dynamically constructs a knowledge graph by combining it with historical fault records, and achieves multi-source information fusion. It uses a weighted time decay model to calculate a health index, improving assessment accuracy. Under abnormal conditions, it performs path reasoning based on the knowledge graph, achieving intelligent fault mode matching and natural language interpretation generation. Relying on an edge-cloud collaborative mechanism, it ensures system real-time performance and continuous knowledge optimization. This method enhances the comprehensiveness of equipment status perception, the intelligence and interpretability of fault diagnosis, and improves the efficiency and reliability of industrial equipment health management. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things, as described in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1

[0021] This invention provides, for example Figure 1 The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things (IoT) includes the following steps:

[0022] Real-time operating data of the device is collected through distributed sensors, and the real-time operating data is subjected to sliding window filtering and smoothing to generate normalized sensor values; further, the formula for sliding window filtering is: ;in, Here is the original data at time t, and n is the window size. This is the filtered output value.

[0023] Based on sensor values ​​and historical fault records, a knowledge graph is constructed that includes equipment entities, component entities, fault mode entities and their relationships. Furthermore, the construction of the knowledge graph includes: extracting fault mode entities from maintenance logs and establishing causal relationship edges between fault mode entities and component entities.

[0024] Based on the entity relationships in the knowledge graph and the current sensor data, the device's health index is calculated; further, the formula for calculating the health index is: ;in, The result is a weighted summation of the sensor values, where λ is the time decay coefficient (ranging from 0.01 to 0.1), and t is the time interval since the last maintenance.

[0025] When the health index is lower than the warning threshold, path reasoning is performed using a knowledge graph to match the current device status with historical fault patterns and generate fault explanation text. Specifically, the generation of fault explanation text includes: extracting nodes in the fault link based on the path reasoning results and generating a natural language description containing maintenance suggestions by combining historical cases.

[0026] Furthermore, path reasoning includes: using breadth-first search to find the shortest path from the current node to the high-probability failure mode, with the weight of each edge determined by the historical co-occurrence frequency.

[0027] Through an edge computing node and cloud collaboration mechanism (transmitting normalized sensor values ​​and health indices via the MQTT protocol and employing a differential upload strategy to reduce data transmission volume), the health indices and fault explanations are uploaded to the cloud, and the knowledge graph weights are optimized based on the cloud knowledge graph update mechanism.

[0028] Furthermore, the fault mode matching uses the cosine similarity calculation formula: ;in, For the current feature vector, Let be the feature vector of the k-th fault mode.

[0029] Furthermore, the knowledge graph update mechanism includes: when the similarity S of a new fault mode is greater than or equal to τ (τ is a set threshold), the weight of the existing node is updated; otherwise, a new node is added to the knowledge graph.

[0030] This method collects and preprocesses equipment operation data through distributed sensors, dynamically constructs a knowledge graph based on historical fault records, and achieves multi-source information fusion. A weighted time decay model is used to calculate a health index, improving assessment accuracy. Under abnormal conditions, path reasoning is performed based on the knowledge graph, enabling intelligent fault mode matching and natural language interpretation generation. An edge-cloud collaborative mechanism ensures system real-time performance and continuous knowledge optimization. This approach enhances the comprehensiveness of equipment status perception, the intelligence and interpretability of fault diagnosis, and improves the efficiency and reliability of industrial equipment health management.

[0031] Example 2

[0032] This embodiment proposes an IoT-based equipment operation knowledge graph construction and health status assessment system, including:

[0033] The data acquisition and preprocessing module is used to acquire real-time operating data of the device through distributed sensors, and to perform sliding window filtering and smoothing on the real-time operating data to generate normalized sensor values.

[0034] The knowledge graph construction module is used to build a knowledge graph containing equipment entities, component entities, fault mode entities and their relationships based on sensor values ​​and historical fault records.

[0035] The health status assessment module is used to calculate the device's health index based on the entity relationships in the knowledge graph and current sensor data;

[0036] The fault intelligent matching and explanation module is used to perform path reasoning using a knowledge graph when the health index is lower than the warning threshold, match the current device status with historical fault patterns, and generate fault explanation text.

[0037] The edge-cloud collaborative update module is used to upload the health index and fault explanation to the cloud through the collaborative mechanism between edge computing nodes and the cloud, and optimize the knowledge graph weight based on the cloud knowledge graph update mechanism.

[0038] In addition, the above modules are also used to implement other steps of an IoT-based equipment operation knowledge graph construction and health status assessment method, as follows:

[0039] 1. Data Acquisition and Preprocessing Module

[0040] This system employs a distributed sensor network architecture, deployed across various industrial equipment, encompassing key mechanical components such as motors, gearboxes, and hydraulic pumps. Sensor types include temperature sensors (DS18B20), triaxial accelerometers (MPU-9250), pressure transmitters (Honeywell PPT0010), and speed measurement devices (photoelectric encoders). All sensors are connected to local edge computing nodes via the LoRaWAN low-power wide-area network protocol, enabling long-distance, low-latency data transmission.

[0041] To improve data stability and reliability, the system introduces a multi-level filtering mechanism. First, a sliding window mean filter is applied to the original signal: ;in, This represents the original data at time t, where n is the window size, ranging from 5 to 10. This step can effectively suppress high-frequency noise interference.

[0042] Subsequently, exponential smoothing was used for further optimization. Where α∈(0,1] is the smoothing coefficient, used to adjust the degree of influence of historical data on the current output. When the equipment is operating stably, α takes a smaller value (e.g., 0.3) to enhance the trend prediction capability; while when the system is in a fluctuating phase, α increases to 0.7-0.9 to respond quickly to changes.

[0043] Normalization unifies the dimensions of different sensors, as follows: Where μ is the historical mean of a certain type of sensor, and σ is its standard deviation. This method ensures that all physical quantities are compared on the same scale, which facilitates weight allocation in subsequent knowledge graph construction.

[0044] In addition, the system also introduces an anomaly detection mechanism, using the Z-score method to identify outliers: If Z>3, the data point is considered abnormal, an alarm is triggered, and a log is recorded for manual review.

[0045] 2. Knowledge Graph Construction Module

[0046] Knowledge graphs consist of entities, relations, and attributes. Their core objective is to abstract the operational state of devices into a semantic network, supporting dynamic updates and reasoning.

[0047] (1) Entity extraction and classification

[0048] The system extracts entities from three sources: real-time sensor data, such as "bearing temperature" and "motor vibration amplitude"; maintenance logs and fault reports, such as "spindle breakage" and "hydraulic leakage"; and technical manuals and expert experience documents, such as "insufficient lubrication leading to wear" and "voltage overload causing burnout".

[0049] Entities are categorized into four types: Equipment: such as "Motor M1" and "Gear G2"; Component: such as "Bearing A3" and "Gear B4"; Failure Mode: such as "Overheating" and "Abnormal Noise"; and Maintenance Action: such as "Replacing Components" and "Cleaning Oil Circuit".

[0050] (2) Relation extraction and modeling

[0051] The system employs a combination of rule-based and statistical methods to extract relationships between entities. For example, extracting "due to decreased cooling water flow, the motor overheated" from a maintenance report establishes the relationship: [decreased cooling water flow] → [motor overheating]. Simultaneously, the system also utilizes co-occurrence frequency analysis to discover potential associations. Let C(e1,e2) be the number of times two events e1 and e2 co-occur in historical records, and C(e1) and C(e2) be the number of times they occur individually, respectively. Then, their correlation strength is defined as:

[0052] When R(e1,e2)≥θ (θ is a threshold, such as 0.2), it is determined that there is a causal or co-occurring relationship between the two, and an edge is added to the knowledge graph.

[0053] (3) Knowledge graph dynamic update mechanism

[0054] Whenever a new fault occurs, the system extracts the feature vector of that event. The similarity is then compared with existing fault modes in the knowledge graph. The similarity calculation formula is as follows:

[0055] ,in, and These represent the i-th eigenvalues ​​in the old and new eigenvectors, respectively.

[0056] If the similarity S ≥ τ (τ is a set threshold, such as 0.7), it is determined to be a known fault mode, and only the weights of existing nodes are updated; otherwise, it is considered a new fault mode, and a new node is added to the graph.

[0057] The weight update formula is as follows: , where γ is the learning rate, which determines the update magnitude, and takes a value of 0.1-0.5.

[0058] 3. Health Status Assessment Module

[0059] The health status assessment module calculates the device health index based on entities and relationships in a knowledge graph, combined with real-time sensor data. This process consists of two steps: local health scoring and global comprehensive assessment.

[0060] (1) Local health score

[0061] For each sensor node, the system assigns an initial weight based on its historical performance and its correlation with faults. Combined with the current normalized values Calculate the score Weight The determination method is as follows: ,in, This represents the correlation strength between the j-th sensor and the k-th fault mode (from the knowledge graph). It is the probability of the occurrence of the kth failure mode (based on historical statistics).

[0062] (2) Calculation of global health index

[0063] The system performs a weighted summation of the scores from all sensors to obtain the overall health index of the equipment. To reflect the impact of time on health status, a time decay factor λ is introduced: Where t represents the time interval since the last maintenance (in hours), and λ controls the decay rate, with a typical value of 0.01-0.1.

[0064] Ultimately, health status is divided into four levels: normal ( ≥0.8); Warning (0.6≤ <0.8); Abnormal (0.4≤ <0.6); Critical fault ( <0.4). A corresponding warning signal will be output, and maintenance actions will be recommended.

[0065] 4. Fault Intelligent Matching and Explanation Module (Intelligent Matching and Fault Explanation Mechanism)

[0066] When the system detects that a device has entered a "warning" or "abnormal" state, it initiates an intelligent matching process to find the most likely fault mode. The matching process is based on a path search algorithm in a knowledge graph, employing an improved Dijkstra's algorithm. The cost function for each edge is defined as: cost(edge) = 1 - R(s,f), where R(s,f) represents the similarity between the sensor value and the fault feature, and is calculated in the same way as above.

[0067] Once the optimal path is found, the system generates a corresponding fault explanation text, such as: "Based on the current increase in vibration amplitude of bearing A3 (0.8g), combined with knowledge graph analysis, the most likely cause of the fault is insufficient lubrication. It is recommended to check the lubrication system." The explanation information not only includes possible causes, but also suggested maintenance actions, the location of relevant components, links to similar historical cases, and other content.

[0068] (1) Fault feature extraction and pattern coding

[0069] The system extracts key features from the data collected by the sensors, including: the deviation between the current value and the historical mean (ΔV); the time series trend (such as the upward / downward slope); the fluctuation amplitude (standard deviation σ); the covariance relationship with other sensors; and the energy proportion of specific frequency components (FFT analysis results).

[0070] For example, the amplitude of vibration in the X-axis direction of a triaxial accelerometer can be expressed as: ,in It is the original signal within a certain time window. If A reading significantly higher than the normal range may indicate bearing wear or imbalance.

[0071] To facilitate subsequent matching, the system vectorizes the extracted features and performs standardization: ; where μ_j and σ_j are the historical mean and standard deviation of the j-th feature, respectively.

[0072] (2) Fault mode representation in knowledge graph

[0073] Each fault mode in the knowledge graph consists of a set of feature vectors and semantic descriptions. For example, the fault mode "bearing inner ring wear" may be associated with the following features: high-frequency energy concentration in the vibration spectrum; slow temperature rise; increased peak-to-peak value of vibration; and intensified hydraulic pressure fluctuations.

[0074] The system encodes these features into a structured fault template: ,in: It is the feature vector of the k-th fault mode; It is a natural language description used to generate the final explanatory text.

[0075] (3) Fault matching algorithm

[0076] To achieve accurate matching, the system uses cosine similarity as a metric:

[0077] The formula calculates the similarity between the current device state and a known fault mode. If Sim ≥ θ (θ is a set threshold, ranging from 0.7 to 0.8), the current state is considered to be highly similar to the fault mode.

[0078] The system also incorporates a weight adjustment mechanism. Certain features are more sensitive to specific faults and are therefore assigned higher weights. The improved similarity formula is:

[0079] ;in" " indicates element-wise multiplication.

[0080] (4) Path reasoning and fault explanation generation

[0081] Once the most likely failure mode is identified, the system uses a knowledge graph to perform path reasoning, finding the root cause link from the current state to the failure. For example: [Increased motor current] → [Increased load] → [Bearing overheating] → [Lubrication failure] → [Increased wear].

[0082] The system employs an improved Breadth-First Search (BFS) algorithm to find the shortest path from the current node to a high-probability failure mode. The weight of each edge is determined by its historical co-occurrence frequency. Where C(e1,e2) is the number of times the two events occur together, and C(e1) is the number of times event e1 occurs alone.

[0083] Finally, the system generates structured explanatory text based on the path information, such as: "Currently, an increase in motor M1 current is detected. Based on historical data analysis, the most likely cause is a sudden increase in load leading to a rise in bearing temperature. It is recommended to check whether the transmission components are stuck and to confirm whether the lubrication system is working properly."

[0084] In addition, the system will also list relevant historical case numbers, maintenance record links, and recommended spare parts lists to improve diagnostic efficiency and accuracy.

[0085] 5. Edge-Cloud Collaborative Update Module (Edge Computing and Cloud Collaboration Mechanism)

[0086] Considering the limited communication bandwidth in industrial settings, the system adopts an edge-cloud collaborative architecture. Edge nodes are responsible for real-time data processing and preliminary health assessments, uploading only critical data to the cloud to reduce communication load.

[0087] (1) Functional module division of edge nodes

[0088] Edge nodes are deployed in industrial sites and have certain computing resources. Their main functions include: data preprocessing and feature extraction; local knowledge base query; preliminary health scoring and alarm generation; voice interaction and local display.

[0089] A lightweight model is used for health scoring at the edge, and its calculation formula is as follows: ;in: These are the normalized sensor values; is the knowledge graph weight corresponding to the sensor; m is the number of sensors participating in the evaluation.

[0090] like < (e.g., 0.6), then a local alarm will be triggered and indicated by a buzzer or LED light; if < If the value is 0.4, the event will be uploaded to the cloud via the MQTT protocol for further analysis.

[0091] (2) Cloud-based functions and knowledge graph management

[0092] The cloud platform possesses enhanced computing power and is responsible for the following tasks: knowledge graph updates and optimization; multi-device collaborative analysis; long-term trend prediction; user access control and remote access; the cloud platform uses the Neo4j graph database to store the knowledge graph, and its core operations include:

[0093] Insert a new entity: CREATE(:Equipment{id:"M1",name:"Motor"});

[0094] Establishing relationships:

[0095] MATCH(a:Sensor),(b:FailureMode)CREATE(a)-[:CAUSE]->(b);

[0096] Query path:

[0097] MATCHp=(s:Sensor)-[*..5]->(f:FailureMode)RETURNpLIMIT10;

[0098] Each time new device anomaly data is received, the cloud performs an incremental learning process, updating the weights of relevant nodes in the knowledge graph: ; where γ is the learning rate, which controls the update intensity.

[0099] (3) Communication protocol and data synchronization strategy

[0100] Low-overhead communication between the edge and the cloud is achieved using the MQTT protocol, and the transmitted data is in JSON format, mainly including: device ID; timestamp; normalized sensor value; local health score; and fault mode matching result.

[0101] To reduce bandwidth consumption, the system employs a differential upload strategy to minimize data transmission, uploading data only under the following conditions: health score falls below a set threshold; a sensor value exceeds the historical 95th percentile; a new fault mode is identified; or the user actively requests an upload. Furthermore, the system incorporates a heartbeat mechanism, sending status reports at regular intervals to ensure connection stability.

[0102] In another embodiment, it also includes a security and access control module (security and access control mechanism), which is designed to ensure the security and data privacy of the system. The system has a complete security mechanism covering multiple aspects such as identity authentication, data encryption, access control, and log auditing.

[0103] (1) Identity authentication and access control

[0104] The system uses the OAuth 2.0 authorization protocol to implement user authentication. The user login process is as follows:

[0105] 1. The user enters their username and password;

[0106] 2. The system calls the authentication service to verify credentials;

[0107] 3. Upon successful authentication, a JWT token is returned;

[0108] 4. Subsequent API requests should include this token to complete authentication.

[0109] Role Permission Description administrator Users can be added / deleted, permission configurations modified, and all devices can be viewed. engineer Users can view equipment status, submit maintenance work orders, and access the knowledge graph. Regular users Only device status and basic alarm information can be viewed.

[0110] Permission information is stored in a MySQL database and access efficiency is improved by caching it with Redis.

[0111] (2) Data encryption and transmission security

[0112] The system employs the TLS 1.3 protocol for network communication encryption to ensure that data is not eavesdropped on or tampered with during transmission. Specific measures include: using the HTTPS protocol to access the web interface; enabling SSL / TLS encryption for MQTT communication; using encrypted channels for database connections; and storing sensitive fields (such as passwords) using AES-256 encryption.

[0113] For data exchange between edge nodes and the cloud, the system uses the PBKDF2 algorithm for key derivation to increase decryption difficulty: key=PBKDF2(password,salt,iterations,key_length), where: password is the user password; salt is a random salt value; iterations controls the number of iterations (recommended ≥10000); key_length is the desired key length.

[0114] (3) Operation logs and auditing mechanisms

[0115] The system records all user actions, including login, data access, parameter modification, and maintenance operations.

[0116] (4) API interface security control

[0117] All RESTful API interfaces provided by the system require token authentication. Additionally, rate limiting is implemented to prevent DDoS attacks.

[0118] Maximum number of requests per minute: ≤50 for regular users, ≤200 for engineers;

[0119] IP blacklist mechanism: Automatically blocks IP addresses that frequently fail to log in;

[0120] Request signature verification: to prevent replay attacks.

[0121] This embodiment provides a complete IoT-based equipment operation knowledge graph construction and health status assessment system, covering the entire process from data acquisition, preprocessing, knowledge graph construction, health assessment, fault interpretation to visual interaction. The system achieves dynamic perception and intelligent diagnosis of equipment status by integrating real-time sensor data and historical knowledge. By introducing various mathematical models and algorithms (such as moving average, exponential smoothing, Z-score, Dijkstra's path search, and time decay factor), the system's adaptability and accuracy are enhanced. Simultaneously, the system possesses good scalability and security, making it suitable for equipment health management needs in various complex industrial scenarios.

[0122] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing an equipment operation knowledge graph and assessing its health status based on the Internet of Things, characterized in that, Includes the following steps: Real-time operating data of the device is collected by distributed sensors, and the real-time operating data is subjected to sliding window filtering and smoothing to generate normalized sensor values. Based on sensor values ​​and historical fault records, a knowledge graph is constructed that includes equipment entities, component entities, fault mode entities and their relationships. Calculate the device's health index based on the entity relationships in the knowledge graph and current sensor data; When the health index is lower than the warning threshold, the knowledge graph is used to perform path reasoning, match the current device status with historical fault patterns, and generate fault explanation text; Through the collaboration mechanism between edge computing nodes and the cloud, the health index and fault explanations are uploaded to the cloud, and the knowledge graph weights are optimized based on the cloud knowledge graph update mechanism.

2. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The formula for the sliding window filtering is: ； in, Here is the original data at time t, and n is the window size. This is the filtered output value.

3. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The formula for calculating the health index is as follows: ; in, The result is a weighted summation of the sensor values, where λ is the time decay coefficient and t is the time interval since the last maintenance.

4. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 3, characterized in that: The time decay coefficient λ ranges from 0.01 to 0.

1.

5. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The construction of the knowledge graph includes: Extract fault mode entities from maintenance logs and establish causal relationship edges between fault mode entities and component entities.

6. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The generation of the fault explanation text includes: Based on the path reasoning results, nodes in the faulty link are extracted, and natural language descriptions containing maintenance suggestions are generated by combining historical cases.

7. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The edge computing node and cloud collaboration mechanism includes: Normalized sensor values ​​and health indices are transmitted via the MQTT protocol, and a differential upload strategy is used to reduce data transmission volume.

8. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The fault mode matching uses the cosine similarity calculation formula: ； For the current feature vector, Let be the feature vector of the k-th fault mode.

9. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The knowledge graph update mechanism includes: When the similarity S of a new fault mode is greater than or equal to τ, where τ is a set threshold, the weights of existing nodes are updated; otherwise, a new node is added to the knowledge graph.

10. The method for constructing an equipment operation knowledge graph and assessing health status based on the Internet of Things as described in claim 1, characterized in that: The path reasoning includes: A breadth-first search is used to find the shortest path from the current node to the high-probability failure mode, and the weight of each edge is determined by the historical co-occurrence frequency.

Images