Safety valve performance online monitoring method based on industrial network

Abstract

The invention relates to a safety valve performance online monitoring method based on an industrial network. The invention discloses a safety valve performance online monitoring method based on an industrial network. The safety valve performance online monitoring method comprises the following steps: S10, deploying hardware and a network system with multiple levels; s20, carrying out multi-source data acquisition, and carrying out fusion association processing on the acquired multi-source data; s30, monitoring and early warning of local data are carried out, and the monitored local early warning data are uploaded to the cloud; after receiving the local data, the cloud compares the local data with historical data and similar equipment cluster data to verify the local judgment accuracy; and S40, data consistency verification, parameter optimization and dynamic threshold calibration are respectively carried out, and results are fed back to the step S20 or the step S30. The safety valve performance online monitoring method based on the industrial network has the advantages of real-time performance, reliability and self-adaptability.

Description

Technical Field

[0001] The present application relates to the field of industrial network operation technology, and in particular to an online monitoring method for safety valve performance based on an industrial network. Background Art

[0002] In the fields of industrial automation and intelligent manufacturing, safety valves serve as the "last line of defense" for critical equipment such as pressure vessels and pipelines. Their operational reliability is directly related to production safety and equipment lifespan. However, current safety valve monitoring technology in industrial sites still faces the following core issues.

[0003] 1) Limitations of Traditional Manual Inspection. Traditional safety valve monitoring relies on regular manual inspections and offline calibration, which creates blind spots. For example, hidden faults such as valve sticking and internal leakage are difficult to detect during the non-opening and closing phases. Furthermore, manual inspections are unable to capture key parameters such as high-frequency vibration and transient temperature changes in real time. The result is delayed fault response, potentially leading to major accidents such as leaks and explosions. Statistics show that over 70% of accidents caused by safety valve failure in the petrochemical industry are due to delayed manual inspections.

[0004] 2) The singleness of existing monitoring technologies and the risk of false alarms. Existing technologies mostly rely on a single parameter (such as vibration spectrum or pressure fluctuations) to judge the valve status and lack the ability to fuse multi-source data. For example, vibration signals in high-temperature environments are susceptible to electromagnetic interference, and relying solely on vibration threshold alarms can result in a false alarm rate of over 30%. Existing algorithms (such as fixed threshold methods) cannot dynamically adapt to scenarios such as equipment aging and environmental changes. For example, frictional heat generated by wear of valve seals can cause abnormal coupling of temperature and vibration, but traditional models find it difficult to distinguish such operating conditions from real faults.

[0005] 3) Data silos and lack of collaborative analysis. Industrial field equipment data is often stored locally, lacking cross-device and cross-system collaborative analysis mechanisms. For example, it's difficult to identify clustered failure modes within a batch of safety valves (e.g., batch defects in sealing materials) using data from a single device. This results in delayed failure warnings and the inability to implement predictive maintenance.

[0006] 4) Insufficient reliability in complex environments. Harsh operating conditions such as high temperature, high pressure, and explosion-proofing place extremely high demands on the adaptability of sensors and network equipment. A technical shortcoming is the lack of redundant design tailored to the characteristics of safety valves.

[0007] 5) System response delays and lack of closed-loop optimization. Existing technologies use centralized cloud-based processing, resulting in data transmission and analysis delays of seconds, which cannot meet the millisecond-level response requirements of safety valves. The lack of a dynamic optimization mechanism that collaborates locally with the cloud prevents automatic adjustments based on equipment aging or environmental changes, resulting in a year-on-year increase in false alarm rates. Summary of the Invention

[0008] Based on this, the purpose of this application is to provide an online monitoring method for safety valve performance based on an industrial network, which is real-time, reliable and adaptable, and provides the advantage of a technological leap from "passive manual" to "active intelligent" for industrial safety valve monitoring.

[0009] In one aspect of the present application, a method for online monitoring of safety valve performance based on an industrial network is provided, comprising the steps of:

[0010] S10, deploying multi-layered hardware and network systems;

[0011] S20, collecting multi-source data and performing fusion and correlation processing on the collected multi-source data;

[0012] S30: Monitor and issue warnings for local data, and upload the original data of the monitored local warning data to the cloud. After receiving the local data, the cloud compares it with historical data and data from similar device clusters to verify the accuracy of the local judgment.

[0013] S40, respectively perform data consistency verification, parameter optimization, and dynamic threshold calibration, and feed the results back to step S20 or S30.

[0014] The industrial network-based online monitoring method for safety valve performance described in this application fuses and correlates multi-source data. In this step, connections are established between different types of data, eliminating the need for single-source processing and judgment, providing a basis for subsequent comprehensive judgment. Furthermore, by correlating multiple parameters, the problem of false alarms for single parameters is resolved. When performing early warning judgment, local processing and cloud processing are performed separately. In this monitoring method, the cloud-processed data only includes the early warning data from the local data. This reduces network bandwidth requirements, reduces the amount of data transmitted over the network, and reduces the possibility of packet loss during data transmission. Furthermore, when performing early warning monitoring of data from the cloud, the local early warning monitoring data can be verified, further ensuring the accuracy of the early warning monitoring. Finally, the data is calibrated and verified, and the verification results are fed back to other steps, forming a closed-loop control scheme. Thus, the online monitoring method of this application not only correlates data to ensure its relevance and integrity, but also compares and judges the early warning data through the cloud, ensuring its accuracy. Furthermore, the data is verified and calibrated, improving the accuracy of the early warning data.

[0015] Compared with the existing technology, this application solves the problem of single parameter false alarm, realizes collaborative closed-loop control between the edge and the cloud, ensures the signal transmission reliability of the industrial network, and guarantees the data integrity. It provides a technical foundation for safety valve monitoring under the industrial network from passive manual to active intelligent, and has promotion value.

[0016] Furthermore, step S20 includes:

[0017] S21, collecting vibration signals and temperature signals respectively;

[0018] S22, establish a coupled model of temperature and vibration

[0019] △T=k1·a 2 ·△t+k2·△P;

[0020] S23, outputs a robust temperature signal through the Kalman filter fusion algorithm to fuse multi-source data;

[0021] S24, according to the result of step S23, dynamic threshold adjustment of the vibration threshold

[0022]

[0023] Furthermore, step S30 includes:

[0024] S31, establish monitoring and early warning rules for local data, the monitoring and early warning rules are:

[0025] If T fusion >T threshold And Vib g >V threshold , then a level one alarm is triggered and immediate action is required;

[0026] If T fusion >T warning or Vib g >V warning , then a secondary alarm is triggered, requiring attention and short-term observation;

[0027] Among them, T fusion and Vib g respectively obtained by step S20;

[0028] S32, establish a cloud-based statistical verification model, which is:

[0029]

[0030] S33, when the local monitoring warning result is abnormal and the cloud verification result is <3σ, the rule modification is triggered;

[0031] Rule amendments are handled as follows:

[0032]

[0033] The obtained temperature coefficient α is used to update the temperature coefficient in step S20.

[0034] Furthermore, step S40 includes:

[0035] S41: Perform data consistency verification. The data consistency verification rules are as follows:

[0036]

[0037] If D KL >0.5, then the optimization of the temperature coefficient α is triggered;

[0038] S42, recalibrate the dynamic threshold so that the local rule approaches the optimal value on the cloud side; the rule is:

[0039] a new =a cloud (1-D KL +∈);

[0040] S43, through Bayesian optimization, the temperature coefficient α is forced to be aligned locally and in the cloud to ensure global consistency. The Bayesian optimization rule is:

[0041] Minimize L = α·false alarm rate + β·missing alarm rate + γ·||α-α cloud || 2 ;

[0042] The temperature coefficient α obtained in step S43 is used to update the temperature coefficient in step S20.

[0043] Furthermore, the rules of the Kalman filter fusion algorithm in step S23 are:

[0044]

[0045] Furthermore, step S10 includes:

[0046] S11, building multi-layer hardware and networks, including the perception layer, transport layer, edge layer, and cloud layer;

[0047] The perception layer includes vibration sensors, pressure sensors, and temperature sensors, which are connected to the network through industrial buses respectively;

[0048] The real-time data of the transport layer is transmitted via the TSN network, and its non-real-time data is transmitted via industrial Ethernet;

[0049] The edge gateway of the edge layer performs data preprocessing, protocol conversion, and local data storage;

[0050] The cloud stores historical data and provides visualization and early warning management;

[0051] S12, encrypt data and implement hierarchical control of permissions;

[0052] S13, perform time synchronization processing of the data. The algorithm of the time synchronization processing is:

[0053] t sync =t master +△t network ;

[0054] S14, through the dual-ring network redundancy design, ensures the reliability of key data transmission. Its redundant network protocol is

[0055]

[0056] Furthermore, in step S40, sampling checks are performed regularly to determine the consistency between local data and cloud alarm results;

[0057] The sampling data selects 10%-20% of the total alarm data.

[0058] Furthermore, the vibration sensor is mounted on the valve stem of the safety valve to collect high-frequency vibration signals;

[0059] The temperature sensor is installed on the surface of the safety valve to detect internal leakage or frictional heat generation in the valve;

[0060] The pressure sensor is used to monitor the inlet and outlet pressure difference of the discharge end of the safety valve;

[0061] The real-time data in S11 includes vibration signals, temperature signals, and pressure signals; the non-real-time data includes device information;

[0062] The vibration sensor, the temperature sensor, the pressure sensor and the network transmission equipment respectively comply with explosion-proof safety certification to ensure safe operation in a flammable and explosive environment.

[0063] Furthermore, in step S31, T threshold >T warning , V threshold >V warning ;

[0064] V warning With V threshold Establish proportional relationships to ensure logical consistency between warnings and alarms;

[0065] When T threshold and V threshold When dynamic adjustment occurs and decreases, T warning Based on T threshold Dynamic floating, V warning Based on V thresholdDynamic downward floating, the dynamic downward floating ratio is 70%-80%, in order to provide early warning of equipment degradation trends.

[0066] For better understanding and implementation, the present application is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flowchart of an exemplary industrial network-based online monitoring method for safety valve performance of this application;

[0068] Figure 2 This is a closed-loop logic diagram of an exemplary industrial network-based online monitoring method for safety valve performance in this application. DETAILED DESCRIPTION

[0069] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended only to facilitate the description of this application and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be understood as limiting this application. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0070] See also Figure 1 and Figure 2 , an exemplary method for online monitoring of safety valve performance based on an industrial network of the present application comprises the following steps:

[0071] S10, deploying multi-layered hardware and network systems;

[0072] S20, collecting multi-source data and performing fusion and correlation processing on the collected multi-source data;

[0073] S30: Monitor and issue warnings for local data, and upload the original data of the monitored local warning data to the cloud. After receiving the local data, the cloud compares it with historical data and data from similar device clusters to verify the accuracy of the local judgment.

[0074] S40, respectively perform data consistency verification, parameter optimization, and dynamic threshold calibration, and feed the results back to step S20 or S30.

[0075] The industrial network-based online monitoring method for safety valve performance described in this application fuses and correlates multi-source data. In this step, connections are established between different types of data, eliminating the need for single-source processing and judgment, providing a basis for subsequent comprehensive judgment. Furthermore, by correlating multiple parameters, the problem of false alarms for single parameters is resolved. When performing early warning judgment, local processing and cloud processing are performed separately. In this monitoring method, the cloud-processed data only includes the early warning data in the local data. This reduces network bandwidth requirements, reduces the amount of data transmitted over the network, and reduces the possibility of packet loss during data transmission. Furthermore, when performing early warning monitoring of data from the cloud, the local early warning monitoring data can be verified, further ensuring the accuracy of the early warning monitoring. Finally, the data is calibrated and verified, and the verification results are fed back to other steps, forming a closed-loop control system. Thus, the online monitoring method of this application not only correlates data to ensure its relevance and integrity, but also compares and judges the early warning data through the cloud, ensuring its accuracy. Furthermore, the data is verified and calibrated, improving the accuracy of the early warning data.

[0076] Compared with the existing technology, this application solves the problem of single parameter false alarm, realizes collaborative closed-loop control between the edge and the cloud, ensures the signal transmission reliability of the industrial network, and guarantees the data integrity. It provides a technical foundation for safety valve monitoring under the industrial network from passive manual to active intelligent, and has promotion value.

[0077] In some preferred embodiments, step S10 includes:

[0078] S11, building multi-layer hardware and networks, including the perception layer, transport layer, edge layer, and cloud layer;

[0079] The perception layer includes vibration sensors, pressure sensors, and temperature sensors, which are connected to the network through industrial buses respectively;

[0080] The real-time data of the transport layer is transmitted via the TSN network, and its non-real-time data is transmitted via industrial Ethernet;

[0081] The edge gateway of the edge layer performs data preprocessing, protocol conversion, and local data storage;

[0082] The cloud stores historical data and provides visualization and early warning management.

[0083] In some preferred embodiments, the vibration sensor is mounted on the valve stem of the safety valve to collect high-frequency vibration signals;

[0084] The temperature sensor is installed on the surface of the safety valve to detect internal leakage or frictional heat generation in the valve;

[0085] The pressure sensor is used to monitor the inlet and outlet pressure difference of the discharge end of the safety valve;

[0086] The real-time data in S11 includes vibration signals, temperature signals, and pressure signals; the non-real-time data includes device information;

[0087] The vibration sensor, the temperature sensor, the pressure sensor and the network transmission equipment respectively comply with explosion-proof safety certification to ensure safe operation in a flammable and explosive environment.

[0088] Specifically:

[0089] For the perception layer, a high-frequency vibration sensor (sampling rate ≥ 10kHz) and a high-temperature resistant optical fiber temperature sensor are installed on the safety valve body to adapt to the high-frequency mechanical movement and high-temperature and high-pressure environment of the valve.

[0090] For the transport layer, real-time data (vibration, temperature, pressure, etc.) is transmitted through the TSN network to ensure microsecond-level synchronization; non-real-time data (such as device information and device files) is transmitted through Industrial Ethernet.

[0091] For the edge layer, deploy an edge gateway to complete data preprocessing (filtering, timing alignment) and local storage.

[0092] S12, encrypt data and implement hierarchical control of permissions.

[0093] S13, perform time synchronization processing of the data. The algorithm of the time synchronization processing is:

[0094] t sync =t master +△t network .

[0095] In the above formula:

[0096] t master : Master clock timestamp (provided by the master clock in the TSN network).

[0097] Δt network : Network transmission delay (dynamically compensated through PTP protocol).

[0098] The function of step S13 is to ensure that the vibration signal (t vib ) and temperature signal (t temp ) to avoid misjudgment due to time offset.

[0099] S14, through the dual-ring network redundancy design, ensures the reliability of key data transmission. Its redundant network protocol is

[0100]

[0101] Packet loss: The number of data packets lost during transmission.

[0102] Total number of packets: The total number of data packets sent.

[0103] The purpose of step S14 is to ensure the reliability of key data (such as vibration spectrum) transmission through the dual-ring network redundancy design to meet the millisecond-level response requirements of the safety valve.

[0104] The significance of step S10 is:

[0105] A1. Network layered deployment.

[0106] Sensing layer: A high-frequency vibration sensor (sampling rate ≥ 10kHz) and a high-temperature fiber optic temperature sensor are installed on the safety valve body to adapt to the valve's high-frequency mechanical movement and high-temperature and high-pressure environment. A pressure sensor is also installed to monitor the inlet and outlet pressure difference at the discharge end of the safety valve.

[0107] Transport layer: Real-time data (vibration, temperature, pressure) is transmitted via the TSN network, ensuring microsecond-level synchronization; non-real-time data (device files) is transmitted via Industrial Ethernet.

[0108] Edge layer: Deploy edge gateways to complete data preprocessing (filtering, time alignment) and local storage.

[0109] Cloud: Stores historical data and provides visualization and alarm management.

[0110] A2. Requirements and functions of safety and reliability.

[0111] Sensors and network equipment must comply with explosion-proof certification (such as ATEX / IECEx) to ensure safe operation in flammable and explosive environments.

[0112] The Redundant Network Protocol (PRP) can ensure uninterrupted transmission of critical data and avoid monitoring failures due to network failures.

[0113] In step S20, multi-source data is collected and fused and associated with the collected multi-source data. The goal of this step is to establish a physical-data hybrid model of temperature and vibration and dynamically modify the abnormality determination threshold.

[0114] When a safety valve experiences internal leakage, frictional heat generated on the valve sealing surface causes a temperature rise, which in turn triggers abnormal harmonics in the vibration spectrum. To suppress ambient noise, an adaptive filtering algorithm is employed to eliminate electromagnetic interference. Specifically, a Kalman filter fusion algorithm is employed.

[0115] In some preferred embodiments, step S20 includes:

[0116] S21 , respectively collecting vibration signals and temperature signals, which are obtained and acquired by the hardware in step S10 .

[0117] S22, establish a coupled model of temperature and vibration

[0118] △T=k1·a 2 ·△t+k2·△P;

[0119] In the above formula:

[0120] ΔT: Valve surface temperature change (unit: ℃).

[0121] a: Vibration acceleration (unit: g), from the accelerometer.

[0122] Δt: Valve opening and closing cycle time (unit: s), collected by the motion sensor.

[0123] ΔP: Pressure fluctuation (unit: %), from the pressure transmitter.

[0124] k1, k2: experimentally calibrated material and structural parameters (e.g., k1 = 0.05 means that the product of the square of the acceleration and the time per unit time results in a temperature increase of 0.05°C).

[0125] The purpose of step S22 is to predict the temperature trend through vibration acceleration and pressure changes to solve the problem of single parameter false alarm (such as false vibration signal caused by high temperature).

[0126] S23 outputs a robust temperature signal through the Kalman filter fusion algorithm to fuse multi-source data.

[0127] In some preferred embodiments, the rules of the Kalman filter fusion algorithm in step S23 are:

[0128]

[0129] In the above formula:

[0130] Temperature predictions based on thermodynamic models.

[0131] F: State transfer matrix (derived from the thermodynamic model, reflecting the physical laws of temperature change).

[0132] K k : Kalman gain (dynamically adjust filter weight, 0 <K k <1).

[0133] T meas : Measured temperature value.

[0134] The function of step S23 is to fuse the temperature prediction value and the measured value and output a robust temperature signal T fusion , eliminating electromagnetic noise interference.

[0135] S24, according to the result of step S23, dynamic threshold adjustment of the vibration threshold

[0136]

[0137] In the above formula:

[0138] T base : Basic vibration threshold (such as 5g), or basic temperature value (unit: ℃).

[0139] α: Temperature coefficient (initial value 0.15 / °C, subsequently optimized in step 4).

[0140] ΔT: Real-time temperature change (unit: °C).

[0141] T max : The upper temperature limit of the valve material (such as 400℃).

[0142] The function of step S24 is to automatically relax the vibration threshold when the temperature rises to avoid false alarms caused by high temperature (for example, when ΔT=50° C., the vibration threshold is adjusted to 5g×(1+0.15×0.125)=5.09g).

[0143] Among them, the dynamic threshold refers to the vibration threshold and temperature threshold. When calculating the vibration threshold, as shown in the following formula, T base Indicates the basic vibration threshold (such as 5g);

[0144]

[0145] Content and significance of step S20:

[0146] B1. Data Collection:

[0147] The vibration sensor collects high-frequency vibration signals (10kHz sampling rate) to capture transient events such as valve sticking and seal failure.

[0148] Temperature sensors monitor valve surface temperature and correlate internal leakage or frictional heating.

[0149] B2. Dynamic Association Modeling:

[0150] The thermodynamic-vibration coupling model predicts temperature changes through vibration acceleration and pressure fluctuations, solving the problem of single parameter false positives.

[0151] Kalman filtering fuses predicted temperature with measured temperature to improve data reliability.

[0152] B3. Adaptive threshold adjustment:

[0153] The dynamic threshold formula adjusts the vibration alarm threshold based on the real-time temperature, reducing the false alarm rate in high temperature environments.

[0154] In S30, local data is monitored and alerted, and the original local alert data is uploaded to the cloud. After receiving the local data, the cloud compares it with historical data and data from similar device clusters to verify the accuracy of the local judgment. The goal of step S30 is to quickly respond to anomalies locally and conduct global verification and model optimization in the cloud.

[0155] In step S30, the edge gateway must complete fault diagnosis within 100ms to achieve local real-time performance. It also performs cloud-based group analysis, aggregating multi-device data to identify abnormal patterns (such as batch seal failures).

[0156] In some preferred embodiments, step S30 includes:

[0157] S31, establish monitoring and early warning rules for local data, the monitoring and early warning rules are:

[0158] If T fusion >T threshold And Vib g >V threshold , then a level one alarm is triggered and immediate action is required;

[0159] If T fusion >T warning or Vib g >V warning , then a secondary alarm is triggered, requiring attention and short-term observation;

[0160] Among them, T fusion and Vib g respectively obtained by step S20;

[0161] In this rule:

[0162] T threshold : The alarm threshold calculated by the dynamic threshold formula in step 2.

[0163] V threshold : Basic vibration threshold (5g).

[0164] T warning : Temperature warning threshold (such as ΔT = 30°C).

[0165] In some preferred embodiments, in step S31, T threshold >T warning , V threshold >V warning ;

[0166] V warning With V threshold Establish proportional relationships to ensure logical consistency between warnings and alarms;

[0167] When T threshold and V threshold When dynamic adjustment occurs and decreases, T warning Based on T threshold Dynamic floating, V warning Based on V threshold Dynamic downward floating, the dynamic downward floating ratio is 70%-80%, in order to provide early warning of equipment degradation trends.

[0168] The function of step S31 is to combine the local temperature-vibration correlation model with the dynamic threshold value to output a preliminary fault level (level 1 alarm or level 2 warning).

[0169] S32, establish a cloud-based statistical verification model, which is:

[0170]

[0171] In this formula:

[0172] μ cloud : Average vibration value of similar valves in the cloud;

[0173] σ cloud : Cloud standard deviation.

[0174] The function of step S32 is to trigger rule modification (e.g., reducing the temperature coefficient α) if the local determination is abnormal but the cloud score is less than 3σ.

[0175] S33, when the local monitoring warning result is abnormal and the cloud verification result is <3σ, the rule modification is triggered;

[0176] Rule amendments are handled as follows:

[0177]

[0178] In this formula:

[0179] η: learning rate (e.g. 0.01).

[0180] L: Verification error (the proportion of samples with inconsistent judgments between local and cloud).

[0181] The purpose of step S31 is to adjust the temperature coefficient α according to the cloud verification result to reduce the false alarm rate.

[0182] The obtained temperature coefficient α is used to update the temperature coefficient in step S20.

[0183] Content and significance of step S30:

[0184] C1. Local real-time judgment:

[0185] The edge gateway makes millisecond-level fault judgments based on dynamic threshold formulas and early warning decision rules.

[0186] If the vibration amplitude exceeds the threshold and the temperature is abnormal, a level 1 alarm will be triggered immediately and the original data will be uploaded to the cloud.

[0187] C2. Cloud-based global verification:

[0188] The cloud aggregates data from multiple devices and verifies the accuracy of local judgments using the 3σ principle.

[0189] If the local and cloud conclusions conflict (for example, an alarm is triggered locally but no anomaly is detected on the cloud), a correction instruction (such as adjusting α) is issued.

[0190] C3. Dynamic optimization of rules:

[0191] The gradient descent algorithm adjusts α according to the cloud verification error to improve the global consistency of the system.

[0192] In step S40, data consistency verification, parameter optimization, and dynamic threshold calibration are performed, and the results are fed back to step S20 or S30. The goal of step S40 is to achieve system self-adaptation through data consistency verification and parameter optimization.

[0193] In step S40, scenarios such as valve sticking and internal leakage are simulated to test system consistency to achieve fault reproducibility verification; and the robustness of the algorithm is verified in a high temperature and high humidity environment to achieve environmental adaptability verification.

[0194] In some preferred embodiments, step S40 includes:

[0195] S41: Perform data consistency verification. The data consistency verification rules are as follows:

[0196]

[0197] If D KL >0.5, then the optimization of the temperature coefficient α is triggered;

[0198] In this formula:

[0199] P: Local alarm distribution (e.g., level 1 alarms account for 30%).

[0200] Q: Distribution of cloud alarms (e.g., level 1 alarms account for 25%).

[0201] The function of step S41 is to: KL>0.5, triggers parameter optimization (such as adjusting α).

[0202] S42, recalibrate the dynamic threshold so that the local rule approaches the optimal value on the cloud side; the rule is:

[0203] a new =a cloud (1-D KL +∈);

[0204] In this formula:

[0205] ∈: A small constant (such as 0.001) to prevent division by zero.

[0206] The function of step S42 is to dynamically adjust α according to the KL divergence so that the local rule approaches the optimal value in the cloud.

[0207] S43, through Bayesian optimization, the temperature coefficient α is forced to be aligned locally and in the cloud to ensure global consistency. The Bayesian optimization rule is:

[0208] Minimize L = α·false alarm rate + β·missing alarm rate + γ·||α-α cloud || 2 ;

[0209] The temperature coefficient α obtained in step S43 is used to update the temperature coefficient in step S20.

[0210] In this formula:

[0211] α cloud : Temperature coefficient after cloud optimization;

[0212] γ: Alignment weight (e.g. 0.5).

[0213] The purpose of step S43 is to force the local α to align with the cloud to ensure global consistency.

[0214] In some preferred embodiments, in step S40 , sampling checks are performed regularly to determine the consistency between local data and cloud alarm results;

[0215] The sampling data selects 10%-20% of the total alarm data.

[0216] Content and significance of step S40:

[0217] D1. Consistency test:

[0218] Regularly sample and check the consistency of local and cloud alarm results (such as randomly sampling 10% of alarm events).

[0219] If the KL divergence exceeds a threshold (such as 0.5), the parameter optimization process is triggered.

[0220] D2. Model re-optimization:

[0221] Bayesian optimization adjusts α and synchronizes it to all edge nodes by minimizing the loss function L.

[0222] D3. Dynamic threshold update:

[0223] The dynamic threshold recalibration formula adjusts α according to the KL divergence to ensure that the local rules are consistent with the global model on the cloud.

[0224] The following is a description of the industrial network-based online monitoring method for safety valve performance of the present application by way of example, and provides the following two examples.

[0225] Example 1: Safety valve internal leakage condition (high temperature and high pressure environment).

[0226] Scenario: A high-temperature, high-pressure steam safety valve at a refinery (rated pressure 10 MPa, medium superheated steam, normal operating temperature 380°C) experiences internal leakage due to seal surface wear. The system requires real-time monitoring of vibration, temperature, and pressure signals, triggering alarms and optimizing parameters. The parameter values ​​in this example are derived from historical data or based on basic settings.

[0227] Step S10: System architecture design and network deployment.

[0228] 1.1 Time synchronization algorithm (IEEE 1588PTP).

[0229] Formula: t sync =t master +Δt network

[0230] Parameter calculation:

[0231] Master clock timestamp t master =1000ms;

[0232] Network transmission delay Δt network =50μs (dynamic compensation through PTP protocol);

[0233] Timestamp after synchronization:

[0234] t sync =1000ms+50μs=1000.00005ms.

[0235] 1.2 Redundant network protocol (PROFINET PRP).

[0236] Data packet arrival rate calculation:

[0237] Total number of packages = 10,000;

[0238] Number of packet losses = 0 (redundant dual-ring network design);

[0239] Reach rate: Reach rate = 1-10,0000 = 100% ≥ 99.999%.

[0240] Step S20: Multi-source data collection and fusion.

[0241] 2.1 Thermodynamic-vibration coupling model.

[0242] Formula: ΔT = k1·a 2 ·Δt+k2·ΔP

[0243] Parameter values:

[0244] Vibration acceleration a = 6.2g (measured);

[0245] Pressure fluctuation ΔP = 2% (measured);

[0246] Valve opening and closing cycle Δt=0.5s;

[0247] Material parameter k1=0.05℃\cdotps / g 2 , k2=0.1℃ / %;

[0248] Temperature change calculation:

[0249] ΔT=0.05*6.2 2 *0.5+0.1*2=0.05*38.44*0.5+0.2=1.161.

[0250] 2.2 Kalman filter fusion algorithm.

[0251] formula:

[0252]

[0253] Known parameters:

[0254] T k-1 =395.2℃ (fusion temperature at the previous moment);

[0255] ΔT pred = 2.122°C (temperature change within 1 second predicted by the thermodynamic model);

[0256] F = 1, (state transfer matrix, here is scalar 1, because the temperature change is linear);

[0257] K k = 0.7 (Kalman gain, assumed to be calculated from the system noise covariance);

[0258] Tmeas =415.3℃ (actual temperature measured by the sensor).

[0259] Step-by-step calculation:

[0260] 1. Prediction stage:

[0261] Predicted temperature Add the temperature change predicted by the model to the fusion temperature at the previous moment:

[0262]

[0263] 2. Update phase:

[0264] Fusion temperature The Kalman gain K is obtained by weighted average of the predicted value and the measured value. k =0.7, indicating more trust in the measured value):

[0265] 2.3 Dynamic threshold adjustment.

[0266] About the temperature dynamic threshold formula:

[0267]

[0268] Known parameters:

[0269] T base =380°C (basic threshold, the initial threshold of the system, usually set based on historical data or standard operating conditions); α = 0.15 (temperature sensitivity coefficient, reflecting the sensitivity of the threshold to temperature changes, dynamically adjusted according to step S40); ΔT = 20°C (current temperature change, the difference between the real-time temperature and the reference temperature);

[0270] T max =400℃ (maximum allowable temperature, critical temperature for safe operation of the system);

[0271] Then, the temperature dynamic threshold is calculated as:

[0272] Temperature dynamic threshold = 380*(1+0.15*(20 / 400)) = 380*(1+0.0075) = 382.85.

[0273] About the vibration dynamic threshold formula:

[0274]

[0275] T base =5g (basic threshold, the initial threshold of the system, usually set based on historical data or standard operating conditions);

[0276] α = 0.15 (temperature sensitivity coefficient, reflecting the sensitivity of the threshold to temperature changes, dynamically adjusted according to step S40); ΔT = 20°C (current temperature change, the difference between the real-time temperature and the reference temperature);

[0277] T max =400℃ (maximum allowable temperature, critical temperature for safe operation of the system);

[0278] Then, the vibration dynamic threshold is calculated as:

[0279] Vibration dynamic threshold = 5*(1+0.15*(20 / 400)) = 5*(1+0.0075) = 5.0375≈5.04.

[0280] Step S30: Local and cloud collaborative monitoring.

[0281] 3.1 Local warning decision-making rules.

[0282] rule:

[0283] If T fusion >T threshold And Vib g >V threshold , then a level one alarm is triggered and immediate action is required;

[0284] If T fusion >T warning or Vib g >V warning , then a secondary alarm is triggered, requiring attention and short-term observation. Parameter values:

[0285] Fusion temperature T fusion =409.618℃;

[0286] Temperature dynamic threshold T threshold =382.85°C (based on S20 of 2.3);

[0287] Vibration effective value Vibg = 6.1g;

[0288] Vibration dynamic threshold V threshold＝ 5.04 (based on S20's 2.3);

[0289] Judgment result:

[0290] 409.26>382.85 and 6.1>

[0291] 3.2 Cloud-based group anomaly detection.

[0292] formula:

[0293]

[0294] Parameter values:

[0295] Local vibration amplitude Vib local =6.1g;

[0296] Cloud mean μ cloud =5.5g;

[0297] Cloud standard deviation σ cloud =0.8g;

[0298] Anomaly score:

[0299]

[0300] 3.3 Rule correction algorithm (gradient descent).

[0301] formula:

[0302]

[0303] Parameter values:

[0304] Current temperature coefficient a t =0.15 / ℃;

[0305] The sum of the weights of false positive samples ∑w i =0.2;

[0306] Learning rate η = 0.01;

[0307] Parameter update:

[0308] a t+1 =0.15-0.01·0.2=0.148 / ℃;

[0309] The obtained α is used to update the temperature coefficient of 2.3 in step S20.

[0310] Step S40: Data verification and closed-loop control.

[0311] 4.1 Consistency Test (KL Divergence)

[0312] formula:

[0313]

[0314] Parameter values:

[0315] The proportion of local level 1 alarms is P = 30%;

[0316] The proportion of cloud-based level 1 alarms is Q = 25%;

[0317] KL divergence calculation:

[0318] D KL =0.3log(0.3 / 0.25)+0.7log(0.7 / 0.75)≈0.3*0.1823+0.7*(-0.069)=0.0064<0.5. Conclusion: The local and cloud data are well consistent, and no parameter adjustment is required.

[0319] 4.2 Dynamic Threshold Recalibration

[0320] formula:

[0321] a new =a cloud (1-D KL +∈).

[0322] Parameter values:

[0323] Cloud temperature coefficient a cloud =0.148;

[0324] KL divergence D KL =0.064;

[0325] Small constant ∈ = 0.001;

[0326] New threshold calculation:

[0327] a new =0.148·(1-0.064+0.001)=0.148·0.938≈0.135 / °C.

[0328] The obtained α is used to update the temperature coefficients of 2.3 in step S20 and 3.3 in step S30.

[0329] Example 2: Safety valve stuck condition (mechanical failure).

[0330] Scenario Description: A safety valve at a refinery cannot be fully closed due to mechanical obstruction, resulting in high-frequency harmonics in the vibration spectrum. Step S10: System Architecture and Network Deployment.

[0331] Sensor deployment:

[0332] Vibration sensor (sampling rate 20kHz): captures high-frequency harmonics.

[0333] Temperature sensor: monitors valve surface temperature (frictional heat generated by sticking).

[0334] Network Configuration:

[0335] The TSN network ensures microsecond-level synchronization of vibration signals, and 5G transmits remote diagnostic data.

[0336] Step S20: Multi-source data collection and fusion.

[0337] Input data:

[0338] Vibration acceleration a = 6.5g (high frequency component accounts for 30%);

[0339] Temperature change ΔT = 18°C;

[0340] Pressure fluctuation ΔP = 8%;

[0341] Formula calculation:

[0342] Thermodynamic-vibration coupling model: ΔT = 0.05·(6.5) 2 ·1.0+0.1·8=2.11+0.8=2.91℃。

[0343] Contradiction: The measured ΔT = 18°C, indicating that mechanical jamming causes abnormal frictional heating.

[0344] Kalman filter fusion:

[0345] Predicted temperature T pred =40℃, measured T meas =55℃→T after fusion fusion =50℃.

[0346] Dynamic threshold adjustment: vibration threshold = 5·(1+0.14·50 / 400)=5·1.0175=5.09 g.

[0347] Temperature threshold = 50·(1+0.14·50 / 400) = 50·1.0175 = 50.9°C.

[0348] Step S30: Local and cloud collaborative monitoring.

[0349] Local judgment:

[0350] Rules Engine:

[0351] If T fusion >T threshold And Vib g >V threshold , then a level one alarm is triggered and immediate action is required;

[0352] If T fusion >T warning or Vib g >V warning , then a secondary alarm is triggered, requiring attention and short-term observation;

[0353] Current Condition: Vib g >V warning →Trigger the second level warning.

[0354] Cloud Verification:

[0355] Group data comparison: In similar jamming events, the proportion of high-frequency vibration is greater than 30%, and the current value meets the characteristics.

[0356] 3σ test: Local vibration 6.5g vs. cloud average 6.0g (σ = 0.5g) → | 6.5 - 6.0 | / 0.5 = 1 < 3σ →

[0357] Maintain the alert level.

[0358] Rule correction: No adjustment is required, the jamming characteristics are clear.

[0359] Step S40: Data verification and closed-loop control.

[0360] Consistency test: KL divergence D KL =0.1, no optimization required.

[0361] Result: The system maintained the second-level warning. Manual intervention inspection found a jam, and the system recovered after replacing the valve spring.

[0362] The advantages and effects of the online monitoring method for safety valve performance based on industrial network in this application are described.

[0363] 1. Summary of the technical advantages of the online monitoring method of this application.

[0364] 1) The physical meaning of the parameters is clear:

[0365] k1 and k2 in the thermodynamic model reflect the frictional heat generation mechanism, avoiding the overfitting risk of pure data-driven models.

[0366] Kalman filter K k Dynamically adjust filter weights to balance predicted and measured data.

[0367] Dynamic parameter alignment: Local α is aligned to the cloud through gradient descent and Bayesian optimization, eliminating data silos.

[0368] 2).Logical closed loop between steps:

[0369] The dynamic threshold formula in step S20 provides input for the local decision in step S30. The cloud-based verification result in step S30 reversely optimizes the parameters of step S20 through gradient descent. The KL divergence test in step S40 enforces global consistency. The final verification result is fed back to the aforementioned steps to achieve a closed loop.

[0370] 3). Industrial-grade reliability:

[0371] The TSN network and explosion-proof design meet the requirements of high-risk safety valve scenarios.

[0372] Dynamic threshold and Kalman filtering reduce false alarm rate (target <5%) and improve system robustness.

[0373] 4) Adaptability:

[0374] Bayesian optimization and KL divergence enable parameter self-update to adapt to equipment aging and environmental changes.

[0375] 2. The core advantages of the industrial network-based online monitoring method for safety valve performance of this application.

[0376] 1. High real-time performance and reliability.

[0377] TSN network: Through the IEEE 1588PTP time synchronization algorithm (microsecond accuracy) and PROFINET PRP redundancy protocol (99.999% data packet arrival rate), it ensures the synchronous transmission of vibration and temperature signals, avoiding misjudgment caused by time offset.

[0378] Edge computing: Local real-time data processing (such as Kalman filtering and dynamic threshold adjustment) reduces cloud dependency and meets millisecond-level response requirements for safety valves.

[0379] 2. Low false alarm rate and high robustness.

[0380] Multi-source data fusion: The thermodynamic-vibration coupling model is combined with Kalman filtering to eliminate single parameter noise interference (such as vibration spikes caused by electromagnetic interference).

[0381] Dynamic threshold adjustment: Automatically relaxes the vibration threshold as the temperature rises to avoid false alarms in high temperature environments (for example, when the temperature fluctuates by ±50°C, the vibration threshold can be adjusted by up to ±15%).

[0382] 3. Adaptation and continuous optimization.

[0383] Cloud-based group verification: Using the 3σ principle and gradient descent algorithm, local model parameters (such as the temperature coefficient α) are corrected based on multi-device data to reduce the false alarm rate (target <5%).

[0384] Closed-loop control: KL divergence and Bayesian optimization enable dynamic alignment of local and cloud model parameters to adapt to device aging and environmental changes.

[0385] 4. Industrial-grade security and compatibility.

[0386] Complies with explosion-proof certification (ATEX / IECEx) and supports high temperature and high pressure environments.

[0387] The layered architecture (perception layer - transport layer - edge layer - cloud) is compatible with existing industrial networks and is easy to deploy.

[0388] 3. Compared with the existing technology, the main innovation of the online monitoring method of safety valve performance based on industrial network in this application.

[0389] 1. Physical-data hybrid modeling.

[0390] The thermodynamic model (k1, k2 parameters) is combined with data-driven methods (Kalman filtering, machine learning) to retain the interpretability of the physical mechanism while dynamically correcting model deviations through data.

[0391] 2. Dynamic thresholds work in conjunction with the rules engine.

[0392] Decision tree rules (level one alarm, level two warning) are linked with dynamic thresholds to achieve graded responses (e.g., temperature anomaly + vibration exceeding the standard triggers a level one alarm, while a single parameter anomaly only triggers a warning).

[0393] 3. Closed-loop self-learning mechanism.

[0394] Through cloud-based statistical verification (KL divergence) and gradient descent, adaptive optimization of parameters (such as α) is achieved, forming a closed loop of "local monitoring → cloud-based verification → model optimization".

[0395] 4. Technical significance of the industrial network-based online monitoring method for safety valve performance of this application.

[0396] 1. Improve equipment safety.

[0397] Early identification of high-risk faults such as internal leakage and seal failure can avoid media leakage or explosion accidents.

[0398] 2. Reduce operation and maintenance costs.

[0399] Reduce the frequency of manual inspections and extend equipment life through predictive maintenance (e.g., extend seal replacement cycles by more than 30%).

[0400] 3. Promote industrial intelligence.

[0401] Provide a reusable "mechanism + data" fusion paradigm for monitoring complex electromechanical systems (such as compressors and pumps).

[0402] 5. Supplementary explanation: In the industrial network-based online monitoring method for safety valve performance of this application, vibration-temperature correlation is performed to compare the advantages of temperature-pressure correlation.

[0403] 1. Why choose vibration-temperature correlation?

[0404] Stronger complementarity:

[0405] Vibration signal: directly reflects the mechanical state (such as friction, blocking, and imbalance) and is closely related to local high temperature caused by internal leakage.

[0406] Temperature signal: reflects energy dissipation (frictional heat) and environmental thermal interference.

[0407] The combination of the two can distinguish the type of fault (such as mechanical wear vs. medium temperature fluctuation).

[0408] Better anti-interference ability:

[0409] Vibration is sensitive to mechanical failure, and temperature is sensitive to the environment. Dynamic threshold adjustment can filter out interference from ambient temperature fluctuations, while pressure signals are easily affected by external factors such as medium flow and valve opening.

[0410] 2. Compare the limitations of temperature-pressure correlation.

[0411] Applicable scenarios for temperature-pressure correlation:

[0412] It is more suitable for detecting abnormal fluid conditions (such as blockage, cavitation) or macro pressure anomalies (such as overpressure).

[0413] 3. In addition, the unique advantages of vibration-temperature correlation:

[0414] More accurate fault location: Vibration signals can locate the fault location (such as the location of valve core wear), while pressure signals are difficult to achieve spatial positioning.

[0415] Stronger early warning capabilities: Vibration anomalies (such as high-frequency harmonics) often occur earlier than significant temperature changes. Combining the two can shorten fault response time.

[0416] 6. Summary of the technical value of the online monitoring method of this application.

[0417] This application solves the core defects of existing technologies in real-time, reliability and adaptability through multi-source data fusion, physical model drive and closed-loop self-optimization mechanism, and provides a technological leap from "passive manual" to "active intelligent" for industrial safety valve monitoring, which has significant industry promotion value.

[0418] This application's industrial network-based online safety valve performance monitoring method achieves "fast, accurate, and stable" safety valve monitoring through the deep integration of physical mechanisms and data science, dynamic threshold adaptation, and cloud-on-premises closed-loop collaboration. The advantages of vibration-temperature correlation lie in its complementarity, interference resistance, and fault location capabilities. Compared to temperature-pressure correlation, it is more suitable for early identification of mechanical faults and preventive maintenance.

[0419] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present application, and such modifications and improvements are all within the scope of protection of the present application.

Claims

1. A safety valve performance online monitoring method based on industrial network, characterized in that: Including steps: S10, deploying multi-layered hardware and network systems; S20, collecting multi-source data and performing fusion and correlation processing on the collected multi-source data; S30, monitor and warn local data, and upload the original data of the monitored local warning data to the cloud; After receiving local data, the cloud compares it with historical data and data from similar device clusters to verify the accuracy of local judgments; S40, respectively perform data consistency verification, parameter optimization, and dynamic threshold calibration, and feed the results back to step S20 or S30.

2. The method for online monitoring of safety valve performance based on industrial network according to claim 1 is characterized in that: Step S20 includes: S21, collecting vibration signals and temperature signals respectively; S22, establish a coupled model of temperature and vibration ΔT=k1·a 2 .Δt+k2·ΔP; S23, outputs a robust temperature signal through the Kalman filter fusion algorithm to fuse multi-source data; S24, according to the result of step S23, dynamic threshold adjustment of the vibration threshold 3. The method for online monitoring of safety valve performance based on industrial network according to claim 2, characterized in that: Step S30 includes: S31, establish monitoring and early warning rules for local data, the monitoring and early warning rules are: If T fusion >T threshold And Vib g >V threshold , then a level one alarm is triggered and immediate action is required; If T fusion >T warning or Vib g >V warning , then a secondary alarm is triggered, requiring attention and short-term observation; Among them, T fusion and Vib g respectively obtained by step S20; S32, establish a cloud-based statistical verification model, which is: S33, when the local monitoring warning result is abnormal and the cloud verification result is <3σ, the rule modification is triggered; Rule amendments are handled as follows: The obtained temperature coefficient α is used to update the temperature coefficient in step S20.

4. The method for online monitoring of safety valve performance based on industrial network according to claim 3 is characterized in that: Step S40 includes: S41: Perform data consistency verification. The data consistency verification rules are as follows: If D KL >0.5, then the optimization of the temperature coefficient α is triggered; S42, recalibrate the dynamic threshold so that the local rule approaches the optimal value on the cloud side; the rule is: a new =a cloud ·(1-D KL +∈); S43, through Bayesian optimization, the temperature coefficient α is forced to be aligned locally and in the cloud to ensure global consistency. The Bayesian optimization rule is: Minimize L = α·false alarm rate + β·missing alarm rate + γ·||α-α cloud || 2 ; The temperature coefficient α obtained in step S43 is used to update the temperature coefficient in step S20.

5. The method for online monitoring of safety valve performance based on industrial network according to claim 4 is characterized in that: The rules of the Kalman filter fusion algorithm in step S23 are:

6. The method for online monitoring of safety valve performance based on an industrial network according to any one of claims 2 to 5, characterized in that: Step S10 includes: S11, building multi-layer hardware and networks, including the perception layer, transport layer, edge layer, and cloud layer; The perception layer includes vibration sensors, pressure sensors, and temperature sensors, which are connected to the network through industrial buses respectively; The real-time data of the transport layer is transmitted via the TSN network, and its non-real-time data is transmitted via industrial Ethernet; The edge gateway of the edge layer performs data preprocessing, protocol conversion, and local data storage; The cloud stores historical data and provides visualization and early warning management; S12, encrypt data and implement hierarchical control of permissions; S13, perform time synchronization processing of the data. The algorithm of the time synchronization processing is: t sync =t master +Δt network ; S14, through the dual-ring network redundancy design, ensures the reliability of key data transmission. Its redundant network protocol is 7. The method for online monitoring of safety valve performance based on industrial network according to claim 6, characterized in that: In step S40, regular sampling checks are performed to determine the consistency between local data and cloud alarm results; The sampling data selects 10%-20% of the total alarm data.

8. The method for online monitoring of safety valve performance based on industrial network according to claim 6, characterized in that: The vibration sensor is installed on the valve stem of the safety valve and is used to collect high-frequency vibration signals; The temperature sensor is installed on the surface of the safety valve to detect internal leakage or frictional heat generation in the valve; The pressure sensor is used to monitor the inlet and outlet pressure difference of the discharge end of the safety valve; The real-time data in S11 includes vibration signals, temperature signals, and pressure signals; the non-real-time data includes device information; The vibration sensor, the temperature sensor, the pressure sensor and the network transmission equipment respectively comply with explosion-proof safety certification to ensure safe operation in a flammable and explosive environment.

9. The method for online monitoring of safety valve performance based on industrial network according to claim 3, characterized in that: In step S31, T threshold >T warning , V threshold >V warning ; V warning With V threshold Establish proportional relationships to ensure logical consistency between warnings and alarms; When T threshold and V threshold When dynamic adjustment occurs and decreases, T warning Based on T threshold Dynamic floating, V warning Based on V threshold Dynamic downward floating, the dynamic downward floating ratio is 70%-80%, in order to provide early warning of equipment degradation trends.

Images