Online monitoring and intelligent diagnosis method for sealing performance of pilot operated safety valve

Abstract

The invention discloses an on-line monitoring and intelligent diagnosis method for the sealing performance of a pilot operated safety valve, and relates to the technical field of nuclear power station safety operation and maintaining.The method comprises the steps that multi-mode sensors are arranged at the inlet end of a main valve, a pilot valve control cavity, a main valve seat, a pilot valve seat, the inlet and outlet pipe wall of the main valve and a valve clack ejector rod; synchronously acquiring pressure fluctuation, acoustic emission stress waves, temperature difference changes and axial micro-displacement signals; filtering, de-noising and normalizing the multi-physical field signal through an edge calculation unit, and extracting time domain statistics and frequency domain transformation features to generate a preliminary feature vector; inputting the time-space attention multi-modal fusion diagnosis model, and outputting a multi-dimensional diagnosis result through multi-channel coding, modal weighting and time sequence focusing; and a health report, a trend curve and a maintenance suggestion are generated in combination with historical data, and a sealing comprehensive health index is dynamically calculated to quantitatively evaluate the state of the valve. Quantitative management and life cycle prediction of the health state of the valve are realized.

Description

Technical Field

[0001] This invention relates to the field of nuclear power plant safety operation and maintenance technology, and in particular to an online monitoring and intelligent diagnostic method and device for the sealing performance of a pilot-operated safety valve. Background Technology

[0002] Pilot-operated safety valves are core overpressure protection devices for pressure pipelines and pressure vessel systems in high-risk industries such as petroleum, chemical, power, and natural gas. Their structure mainly consists of two functional units: a main valve and a pilot valve. This type of safety valve controls the opening and closing of the main valve through the opening and closing action of the pilot valve, achieving a precise response to changes in system pressure and thus completing the automatic pressure relief protection function. Under normal operating conditions, the system pressure acts on the piston of the main valve through the conduit, and the pressure of the medium itself keeps the main valve tightly closed. Therefore, its sealing performance has a significant advantage over traditional spring-loaded safety valves, especially under operating conditions close to the set pressure, achieving a zero-leakage or minimal-leakage sealing effect.

[0003] Sealing performance is a key factor in ensuring the reliable operation of pilot-operated safety valves. Sealing failure directly leads to media leakage, causing not only severe economic losses and environmental pollution but also potentially triggering major safety accidents. Analysis shows that the main causes of sealing failure include wear, scratches, and corrosion on the sealing surfaces of the main valve seat and pilot valve seat, as well as aging and damage to the internal piston seal ring.

[0004] Currently, the testing methods for the sealing performance of safety valves mainly rely on offline calibration and periodic maintenance. These methods have significant drawbacks: firstly, they cannot achieve continuous monitoring of the safety valve's operating status, making it difficult to effectively identify and warn of gradual seal degradation or sudden leakage events during operation; secondly, maintenance requires shutting down the equipment and disassembling the valve, resulting in high maintenance costs and severely impacting the continuous operation of the production system. Existing online monitoring equipment often employs single-physical-parameter detection schemes, such as installing temperature sensors at the valve outlet to utilize the Joule-Thomson effect caused by the throttling effect of the leaking medium to determine the temperature drop, or using acoustic sensors to capture high-frequency noise signals generated by the leak for leak detection. However, these single-parameter monitoring methods have significant limitations. Their detection results are easily affected by external environmental factors, resulting in low reliability. They can only qualitatively determine whether a safety valve leaks, but cannot accurately locate the leak source, diagnose the root cause of seal failure, or quantitatively assess the severity of the leak.

[0005] Based on the shortcomings of the existing technology, there is an urgent need in the field for a monitoring device that can perform real-time, continuous, and high-precision online monitoring of pilot-operated safety valves and can intelligently diagnose the location of leaks, failure types, and development trends. Summary of the Invention

[0006] The main objective of this invention is to provide an online monitoring and intelligent diagnostic method for the sealing performance of a pilot-operated safety valve.

[0007] Another objective of this invention is to provide an online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve.

[0008] The third objective of this invention is to provide an electronic device.

[0009] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0010] To achieve the above objectives, a first aspect of the present invention provides a method for online monitoring and intelligent diagnosis of the sealing performance of a pilot-operated safety valve, comprising: S1, deploy multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls and valve disc top rod, and simultaneously collect pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals and axial micro-displacement signals; S2, based on the edge computing unit, performs filtering, denoising and normalization processing on the acquired multi-physics field signals, and extracts time-domain statistical features and frequency-domain transform features respectively to generate preliminary feature vectors; S3, input the preliminary feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extract deep spatial features through multi-channel feature encoding, and determine the dynamic weight of each signal mode by using intermodal attention weighting. After temporal feature encoding and temporal attention focusing, output multidimensional diagnostic results. S4 uploads multidimensional diagnostic results to the cloud-based intelligent management platform, combines historical data to generate a visualized health status report, trend analysis curves, and predictive maintenance suggestions, and dynamically calculates the comprehensive sealing health index to quantitatively assess the valve's health status.

[0011] Optionally, the deployment of multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls, and valve disc push rod to simultaneously acquire pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals, and axial micro-displacement signals further includes: The multimodal sensor consists of at least two high-frequency pressure sensors, at least two broadband acoustic emission sensors, at least two high-precision temperature sensors, and at least one micro-displacement sensor. The high-frequency pressure sensors are deployed at the main valve inlet and the pilot valve control chamber, the broadband acoustic emission sensors are attached to the valve body near the main valve seat and the pilot valve seat, the high-precision temperature sensors are deployed on the main valve inlet pipe wall and the outlet pipe wall, and the micro-displacement sensor is aligned with the main valve disc rod in a non-contact manner. The sensors synchronously collect pressure fluctuation signals, high-frequency acoustic emission stress wave signals from leakage, temperature difference change signals from leakage throttling effect, and axial micro-displacement signals of the valve disc, all related to the sealing state of the safety valve.

[0012] Optionally, the edge computing and diagnostic unit is electrically connected to the sensing subsystem and is an embedded hardware platform based on a high-performance SoC. Its core integrates a neural network processing unit. This unit synchronously acquires and preprocesses multimodal sensor data through a high-speed multi-channel ADC and digital interface, and runs a lightweight anomaly detection algorithm and core intelligent diagnostic model in real time at the edge. It also integrates multiple industrial communication models to achieve data interaction with the cloud.

[0013] Optionally, the step of extracting time-domain statistical features and frequency-domain transform features to generate a preliminary feature vector further includes: Perform time-frequency domain transformation on the preprocessed multiphysics field signal to generate a time-frequency spectrum characterizing the time-frequency distribution of the signal; Based on the time-frequency spectrum, the time-domain statistical features and frequency-domain transformation features of the preprocessed multi-physics field signal are extracted to form the preliminary feature vector.

[0014] Optionally, the step of inputting the feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extracting deep spatial features through multi-channel feature encoding, determining the dynamic weights of each signal mode using intermodal attention weighting, and outputting multidimensional diagnostic results after temporal feature encoding and temporal attention focusing, further includes: Multi-channel feature encoding is performed on each modality of data, and deep spatial features are extracted from each modality of data using parallel convolutional neural network branches. Perform intermodal attention weighting, dynamically learn and assign importance weights to different modal features under the current working conditions through an attention mechanism; Perform temporal feature encoding, input the weighted and fused feature sequence into a recurrent neural network, and capture the time dependence of fault evolution; The execution time attention focus applies an attention mechanism to the output of the timing encoder again, focusing on the critical time steps before and after the fault occurs; The diagnostic results are output, and a multi-dimensional diagnostic vector is generated through the output layer. The multi-dimensional diagnostic vector includes at least sealing status determination, leakage location location, and root cause inference. The sealing status determination results include normal, minor leakage, moderate leakage, and severe leakage. The leakage location location results include the main valve seat, pilot valve seat, and piston seal. The root cause inference results include wear, scratches, and foreign object blockage.

[0015] Optionally, the cloud-based intelligent management platform further includes: The edge computing unit receives and stores the raw data, preprocessed data, multimodal feature vectors, and diagnostic results of multiphysics field signals uploaded by the edge computing unit, thereby enabling the storage and management of massive historical data. Data analysis is conducted based on the massive historical data to form a data source support for the training and optimization of deep learning models; The trained and optimized deep learning model is delivered online to the edge computing unit to achieve model updates and iterations; Configure an integrated user interface that includes data visualization, equipment health status assessment, fault trend analysis, and predictive maintenance recommendations, allowing users to view and operate the interface.

[0016] To achieve the above objectives, a second aspect of the present invention provides an online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve, comprising: The multimodal signal acquisition module is used to deploy multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls, and valve disc top rod to simultaneously acquire pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals, and axial micro-displacement signals. The signal preprocessing and preliminary feature extraction module is used to filter, denoise and normalize the acquired multi-physics field signals based on the edge computing unit, and extract time-domain statistical features and frequency-domain transform features to generate preliminary feature vectors. The multimodal fusion diagnostic reasoning module is used to input the preliminary feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extract deep spatial features through multi-channel feature encoding, determine the dynamic weight of each signal mode by using intermodal attention weighting, and output multidimensional diagnostic results after temporal feature encoding and temporal attention focusing. The cloud-based data processing and health assessment module is used to upload multidimensional diagnostic results to the cloud-based intelligent management platform, combine historical data to generate visualized health status reports, trend analysis curves and predictive maintenance suggestions, and dynamically calculate the comprehensive sealing health index to quantitatively assess the valve's health status.

[0017] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing an online monitoring and intelligent diagnosis method for the sealing performance of a pilot-operated safety valve as described in the first aspect embodiment.

[0019] To achieve the above objectives, the fourth aspect of this application proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements an online monitoring and intelligent diagnosis method for the sealing performance of a pilot-operated safety valve as described in the first aspect embodiment.

[0020] The embodiments of the present invention have the following beneficial effects: 1. Significantly improved monitoring accuracy and reliability. This invention leverages the redundancy and complementarity of four modal information—pressure, acoustic emission, temperature, and displacement—combined with an intelligent fusion algorithm. This effectively overcomes the technical limitations of single-sensor methods being susceptible to environmental interference, greatly enhancing the sensitivity and reliability of detecting minute leaks. It breaks through the limitations of traditional technologies that can only qualitatively determine "whether there is a leak," accurately locating leaking components, diagnosing the root cause of the fault, and quantifying the severity of the leak, providing users with clear and actionable maintenance guidance.

[0021] 2. Achieve predictive maintenance and full lifecycle management. Acoustic emission and micro-displacement sensing are highly sensitive to structural micro-damage and valve disc micro-movements. Combined with the pattern recognition capabilities of artificial intelligence models for weak features, early warnings can be issued before significant changes in macroscopic parameters of leakage occur, achieving true predictive maintenance. At the same time, based on the dynamic calculation of the "comprehensive sealing health index" based on fused diagnostic results, quantitative management of valve health status and lifecycle prediction can be achieved.

[0022] 3. Adaptable to large-scale industrial deployment needs. This invention adopts an edge-cloud collaborative computing architecture, which balances real-time response to monitoring data with deep big data analysis capabilities, effectively optimizing network bandwidth utilization efficiency and meeting the deployment and application requirements of large-scale industrial sites. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating an online monitoring and intelligent diagnosis method for the sealing performance of a pilot-operated safety valve, provided in an embodiment of the present invention; Figure 2 A flowchart of the intelligent diagnostic method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the sensor installation location structure provided in an embodiment of the present invention; Figure 4 This is a structural diagram of an online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve, provided in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] The following description, with reference to the accompanying drawings, describes an online monitoring and intelligent diagnostic method and apparatus for the sealing performance of a pilot-operated safety valve according to an embodiment of the present invention.

[0027] Example 1 This invention provides an online monitoring and intelligent diagnostic method for the sealing performance of a pilot-operated safety valve. Figure 1 This is a schematic flowchart illustrating an online monitoring and intelligent diagnostic method for the sealing performance of a pilot-operated safety valve provided in an embodiment of the present invention. Figure 2 A flowchart illustrating the intelligent diagnostic method provided in an embodiment of the present invention. Figure 1 , Figure 2 As shown, the method includes the following steps: S1 deploys multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls, and valve disc top rod to simultaneously collect pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals, and axial micro-displacement signals.

[0028] In this embodiment, the multimodal sensing acquisition subsystem needs to be precisely integrated and deployed in the key parts of the pilot-operated safety valve in order to achieve comprehensive and synchronous acquisition of multi-physical field signals related to the sealing state.

[0029] The specific deployment method is as follows: (e.g.) Figure 3As shown, at least two high-frequency pressure sensors are installed, one at the inlet of the main valve and the other in the control chamber of the pilot valve. These sensors are used to monitor system pressure fluctuations and abnormal pressure changes in the pilot valve control chamber in real time, providing pressure-based data support for determining whether the seal has failed. At least two broadband acoustic emission sensors are also installed, one close to the main valve body near the main valve seat and the other close to the pilot valve body near the pilot valve seat. They are preferably fixed with magnetic mounts to ensure a tight fit between the sensors and the valve body, thus efficiently capturing high-frequency stress wave signals generated by leakage. This signal accurately reflects minor damage to the sealing surface and subtle changes in the early stages of leakage. At least two high-precision temperature sensors, using PT100 temperature sensors, are installed on the inlet and outlet pipe walls of the main valve, respectively. These sensors monitor temperature differences caused by the throttling effect of leakage, indirectly sensing leakage phenomena using the Joule-Thomson effect. At least one micro-displacement sensor, employing a laser micro-displacement sensor, is installed non-contactly aligned with the main valve disc rod to monitor the axial micro-displacement of the valve disc in real time, accurately capturing abnormal floating or vibration of the valve disc. These minute movements are often early signs of sealing performance degradation.

[0030] In this embodiment, the types of signals acquired by each sensor are different, and the corresponding acquisition parameters are also specifically set: the analog voltage signals of the pressure sensor and displacement sensor are acquired by a multi-channel, 24-bit Sigma-Delta ADC with a sampling rate of up to 256kSPS; the signal of the acoustic emission sensor is first amplified by 40dB by a 1MHz bandwidth preamplifier, and then acquired by a high-speed 12-bit ADC at a rate of 2MSPS; the temperature sensor uses a dedicated PT100 conversion chip to interact with the subsequent edge computing unit via an I2C bus interface, ultimately achieving synchronous acquisition of data from all multimodal sensors.

[0031] S2, based on the edge computing unit, performs filtering, denoising and normalization processing on the acquired multi-physics field signals, and extracts time-domain statistical features and frequency-domain transform features to generate preliminary feature vectors.

[0032] In this embodiment, the edge computing unit is electrically connected to the multimodal sensing and acquisition subsystem. This unit is an embedded hardware platform based on a high-performance SoC. The core is the Rockchip RK3568 SoC, which contains a quad-core ARM Cortex-A55 CPU and an NPU with a computing power of 1 TOPS, which can efficiently support data processing and the operation of subsequent intelligent algorithms.

[0033] Furthermore, the acquired multiphysics field raw signals are first transmitted to the edge computing unit, where targeted preprocessing operations are performed. Specifically, the preprocessing includes filtering, denoising, and normalization: filtering removes high-frequency interference noise (such as electromagnetic interference and mechanical vibration interference in industrial settings); denoising algorithms further purify the signal and improve its signal-to-noise ratio; and normalization unifies the modal signals with different dimensions and amplitude ranges to the same numerical range, avoiding the impact of large differences in signal amplitude on subsequent feature extraction and model calculation.

[0034] After signal preprocessing, the edge computing unit performs feature extraction operations on each modal signal: on the one hand, it extracts time-domain statistical features, including parameters that reflect the signal's time-domain distribution characteristics such as peak value, mean, variance, kurtosis, and skewness; on the other hand, it performs frequency-domain transformation (such as Fourier transform, wavelet transform, etc.) on the preprocessed signal to generate a time-frequency spectrum, and extracts frequency-domain transformation features from it, including parameters that reflect the signal's frequency distribution characteristics such as dominant frequency, band energy, and spectral peak value. Finally, in this embodiment, the extracted time-domain statistical features and frequency-domain transformation features are fused to generate a preliminary feature vector that can comprehensively characterize the sealing state of the safety valve, providing a data foundation for subsequent intelligent diagnosis.

[0035] In addition, in this embodiment, the edge computing unit is also equipped with a storage unit (including DDR4 memory and eMMC flash memory) for temporarily storing intermediate data, extracted preliminary feature vectors and related applications during the preprocessing process. It is also equipped with a power management unit, which uses a dedicated PMIC chip to provide stable and efficient multi-channel voltage rails for the entire unit and supports wide voltage input to ensure stable operation of the edge computing unit in complex industrial environments.

[0036] S3 inputs the preliminary feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extracts deep spatial features through multi-channel feature encoding, and determines the dynamic weight of each signal mode by using intermodal attention weighting. After temporal feature encoding and temporal attention focusing, it outputs multidimensional diagnostic results.

[0037] In this embodiment, the edge computing unit inputs the preliminary feature vector generated in step S2 into a pre-deployed spatiotemporal attention multimodal fusion diagnostic model (i.e., the STAMF-DN model) for intelligent diagnostic inference. This model is trained and optimized in the cloud using a large amount of laboratory simulated fault data (such as labeled data under various typical fault conditions like main valve seat scratches, pilot valve seat wear, piston seal aging, and foreign object jamming), and then deployed and run on the edge side to ensure the model has good fault diagnosis performance.

[0038] The specific model inference process includes the following steps: First, multi-channel feature encoding: the model uses parallel convolutional neural network branches to perform deep mining on the preliminary feature vectors of each modality, extracting the deep spatial features corresponding to each modality data, and fully mining the key information related to sealing faults hidden in different modal signals; next, inter-modal attention weighting is performed, dynamically learning and assigning importance weights of different modal features under the current working conditions through an attention mechanism, utilizing the redundancy and complementarity of four modal information (pressure, acoustic emission, temperature, and displacement) to effectively overcome the shortcomings of single sensing methods being susceptible to interference and improve the reliability of diagnosis; then, temporal feature encoding is performed, inputting the weighted and fused feature sequence into a recurrent neural network to capture the time dependence in the fault evolution process and accurately reflect the temporal change law from the initiation to the development of the fault; next, temporal attention focusing is performed, applying the attention mechanism again to the output of the temporal encoder to focus on the key time steps before and after the fault occurs, strengthening the role of key fault information and improving diagnostic accuracy; finally, a multi-dimensional diagnostic result vector is generated through the output layer.

[0039] The multidimensional diagnostic results include at least three core dimensions: first, sealing status assessment, which is specifically divided into four levels: normal, minor leakage, moderate leakage, and severe leakage, to achieve a quantitative assessment of the severity of leakage; second, leakage location, which can accurately locate the specific leaking components such as the main valve seat, pilot valve seat, and piston seal; and third, root cause inference, which clarifies the specific types of faults such as wear, scratches, foreign object blockage, and aging of the sealing ring.

[0040] It is understood that, in the embodiments of this application, acoustic emission and micro-displacement sensing are extremely sensitive to structural micro-damage and valve disc micro-movement. Combined with the model's ability to recognize patterns of weak features, fault warnings can be achieved before significant changes in macroscopic parameters of leakage occur, providing support for predictive maintenance.

[0041] S4 uploads multidimensional diagnostic results to the cloud-based intelligent management platform, combines historical data to generate a visualized health status report, trend analysis curves, and predictive maintenance suggestions, and dynamically calculates the comprehensive sealing health index to quantitatively assess the valve's health status.

[0042] In this embodiment, the edge computing unit uploads the multidimensional diagnostic results output in step S3, along with some key raw data fragments, to the cloud-based intelligent management platform via an integrated communication module. The communication module integrates a gigabit Ethernet port and is extended with a 5G communication module and a LoRaWAN module via an M.2 interface. It can flexibly select the communication method according to the actual network conditions of the industrial site, taking into account the real-time performance and stability of data transmission, while optimizing network bandwidth utilization, making it suitable for large-scale industrial site deployment.

[0043] It is clear that the cloud-based intelligent management platform is a software system deployed on a cloud server. After receiving data uploaded from one or more edge computing units, it performs in-depth analysis by combining this data with the massive historical diagnostic data stored on the platform. Specifically, the cloud platform dynamically calculates the "Comprehensive Sealing Health Index" based on the fused diagnostic results, enabling quantitative management and full lifecycle prediction of valve health status. It also generates visualized health status reports and trend analysis curves, clearly presenting the changing trends of valve sealing performance, and providing targeted predictive maintenance suggestions based on fault type and severity. When the "Comprehensive Sealing Health Index" falls below a preset threshold, the cloud-based intelligent management platform automatically generates a maintenance work order and pushes it to the user's terminal, providing clear and actionable maintenance guidance and avoiding unnecessary downtime for repairs.

[0044] Furthermore, in this embodiment, the cloud-based intelligent management platform also undertakes the functions of continuous training, optimization, and online updates of the deep learning model. By continuously incorporating new fault data to optimize model parameters, it improves the diagnostic performance of the entire system and ensures the long-term stable and reliable operation of the system.

[0045] In the application of one embodiment of the present invention, the implementation process is as follows: In the laboratory, a valve failure simulation platform was built. Taking a pilot-operated safety valve with a nominal pressure of 10 MPa and nitrogen as the medium as an example: Normal sample collection: Data was collected after stable operation at 9MPa pressure for 1 hour and labeled as {normal, none, none}. Slight leakage-wear sample collection: A slightly worn main valve seat was replaced (surface roughness Ra increased by 0.8μm), and data was collected at 9MPa. At this time, the leakage rate was 5 bubbles / min, labeled as {slight leakage, main valve seat, wear}. Severe leakage-jamming sample collection: A 50μm diameter metal particle was placed between the sealing surfaces of a normal valve seat, and data was collected at 9MPa. At this time, the leakage rate was greater than 50 bubbles / min, and the pressure dropped significantly, labeled as {severe leakage, main valve seat, foreign object jamming}. By combining different fault types, locations, and severity levels, we constructed a massive dataset containing hundreds of thousands of samples for supervised learning training of the STAMF-DN model. The edge computing unit uploads diagnostic results (such as "moderate leakage, valve seat, scratches") to the cloud platform 500 via a 5G network, where they can be displayed on the local HMI interface. On the web interface of the cloud platform 500, users can see real-time curves of the valve health index, historical alarm records, fault diagnosis reports, and remaining life predictions and maintenance recommendations based on trend analysis.

[0046] Example 2 This invention provides an online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve. Figure 4 This is a schematic flowchart of an online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve, provided in an embodiment of the present invention. Figure 4 As shown, the device includes: The multimodal signal acquisition module 100 is used to deploy multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls and valve disc top rod to simultaneously acquire pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals and axial micro-displacement signals. The signal preprocessing and preliminary feature extraction module 200 is used to filter, denoise and normalize the acquired multi-physics field signals based on the edge computing unit, and extract time-domain statistical features and frequency-domain transform features respectively to generate preliminary feature vectors. The multimodal fusion diagnostic reasoning module 300 is used to input the preliminary feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extract deep spatial features through multi-channel feature encoding, determine the dynamic weight of each signal mode by using intermodal attention weighting, and output multidimensional diagnostic results after temporal feature encoding and temporal attention focusing. The cloud-based data processing and health assessment module 400 is used to upload multidimensional diagnostic results to the cloud-based intelligent management platform, combine historical data to generate visualized health status reports, trend analysis curves and predictive maintenance suggestions, and dynamically calculate the comprehensive sealing health index to quantitatively assess the valve's health status.

[0047] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0048] Example 3 To implement the methods of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0049] Example 4 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0051] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0052] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for online monitoring and intelligent diagnosis of the sealing performance of a pilot-operated safety valve, characterized in that, include: S1, deploy multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls and valve disc top rod, and simultaneously collect pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals and axial micro-displacement signals; S2, based on the edge computing unit, performs filtering, denoising and normalization processing on the acquired multi-physics field signals, and extracts time-domain statistical features and frequency-domain transform features respectively to generate preliminary feature vectors; S3. The initial feature vector is input into the spatiotemporal attention multimodal fusion diagnostic model. Deep spatial features are extracted through multi-channel feature encoding, and the dynamic weights of each signal mode are determined by intermodal attention weighting. After temporal feature encoding and temporal attention focusing, multidimensional diagnostic results are output. S4 uploads multidimensional diagnostic results to the cloud-based intelligent management platform, combines historical data to generate a visualized health status report, trend analysis curves, and predictive maintenance suggestions, and dynamically calculates the comprehensive sealing health index to quantitatively assess the valve's health status.

2. The method according to claim 1, characterized in that, The method of deploying multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls, and valve disc rod to simultaneously acquire pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals, and axial micro-displacement signals also includes: The multimodal sensor consists of at least two high-frequency pressure sensors, at least two broadband acoustic emission sensors, at least two high-precision temperature sensors, and at least one micro-displacement sensor. The high-frequency pressure sensors are deployed at the main valve inlet and the pilot valve control chamber, the broadband acoustic emission sensors are attached to the valve body near the main valve seat and the pilot valve seat, the high-precision temperature sensors are deployed on the main valve inlet pipe wall and the outlet pipe wall, and the micro-displacement sensor is aligned with the main valve disc rod in a non-contact manner. The sensors synchronously collect pressure fluctuation signals, high-frequency acoustic emission stress wave signals from leakage, temperature difference change signals from leakage throttling effect, and axial micro-displacement signals of the valve disc, all related to the sealing state of the safety valve.

3. The method according to claim 2, characterized in that, The edge computing and diagnostic unit is electrically connected to the sensing subsystem and is an embedded hardware platform based on a high-performance SoC. Its core integrates a neural network processing unit. This unit synchronously acquires and preprocesses multimodal sensor data through a high-speed multi-channel ADC and digital interface, and runs a lightweight anomaly detection algorithm and core intelligent diagnostic model in real time at the edge. It also integrates multiple industrial communication models to achieve data interaction with the cloud.

4. The method according to claim 3, characterized in that, The step of extracting time-domain statistical features and frequency-domain transform features respectively to generate a preliminary feature vector also includes: Perform time-frequency domain transformation on the preprocessed multiphysics field signal to generate a time-frequency spectrum characterizing the time-frequency distribution of the signal; Based on the time-frequency spectrum, the time-domain statistical features and frequency-domain transformation features of the preprocessed multi-physics field signal are extracted to form the preliminary feature vector.

5. The method according to claim 4, characterized in that, The method of inputting feature vectors into a spatiotemporal attention multimodal fusion diagnostic model, extracting deep spatial features through multi-channel feature encoding, determining the dynamic weights of each signal mode using intermodal attention weighting, and outputting multidimensional diagnostic results after temporal feature encoding and temporal attention focusing, further includes: Multi-channel feature encoding is performed on each modality of data, and deep spatial features are extracted from each modality of data using parallel convolutional neural network branches. Perform intermodal attention weighting, dynamically learn and assign importance weights to different modal features under the current working conditions through an attention mechanism; Perform temporal feature encoding, input the weighted and fused feature sequence into a recurrent neural network, and capture the time dependence of fault evolution; The execution time attention focus applies an attention mechanism to the output of the timing encoder again, focusing on the critical time steps before and after the fault occurs; The diagnostic results are output, and a multi-dimensional diagnostic vector is generated through the output layer. The multi-dimensional diagnostic vector includes at least sealing status determination, leakage location location, and root cause inference. The sealing status determination results include normal, minor leakage, moderate leakage, and severe leakage. The leakage location location results include the main valve seat, pilot valve seat, and piston seal. The root cause inference results include wear, scratches, and foreign object blockage.

6. The method according to claim 5, characterized in that, The cloud-based intelligent management platform also includes: The edge computing unit receives and stores the raw data, preprocessed data, multimodal feature vectors, and diagnostic results of multiphysics field signals uploaded by the edge computing unit, thereby enabling the storage and management of massive historical data. Data analysis is conducted based on the massive historical data to form a data source support for the training and optimization of deep learning models; The trained and optimized deep learning model is delivered online to the edge computing unit to achieve model updates and iterations; Configure an integrated user interface that includes data visualization, equipment health status assessment, fault trend analysis, and predictive maintenance recommendations, allowing users to view and operate the interface.

7. An online monitoring and intelligent diagnostic device for the sealing performance of a pilot-operated safety valve, characterized in that, include: The multimodal signal acquisition module is used to deploy multimodal sensors at the main valve inlet, pilot valve control chamber, main valve seat, pilot valve seat, main valve inlet and outlet pipe walls, and valve disc top rod to simultaneously acquire pressure fluctuation signals, acoustic emission stress wave signals, temperature difference change signals, and axial micro-displacement signals. The signal preprocessing and preliminary feature extraction module is used to filter, denoise and normalize the acquired multi-physics field signals based on the edge computing unit, and extract time-domain statistical features and frequency-domain transform features to generate preliminary feature vectors. The multimodal fusion diagnostic reasoning module is used to input the preliminary feature vector into the spatiotemporal attention multimodal fusion diagnostic model, extract deep spatial features through multi-channel feature encoding, determine the dynamic weight of each signal mode by using intermodal attention weighting, and output multidimensional diagnostic results after temporal feature encoding and temporal attention focusing. The cloud-based data processing and health assessment module is used to upload multidimensional diagnostic results to the cloud-based intelligent management platform, combine historical data to generate visualized health status reports, trend analysis curves and predictive maintenance suggestions, and dynamically calculate the comprehensive sealing health index to quantitatively assess the valve's health status.

8. The apparatus according to claim 7, characterized in that, The multimodal fusion diagnostic reasoning module is also used for: Multi-channel feature encoding is performed on each modality of data, and deep spatial features are extracted from each modality of data using parallel convolutional neural network branches. Perform intermodal attention weighting, dynamically learn and assign importance weights to different modal features under the current working conditions through an attention mechanism; Perform temporal feature encoding, input the weighted and fused feature sequence into a recurrent neural network, and capture the time dependence of fault evolution; The execution time attention focus applies an attention mechanism to the output of the timing encoder again, focusing on the critical time steps before and after the fault occurs; The diagnostic results are output, and a multi-dimensional diagnostic vector is generated through the output layer. The multi-dimensional diagnostic vector includes at least sealing status determination, leakage location location, and root cause inference. The sealing status determination results include normal, minor leakage, moderate leakage, and severe leakage. The leakage location location results include the main valve seat, pilot valve seat, and piston seal. The root cause inference results include wear, scratches, and foreign object blockage.

9. An electronic device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images