Valve fault positioning method, positioning system and valve

Abstract

The invention relates to the technical field of valves. The invention discloses a valve fault positioning method and system and a valve, and the method comprises the steps: collecting the visible light, ultrasonic echo and pressure pulse data of a valve body through a detection device, and synchronously recording and associating the space coordinates of a sampling point; after the three kinds of data are preprocessed, the three kinds of data are registered and fused into three-modal data under a unified coordinate system according to valve body positioning marks; carrying out clustering segmentation to obtain analysis sub-regions; comparing threshold values to identify three types of abnormal regions; analyzing abnormal space association; calculating a comprehensive score according to abnormal conditions and scenes; and comparing threshold values to mark fault areas, output positions, types and maintenance priorities. According to the invention, surface, internal and functional faults of the valve are covered by means of three-mode collection and fusion, and the recognition comprehensiveness and accuracy are improved. Through clustering segmentation and scene weighted scoring, a fault area is accurately positioned, a single or composite fault type and a maintenance priority are clarified, and empirical misjudgment is avoided.

Description

Technical Field

[0001] This invention relates to the field of valve technology, and in particular to a valve fault location method, a location system, and a valve. Background Technology

[0002] In industrial sectors such as petrochemicals, water treatment, and energy transmission, valves, as core components for fluid control and pressure regulation, directly impact the safety and stability of the entire process. During long-term use, valves are prone to various malfunctions due to factors such as media corrosion, pressure fluctuations, and mechanical wear. These include valve body surface corrosion, internal cracks, internal leakage caused by seal failure, and switching malfunctions due to valve seat wear. Failure to detect and locate these malfunctions in a timely manner can not only lead to media leakage and energy waste but also potentially cause equipment downtime and safety accidents. Therefore, accurate detection and location of valve malfunctions are of significant practical importance.

[0003] Current methods for detecting valve faults primarily rely on single-modal testing, which fails to achieve comprehensive coverage. For example, visible light testing can only identify surface defects such as corrosion and coating peeling, but cannot detect hidden internal cracks or valve seat wear. While ultrasonic testing can examine the internal structure of the valve body, its accuracy in identifying minute surface corrosion is low, and it is easily affected by environmental vibrations. Pressure pulse testing can only determine the presence of internal leaks through changes in fluid pressure, but it cannot pinpoint the location of the leak or associated structural damage. Furthermore, existing detection methods often lack effective integration of different types of test data. Each test data is stored independently with inconsistent spatial coordinates, making it difficult to establish a correspondence between fault characteristics and spatial location. This results in low fault location accuracy and high rates of missed and false detections.

[0004] Furthermore, existing valve fault detection systems suffer from poor module coordination and low data processing efficiency. In some systems, there is a lack of stable synchronization mechanisms between the data acquisition and data processing modules, resulting in insufficient reliability of the correlation between the spatial coordinates of sampling points and the detection data, which affects the accuracy of subsequent data analysis. Simultaneously, detection results are often output as a single fault type, failing to classify the severity of the fault based on the actual application scenario of the valve, and failing to clearly define the fault propagation path and scope of impact. This makes it difficult for maintenance personnel to quickly determine maintenance priorities, increasing equipment maintenance costs and downtime. In traditional valve structures, the design of the cooperation between the sealing components and the elastic mechanism is not optimized, making them prone to failure due to seal failure. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention discloses a valve fault location method, a location system, and a valve that covers valve surface, internal, and functional faults through three-modal data acquisition and fusion, clustering and segmentation, and dynamic weighted scoring.

[0006] This invention discloses a valve comprising:

[0007] The valve body is a tubular shell structure, and includes an inlet and an outlet. The inlet and outlet penetrate the valve body to form a fluid channel.

[0008] The switch adjustment part is fixedly installed inside the liquid inlet of the valve body. The switch adjustment part includes a hollow receiving cavity. A switch inlet is provided through the end of the switch adjustment part near the liquid inlet, and a switch outlet is provided through the end of the switch adjustment part away from the liquid inlet. The switch outlet is recessed inward to form an outlet limiting part. The diameter of the switch inlet is smaller than the inner diameter of the receiving cavity.

[0009] An elastic part is disposed inside the receiving cavity and one end abuts against the outlet limiting part;

[0010] The fixed ball has a diameter smaller than that of the receiving cavity. The fixed ball is located inside the receiving cavity at the end of the elastic part away from the outlet limit part, and the fixed ball cooperates with the switch inlet to form a blockage.

[0011] This invention discloses a valve fault location method, which includes the following steps:

[0012] S1: The detection device is equipped with a visible light acquisition unit, an ultrasonic detection unit and a pressure pulse detection unit. Data is collected along the preset path of the key components of the valve body to obtain visible light image data, ultrasonic echo data and pressure pulse data. The spatial coordinates of the sampling points are recorded and the data is associated simultaneously.

[0013] S2: After distortion correction or noise reduction of the three types of data respectively, spatial correspondence is established based on the valve body positioning mark, and the data are registered and fused into three-modal detection data in a unified coordinate system.

[0014] S3: Cluster and segment the three-modal data, set uniformly distributed initial cluster centers, and delineate initial cluster regions;

[0015] Calculate the spatial distance between the sampling points and the center within the region, as well as the distance to the three-modal features;

[0016] The cluster centers are adjusted based on the sum of the two types of distances, and the regions are redefined. This process is repeated until the sum of the distances no longer changes. The boundaries of the stable cluster regions are used as the sub-regions for analysis.

[0017] S4: Compare and analyze the sub-regions with the preset normal thresholds to identify areas with abnormal appearance, abnormal ultrasound, and abnormal pressure.

[0018] S5: Analyze the spatial correlation of the three abnormal regions, determine whether they overlap, and identify the abnormal propagation path and the scope of impact.

[0019] S6: Set the appearance, ultrasonic, and pressure scores according to the proportion of abnormal area and the degree of feature deviation, assign weights according to the valve application scenario, and calculate the comprehensive score;

[0020] S7: Compare the overall score with the fault level threshold. If the score exceeds the threshold, mark the fault area and output the location, type and maintenance priority.

[0021] Furthermore, the data collected in S1 specifically includes: data collection methods including handheld setup and data collection using fixed rack-mounted equipment;

[0022] Handheld devices collect samples at intervals set according to component size, while fixed-frame devices collect samples continuously at set intervals.

[0023] Furthermore, the registration and fusion in S2 specifically involves:

[0024] Zhang's calibration method was used to correct visible light images, and wavelet filtering was used to reduce noise in ultrasound and pressure data.

[0025] The pixel coordinates of the identification marker and the sampling point coordinates form a reference pair. The coordinate transformation matrix is ​​calculated to align the data, and the three data features are fused using a feature-level fusion method.

[0026] Furthermore, the clustering segmentation in S3 specifically involves:

[0027] Spatial distance is calculated based on the coordinate difference between the sampling point and the center.

[0028] In the three-modal feature distance, the color difference is taken from the difference of RGB channels, the ultrasonic difference is taken from the difference of echo peak value and propagation time, and the pressure difference is taken from the difference of pulse peak value and duration. The three are dynamically weighted and summed according to the valve body application scenario and data quality.

[0029] Repeat the adjustment until the sum of the distances between the two iterations is ≤0.01.

[0030] Furthermore, anomaly detection in S4 specifically involves:

[0031] Normal thresholds include: appearance, ultrasound, and pressure thresholds;

[0032] If the percentage of sampling points exceeding the threshold in a sub-region is ≥10%, it is marked as abnormal.

[0033] Furthermore, the overall score in S6 specifically includes:

[0034] Appearance rating, ultrasound rating, and pressure rating.

[0035] Furthermore, the fault flags in S7 are specifically as follows:

[0036] A single anomaly exceeding the corresponding threshold marks a single fault region; two or more anomalies exceeding the comprehensive threshold mark a composite fault region.

[0037] Output coordinate range and high / medium / low maintenance priority.

[0038] This invention discloses a valve fault location system for implementing any of the valve fault location methods described above, comprising:

[0039] Three-modal data acquisition module: acquires visible light images of the valve body, ultrasonic echoes, and pressure pulse data; uses laser positioning to record the coordinates of sampling points and associates the data.

[0040] Data fusion module: Corrects visible light image distortion, reduces noise in ultrasonic and pressure data, registers the three types of data with valve body positioning marks, and generates three-modal detection data;

[0041] Clustering and segmentation module: Set initial cluster centers, calculate the spatial and feature distances between sampling points and centers, iteratively adjust the centers to divide regions, and obtain sub-regions for analysis;

[0042] Multi-dimensional anomaly identification module: compares features with normal thresholds to identify areas of abnormal appearance, ultrasound, and pressure.

[0043] Anomaly Correlation Analysis Module: Identifies overlapping anomalies, analyzes propagation paths and the scope of impact;

[0044] Dynamic scoring module: Calculates a comprehensive score based on the weight of abnormal situations and scenarios;

[0045] Fault location output module: compares the score with the threshold, marks the fault area and outputs information.

[0046] The three-modal data acquisition module transmits data to the data fusion module via RS485. The data fusion module transmits the data to the clustering and segmentation module via USB3.0. The clustering and segmentation module transmits the data to the multi-dimensional anomaly identification module via Ethernet. The multi-dimensional anomaly identification module transmits the data to the anomaly correlation analysis module via CAN bus. The anomaly correlation analysis module transmits the data to the dynamic scoring module via serial port. The dynamic scoring module transmits the data to the fault location output module via SPI.

[0047] Furthermore, in the three-modal data acquisition module, the visible light unit has a resolution ≥1080P, the ultrasonic unit has a frequency of 2-5MHz, and the pressure unit has a range of 0-10MPa;

[0048] The laser positioning unit supports real-time coordinate uploading.

[0049] The beneficial effects of this invention are:

[0050] This invention utilizes a three-modal data acquisition and fusion technology. A visible light acquisition unit captures the surface appearance features of the valve body, an ultrasonic testing unit identifies internal structural defects, and a pressure pulse detection unit monitors sealing and fluid flow anomalies. Simultaneously, the spatial coordinates of the sampling points are correlated, and after distortion correction, noise reduction, and coordinate registration, three-modal detection data in a unified coordinate system is generated. This method overcomes the limitations of single-mode detection, simultaneously covering potential faults at the valve body surface, internal structure, and functional performance levels. It effectively avoids the problems of overlooking internal cracks due to external anomalies or ignoring surface corrosion during pressure testing, as seen in traditional single-mode detection, thus improving the comprehensiveness and accuracy of valve fault identification.

[0051] This invention employs a clustering segmentation and dynamic weighted scoring mechanism. It achieves region division by uniformly setting initial cluster centers and calculating spatial location and trimodal feature distances, ensuring that the analyzed sub-regions correspond to the actual functional zones of the valve body. Simultaneously, it assigns weights to different detection dimensions based on the valve application scenario, calculating a comprehensive score based on the proportion of abnormal area and the degree of feature deviation, and matching the fault level threshold. This design can accurately locate fault area boundaries, clearly identify single or compound fault types, and the maintenance priority division aligns with actual operating conditions, avoiding misjudgments or priority confusion caused by reliance on experience in traditional fault diagnosis, thus improving the targeting and efficiency of valve fault repair.

[0052] The valve fault location system and method of this invention are highly compatible. Each module achieves efficient data flow through a stable data transmission interface, while the laser positioning unit ensures the accuracy of sampling point coordinates and guarantees reliable data correlation. Simultaneously, the designed valve structure achieves efficient sealing through the cooperation of the elastic part and the fixed ball, and the outlet limiting part of the switching adjustment unit further optimizes sealing performance. This dual-dimensional approach, from both the detection end and the equipment end, ensures the valve's operational reliability. The overall solution not only reduces the missed detection rate and false detection rate of valve faults but also reduces downtime and maintenance costs caused by faults, providing strong support for the long-term stable operation of the valve. Attached Figure Description

[0053] Figure 1 This is a flowchart of a valve fault location method according to an embodiment of this application.

[0054] Figure 2 This is a schematic diagram of the valve structure in one embodiment of this application.

[0055] Figure 3 This is a cross-sectional structural diagram of a valve in an embodiment of this application.

[0056] In the figure: valve 100, valve body 11, switch adjustment part 12, outlet limit part 121, elastic part 13, fixed ball 14. Detailed Implementation

[0057] To enable those skilled in the art to better understand the present invention, the technical solutions in the specific embodiments of the present invention will be clearly and completely described below.

[0058] This invention discloses a valve fault location method, which includes the following steps:

[0059] S1: Data Acquisition Steps:

[0060] The detection device includes a handheld detection device or a fixed frame-mounted detection device, which is equipped with a visible light acquisition unit, an ultrasonic detection unit, and a pressure pulse detection unit.

[0061] The data collection targets are key functional components of the valve body, including the flange sealing surface, the valve stem and valve seat mating area, the pressure-bearing side wall of the valve body and the valve cover sealing interface. The data collection path is planned in advance to cover all key components along the circumference and axis of the valve body, with no sampling areas omitted.

[0062] After the detection device is activated, the visible light acquisition unit captures images of the valve body surface to detect surface corrosion, coating peeling, structural deformation, and water residue. The ultrasonic detection unit emits ultrasonic signals through a probe against the valve body surface and receives echo data to detect internal cracks, valve seat wear, and seal damage. The pressure pulse detection unit contacts preset detection points on the valve body, such as the outer side of the flange sealing surface or the valve body flow channel interface, to collect pressure pulse signals to detect internal leakage, seal failure, and abnormal fluid flow. During the acquisition process, the laser positioning unit on the detection device simultaneously records the spatial coordinates of each sampling point with an accuracy controlled within ±0.1mm, including X, Y, and Z three-dimensional coordinates. The coordinate information is embedded into the headers of the corresponding visible light image data, ultrasonic echo data, and pressure pulse data, achieving one-to-one association storage between the three types of data and spatial location.

[0063] S2: Data registration and fusion steps:

[0064] The three types of data collected by S1 were preprocessed separately. Zhang's calibration method was used to correct lens distortion in the visible light image data. Multiple sets of checkerboard calibration board images were taken to determine the distortion coefficients involved in the lens. Then, the collected valve body image was corrected pixel by pixel to eliminate edge stretching and distortion, ensuring the spatial accuracy of surface appearance features. Wavelet threshold filtering algorithm was used to denoise the ultrasonic echo data and pressure pulse data. The original signal was first decomposed into three layers of wavelet coefficients. The high-frequency noise coefficients were filtered by setting a threshold according to the principle of maximizing the signal-to-noise ratio. Then, the effective signal was recovered by wavelet reconstruction to remove the influence of environmental vibration and electromagnetic interference on the data.

[0065] Using the pre-set positioning marks on the valve body surface, including the center of the flange bolt holes, the valve body scale lines, and the positioning groove at the top of the valve stem, as the registration reference, the pixel coordinates corresponding to the marks are identified in the visible light image, such as the pixel coordinates of the bolt hole center (x1, y1). The three-dimensional spatial coordinates of the sampling points corresponding to the same positioning mark are identified in the ultrasonic echo data and pressure pulse data, such as the spatial coordinates of the bolt hole center (X1, Y1, Z1), forming at least 3 sets of pixel coordinate-spatial coordinate reference point pairs. Based on the reference point pairs, the coordinate transformation matrix is ​​calculated using the least squares method to transform the spatial coordinates of all sampling points in the ultrasonic echo data and pressure pulse data to the pixel coordinate system of the visible light image, realizing a one-to-one correspondence between the three types of data in spatial position.

[0066] To address the feature space heterogeneity of the three types of data, intramodal feature extraction is performed separately. For the corrected visible light image, a convolutional neural network is used to extract features, focusing on capturing the texture features, edge features, and color distribution features of the valve body surface, and outputting a high-dimensional visible light feature vector with uniform dimension.

[0067] For the denoised ultrasonic echo data, time-frequency analysis methods such as wavelet transform are used to extract features, focusing on capturing the spectral peak features, attenuation features, and waveform distortion features of the echo signal, and converting the time-frequency domain information into a fixed-dimensional high-dimensional ultrasonic feature vector; for the denoised pressure pulse data, Fourier transform is used to extract features, focusing on capturing the frequency features, amplitude features, and attenuation time features of the pulse signal, and converting the time-domain signal into a fixed-dimensional high-dimensional pressure feature vector.

[0068] The three extracted high-dimensional feature vectors are subjected to feature alignment processing. Principal component analysis is used to reduce the dimensionality of each modal feature vector, remove redundant information, and map all features to a common feature space of the same dimension. At the same time, based on the spatial correlation of key components of the valve body, semantic correspondence of different modal features is established.

[0069] A feature-level fusion method is adopted to achieve pixel-level fusion. The fusion weights are dynamically configured based on the application scenario of the valve. For example, in the corrosive fluid scenario, the weights of visible light features are 0.4, ultrasonic features are 0.3, and pressure features are 0.3; in the high-pressure gas scenario, the weights of pressure features are 0.45, ultrasonic features are 0.35, and visible light features are 0.2. The key modal features in each scenario are strengthened through an attention mechanism. For example, in the high-pressure scenario, the correlation between pressure features and ultrasonic features is emphasized. The aligned three modal feature vectors are weighted and fused to form a three-modal fusion feature vector of single-modal key features and cross-modal correlation features corresponding to each pixel position.

[0070] Based on the fused feature vectors, three-modal detection data in a unified pixel coordinate system is generated. Each pixel in the data contains comprehensive feature information of the surface appearance, internal structure, and functional performance of the valve body at the corresponding position, providing standardized data support for subsequent S3 clustering segmentation and S4 multi-dimensional anomaly identification.

[0071] S3: Clustering segmentation steps:

[0072] Based on the trimodal detection data generated by S2, a predetermined number of initial cluster centers are evenly set according to the valve body structure. The number of cluster centers is determined according to the valve body size. For example, 30-50 initial cluster centers are set for valve bodies with a diameter ≤200mm to ensure coverage of all key components. The trimodal detection data is divided into several initial cluster regions. Each initial cluster region contains several adjacent pixels and corresponding trimodal feature data.

[0073] To determine the number of initial cluster centers for non-circular valves, the effective detection cross-section of the valve should be clearly defined first. For square valve bodies, the effective detection cross-section is the cross-section containing the flow channel; for eccentric valve bodies, the effective detection cross-section is the eccentric cross-section containing the valve core sealing surface. The baseline number of initial cluster centers is determined based on the actual size of the effective detection cross-section. Fewer initial cluster centers are needed for smaller effective detection cross-sections, and more are needed for larger ones. The baseline number should be controlled between 5 and 100 to avoid either too few centers resulting in coarse clustering and missed minor faults, or too many centers increasing meaningless calculations. Further adjustments to the number of initial cluster centers are made based on the structural characteristics of non-circular valves. For square valve bodies, an additional initial cluster center is added to each of the four corner areas, as corners are stress concentration points and areas prone to faults. For eccentric valve bodies, an additional initial cluster center is added every 10 millimeters along the sealing surface area on the eccentric side of the valve core to ensure sufficiently fine clustering in these high-fault sealing surface areas. Finally, the system parameter interface allows manual fine-tuning of the initial cluster center number within ±20% of the baseline number, further adapting to the actual structure and detection requirements of different non-circular valves, and balancing fault detection accuracy and computational efficiency.

[0074] Using a unified pixel coordinate system as a reference, two types of distances are calculated between all sampling points and the initial cluster center within each initial cluster region. Spatial distance is calculated based on pixel coordinates, while feature distance is calculated based on normalized feature vectors. Spatial distance is calculated based on the difference in three-dimensional coordinates between the sampling point and the cluster center, reflecting the actual geometric distance between the sampling point and the center on the valve body surface. The three-modal feature distance is calculated by separately calculating the visible light color difference between the sampling point and the cluster center, the actual difference in red, green, and blue channel values, the difference in ultrasonic echo characteristics, the difference in echo peak intensity, the difference in ultrasonic propagation time, and the difference in pressure pulse characteristics, the difference in pulse peak value, and the difference in pulse duration.

[0075] In the calculation of the three-modal feature distance, the weight allocation does not use a fixed ratio, but is dynamically adjusted according to the application scenario of the valve body and the real-time data quality: Scenario type adaptation: The system pre-sets multiple scenario weight templates, such as: Corrosive fluid scenario: appearance weight 0.4, ultrasonic weight 0.3, pressure weight 0.3; High-pressure gas scenario: pressure weight 0.45, ultrasonic weight 0.35, appearance weight 0.2; Low-pressure water medium scenario: appearance weight 0.35, ultrasonic weight 0.3, pressure weight 0.35. The system evaluates the signal-to-noise ratio and feature consistency of each modality in real time. If the quality of a certain modality data significantly deteriorates, such as visible light images becoming blurred due to dim ambient light, the system automatically reduces the weight of that modality and increases the weights of other modalities. The final weight is the weighted average of the scenario weight and the real-time reliability weight, ensuring high robustness and accuracy even under complex working conditions.

[0076] After calculating the spatial distance and the three-modal feature distance, the minimum-maximum normalization method is used to normalize both to the [0, 1] interval, obtaining the normalized distance value. The spatial distance and the three-modal feature distance are added together to obtain the total distance between each sampling point and the corresponding cluster center. Based on the total distance, the initial cluster center position is adjusted. The average three-dimensional coordinates of all sampling points in the region and the weighted average of the three-modal features are used as the coordinates and feature values ​​of the new cluster center, and the sampling points are re-divided into the regions corresponding to the cluster centers with the smallest total distance. The steps of adjusting the cluster centers and re-dividing the regions are repeated until the difference between the sum of the total distances of the sampling points in all cluster regions in two adjacent iterations is ≤0.01, that is, the total distance no longer changes significantly. The boundary contour of each stable cluster region is determined as the contour of the analysis sub-region. The boundary contour is a closed contour formed by connecting the edge pixels of the region, and finally forms multiple analysis sub-regions that correspond one-to-one with the actual functional partitions of the valve body, such as the flange sealing area, the valve stem operating area, and the valve body pressure bearing area.

[0077] S4: Multi-dimensional anomaly identification steps:

[0078] Based on experimental data, theoretical calculations, or historical experience, three-modal characteristic thresholds are preset for the valve under normal operating conditions. The normal appearance characteristic thresholds are: red channel value 45-65, green channel value 55-75, blue channel value 75-95, and surface texture entropy value 0.25-0.45. The normal ultrasonic characteristic thresholds are: ultrasonic echo peak intensity 1.8-2.2V, ultrasonic waveform distortion rate ≤5%, and distortion rate being the ratio of the actual waveform to the standard waveform. The normal pressure pulse characteristic thresholds are: pressure pulse peak deviation rate ≤8%, the deviation ratio relative to the valve's rated operating pressure, and pressure pulse decay time 0.8-1.2s. For each analysis sub-region, the average color channel value and average texture entropy value of the visible light image data are extracted and compared with the normal appearance characteristic thresholds. If any parameter exceeds the threshold, and the number of sampling points exceeding the threshold in that sub-region accounts for ≥10% of the total number of sampling points, it is marked as an abnormal appearance region. The average echo peak intensity and average waveform distortion rate of the ultrasound echo data are extracted and compared with the normal ultrasound characteristic thresholds. Areas exceeding the thresholds and meeting the sampling point ratio criteria are marked as ultrasound abnormal regions. The average pulse peak deviation rate and average decay time of the pressure pulse data are extracted and compared with the normal pressure pulse characteristic thresholds. Areas meeting the above conditions are marked as pressure abnormal regions.

[0079] S5: Spatial correlation analysis steps:

[0080] Establish a spatial location correspondence table for abnormal areas in appearance, ultrasound, and pressure. The table records the three-dimensional coordinate range and corresponding anomaly type for each anomaly area. By comparing the coordinate ranges, determine whether two or three anomaly areas overlap at the same valve body location, such as a flange sealing surface area being simultaneously marked as both an appearance and pressure anomaly area. Based on the spatial distribution of the anomaly areas, analyze the propagation path of the anomaly characteristics: for appearance anomaly areas, trace the expansion direction of corrosion and deformation along the valve body structure, such as flange bolt holes and valve stem axial direction. For ultrasound anomaly areas, trace the propagation trend of internal cracks and valve seat wear along the valve body wall thickness direction. For pressure anomaly areas, trace the propagation direction of internal leakage signals along the valve body flow channel. Combining the propagation directions of the three anomalies, determine the comprehensive influence range of the anomaly characteristics, such as whether the pressure anomaly caused by internal leakage covers the entire valve body sealing surface.

[0081] Anomaly propagation path analysis and impact range calculation are performed by establishing a valve body structural topology diagram. Based on the valve body's 3D model or preset structural relationships, a connection diagram between key components is established, such as flange sealing surface - valve body sidewall - valve seat - valve stem. Anomaly area coordinate mapping maps the 3D coordinates of identified appearance, ultrasonic, and pressure anomalies to corresponding nodes in the topology diagram. Propagation path analysis identifies multiple anomalies at the same or adjacent nodes as superimposed anomalies. The spatial continuity of anomaly features is traced along the connection direction of the topology diagram, such as corrosion extending from the flange sealing surface to the valve body sidewall. Combined with a fluid dynamics model, the propagation trend of pressure anomalies along the flow path is predicted. Impact range calculation starts from the anomaly center and extends a certain distance along the propagation direction to form the anomaly impact area. This is then combined with the valve body structural strength model to assess the potential structural risk range. The valve body structure topology diagram includes nodes with geometric centers of key components such as flange sealing surfaces, valve seats, valve stems, and valve body sidewalls. Connections are governed by rules: direct connections are established when the physical distance between two components is ≤3cm, and indirect connections are established when the distance is >3cm but a fluid passage exists. The topology diagram can be automatically generated by importing a 3D valve body model in STEP / IGES format. During topology diagram construction, a 3D valve body model in STEP or IGES format is first imported. The geometric centers of key components such as flange sealing surfaces and valve seats are extracted as topology nodes using a triangular facet segmentation algorithm. Edges are then constructed according to the physical connection relationships of the components: direct connections are established when the physical distance is ≤3cm, and indirect connections are established when the distance is >3cm and a fluid passage exists. For non-standard valve bodies, nodes and connection relationships can be manually added. The topology diagram update frequency is synchronized with the inspection cycle: fixed rack-mounted equipment updates every 5 minutes, while handheld equipment updates automatically after a single inspection. The 3D coordinates of abnormal areas are mapped to a 5mm neighborhood of the corresponding node in the topology diagram via a coordinate transformation matrix, achieving a precise correlation between abnormalities and the valve body structure. For structural anomalies, the shortest path first algorithm is used. If the adjacent nodes of the initial anomaly node show the same or derived anomalies for two consecutive detection cycles, the connecting edge becomes the propagation path. For fluid anomalies, a simplified fluid dynamics model is used, combined with the time difference between the rated flow velocity and pressure anomalies of the medium to infer the propagation path. The propagation distance for structural anomalies is determined according to their severity: 5mm for mild corrosion, 10mm for medium corrosion, 15mm for severe corrosion, and 20% for corrosion covering welds.

[0082] S6: Dynamic weighted scoring steps:

[0083] The scoring values ​​are set according to the actual situation of the abnormal area: Appearance scores are determined by the percentage of abnormal area plus the degree of feature deviation. 20 points are awarded when the percentage of abnormal area is ≤12% and the feature deviation is ≤10%; 50 points are awarded when the percentage of abnormal area is 12%-32% and the feature deviation is 10%-22%; and 80 points are awarded when the percentage of abnormal area is ≥32% and the feature deviation is ≥22%. Ultrasound scores are set similarly: 25 points are awarded when the percentage of abnormal area is ≤10% and the feature deviation is ≤8%; 55 points are awarded when the percentage of abnormal area is 10%-30% and the feature deviation is 8%-20%; and 85 points are awarded when the percentage of abnormal area is ≥30% and the feature deviation is ≥20%. Pressure scores are set similarly: 30 points are awarded when the percentage of abnormal area is ≤8% and the feature deviation is ≤6%; 60 points are awarded when the percentage of abnormal area is 8%-28% and the feature deviation is 6%-18%; and 90 points are awarded when the percentage of abnormal area is ≥28% and the feature deviation is ≥18%. The scoring is based on the degree of impact of the fault. A score of 20 points corresponds to no downtime risk, a score of 50 points corresponds to a fault that may lead to a decrease in efficiency, and a score of 80 points corresponds to a fault that poses a safety hazard.

[0084] Different weights are assigned to the valves based on their application scenarios: when conveying corrosive fluids, the weight of the appearance score is 0.4, the weight of the ultrasonic score is 0.3, and the weight of the pressure score is 0.3; when conveying high-pressure gas (pressure ≥ 1 MPa), the weight of the pressure score is 0.45, the weight of the ultrasonic score is 0.35, and the weight of the appearance score is 0.2; the comprehensive score for each abnormal area is calculated as follows: comprehensive score = appearance score × appearance weight + ultrasonic score × ultrasonic weight + pressure score × pressure weight.

[0085] S7: Fault Location Marking Steps: Preset fault level thresholds, including a visual anomaly threshold of 30 points, an ultrasonic anomaly threshold of 35 points, a pressure anomaly threshold of 40 points, and a comprehensive anomaly threshold of 45 points. Compare the comprehensive score of each anomaly area with the corresponding threshold: If the comprehensive score of a single anomaly area exceeds the corresponding single threshold, it is marked as a single fault area, and the specific fault type is noted, such as a surface corrosion fault area, an internal crack fault area, or an internal leakage fault area. If two or more anomaly areas are superimposed, and the comprehensive score exceeds the comprehensive anomaly threshold, it is marked as a composite fault area, and the superimposed fault type is also noted, such as a surface corrosion + internal leakage composite fault area. The final output includes fault location information, such as the specific location of the fault area on the valve body, like 5mm from the flange sealing surface near bolt hole No. 1, the three-dimensional coordinate range of the fault area (e.g., X: 120.5-120.8mm, Y: 85.2-85.5mm, Z: 50.1-50.4mm), the fault type, and the repair priority. The fault is classified into high priority (>80 points), medium priority (50-80 points), and low priority (<50 points) based on the comprehensive score.

[0086] As one implementation method, the specific operation of data acquisition in S1 is as follows: If a handheld detection device is used, the operator needs to simultaneously attach the lens of the visible light acquisition unit, the probe of the ultrasonic detection unit, and the pressure probe of the pressure pulse detection unit to the corresponding detection point on the valve body, and set the sampling interval according to the size of the valve body components: one sampling point is set every 2cm in the flange sealing surface area, one sampling point is set every 1cm at the valve stem and valve seat mating area, and one sampling point is set every 3cm on the valve body side wall, ensuring that there are no blind spots and no overlap in the acquisition range of adjacent sampling points; if a fixed frame detection device is used, the three acquisition units are fixed in preset positions around the valve body by an adjustable metal bracket. The height and angle of the bracket can be adjusted according to the size of the valve body, so that the acquisition unit is always aligned with the target detection part, and the acquisition frequency is set to acquire data once every 5 minutes to achieve continuous monitoring of the valve body status; during all sampling processes, the laser positioning unit records the three-dimensional coordinates of the sampling points in real time, and embeds the coordinate data into the header information of the corresponding data file through wired transmission, ensuring that the coordinate information can be directly called for association during subsequent data processing.

[0087] As one implementation method, the specific processing steps of registration and fusion in S2 are as follows: When using Zhang's calibration method to correct distortion in visible light image data, multiple sets of checkerboard calibration board images are first captured, the lens intrinsic parameter matrix and distortion coefficients are calculated, and then the acquired valve body image is corrected pixel by pixel based on the distortion coefficients to eliminate image edge stretching and distortion. When using wavelet threshold filtering algorithm to denoise ultrasonic echo data and pressure pulse data, the original signal is first decomposed into wavelets with a decomposition layer of 3. Then, a threshold is set for the high-frequency wavelet coefficients, determined according to the principle of maximizing the signal-to-noise ratio. Noise coefficients smaller than the threshold are filtered out, and finally, the denoised signal is obtained through wavelet reconstruction. When establishing reference point pairs, each positioning mark corresponds to at least 3 sets of pixel coordinate-spatial coordinate data to ensure the accuracy of coordinate transformation matrix calculation. After the coordinate transformation matrix is ​​calculated, each sampling point of the ultrasonic echo data and pressure pulse data is transformed to the pixel coordinate system according to the matrix, and then fused with the corresponding pixel of the visible light image to form three-modal detection data containing 7 feature information per pixel. The 7 feature information include the red, green and blue channel values, ultrasonic echo peak value, ultrasonic propagation time, pressure pulse peak value and pulse duration.

[0088] As one implementation method, the specific calculation details of cluster segmentation in S3 are as follows: When calculating spatial distance, the difference between the three-dimensional coordinates of the sampling point and the cluster center is directly used to reflect the actual physical distance between them on the valve body surface. When calculating the three-modal feature distance, the visible light color difference is taken as the actual difference between the red, green, and blue channel values ​​of the sampling point and the cluster center; the ultrasonic echo feature difference is taken as the average of the difference between the echo peak intensity and the ultrasonic propagation time; and the pressure pulse feature difference is taken as the average of the difference between the pulse peak value and the pulse duration. These three differences are then added together according to their weights. When adjusting the cluster center, the three-dimensional coordinates of the new center are the arithmetic mean of the coordinates of all sampling points within the region, and the three-modal feature value of the new center is the weighted average of the corresponding feature values ​​of all sampling points within the region, with the weights consistent with those used in the feature distance calculation. In addition to the condition that the difference in the total distance sum is ≤0.01, the iteration termination condition also sets a maximum number of iterations of 50. If the total distance sum is still not stable after 50 iterations, the iteration stops and the clustering result of the 50th iteration is taken as the final result. The unit of the difference in the total distance sum ≤0.01 is the normalized distance, with a value range of [0, 1]. After the analysis sub-regions are formed, the corresponding valve body functional parts are labeled for each sub-region, such as the flange sealing sub-region and the valve stem mating sub-region, to facilitate subsequent anomaly identification and fault location.

[0089] In the three-modal feature distance, color difference is taken from the difference in RGB channels, ultrasonic difference from the difference in echo peak value and propagation time, and pressure difference from the difference in pulse peak value and duration. These three are dynamically weighted and then summed. The dynamic weighting is calculated as follows: weight = scene base weight × 60% + data quality weight × 40%. The scene base weight is preset to corrosive fluid scenario and high-pressure gas scenario. The data quality weight is calculated based on the normalized signal-to-noise ratio (SNR) of each modality. Data quality evaluation indicators are: SNR = signal peak value / noise root mean square (RMS), with visible light ≥ 25dB, ultrasonic ≥ 30dB, and pressure ≥ 28dB considered acceptable; feature consistency = variance of feature values ​​within the region / normal threshold range, with ≤ 0.3 indicating consistency; data quality weight = (actual SNR / acceptable SNR) × 0.6 + (1 - feature consistency) × 0.4.

[0090] As one implementation method, the specific judgment criteria for anomaly identification in S4 are as follows: When judging appearance anomalies, if the average red channel value of the analyzed sub-region is <45 or >65, the average green channel value is <55 or >75, the average blue channel value is <75 or >95, or the average texture entropy value is <0.25 or >0.45, and the proportion of sampling points exceeding the threshold is ≥10%, then it is directly marked as an appearance anomaly area, and the anomaly type is further distinguished, such as color exceeding the threshold corresponding to coating peeling, and texture entropy value exceeding the threshold corresponding to surface corrosion; when judging ultrasonic anomalies, if If the average echo peak intensity is <1.8V or >2.2V, or the average waveform distortion rate is >5%, and the sampling point ratio condition is met, it is marked as an ultrasonic abnormal area, and the abnormality type is distinguished. For example, abnormal peak intensity corresponds to valve seat wear, and abnormal waveform distortion rate corresponds to internal cracks. When judging pressure abnormalities, if the average pulse peak deviation rate is >8%, or the average decay time is <0.8s or >1.2s, and the sampling point ratio condition is met, it is marked as a pressure abnormal area, and the abnormality type is distinguished. For example, abnormal peak deviation rate corresponds to internal leakage, and abnormal decay time corresponds to seal failure.

[0091] As one implementation method, the specific scenario adaptation of dynamic weighted scoring in S6 is as follows: Except for corrosive fluid and high-pressure gas scenarios, for low-pressure water medium transportation scenarios, when calculating the scoring value, the abnormal area ratio is calculated as the abnormal area area / the total area of ​​the corresponding valve body functional parts, such as the abnormal area area of ​​the flange sealing surface ÷ the total area of ​​the flange sealing surface. The feature deviation degree is calculated as (actual feature value - normal threshold average value) ÷ normal threshold range, such as the red channel value deviation degree = (actual average value - 55) ÷ 20, where 55 is the normal threshold average value and 20 is the threshold range. After the comprehensive scoring is calculated, one decimal place is retained to ensure scoring accuracy.

[0092] As one implementation method, the specific output requirements for fault location marking in S7 are as follows: the description of the fault area location should be combined with easily identifiable structural features of the valve body, such as bolt holes, scale lines, and valve stem scales, to avoid vague descriptions; the three-dimensional coordinate range should be accurate to one decimal place to facilitate precise location during maintenance; the fault type labeling should be based on the anomaly type judgment results of S4 to ensure correspondence with the anomaly characteristics, such as the superposition of surface corrosion in appearance anomalies and internal leakage in pressure anomalies, which should be labeled as a composite fault of surface corrosion + internal leakage; in addition to being divided according to the comprehensive score, the maintenance priority should also be combined with the importance of the valve, such as the main pipeline valve having a higher priority than the branch pipeline valve, and if the comprehensive scores are the same, the important valve has a higher priority.

[0093] This invention discloses a valve fault location system for implementing any of the valve fault location methods described above, comprising:

[0094] The three-modal data acquisition module consists of a visible light acquisition unit, an ultrasonic detection unit, a pressure pulse detection unit, and a laser positioning unit. The visible light acquisition unit uses an industrial camera with a resolution ≥1080P, equipped with a macro lens and a focal length of 25mm, to capture clear images of the valve body surface. The ultrasonic detection unit uses an ultrasonic probe with a frequency of 2-5MHz, supporting manual or automatic triggering sampling, for transmitting and receiving ultrasonic signals. The pressure pulse detection unit uses a pressure sensor with a range of 0-10MPa and a sampling frequency ≥1kHz to acquire pressure pulse signals. The laser positioning unit uses a laser rangefinder with a positioning accuracy of ±0.1mm, supporting real-time output of three-dimensional coordinates and correlation with data from the three acquisition units. The synchronization accuracy of the three-modal data acquisition module is ≤10ms. The laser positioning unit shares the same clock source with the three acquisition units, ensuring complete matching between the spatial coordinates of the sampling points and the timestamps of the detection data. The data fusion module consists of a distortion correction unit, a noise reduction processing unit, and a coordinate registration unit. The distortion correction unit incorporates Zhang's calibration algorithm for processing visible light image data. The noise reduction unit incorporates a wavelet threshold filtering algorithm to process ultrasonic echo data and pressure pulse data. The coordinate registration unit supports calculating the coordinate transformation matrix based on reference points, achieving spatial alignment and pixel-level fusion of the three types of data, and outputting tri-modal detection data. The clustering and segmentation module consists of an initial center setting unit, a distance calculation unit, an iterative adjustment unit, and a sub-region generation unit. The initial center setting unit supports automatically allocating the number of initial cluster centers according to the valve body size. The distance calculation unit calculates spatial distance and tri-modal feature distance, and outputs the total distance. The iterative adjustment unit adjusts the cluster centers according to the total distance and re-divides the regions until the termination condition is met. The sub-region generation unit extracts the boundary contours of stable cluster regions, labels functional parts, and outputs the analysis sub-regions. The multi-dimensional anomaly identification module consists of a threshold storage unit, a feature extraction unit, and an anomaly determination unit. The threshold storage unit pre-stores normal feature thresholds for different application scenarios. The feature extraction unit extracts the tri-modal feature parameters of the analysis sub-regions. The anomaly determination unit compares the extracted feature parameters with the thresholds, marks three types of abnormal regions (appearance, ultrasound, and pressure), and distinguishes the anomaly type. The anomaly correlation analysis module consists of a spatial correspondence unit, a propagation path analysis unit, and an impact range calculation unit. The spatial correspondence unit establishes the coordinate correspondence between three types of anomaly regions and determines the superposition of anomalies. The propagation path analysis unit tracks the propagation direction of anomaly features along the valve body structure. The impact range calculation unit determines the comprehensive impact range of the anomaly features. The dynamic scoring module consists of a scoring standard storage unit, a weight adaptation unit, and a scoring calculation unit. The scoring standard storage unit stores the scoring numerical rules for different anomaly types; the weight adaptation unit supports automatic or manual weight adjustment based on the valve's application scenario. The scoring calculation unit calculates the comprehensive score for each anomaly region according to the weights and scoring standards.The fault location output module consists of a threshold comparison unit, a fault marking unit, and an information output unit. The threshold comparison unit compares the comprehensive score with the fault level threshold. The fault marking unit marks single or compound fault areas according to the comparison results. The information output unit outputs fault location information in text, image, and table formats, and supports exporting to terminal devices.

[0095] As one implementation method, in the three-modal data acquisition module, the visible light acquisition unit has a frame rate of ≥30fps to ensure rapid capture of valve body surface details. The ultrasonic detection unit uses a piezoelectric ceramic probe with a diameter of 10-15mm to adapt to the detection needs of valve bodies of different sizes; the pressure pulse detection unit uses a diffused silicon pressure sensor with a response time of ≤1ms to ensure capture of instantaneous pressure changes; the laser positioning unit supports the RS485 communication protocol and can be synchronously triggered with other acquisition units to ensure that the sampling time difference of the three types of data is ≤10ms, further guaranteeing the spatial consistency of sampling points. The system is also equipped with a data storage unit that supports local SD cards with a storage capacity of ≥64GB and cloud storage, facilitating historical data retrieval and fault tracing.

[0096] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A valve, characterized in that, include: The valve body is a tubular shell structure, and includes an inlet and an outlet. The inlet and outlet penetrate the valve body to form a fluid channel. The switch adjustment part is fixedly installed inside the liquid inlet of the valve body. The switch adjustment part includes a hollow receiving cavity. A switch inlet is provided through the end of the switch adjustment part near the liquid inlet, and a switch outlet is provided through the end of the switch adjustment part away from the liquid inlet. The switch outlet is recessed inward to form an outlet limiting part. The diameter of the switch inlet is smaller than the inner diameter of the receiving cavity. An elastic part is disposed inside the receiving cavity and one end abuts against the outlet limiting part; The fixed ball has a diameter smaller than that of the receiving cavity. The fixed ball is located inside the receiving cavity at the end of the elastic part away from the outlet limit part, and the fixed ball cooperates with the switch inlet to form a blockage.

2. A valve fault location method for detecting a valve according to claim 1, characterized in that, Includes the following steps: S1: The detection device is equipped with a visible light acquisition unit, an ultrasonic detection unit and a pressure pulse detection unit. Data is collected along the preset path of the key components of the valve body to obtain visible light image data, ultrasonic echo data and pressure pulse data. The spatial coordinates of the sampling points are recorded and the data is associated simultaneously. S2: After distortion correction or noise reduction of the three types of data respectively, spatial correspondence is established based on the valve body positioning mark, and the data are registered and fused into three-modal detection data in a unified coordinate system. S3: Cluster and segment the three-modal data, set uniformly distributed initial cluster centers, and delineate initial cluster regions; Calculate the spatial distance between the sampling points and the center within the region, as well as the distance to the three-modal features; The cluster centers are adjusted based on the sum of the two types of distances, and the regions are redefined. This process is repeated until the sum of the distances no longer changes. The boundaries of the stable cluster regions are used as the sub-regions for analysis. S4: Compare and analyze the sub-regions with the preset normal thresholds to identify areas with abnormal appearance, abnormal ultrasound, and abnormal pressure. S5: Analyze the spatial correlation of the three abnormal regions, determine whether they overlap, and identify the abnormal propagation path and the scope of impact. S6: Set the appearance, ultrasonic, and pressure scores according to the proportion of abnormal area and the degree of feature deviation, assign weights according to the valve application scenario, and calculate the comprehensive score; S7: Compare the overall score with the fault level threshold. If the score exceeds the threshold, mark the fault area and output the location, type and maintenance priority.

3. The valve fault location method according to claim 2, characterized in that: The data collected in S1 specifically includes: data collection methods such as handheld setup and data collection using fixed rack-mounted equipment; Handheld devices collect samples at intervals set according to component size, while fixed-frame devices collect samples continuously at set intervals.

4. The valve fault location method according to claim 2, characterized in that: The specific registration and fusion process in S2 is as follows: Zhang's calibration method was used to correct visible light images, and wavelet filtering was used to reduce noise in ultrasound and pressure data. The pixel coordinates of the identification marker and the sampling point coordinates form a reference pair. The coordinate transformation matrix is ​​calculated to align the data, and the three data features are fused using a feature-level fusion method.

5. The valve fault location method according to claim 2, characterized in that: The specific clustering segmentation in S3 is as follows: Spatial distance is calculated based on the coordinate difference between the sampling point and the center. In the three-modal feature distance, the color difference is taken from the difference of RGB channels, the ultrasonic difference is taken from the difference of echo peak value and propagation time, and the pressure difference is taken from the difference of pulse peak value and duration. The three are dynamically weighted and summed according to the valve body application scenario and data quality. Repeat the adjustment until the sum of the distances between the two iterations is ≤0.

01.

6. The valve fault location method according to claim 2, characterized in that: Anomaly detection in S4 specifically involves: Normal thresholds include: appearance, ultrasound, and pressure thresholds; If the percentage of sampling points exceeding the threshold in a sub-region is ≥10%, it is marked as abnormal.

7. The valve fault location method according to claim 2, characterized in that: The overall score in S6 specifically includes: Appearance rating, ultrasound rating, and pressure rating.

8. A valve fault location method according to claim 2, characterized in that: The fault flags in S7 are as follows: A single anomaly exceeding the corresponding threshold marks a single fault region; two or more anomalies exceeding the comprehensive threshold mark a composite fault region. Output coordinate range and high / medium / low maintenance priority.

9. A valve fault location system, used to implement the valve fault location method according to any one of claims 2-8, characterized in that, include: Three-modal data acquisition module: acquires visible light images of the valve body, ultrasonic echoes, and pressure pulse data; uses laser positioning to record the coordinates of sampling points and associates the data. Data fusion module: Corrects visible light image distortion, reduces noise in ultrasonic and pressure data, registers the three types of data with valve body positioning marks, and generates three-modal detection data; Clustering and segmentation module: Set initial cluster centers, calculate the spatial and feature distances between sampling points and centers, iteratively adjust the centers to divide regions, and obtain sub-regions for analysis; Multi-dimensional anomaly identification module: compares features with normal thresholds to identify areas of abnormal appearance, ultrasound, and pressure. Anomaly Correlation Analysis Module: Identifies overlapping anomalies, analyzes propagation paths and the scope of impact; Dynamic scoring module: Calculates a comprehensive score based on the weight of abnormal situations and scenarios; Fault location output module: compares the score with the threshold, marks the fault area and outputs the information; Data from the three-modal data acquisition module is transmitted to the data fusion module, data from the data fusion module is transmitted to the clustering and segmentation module, data from the clustering and segmentation module is transmitted to the multi-dimensional anomaly identification module, data from the multi-dimensional anomaly identification module is transmitted to the anomaly correlation analysis module, data from the anomaly correlation analysis module is transmitted to the dynamic scoring module, and data from the dynamic scoring module is transmitted to the fault location output module.

10. A valve fault location system according to claim 9, characterized in that: In the three-modal data acquisition module, the visible light unit has a resolution of ≥1080P, the ultrasonic unit has a frequency of 2-5MHz, and the pressure unit has a range of 0-10MPa. The laser positioning unit supports real-time coordinate uploading.

Images