A method for intelligently storing maintenance data of medical equipment

Abstract

The present invention proposes an intelligent storage method for maintenance data of medical equipment, which includes: using sensors to collect the operation data of medical equipment, deploying microprocessors or embedded computing devices at the edge nodes of each medical device; segmenting the data according to the time window, constructing a physical model of the medical equipment to obtain the predicted value, identifying abnormal conditions, and setting the fault threshold of each characteristic value; constructing a fault knowledge graph of the medical equipment; constructing an intelligent semantic retrieval based on the knowledge graph to obtain the required maintenance records and equipment operation data; automatically generating a maintenance plan based on the fault prediction data of the physical model and the fault knowledge graph of the medical equipment; and protecting the privacy of the data in the central storage system. The present invention not only improves the utilization rate and retrieval efficiency of the data, but also enhances the prediction ability of equipment failures, reduces maintenance costs, ensures the security and privacy protection of data, and improves the maintenance efficiency and reliability of medical equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent management of maintenance data, and in particular relates to an intelligent storage method for maintenance data of medical equipment. Background Art

[0002] With the widespread use of medical equipment and technological advances, modern medical equipment plays a vital role in the daily operations of hospitals and clinics. These devices include MRI, CT scanners, electrocardiographs, ultrasound equipment, etc. The reliability and accuracy of their operation directly affect the quality of medical diagnosis and treatment. Therefore, the maintenance and repair of medical equipment is of vital importance. Traditional medical equipment maintenance methods mainly rely on regular inspections and post-failure repairs, which have many defects, such as high maintenance costs, untimely fault detection, and long equipment downtime. At the same time, with the increase in equipment usage and the accumulation of operating data, how to efficiently manage and utilize this data has become a major challenge in medical equipment maintenance.

[0003] Existing methods for storing medical equipment maintenance data are mostly centralized storage, which mainly has the following problems: First, the data storage and management methods lack intelligence, resulting in low data utilization; second, it is impossible to monitor the equipment operation status in real time, lack of predictive maintenance methods, and often maintenance can only be carried out after the equipment fails, affecting the continuity of medical services; third, data retrieval efficiency is low, and it is difficult for maintenance personnel to quickly obtain the required maintenance records and equipment operation data; fourth, data security and privacy protection are insufficient. Medical equipment operation data involves patient privacy and confidential information of equipment operation, which requires strict security protection measures. Summary of the invention

[0004] The purpose of the present invention is to design an intelligent storage method for maintenance data of medical equipment. By combining predictive maintenance, knowledge graph and edge computing technology, the problems existing in data storage and management are solved, and the intelligent storage and efficient management of medical equipment maintenance data are realized. The present invention not only improves the utilization rate and retrieval efficiency of data, but also enhances the ability to predict equipment failures, reduces maintenance costs, ensures data security and privacy protection, and improves the maintenance efficiency and reliability of medical equipment.

[0005] In order to achieve the above object, the present invention provides a method for intelligently storing maintenance data of medical equipment, the method comprising:

[0006] S1. Install sensors on each medical device, use the sensors to collect the operating data of the medical device, and deploy microprocessors or embedded computing devices at the edge nodes of each medical device to process and store the data collected by the sensors in real time. Clean and pre-process the collected data at the edge nodes, and transmit the pre-processed data to the central storage system through the wireless network.

[0007] S2. The data of the central storage system is segmented according to time windows, each time window is of length T. The data in each time window T is subjected to feature extraction, and a physical model of the medical equipment is constructed based on the features to obtain a predicted value. The abnormal situation is identified by comparing the actual change rate with the change rate predicted by the physical model. At the same time, the fault threshold of each characteristic value is set based on historical operation data and empirical knowledge. If a certain characteristic value exceeds the set threshold range, it is determined to be a fault. Then, according to the fault type and characteristic value, combined with the physical model of the equipment and historical maintenance records, fault diagnosis is performed to determine the possible cause of the fault. According to the diagnosis result of the fault cause, a detailed maintenance plan is automatically generated. After the maintenance is completed, the maintenance personnel record the actual maintenance process and results in the maintenance management system and feedback to the central storage system.

[0008] S3. Perform structural processing based on the operation data of medical equipment, the fault prediction data of the physical model, and the historical maintenance records to build a fault knowledge graph of medical equipment;

[0009] S4. Build intelligent semantic retrieval based on knowledge graph to obtain the required maintenance records and equipment operation data;

[0010] S5. Automatically generate maintenance plans based on the fault prediction data of the physical model and the fault knowledge graph of the medical equipment;

[0011] S6. Build a data security and privacy protection mechanism to protect the privacy of data in the central storage system.

[0012] Furthermore, in the central storage system, the received pre-processed data is stored in a structured format; the cleaning and pre-processing include data cleaning, data filtering processing and data compression processing.

[0013] Further, in said S2, said feature extraction includes performing statistical feature extraction on value, maximum value, minimum value, standard deviation, rate of change, kurtosis and skewness;

[0014] Among them, the mean value μ is calculated as follows:

[0015]

[0016] Where N is the number of data points in the time window, x i is the i-th data point;

[0017] The maximum value x max , calculated as follows:

[0018] x max =max(x1,x2,…,x i ,x N )

[0019] The minimum value x min , calculated as follows:

[0020] x min =min(x1,x2,…,x i ,x N )

[0021] The standard deviation σ is calculated as follows:

[0022]

[0023] The rate of change Δ is calculated as follows:

[0024]

[0025] Among them, x N and x1 represent the last and first data points in the time window, respectively;

[0026] The kurtosis K is calculated as follows:

[0027]

[0028] The skewness, S, is calculated as follows:

[0029]

[0030] Further, in said S2, the data structure includes device ID, sensor type, timestamp, data value, fault type, maintenance operation and maintenance result.

[0031] Furthermore, the entity types in the knowledge graph are defined to include equipment, sensors, data, faults, maintenance operations, and maintenance results. The relationships between entities are defined. A graph database is used to build a knowledge graph. Structured data is stored as nodes and edges in the graph database. A node is created for each entity, an edge is created for each relationship, and relevant attributes are attached to the nodes and edges. The knowledge graph is queried and updated through the query language of the graph database, and the knowledge graph is updated regularly.

[0032] Furthermore, the knowledge graph is dynamically updated according to the predicted value of the physical model, and new predicted fault nodes and corresponding maintenance suggestion nodes are added.

[0033] Furthermore, the step S4 specifically includes:

[0034] S401. Use natural language processing technology to build a query semantic model, convert the natural language query of the maintenance personnel into an executable knowledge graph query, and use word embedding technology to convert the vocabulary in the query sentence into a vector representation, represented as w, and use the sentence vectorization model to convert the entire query sentence into a vector representation, represented as q;

[0035] S402: Use the classification model to identify the query vector intent, as shown below:

[0036] intent=SVM(q)

[0037] Among them, SVM represents the classification model, q represents the query vector, and intent represents the identified query intent;

[0038] S403, generating a corresponding knowledge graph query statement according to the identified query intent;

[0039] S404, optimizing the generated query statement; wherein the optimization strategy includes index optimization, query cache and data sharding;

[0040] S405: Perform structured processing on the query results and return the processed results to the maintenance personnel;

[0041] S406: Sort and optimize the query results. Optimize the result sorting according to the query intent and historical query records, and return the most relevant results first. The formula of the feedback mechanism is as follows:

[0042] w new =w old +α(rw old )

[0043] Among them, w new represents the updated query weight, w old represents the original query weight, α represents the learning rate, and r represents the feedback result vector;

[0044] S407, according to the prediction value of the physical model in step S2, optimize the query model, give priority to returning the predicted potential fault information, update the query semantic model and classification model, and update the query process in combination with the latest prediction results and query records;

[0045] S408, dynamically update the new data and feedback results obtained during the query process to the knowledge graph, as shown below:

[0046] KG new =KG old +ΔKG

[0047] Among them, KG new Represents the updated knowledge graph, KG old represents the original knowledge graph, and ΔKG represents the newly added knowledge graph information;

[0048] S409. Use the visualization tool of the graph database to display the knowledge graph.

[0049] Furthermore, the S5 specifically includes:

[0050] S501, extracting data from the fault prediction results in S2 and the knowledge graph constructed in S3, and extracting features from the data; the data includes equipment ID, fault type, fault occurrence time, sensor data and historical maintenance records; the features include failure rate λ, equipment health index HI and mean repair time MTTR; wherein the failure rate λ is calculated as follows:

[0051]

[0052] The equipment health index HI is calculated as follows:

[0053]

[0054] The mean repair time MTTR is calculated as follows:

[0055]

[0056] Where n represents the number of failures and m represents the number of repairs;

[0057] S502, automatically generating detailed maintenance tasks according to the predicted fault type and characteristic data;

[0058] S503. Use linear programming model to optimize maintenance plan to ensure optimal allocation of resources and reasonable arrangement of tasks;

[0059] S504, automatically generate detailed maintenance plan documents;

[0060] S505, assigning the generated maintenance tasks to relevant maintenance personnel through the maintenance management system;

[0061] S506. The maintenance personnel perform maintenance operations according to the assigned tasks and record the actual maintenance process and results in the maintenance management system;

[0062] S507. Continuously update and optimize the prediction model and maintenance plan generation algorithm based on the maintenance feedback record.

[0063] Furthermore, the objective function of the linear programming model is expressed as follows:

[0064]

[0065] Among them, C i represents the cost of maintaining step i, represents the time of step i, and k represents the total number of steps;

[0066] The constraints of the objective function are as follows:

[0067]

[0068] Among them, R i represents the resources required for step i, R max Indicates the maximum amount of available resources;

[0069] The task allocation formula for the maintenance task is as follows:

[0070]

[0071] Among them, L p represents the task load of maintenance personnel p, and P is the set of tasks assigned to maintenance personnel p;

[0072] The optimal allocation task set P is expressed as follows:

[0073]

[0074] Furthermore, the S6 specifically includes:

[0075] S601, generate a key k at the edge computing node, use the key k to perform AES encryption on the data P to obtain C, encrypt the key k using an asymmetric encryption algorithm and transmit it to the central storage system, transmit the encrypted data C to the central storage system, decrypt it using the private key of the asymmetric encryption algorithm in the central storage system to obtain the key k, and use the key k to decrypt the encrypted data C to obtain the original data P;

[0076] S602. Use a role-based access control mechanism to ensure that only authorized users can access and operate specific data;

[0077] S603. Use data anonymization technology to protect sensitive information during data storage and processing;

[0078] S604. During the data analysis process, differential privacy technology is used to protect data privacy;

[0079] S605. Perform regular data backup to ensure rapid recovery in case of data loss or damage;

[0080] S606. During the data access and operation process, all operation logs are recorded.

[0081] The beneficial technical effects of the present invention are at least as follows:

[0082] (1) The present invention addresses the problem of the lack of real-time monitoring and predictive maintenance methods in the prior art. The present invention uses a long short-term memory (LSTM) network and a random forest algorithm to perform time series analysis and fault classification prediction on the operating data of medical equipment. Through these predictive maintenance algorithms, it is possible to identify possible fault points of the equipment in advance, take preventive measures in time, and reduce the unplanned downtime and maintenance costs of the equipment. Advanced machine learning algorithms are used for predictive maintenance, and data storage strategies are dynamically adjusted to ensure priority storage and rapid access to key data, thereby improving the reliability of equipment operation and maintenance efficiency.

[0083] (2) The present invention aims to solve the problem of low data retrieval efficiency and difficulty for maintenance personnel to quickly obtain required information in the prior art. The present invention constructs a medical equipment maintenance knowledge graph to structure the equipment operation data, maintenance records and fault logs. The BERT model is combined with semantic retrieval and intelligent analysis to enable maintenance personnel to quickly find relevant maintenance records and equipment operation data, thereby improving data retrieval efficiency. The knowledge graph is constructed and combined with natural language processing technology to achieve intelligent storage and efficient retrieval of data, automatically generate maintenance suggestions and fault solutions, and improve data utilization and maintenance efficiency.

[0084] (3) The present invention addresses the problem that the data storage method in the prior art lacks intelligence and cannot monitor the operating status of the equipment in real time. The present invention deploys computing resources at the edge nodes of medical equipment to process and store equipment operating data in real time. By performing preliminary screening and preprocessing of data at the edge nodes, only key data is uploaded to the central storage system, which reduces the pressure on central storage and realizes real-time monitoring of equipment status and fault alarm. By using edge computing technology, equipment operating data is processed in real time, and preliminary screening and preprocessing of data are performed, which realizes intelligent data storage and real-time monitoring, and enhances the real-time detection and alarm capabilities of equipment failures.

[0085] (4) This invention combines predictive maintenance, knowledge graph and edge computing technology to address the deficiencies in the prior art and proposes an intelligent storage method for maintenance data of medical equipment. This method not only improves the management and utilization efficiency of data, but also enhances the ability to predict equipment failures and the security of data. It is highly innovative and practical and can effectively improve the maintenance efficiency and reliability of medical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0087] Figure 1 The present invention provides a flow chart of an intelligent storage method for maintenance data of medical equipment. DETAILED DESCRIPTION

[0088] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0089] In one or more embodiments, Figure 1 As shown, a method for intelligently storing maintenance data of medical equipment is disclosed, and the method comprises the following steps S1-S6:

[0090] S1. Install sensors on each medical device and use the sensors to collect the operating data of the medical device. At the same time, deploy microprocessors or embedded computing devices at the edge nodes of each medical device to process and store the data collected by the sensors in real time. Clean and pre-process the collected data at the edge nodes and transmit the pre-processed data to the central storage system via the wireless network.

[0091] Specifically, sensors are installed on each medical device to collect the operating data of the device. These sensors include but are not limited to temperature sensors, pressure sensors, current sensors, vibration sensors, etc. The data collection frequency is determined according to the operating status of the device and the data change rate. For example, for high-frequency changing data, the collection frequency can be set to once per second, and for low-frequency changing data, it can be set to once per minute. The collected data can be expressed as:

[0092] D(t)={d1(t),d2(t),…,d n (t)} (1)

[0093] Where D(t) represents the set of all sensor data collected at time t, d i (t) represents the data value collected by sensor i at time t.

[0094] Furthermore, a microprocessor or embedded computing device (such as Raspberry Pi) is deployed at the edge node of each medical device to process and store the data collected by the sensor in real time.

[0095] Edge computing nodes need to have the following functions:

[0096] Data acquisition interface: connect various sensors to obtain operation data in real time.

[0097] Data storage module: used to temporarily store real-time collected data. It is recommended to use local storage and buffer mechanisms.

[0098] Data processing module: performs preliminary data processing, including data cleaning, filtering and compression.

[0099] Furthermore, the data cleaning and preprocessing includes:

[0100] Data cleaning: Remove abnormal data caused by sensor failure or external interference, and use the sliding window averaging method for smoothing. Sliding window averaging method formula:

[0101]

[0102] in, is the data of sensor i after cleaning at time t, x i (j) is the raw data of sensor i at time j, and n is the window size.

[0103] Data filtering uses a low-pass filter to remove high-frequency noise. Low-pass filter formula:

[0104] y i (t) = αx i (t)+(1-α)y i (t-1) (3)

[0105] Among them, y i (t) is the filtered data of sensor i at time t, x i (t) is the raw data of sensor i at time t, and α is the filter coefficient (usually between 0.1 and 0.5).

[0106] Data compression uses differential coding to reduce the amount of data transmission. Differential coding formula:

[0107] d i (t) = x i (t)-x i (t-1) (4)

[0108] Among them, d i (t) is the compressed data of sensor i at time t, x i (t) is the original data of sensor i at time t.

[0109] Furthermore, the preprocessed data is transmitted to the central storage system via a wireless network (such as Wi-Fi or Bluetooth) to ensure the real-time and integrity of the data. The Transmission Control Protocol (TCP) is used for data transmission, and a retransmission mechanism is set to ensure that the data packet can be resent after it is lost. A data synchronization mechanism is set between the edge computing node and the central storage system to ensure data consistency. Data synchronization is performed at regular intervals (such as 1 minute). The transmitted data can be expressed as:

[0110] T(t)={d1(t),d2(t),…,d n (t)} (5)

[0111] Wherein, T(t) is the data set transmitted at time t.

[0112] Furthermore, the received pre-processed data is stored in a structured format (such as CSV or database) in the central storage system. Each data record should contain fields such as timestamp, sensor ID, data value, etc. to facilitate subsequent data analysis and processing. The following is an example of the database storage structure:

[0113]

[0114] Furthermore, a data backup mechanism is set up. The edge computing nodes and the central storage system perform data backup regularly (e.g., once a day) and back up the data to cloud storage or external storage devices. In the event of data loss or damage, the data can be restored from the backup to ensure data integrity and reliability.

[0115] S2. The data of the central storage system is segmented according to time windows, and the length of each time window is T. The data in each time window T is subjected to feature extraction, and a physical model of the medical equipment is constructed based on the features to obtain a predicted value. Abnormal conditions are identified by comparing the actual rate of change with the rate of change predicted by the physical model. At the same time, the fault threshold of each characteristic value is set based on historical operating data and empirical knowledge. If a characteristic value exceeds the set threshold range, it is determined to be a fault. Then, according to the fault type and characteristic value, combined with the physical model of the equipment and historical maintenance records, fault diagnosis is performed to determine the possible cause of the fault. Based on the diagnosis result of the fault cause, a detailed maintenance plan is automatically generated. After the maintenance is completed, the maintenance personnel record the actual maintenance process and results in the maintenance management system and feedback to the central storage system.

[0116] Specifically, the preprocessed data transmitted from the edge computing node to the central storage system is segmented according to time windows, and the length of each time window is T (such as 10 minutes).

[0117] Extract features from the data in each time window, including statistical features such as mean, maximum, minimum, standard deviation, rate of change, kurtosis and skewness. Statistical feature calculation formula:

[0118] Mean μ:

[0119]

[0120] Where N is the number of data points in the time window, x i is the i-th data point. A change in the mean can reflect the normal working state of the device or a deviation from the normal working state. For example, an increase in the mean of the temperature sensor data may indicate that the device is overheating.

[0121] Maximum x max :

[0122] x max =max(x1,x2,…,x N ) (7)

[0123] Extreme changes in maximum values ​​can reflect sudden failures. For example, an abnormal maximum value in pressure sensor data may indicate that the system pressure is too high and there is a risk of leakage.

[0124] Minimum value x min :

[0125] x min =min(x1,x2,…,x N ) (8)

[0126] Extreme changes in minimum values ​​can reflect abnormal conditions of the device. For example, an abnormal minimum value in voltage sensor data may indicate a power failure.

[0127] Standard deviation σ:

[0128]

[0129] An increase in standard deviation can indicate unstable equipment operation. For example, an increase in standard deviation in vibration sensor data can indicate loose or worn mechanical parts.

[0130] Rate of change Δ:

[0131]

[0132] Among them, x N x and x1 are the last and first data points in the time window, respectively. A sudden change in the rate of change can indicate a fault. For example, a sudden change in the rate of change of the temperature sensor data may indicate a cooling system fault.

[0133] Kurtosis K:

[0134]

[0135] An increase in kurtosis may indicate unusual peaks in the data. For example, high kurtosis in current sensor data may indicate a short or overload in a circuit.

[0136] Skewness S:

[0137]

[0138] Changes in skewness can indicate anomalies in the data distribution. For example, high skewness in pressure sensor data may indicate a bias in the system.

[0139] Furthermore, a physical model of the device is constructed, and the device status analysis is performed based on the extracted feature values, where the features here refer to the features extracted above.

[0140] Taking the temperature sensor as an example, we use the heat conduction equation for analysis:

[0141]

[0142] Where θ is temperature, t is time, α is thermal diffusion coefficient, is the Laplace operator.

[0143] Identify anomalies by comparing the actual temperature change rate with the temperature change rate predicted by the model:

[0144]

[0145] Among them, θ act is the actual temperature, θ mod is the temperature predicted by the model, and ∈ is the threshold.

[0146] Electrical equipment is analyzed using the circuit equation model:

[0147]

[0148] Among them, V(t) is voltage, I(t) is current, R is resistance, L is inductance, and C is capacitance.

[0149] Identify abnormal conditions by comparing actual voltage with model-predicted voltage:

[0150] |V act (t)-V mod (t)|>δ(16)

[0151] Among them, V act is the actual voltage, V mod is the voltage predicted by the model and δ is the threshold.

[0152] Furthermore, the fault threshold of each characteristic value is set based on historical operation data and empirical knowledge. For example, the normal operating range of the temperature sensor is [θ min ,θ max ].

[0153] If a characteristic value exceeds the set threshold range, it is judged as a fault.

[0154] Generate fault alarm information, including fault time, fault type, sensor ID, abnormal characteristic value, etc., and push it to maintenance personnel. Fault judgment formula:

[0155] Temperature fault determination:

[0156]

[0157] Furthermore, based on the fault type and characteristic value, combined with the physical model of the equipment and historical maintenance records, fault diagnosis is performed to determine the possible cause of the fault. Maintenance suggestions are generated, including components to be checked, parts to be replaced, specific maintenance steps, etc., and pushed to maintenance personnel.

[0158] Combining the physical model and the eigenvalue change pattern, the fault type and possible cause can be determined through pattern recognition technology. For example, for mechanical vibration data, the fault types such as bearing wear and balance failure can be identified through frequency domain analysis.

[0159] Preferably, after the maintenance is completed, the maintenance personnel record the actual maintenance process and results in the maintenance management system, including the replaced parts, resolved problems, unresolved problems, etc. This data will be fed back to the central storage system to update the historical maintenance records and be used to improve the fault prediction model and maintenance suggestion generation algorithm. The data record format example is as follows:

[0160]

[0161] Through the above detailed steps, a systematic predictive maintenance analysis process has been established, from data segmentation and feature extraction to fault prediction, alarm generation, fault diagnosis and maintenance suggestion generation, automatic generation and execution of maintenance plans, and then to data recording and feedback, forming a closed-loop fault prediction and maintenance management system. Each step is closely linked, providing technical support for the efficient and reliable maintenance of medical equipment.

[0162] S3. Perform structured processing based on the operation data of medical equipment, the fault prediction data of the physical model, and the historical maintenance records to build a fault knowledge graph for medical equipment.

[0163] Specifically, the medical equipment operation data, fault prediction data, and historical maintenance records collected in step 1 and step 2 are structured.

[0164] Define the data structure, including fields such as device ID, sensor type, timestamp, data value, fault type, maintenance operation, and maintenance result.

[0165] The data example is as follows:

[0166]

[0167] Define entity types in the knowledge graph, including equipment, sensors, data, faults, maintenance operations, maintenance results, etc.

[0168] Furthermore, relationships between entities are defined, for example, equipment includes sensors, sensors generate data, data triggers faults, faults trigger maintenance operations, maintenance operations generate maintenance results, etc.

[0169] Preferably, examples of entities and relationships are as follows:

[0170] entity:

[0171] Equipment: E i ;

[0172] Sensor: S ij ;

[0173] Data: D ijk ;

[0174] Fault: F l ;

[0175] Maintenance operation: M m ;

[0176] Maintenance result: R n ;

[0177] relation:

[0178] The device contains sensors: E i →S ij ;

[0179] Sensor generates data: S ij →D ijk ;

[0180] Data caused the failure: D ijk →F l ;

[0181] Fault triggers maintenance action: F l →M m ;

[0182] Maintenance operation produces maintenance result: M m →R n .

[0183] Furthermore, we use a graph database (such as Neo4j) to build a knowledge graph and store structured data as nodes and edges in the graph database. We create nodes for each entity, edges for each relationship, and attach relevant attributes to nodes and edges. The graph database construction process is as follows:

[0184] Create a device node:

[0185] CREATE(e:E{id:'D1'})

[0186] Create a sensor node and associate it with the device node:

[0187] CREATE(s:S{id:'Temp1',type:'Temp'})-[:BELONGS_TO]->(e)

[0188] Create a data node and associate it with the sensor node:

[0189] CREATE(d:D{value:36.5,timestamp:'2024-07-23 10:00'})-[:GENERATED_BY]->(s)

[0190] Create a faulty node and associate it with the data node:

[0191] CREATE(f:F{type:'Overheat'})-[:CAUSED_BY]->(d)

[0192] Create a maintenance operation node and associate it with the faulty node:

[0193] CREATE(m:M{operation:'Replace Fan'})-[:TRIGGERED_BY]->(f)

[0194] Create a maintenance result node and associate it with the maintenance operation node:

[0195] CREATE(r:R{result:'Solved'})-[:RESULTED_IN]->(m)

[0196] Furthermore, the query and update of the knowledge graph is realized through the query language of the graph database (such as Cypher). The knowledge graph is updated regularly to add new data nodes, fault nodes, maintenance operation nodes, and maintenance result nodes. Query example: Find all fault and maintenance records of a specific device:

[0197] MATCH(e:E{id:'D1'})-[:BELONGS_TO]->(s:S)-[:GENERATED_BY]->(d:D)-

[0198] [:CAUSED_BY]->(f:F)-[:TRIGGERED_BY]->(m:M)-[:RESULTED_IN]->(r:R)

[0199] RETURN e,s,d,f,m,r

[0200] Preferably, according to the fault prediction result in step 2, the knowledge graph is dynamically updated to add new predicted fault nodes and corresponding maintenance suggestion nodes. Dynamic update example: adding predicted fault and maintenance suggestion nodes:

[0201] CREATE(pf:F{type:'Predicted Overheat'})-[:PREDICTED_BY]->(pd:D{value:38.0,timestamp:'2024-07-23 11:00'})

[0202] CREATE(ma:M{operation:'Check Cooling System'})-[:RECOMMENDED_FOR]->(pf)

[0203] Preferably, a graph database visualization tool (such as Neo4j Bloom) is used to display the knowledge graph to help maintenance personnel intuitively view the operating status, fault records, and maintenance history of the equipment.

[0204] Implement analytical functions based on knowledge graphs, such as identifying failure modes and evaluating maintenance effects, to help optimize maintenance strategies. Visualization and analysis example: Displaying the knowledge graph of a specific device:

[0205] MATCH(e:E{id:'D1'})-[:BELONGS_TO]->(s:S)-[:GENERATED_BY]->(d:D)-

[0206] [:CAUSED_BY]->(f:F)-[:TRIGGERED_BY]->(m:M)-[:RESULTED_IN]->(r:R)

[0207] RETURN e,s,d,f,m,r

[0208] S4. Build intelligent semantic retrieval based on knowledge graph to obtain the required maintenance records and equipment operation data, including:

[0209] S401. Use natural language processing technology to build a query semantic model, convert the natural language query of the maintenance personnel into an executable knowledge graph query, and use word embedding technology to convert the vocabulary in the query sentence into a vector representation, represented as w, and use the sentence vectorization model to convert the entire query sentence into a vector representation, represented as q, as shown below:

[0210] q=BERT(query) (18)

[0211] Among them, BERT is a sentence vectorization model, query is the query statement of the maintenance personnel, and q is the query vector.

[0212] S402: Use a classification model (such as a support vector machine, SVM) to perform intent recognition on the query vector, as shown below:

[0213] intent=SVM(q)

[0214] Among them, SVM represents the classification model, q represents the query vector, and intent represents the identified query intent;

[0215] S403. Generate a corresponding knowledge graph query statement (such as a Cypher query statement) according to the identified query intent; for example, for the query intent of querying the device failure history records, generate the following Cypher query statement:

[0216] MATCH (e: device)-[: include]->(s: sensor)-[: generate]->(d: data)-[: trigger]->(f: fault)

[0217] WHERE e.id='D1'

[0218] RETURN e,s,d,f;

[0219] S404, optimizing the generated query statement; wherein the optimization strategy includes index optimization, query cache and data sharding; optimization formula:

[0220] Q opt =Optimize(Q) (20)

[0221] Among them, Q opt is the optimized query statement, and Optimize is the optimization function. Execute the optimized query statement in the graph database to obtain the query result.

[0222] S405: Perform structured processing on the query results and return the processed results to the maintenance personnel; the result example is as follows:

[0223]

[0224] The processed results are returned to the maintenance personnel.

[0225] S406: Sort and optimize the query results. Optimize the result sorting according to the query intent and historical query records, and return the most relevant results first. The formula of the feedback mechanism is as follows:

[0226] w new =w old +α(rw old )(twenty one)

[0227] Among them, w new represents the updated query weight, w old represents the original query weight, α represents the learning rate, and r represents the feedback result vector;

[0228] S407. According to the predicted value of the physical model in step S2, the query model is optimized, the predicted potential fault information is returned preferentially, the query semantic model and the classification model are updated, and the query process is updated in combination with the latest prediction results and query records; the optimization formula is:

[0229] q opt =Optimize(q,p) (22)

[0230] Among them, q opt is the optimized query vector, q is the original query vector, p is the prediction result vector, and Optimize is the optimization function.

[0231] S408, dynamically update the new data and feedback results obtained during the query process to the knowledge graph, as shown below:

[0232] KG new =KG old +ΔKG (23)

[0233] Among them, KG new Represents the updated knowledge graph, KG old represents the original knowledge graph, and ΔKG represents the newly added knowledge graph information;

[0234] S409. Use graph database visualization tools to display knowledge graphs. Specifically, use graph database visualization tools (such as Neo4j Bloom) to display knowledge graphs to help maintenance personnel intuitively view the operating status, fault records, and maintenance history of the equipment. Implement analysis functions based on knowledge graphs, such as identifying fault modes, evaluating maintenance effects, etc., to help optimize maintenance strategies.

[0235] Preferred, visualization and analysis example: Displaying the knowledge graph for a specific device:

[0236] MATCH(e:device{id:'D1'})-[:contains]->(s:sensor)-[:generates]->(d:data)-[:triggers]->(f:fault)-[:trigger]->(m:maintenance operation)-[:produces]->(r:maintenance result)

[0237] RETURN e,s,d,f,m,r.

[0238] S5. Automatically generate maintenance plans based on the fault prediction data of the physical model and the fault knowledge graph of the medical equipment, including:

[0239] S501, extracting data from the fault prediction results in S2 and the knowledge graph constructed in S3, and extracting features from the data; the data includes equipment ID, fault type, fault occurrence time, sensor data and historical maintenance records; the features include failure rate λ, equipment health index HI and mean repair time MTTR; wherein the failure rate λ is calculated as follows:

[0240]

[0241] The equipment health index HI is calculated as follows:

[0242]

[0243] The mean repair time MTTR is calculated as follows:

[0244]

[0245] Where n represents the number of failures and m represents the number of repairs;

[0246] S502. Automatically generate detailed maintenance tasks based on the predicted fault type and characteristic data; Automatically generate detailed maintenance tasks based on the predicted fault type and characteristic data, including maintenance steps, required materials, estimated time, etc.

[0247] Formula to generate maintenance tasks:

[0248] Maintenance step set M:

[0249] M={m1,m2,…,m k} (27)

[0250] Description and time estimate for each maintenance step:

[0251]

[0252] in, is the estimated time of step i, and f is the function for time estimation based on fault type and historical data.

[0253] S503. Use a linear programming model to optimize the maintenance plan to ensure optimal allocation of resources and reasonable arrangement of tasks. Use a linear programming model to optimize the maintenance plan to ensure optimal allocation of resources and reasonable arrangement of tasks.

[0254] Objective function:

[0255]

[0256] Among them, C i is the cost of maintaining step i, is the time of step i.

[0257] Constraints:

[0258]

[0259] Among them, R i is the resource required for step i, R max The maximum available resource.

[0260] S504, automatically generate detailed maintenance plan documents;

[0261] Specifically, a detailed maintenance plan document is automatically generated, including equipment information, maintenance steps, required materials, estimated time, maintenance personnel, etc.

[0262] Document structure:

[0263] Device information: device ID, model, running time, etc.

[0264] Fault information: fault type, occurrence time, prediction basis, etc.

[0265] Maintenance steps: detailed description of each step, tools and materials required, estimated time, etc.

[0266] Maintenance personnel: Assign to specific maintenance personnel and indicate the tasks they are responsible for.

[0267] S505, assigning the generated maintenance tasks to relevant maintenance personnel through the maintenance management system;

[0268] Specifically, the generated maintenance tasks are assigned to relevant maintenance personnel through the maintenance management system (CMMS), and the tasks are tracked and managed through the system.

[0269] Task allocation formula:

[0270] Maintenance personnel task load L p :

[0271]

[0272] Among them, L p is the task load of maintenance personnel p, and P is the set of tasks assigned to maintenance personnel p.

[0273] Optimize the task set P:

[0274]

[0275] Ensure balanced task distribution by minimizing the maximum task load.

[0276] S506. The maintenance personnel perform maintenance operations according to the assigned tasks and record the actual maintenance process and results in the maintenance management system;

[0277] Specifically, maintenance personnel perform maintenance operations according to the assigned tasks and record the actual maintenance process and results in the maintenance management system, including the completion time of each step, problems encountered, solutions, etc. Maintenance feedback record formula:

[0278] Actual repair time T actual :

[0279]

[0280] Fault resolution rate R s :

[0281]

[0282] Maintenance effect evaluation:

[0283]

[0284] in, is the actual completion time of each maintenance step, E m To evaluate the maintenance effect.

[0285] S507. Continuously update and optimize the prediction model and maintenance plan generation algorithm based on the maintenance feedback record;

[0286] Specifically, based on maintenance feedback records, continuously update and optimize the prediction model and maintenance plan generation algorithm, and continuously improve the formula:

[0287] Update failure rate λ:

[0288] new

[0289]

[0290] Update the device health index HI new :

[0291]

[0292] Among them, n new is the number of new faults and the severity of the fault inew is the new fault severity.

[0293] S6. Building a data security and privacy protection mechanism to protect the privacy of data in the central storage system. S6 specifically includes:

[0294] S601, generate a key k at the edge computing node, use the key k to perform AES encryption on the data P to obtain C, encrypt the key k using an asymmetric encryption algorithm and transmit it to the central storage system, transmit the encrypted data C to the central storage system, decrypt it using the private key of the asymmetric encryption algorithm in the central storage system to obtain the key k, and use the key k to decrypt the encrypted data C to obtain the original data P;

[0295] Specifically, in the process of transmitting data from the edge computing node to the central storage system, a symmetric encryption algorithm (such as AES) is used to encrypt the data to ensure the security of the data during transmission.

[0296] Encryption formula:

[0297] C=E k (P)(38)

[0298] Among them, C is the encrypted data, E k is the encryption function using key k, and P is the original data.

[0299] Decryption formula:

[0300] P=D k (C)(39)

[0301] Among them, D k is the decryption function using key k.

[0302] S602. Use a role-based access control mechanism to ensure that only authorized users can access and operate specific data;

[0303] Specifically, in the central storage system, a role-based access control (RBAC) mechanism is adopted to ensure that only authorized users can access and operate specific data.

[0304] Role Definition:

[0305] Administrator: Has all permissions.

[0306] Maintenance personnel: can only view and edit maintenance-related data.

[0307] Data Analyst: Can only view data, not edit it.

[0308] Access Control Matrix:

[0309]

[0310] Access control formula:

[0311]

[0312] Among them, A is the access control function, u is the user, o is the object, and p is the permission.

[0313] S603. Use data anonymization technology to protect sensitive information during data storage and processing;

[0314] Specifically, during data storage and processing, data anonymization technology (such as k-anonymity) is used to protect sensitive information.

[0315] k-anonymity model:

[0316]

[0317] Among them, D is the dataset and k is the anonymization parameter.

[0318] k-anonymization steps:

[0319] Select the attribute to anonymize (such as patient ID).

[0320] These attributes are generalized so that each attribute value appears at least k times in the dataset.

[0321] Generalization formula:

[0322]

[0323] Among them, A i is the attribute to be anonymized and Gen is the generalization operation.

[0324] S604. During the data analysis process, differential privacy technology is used to protect data privacy;

[0325] Specifically, during the data analysis process, differential privacy technology is used to protect data privacy and ensure that the analysis results do not leak personal privacy information.

[0326] Differential privacy formula:

[0327] ∈-Differential Privacy:P(A(D)=x)≤e ∈P(A(D′)=x) (43)

[0328] Among them, A is the algorithm, D and D′ are adjacent data sets, and ∈ is the privacy parameter.

[0329] Steps to add noise:

[0330] A. Calculate the query results.

[0331] B. Add noise according to the Laplace mechanism:

[0332]

[0333] Among them, λ is the noise scale parameter.

[0334] Furthermore, differential privacy optimization: select an appropriate noise scale λ to achieve a balance between privacy protection and data accuracy. The optimal λ value is selected through empirical methods or automatic parameter tuning algorithms.

[0335] S605. Perform regular data backup to ensure rapid recovery in case of data loss or damage;

[0336] Specifically, regularly back up data to ensure quick recovery in the event of data loss or damage.

[0337] Data backup strategy:

[0338] Full backup: Perform a full data backup once a week.

[0339] Incremental backup: Perform incremental data backup once a day.

[0340] Data recovery steps:

[0341] Restore your data from the most recent full backup.

[0342] Apply all incremental backups to restore to the most recent state.

[0343] Backup verification formula:

[0344] H backup =Hash(D backup ) (45)

[0345] Among them, H backup is the hash value of the backup data, which is used to verify the integrity of the backup data. backup To back up data.

[0346] S606. During the data access and operation process, all operation logs are recorded.

[0347] Specifically, during data access and operation, all operation logs are recorded to ensure that each operation can be tracked and audited.

[0348] Log record content: operation time: t; operation user: u; operation type: op; operation object: o.

[0349] Logging formula:

[0350] L=(t,u,op,o) (46)

[0351] Log audit steps:

[0352] Check the operation log regularly to detect abnormal operations.

[0353] Trace the source of abnormal operations through logs and take corresponding security measures.

[0354] Anomaly detection formula:

[0355]

[0356] The data security and privacy protection process forms a complete data security and privacy protection system from data encryption transmission, access control mechanism, data anonymization processing, differential privacy protection, data backup and recovery, to log audit and monitoring. Each step is closely linked, providing technical support for the efficient and secure management of medical device data.

[0357] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for intelligently storing maintenance data of medical equipment, characterized in that: The method comprises: S1. Install sensors on each medical device, use the sensors to collect the operating data of the medical device, and deploy microprocessors or embedded computing devices at the edge nodes of each medical device to process and store the data collected by the sensors in real time. Clean and pre-process the collected data at the edge nodes, and transmit the pre-processed data to the central storage system through the wireless network. S2. The data of the central storage system is segmented according to time windows, each time window is of length T. The data in each time window T is subjected to feature extraction, and a physical model of the medical equipment is constructed based on the features to obtain a predicted value. The abnormal situation is identified by comparing the actual change rate with the change rate predicted by the physical model. At the same time, the fault threshold of each characteristic value is set based on historical operation data and empirical knowledge. If a certain characteristic value exceeds the set threshold range, it is determined to be a fault. Then, according to the fault type and characteristic value, combined with the physical model of the equipment and historical maintenance records, fault diagnosis is performed to determine the possible cause of the fault. According to the diagnosis result of the fault cause, a detailed maintenance plan is automatically generated. After the maintenance is completed, the maintenance personnel record the actual maintenance process and results in the maintenance management system and feedback to the central storage system. S3. Perform structural processing based on the operation data of medical equipment, the fault prediction data of the physical model, and the historical maintenance records to build a fault knowledge graph of medical equipment; S4. Build intelligent semantic retrieval based on knowledge graph to obtain the required maintenance records and equipment operation data; S5. Automatically generate maintenance plans based on the fault prediction data of the physical model and the fault knowledge graph of the medical equipment; S6. Build a data security and privacy protection mechanism to protect the privacy of data in the central storage system; In S2, the feature extraction includes statistical feature extraction of mean, maximum, minimum, standard deviation, rate of change, kurtosis and skewness; Among them, the mean value μ is calculated as follows: Where N is the number of data points in the time window, x i is the i-th data point; The maximum value x max , calculated as follows: x max =max(x1,x2,…,x i ,x N ) The minimum value x min , calculated as follows: x min =min(x1,x2,…,x i ,x N ) The standard deviation σ is calculated as follows: The rate of change Δ is calculated as follows: Among them, x N and x1 represent the last and first data points in the time window, respectively; The kurtosis K is calculated as follows: The skewness, S, is calculated as follows: The step S4 specifically includes: S401. Use natural language processing technology to build a query semantic model, convert the natural language query of the maintenance personnel into an executable knowledge graph query, and use word embedding technology to convert the vocabulary in the query sentence into a vector representation, represented as w, and use the sentence vectorization model to convert the entire query sentence into a vector representation, represented as q; S402: Use the classification model to identify the query vector intent, as shown below: intent=SVM(q) Among them, SVM represents the classification model, q represents the query vector, and intent represents the identified query intent; S403, generating a corresponding knowledge graph query statement according to the identified query intent; S404, optimizing the generated query statement; wherein the optimization includes index optimization, query cache and data sharding; S405, perform structured processing on the query results, and return the processed results to the maintenance personnel; S406: Sort and optimize the query results. Optimize the result sorting according to the query intent and historical query records, and return the most relevant results first. The formula of the feedback mechanism is as follows: w new =w old +α(r-w old ) Among them, w new represents the updated query weight, w old represents the original query weight, α represents the learning rate, and r represents the feedback result vector; S407, according to the prediction value of the physical model in step S2, optimize the query model, give priority to returning the predicted potential fault information, update the query semantic model and classification model, and update the query process in combination with the latest prediction results and query records; S408, dynamically update the new data and feedback results obtained during the query process to the knowledge graph, as shown below: KG new =KG old +ΔKG Among them, KG new Represents the updated knowledge graph, KG old represents the original knowledge graph, and ΔKG represents the newly added knowledge graph information; S409, using a visualization tool of a graph database to display a knowledge graph; The S5 specifically includes: S501, extracting data from the fault prediction results in S2 and the knowledge graph constructed in S3, and extracting features from the data; the data includes equipment ID, fault type, fault occurrence time, sensor data and historical maintenance records; the features include failure rate λ, equipment health index HI and mean repair time MTTR; wherein the failure rate λ is calculated as follows: The equipment health index HI is calculated as follows: The mean repair time MTTR is calculated as follows: Where n represents the number of failures and m represents the number of repairs; S502, automatically generating detailed maintenance tasks according to the predicted fault type and characteristic data; S503. Use linear programming model to optimize maintenance plan to ensure optimal allocation of resources and reasonable arrangement of tasks; S504, automatically generate detailed maintenance plan documents; S505, assigning the generated maintenance tasks to relevant maintenance personnel through the maintenance management system; S506. The maintenance personnel perform maintenance operations according to the assigned tasks and record the actual maintenance process and results in the maintenance management system; S507. Continuously update and optimize the prediction model and maintenance plan generation algorithm based on the maintenance feedback record.

2. The intelligent storage method for maintenance data of medical equipment according to claim 1, characterized in that: In the central storage system, the received pre-processed data is stored in a structured format; the cleaning and pre-processing include data cleaning, data filtering and data compression.

3. The intelligent storage method for maintenance data of medical equipment according to claim 1, characterized in that: In said S2, the data structure includes device ID, sensor type, timestamp, data value, fault type, maintenance operation and maintenance result.

4. The intelligent storage method for maintenance data of medical equipment according to claim 3, characterized in that: Define the entity types in the knowledge graph, including equipment, sensors, data, faults, maintenance operations, and maintenance results. Define the relationships between entities. Use a graph database to build a knowledge graph. Store structured data as nodes and edges in the graph database. Create a node for each entity and an edge for each relationship. Attach relevant attributes to the nodes and edges. Query and update the knowledge graph through the query language of the graph database. Update the knowledge graph regularly.

5. The intelligent storage method for maintenance data of medical equipment according to claim 4, characterized in that: The knowledge graph is dynamically updated according to the predicted value of the physical model, and new predicted fault nodes and corresponding maintenance suggestion nodes are added.

6. The intelligent storage method for maintenance data of medical equipment according to claim 1, characterized in that: The objective function of the linear programming model is expressed as follows: Among them, C i represents the cost of maintenance step i, T mi represents the time of step i, and k represents the total number of steps; The constraints of the objective function are as follows: Among them, R i represents the resources required for step i, R max Indicates the maximum amount of available resources; The task allocation formula for the maintenance task is as follows: Among them, L p represents the task load of maintenance personnel p, and P is the set of tasks assigned to maintenance personnel p; The optimal allocation task set P is expressed as follows:

7. The intelligent storage method for maintenance data of medical equipment according to claim 1, characterized in that: The S6 specifically includes: S601, generate a key k at the edge computing node, use the key k to perform AES encryption on the data P to obtain C, encrypt the key k using an asymmetric encryption algorithm and transmit it to the central storage system, transmit the encrypted data C to the central storage system, decrypt it using the private key of the asymmetric encryption algorithm in the central storage system to obtain the key k, and use the key k to decrypt the encrypted data C to obtain the original data P; S602. Use a role-based access control mechanism to ensure that only authorized users can access and operate specific data; S603. Use data anonymization technology to protect sensitive information during data storage and processing; S604. During the data analysis process, differential privacy technology is used to protect data privacy; S605. Perform regular data backup to ensure rapid recovery in case of data loss or damage; S606. During the data access and operation process, all operation logs are recorded.

Images