A kind of columnar membrane wet state press defect detection and automatic plugging system and method

Abstract

A kind of column type membrane wet state pressurization defect detection and automatic plugging system and method, belong to membrane module integrity detection technical field.Preset pressure fluid is applied to membrane module under the condition that membrane module end face is immersed in liquid medium, make the end of leaky membrane wire produce bubble, continuously collect bubble image sequence of preset observation area on membrane module end face or above, carry out bubble detection, continuous frame time sequence consistency judgment, bubble centroid trajectory extraction to bubble image sequence, and based on multiple frames bubble centroid trajectory, deduce leak source coordinates, map leak source coordinates to membrane end face physical coordinate system, and match with preset membrane hole template coordinates, to determine target leaky membrane wire, according to the position coordinates of target leaky membrane wire, execute plugging action.

Description

Technical Field

[0001] This invention belongs to the field of membrane module integrity testing technology, and relates to a system and method for detecting and automatically sealing defects in a column membrane under wet pressure. Background Technology

[0002] After production, column-type membrane modules typically require wet pressure testing to determine if there are defects such as membrane fiber breakage, pinholes, or seal failure. Current technology often relies on visual inspection of the bubbles generated during the pressure testing process, using the location and number of bubbles to assess leakage.

[0003] However, existing detection methods have the following problems: (1) Poor stability of manual identification. The detection results depend heavily on the operator's experience. Long-term detection can easily lead to visual fatigue, resulting in missed detections, false detections, and poor consistency. (2) Insufficient anti-interference ability. During wet pressure testing, it is easily affected by factors such as liquid surface fluctuation, light reflection, water flow disturbance and random bubbles. Existing methods are difficult to effectively distinguish between real leakage bubbles and transient interference bubbles, resulting in a decrease in detection accuracy; (3) Low accuracy in locating the leak source. Most existing image processing methods use simple threshold segmentation or single-frame analysis, which makes it difficult to continuously track the movement of bubbles. Under conditions of bubble overlap, drift, or complex background, target loss or positioning deviation is likely to occur. (4) Low level of automation in sealing. In the existing technology, leakage defects usually rely on manual location and sealing, which is inefficient and prone to problems such as inaccurate location, uneven sealing and repeated repairs; (5) Lack of linkage between detection and sealing. The existing detection system cannot transmit the leak location information to the sealing mechanism in real time, and the defect location still needs to be confirmed manually, which is difficult to meet the needs of automated production.

[0004] Columnar hollow fiber membrane modules are widely used in water treatment, gas separation, and biomedicine. The integrity of the membrane module directly affects its filtration performance and service life; therefore, integrity testing is necessary during production. Currently, the wet pressure bubble method is commonly used in industrial production for leak detection. This method involves introducing gas at a certain pressure into the columnar membrane module, immersing the membrane end face in water, and observing whether bubbles are generated in the water to determine if the membrane fibers are damaged.

[0005] N117805108A discloses a non-contact hollow fiber membrane integrity testing device. This device covers the end face of the membrane module with a thin film and uses a linear laser and an industrial camera to detect localized protrusions in the film under ventilation pressure to obtain the location of defective filaments. This method is an indirect detection approach based on membrane deformation. Before testing, a thin film needs to be attached to the end face of the membrane element. The test results are affected by the uniformity of film attachment, tension, and the local deformation transmission characteristics of the film. Furthermore, due to the use of a film-covering technique, it can only indirectly record defect locations. After testing, the film is removed, the specific defective membrane filament is manually located, manually sealed, the film is reapplied, and the test is repeated until the test is passed. It can be seen that this technical solution is complex to operate, has poor accuracy in test results, cannot automatically seal or perform closed-loop re-inspection after sealing, and has low operability.

[0006] Existing detection methods primarily rely on manual observation, which has several drawbacks: ① High dependence on manual intervention: Bubbles on the membrane element end face are characterized by small diameter (<2mm), high density (tens of membrane filaments per square centimeter), and overlapping targets. Applying traditional underwater bubble detection methods directly to the membrane element end face can lead to serious missed and false detections. ② Low detection efficiency, making it difficult to meet the needs of large-scale production. ③ Insufficient accuracy: Different operators have different evaluation criteria, which can easily cause batch variations. This is especially true for columnar membrane modules, whose end faces typically contain hundreds or thousands of membrane filament pores. Once a leak occurs, it is difficult for humans to quickly and accurately locate the specific pore. Relying on membrane adhesion and changes in optical signals for detection is an indirect method and cannot directly reflect the actual leakage situation of the membrane filaments. For minor leaks, the signal changes are not obvious, resulting in low detection sensitivity. Indirectly inferring the location of defective membrane filaments through membrane deformation can only assess the condition of a local area of ​​the membrane module and cannot achieve precise positioning at the level of a single membrane filament, making subsequent targeted repair difficult. Detection results are easily affected by the quality of membrane adhesion, lighting environment, and membrane surface condition, posing a risk of misjudgment. It only has detection capabilities. If subsequent defect repair and re-inspection are required, the film must be removed, manually identified, manually repaired, re-coated, and re-inspected until it finally passes. The automation level is low and the efficiency is not high. Summary of the Invention

[0007] This invention addresses the problems of existing technologies by providing a system and method for detecting and automatically sealing defects in columnar membranes under wet pressure. It is a system for detecting and repairing the integrity of columnar membrane elements that does not require end-face covering with a thin film, can directly identify leaking bubbles under wet pressure conditions, directly locate and automatically seal defective membrane fibers in real time, and can quickly re-inspect closed loops.

[0008] A column-type membrane wet pressure testing defect detection and automatic sealing system includes a wet pressure testing module, an image acquisition module, a central control unit, a membrane pore template positioning module, and a mechanical sealing execution unit; The wet pressure module is used to apply a preset pressure fluid to the membrane module while the end face of the membrane module is immersed in the liquid medium, so as to generate bubbles at the end of the leaking membrane filament; The image acquisition module is used to continuously acquire bubble image sequences from the end face of the membrane module or a preset observation area above the end face; The central control unit is used to perform bubble detection, consecutive frame temporal consistency judgment, bubble centroid trajectory extraction on bubble image sequences, and deduce the coordinates of the leakage source based on the multi-frame bubble centroid trajectory. The membrane pore template positioning module is used to map the coordinates of the leakage source to the physical coordinate system of the membrane end face and match them with the preset membrane pore template coordinates to determine the target leaking membrane filament; The mechanical plugging actuator is used to perform plugging actions based on the position coordinates of the target leaking membrane filament; The central control unit is also used to maintain the pressure state after sealing and to acquire image sequences again to determine whether the sealing was successful.

[0009] A method for detecting and automatically sealing defects in a column membrane under wet pressure testing includes the following steps: The wet pressure test step involves applying a preset pressure fluid to the membrane module while the end face of the membrane module is immersed in a liquid medium, causing bubbles to form at the end of the leaking membrane fibers.

[0010] The image acquisition step involves continuously acquiring a sequence of bubble images from the end face of the membrane module or a preset observation area above the end face.

[0011] The central control step involves bubble detection, temporal consistency judgment of consecutive frames, extraction of bubble centroid trajectory, and deduction of leakage source coordinates based on the multi-frame bubble centroid trajectory.

[0012] The membrane pore template positioning step maps the coordinates of the leakage source to the physical coordinate system of the membrane end face and matches them with the preset membrane pore template coordinates to determine the target leaking membrane filament.

[0013] The mechanical plugging procedure involves performing the plugging action based on the location coordinates of the target leaking membrane filament.

[0014] The central control procedure also includes maintaining the pressure state after the blockade and acquiring image sequences again to determine whether the blockade was successful.

[0015] The advantages of this invention are: it does not simply combine existing detection technologies with automated actuators, but rather addresses the technical problem of difficulty in locating leak sources under bubble diffusion conditions by proposing a location method based on bubble temporal consistency analysis and centroid sequence back-inference. This method achieves reverse inference from bubble diffusion phenomena to the location of the leak source by jointly modeling the spatial distribution and temporal evolution characteristics of bubbles, and then combines template matching to achieve precise location at the single-filament level. Based on this, an automatic sealing and re-inspection closed-loop control process is further constructed, thereby realizing integrated automatic operation of detection and repair.

[0016] A columnar membrane defect detection method capable of stable identification of bubble defects, precise location of leakage sources, and automatic sealing during wet pressurization is presented. Compared with the non-contact hollow fiber membrane integrity detection device disclosed in CN117805108A, this existing solution involves covering the membrane module end face with a thin film and using a line laser and industrial camera to detect local protrusions of the film under ventilation pressure to obtain the location of defect filaments, which is an indirect detection method based on film deformation. This solution requires attaching a thin film to the membrane end face, and its detection results are easily affected by the uniformity of film attachment, tension, and deformation transmission characteristics. Furthermore, it lacks automatic sealing and closed-loop control for post-sealing re-inspection. Therefore, this invention directly identifies leakage phenomena through wet bubble time-series images and links the leakage location results with mechanical sealing actions, enabling integrated automatic processing of detection, location, sealing, and re-inspection.

[0017] Compared with the non-contact hollow fiber membrane integrity testing device disclosed in CN117805108A, the present invention has the following technical advantages: (1) Improve the accuracy of leak defect detection This invention directly utilizes the continuous leakage bubbles generated during wet pressure testing as the detection object. Defect identification is achieved through multi-frame image acquisition and bubble timing analysis. There is no need to attach an additional detection film to the end face of the membrane module, which avoids detection deviations caused by uneven film thickness, differences in adhesion tightness, local wrinkles, and deformation transmission errors, thereby improving the authenticity and accuracy of defect detection results.

[0018] (2) Improve the accuracy of leak source location

[0019] This invention establishes a leakage source inverse model based on the centroid trajectory of bubbles across multiple consecutive frames, and combines this with a membrane pore template matching mechanism to locate defective membrane filaments. Compared to methods that rely solely on the location of membrane bulges for indirect judgment, this invention utilizes the timing information of bubble movement to solve for the leakage source in reverse, thereby achieving precise positioning at the single membrane filament level and improving positioning accuracy and stability.

[0020] (3) Achieve closed-loop automation of detection and repair

[0021] This invention directly converts defect detection results into mechanical plugging execution coordinates, achieving closed-loop control of detection, positioning, automatic plugging, and re-inspection verification. Unlike existing technologies that can only perform defect detection, this invention can automatically complete the defective membrane fiber repair process, reducing manual intervention and improving the level of automation.

[0022] (4) Improve production efficiency and reduce labor costs

[0023] This invention enables online detection and automatic sealing during membrane module production, eliminating the need for manual searching of defective membrane fibers and manual repair, significantly shortening detection time, increasing production cycle time, and reducing labor intensity and production costs.

[0024] (5) Improve product consistency and finished product qualification rate

[0025] By accurately identifying and automatically sealing defective membrane fibers, and combining this with a post-sealing re-inspection and confirmation mechanism, missed detections and sealing failures can be effectively avoided, improving the reliability of membrane module integrity testing, and enhancing product consistency and final product qualification rate.

[0026] Therefore, this invention can achieve high-precision detection, accurate positioning, automatic repair and closed-loop verification of defects in column membrane modules under wet pressure conditions. It has the advantages of high detection accuracy, high degree of automation, strong anti-interference ability and suitability for industrial continuous production. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. As shown in the figures: Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart of the present invention; The same bubble candidate target is detected within T consecutive frames, where T is the number of consecutive effective detection frames, preferably T ≥ 5; This represents the maximum number of iterations. For membrane regions; The pore matching radius is denoted by . Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1: As Figure 1 and Figure 2 As shown, a column-type membrane wet pressure testing defect detection and automatic sealing system and method can be used for integrity detection, leak location and repair in the membrane module production process. It is a technical solution that can automatically detect and accurately locate leaks and achieve repeatable detection.

[0030] A method for detecting and automatically sealing defects in a column membrane under wet pressure testing includes the following steps: 1. Membrane module loading and wet pressure preparation: Fix the column membrane module to be tested at the testing station, immerse the end face of the membrane module in the liquid medium, and complete the reading of the membrane module number, binding of identity information and initialization of test parameters.

[0031] 2. Wet pressure test: The central control unit controls the wet pressure test module to apply a preset pressure fluid to the inside of the membrane module. When there is damage, pinholes or seal failure in the membrane fibers, gas escapes from the defect location to form continuous bubbles.

[0032] 3. Image acquisition: The image acquisition module uses an industrial camera to continuously acquire image sequences of the membrane module end face or the preset observation area above the end face, and stores the images synchronously with timestamps and pressure parameters.

[0033] 4. Bubble detection and time series analysis: The central control unit uses a target detection algorithm to identify bubble targets in the image and tracks and analyzes bubbles in continuous frames. It judges effective leaking bubbles based on conditions such as the existence of continuous frames, the rate of change of area, the centroid displacement, and the detection confidence level.

[0034] 5. Leakage source reverse location: Based on the centroid trajectory of the bubble in multiple consecutive frames, combined with the bubble's rising velocity, lateral drift velocity, and the position of the observation plane, a leakage source reverse location model is established, and the spatial coordinates of the leakage source are calculated.

[0035] 6. Membrane pore template matching: Map the leakage source coordinates obtained by reverse calculation to the physical coordinate system of the membrane module end face, and match them with the preset membrane pore template coordinates to determine the location and number of the target leakage membrane filament.

[0036] 7. Automatic sealing: The mechanical sealing unit automatically moves to the designated position according to the target membrane wire coordinates and performs sealing operations such as inserting the plug, applying adhesive, hot-melt sealing, or injecting sealing material.

[0037] 8. Re-inspection and confirmation: After the sealing is completed, continue to maintain the pressure state, and perform image acquisition and bubble detection again to determine whether there are still continuous leakage bubbles at the target location.

[0038] 9. Result output: If no continuous stable bubbles are detected, the sealing is considered successful and the detection result is output; if leakage still exists, re-sealing, manual verification, or scrapping is performed.

[0039] The above steps form a closed-loop control process of detection, location, sealing, and re-inspection.

[0040] like Figure 2 As shown, in this invention, the central control unit first receives the bubble detection image sequence and the bubble observation position set. And related calculation parameters, including the maximum allowable drift speed. Iterative convergence threshold Maximum permissible back-calculation error Matching threshold with membrane pores The reverse algorithm was then initialized, and the number of valid observation frames was set. Initial location of the leak source lateral drift speed of bubbles and number of iterations This provides initial conditions for subsequent calculations to locate the leakage source.

[0041] During image acquisition, the central control unit first determines the valid observation data. When the number of valid bubble observation frames used in the calculation meets the requirements... At this point, it is considered that the basic conditions for trajectory fitting and position inversion have been met; if the number of effective frames is insufficient, images are continuously acquired and the bubble trajectory information is updated until the calculation requirements are met. Subsequently, the reasonableness of the bubble's lateral drift velocity is verified to determine whether it meets the requirements. If the drift velocity exceeds a preset threshold, the current observation result is considered to be affected by liquid disturbance, local flow field anomaly, or bubble interference. The system automatically removes abnormal observation data, re-estimates the drift velocity, and makes a judgment again.

[0042] Once the observation data meets the requirements, the central control unit establishes a bubble motion observation model based on the positions of the bubble centroids across multiple consecutive frames, and constructs a back-calculation objective function for the leakage source by combining the dynamic laws of bubble buoyancy. Subsequently, a nonlinear least squares optimization algorithm was used to iteratively solve for the location of the leakage source to obtain the optimal estimated coordinates of the leakage source. After each iteration, the system calculates the parameter update amount. And update the iteration count to .

[0043] During the optimization process, the system continuously monitors the iterative convergence status. When the parameter increment satisfies... If the optimization process has reached convergence, it is considered to have reached convergence, and the subsequent verification steps are initiated; otherwise, if convergence has not been reached and the number of iterations has not exceeded the maximum number of iterations... If the iteration count reaches its limit and convergence is still not achieved, the system will continue to perform LM iterative optimization. If the iteration count reaches its limit and convergence is still not achieved, the system will automatically expand the search range, reinitialize the optimization parameters, and re-execute the back-calculation to avoid getting trapped in a local optimum.

[0044] After obtaining the location of the leak source, the central control unit further verifies the spatial rationality of the results and determines the leak source coordinates obtained by reverse calculation. Is it located in the effective area of ​​the membrane fiber end face? If the reversed position exceeds the effective area, the current result is determined to be affected by abnormal drift or incorrect matching. The system uses prior information on the membrane filament arrangement to reconstrain the solution process and returns to the aforementioned steps to recalculate, so as to ensure that the reversed result conforms to the actual physical structure.

[0045] For the results that pass the spatial rationality verification, the system further calculates the back-calculation error. The leakage intensity, leakage pressure, and other relevant parameters are then retrieved by combining the bubble trajectory fitting results. Subsequently, it is determined whether the back-calculation error meets the requirements. If the error exceeds the allowable range, it is considered that there may be abnormalities such as bubble aggregation, bubble splitting, target tracking errors, or image detection errors during the observation process. The system automatically removes abnormal observation samples and re-executes optimization calculations to improve the accuracy and stability of the positioning results.

[0046] Once the back-reasoning results meet the error requirements, the central control unit uses a pre-established membrane pore template to process the back-reasoned leak source coordinates. The distance to the center position of each membrane pore in the template is calculated to obtain the corresponding distance. The membrane pore with the smallest distance is selected as the candidate leakage membrane pore, and its index is denoted as . Then it is determined whether the minimum matching distance is satisfied. If the matching conditions are met, the location of the leak membrane pore is considered successfully determined, and the pore number is output. Leakage source coordinates Back-calculation error And location status information.

[0047] If the minimum matching distance exceeds the preset threshold If the current reverse-engineering result cannot reliably correspond to the existing membrane pore template, the system marks the location as a suspected leak point and records the relevant detection data for subsequent manual review or further detection and analysis. This completes the entire process of leak source reverse-engineering and membrane pore location.

[0048] Example 2: Figure 1 and Figure 2 As shown, a column membrane wet pressure testing defect detection and automatic sealing system adopts optical non-contact measurement, which is faster, more accurate, and saves detection and maintenance costs. It has a large measurement range and is suitable for end face inspection of various column membrane modules. It does not damage the membrane fibers, has high versatility and strong applicability, and is suitable for defect detection, leak location and automatic repair in the production and assembly process of hollow fiber column membrane modules.

[0049] A column-type membrane wet pressure testing defect detection and automatic plugging system includes a wet pressure testing module, an image acquisition module, a central control unit, a membrane pore template positioning module, and a mechanical plugging execution unit. The wet pressure testing module applies a preset pressure fluid to the membrane module while the membrane module end face is immersed in a liquid medium, causing bubbles to form at the ends of leaking membrane filaments. The image acquisition module continuously acquires bubble image sequences from the membrane module end face or a preset observation area above the end face. The central control unit performs bubble detection, consecutive frame temporal consistency judgment, and bubble centroid trajectory extraction on the bubble image sequences, and deduces the coordinates of the leakage source based on the multi-frame bubble centroid trajectories. The membrane pore template positioning module maps the leakage source coordinates to the membrane end face physical coordinate system and matches them with the preset membrane pore template coordinates to determine the target leaking membrane filament. The mechanical plugging execution unit performs a plugging action based on the position coordinates of the target leaking membrane filament. The central control unit also maintains the pressure test after plugging and acquires image sequences again to determine whether the plugging was successful.

[0050] Before acquiring images of the end face, the image acquisition module first calibrates the camera's field of view and establishes a mapping relationship between the pixel coordinate system and the physical coordinate system of the membrane end face. At the same time, it performs noise reduction, reflection suppression, and contrast enhancement on the acquired images to improve the accuracy of bubble boundary recognition in a wet environment.

[0051] After fixing the column-type hollow fiber membrane module in the wet pressure testing area, the identification of the membrane module under test is collected. Preferably, the membrane module number can be obtained through a QR code, barcode, or coded label set on the bottom or side of the membrane module, and the number is associated with the collected images, pressure data, and sealing results. Subsequently, the image acquisition module performs the first image acquisition on the end face of the membrane module or the preset observation area above the end face, and completes the initial state recording. Both the image acquisition module and the mechanical sealing execution unit maintain a preset working distance relative to the end face of the membrane module, and the field of view of the image acquisition module covers the entire area to be inspected or covers the entire area to be inspected through partitioned acquisition.

[0052] During the image acquisition phase, the industrial camera continuously images the end face of the cylindrical membrane immersed in the liquid medium to obtain images of bubble diffusion generated during membrane filament leakage. Due to problems such as liquid surface reflection, light scattering, refractive distortion, and low contrast in the liquid medium environment, the system constructs an underwater complex optical environment adaptation module to improve the recognition accuracy and positioning stability of tiny bubbles.

[0053] The image acquisition module includes an industrial camera and a multi-angle light source system. The light source system adopts a combination of ring light and side supplementary lighting. The ring light is used to improve the overall illumination uniformity of the membrane fiber end face, while the side supplementary lighting is used to enhance the scattering characteristics of the bubble edges, thereby improving the imaging contrast of the submerged microbubble region.

[0054] To reduce the interference of specular reflection and bright areas of the liquid medium on the detection results, the system preprocesses the acquired images. The preprocessing includes polarization filtering, specular suppression, adaptive contrast enhancement, and noise filtering. Median filtering and Gaussian filtering are combined to improve the clarity of bubble boundaries.

[0055] To address the issue of light path offset caused by liquid media, the system incorporates liquid refractive index parameters to compensate for refraction errors in image coordinates, thereby improving the accuracy of spatial positioning of leaking membrane filaments.

[0056] After receiving the membrane module identification information and image data uploaded by the image acquisition module, the central control unit numbers, initializes the coordinate system and status of the hollow fiber membrane module under test, and controls the pressure fluid input system to supply pressure fluid to the test unit. The central control unit controls the pressure fluid (such as compressed air) to be input into the shell side of the membrane module through the air inlet, while simultaneously opening or connecting the other end of the membrane module to the atmosphere to establish a transmembrane pressure difference from the outside to the inside of the membrane fibers. When defective membrane fibers are present, gas will penetrate the membrane fiber wall from the shell side into the tube side (i.e., the inner cavity of the membrane fiber) and escape at the opening of the membrane fiber end face immersed in the liquid, forming bubbles.

[0057] Before the pressure test, liquid medium is injected into the housing of the hollow fiber membrane module to be tested and the end face of the column membrane to fully wet the membrane fibers and complete the pre-wetting treatment, so as to eliminate the influence of gas retention and local adhesion under dry test conditions. Then, the central control unit controls the compressed air or other pressure medium to gradually increase the pressure and execute continuous loading according to the preset pressure steps.

[0058] During the pressurization process, the image acquisition module continuously acquires images of the end face of the membrane module immersed in the liquid medium or the preset observation area above the end face at a preset frame rate. This is used to monitor in real time whether there is continuous bubble generation caused by membrane fiber leakage. The acquired images are synchronously cached by the central control unit and stored in association with the acquisition time, pressure value, membrane module number and camera calibration parameters. A bubble detection image sequence is constructed in chronological order.

[0059] To achieve automatic identification of bubble regions, the central control unit performs frame-by-frame target detection and segmentation on the bubble detection image sequence, extracting the spatial location, contour boundary, area, centroid, and detection confidence of the bubble region. Preferably, bubble detection can employ a small target detection model based on deep learning; the small target detection model is built on the YOLO target detection network and improved to address the characteristics of small bubble target size, low contrast, blurred edges, and susceptibility to interference from liquid surface reflection in wet pressure testing scenarios.

[0060] Specifically, a Deformable Convolution (DCN) module is introduced into the backbone feature extraction network. By learning the convolution sampling position offset, the convolution kernel can adaptively adjust the receptive field according to the actual shape of the bubble, thereby enhancing the model's feature extraction capabilities for irregularly shaped bubbles, locally deformed bubbles, and bubbles with blurred edges, and improving the detection accuracy of tiny bubble targets in complex liquid environments.

[0061] Simultaneously, an attention mechanism module is introduced into the feature fusion network to adaptively weight the channel-dimensional and spatial-dimensional features. Preferably, the attention mechanism adopts the CBAM attention module, including a channel attention submodule and a spatial attention submodule. Channel attention is used to enhance the feature response related to the bubble target and suppress background noise and interference from the liquid surface reflection area; spatial attention is used to highlight the features of the bubble edge and center region, improving the bubble target localization capability.

[0062] Furthermore, to enhance the detection capability for tiny bubble targets, a high-resolution detection layer is added to the feature pyramid structure, and shallow texture features and deep semantic features are fused to improve the recognition capability for bubble targets with a diameter of less than 2 mm.

[0063] The model outputs the bounding box, segmentation mask, and detection confidence information of the bubble target. The central control unit calculates the bubble region area, contour boundary, and centroid coordinates based on the detection results, and performs trajectory tracking and temporal consistency analysis by combining multiple consecutive frames of images, thereby achieving effective leakage bubble identification and subsequent leakage source inference and localization.

[0064] By introducing deformable convolution modules and attention mechanism modules, the detection accuracy, anti-interference ability, and positioning stability of microbubble targets in complex humid environments can be effectively improved, while reducing the false detection rate and missed detection rate, thus providing a reliable data foundation for the subsequent precise positioning of leaked membrane filaments.

[0065] In one implementation, the model incorporates deformable convolutional modules and an attention mechanism to enhance its ability to identify low-contrast, small-sized, and morphologically variable bubbles. The aforementioned deep learning model is used to improve the accuracy of bubble candidate region extraction, but leakage determination is jointly based on the temporal consistency of consecutive frames, trajectory constraints, and the back-inference results from the source point.

[0066] To improve the stability and anti-interference capability of the detection results, the central control unit performs temporal consistency analysis on the candidate bubble regions in consecutive frames. When the same candidate bubble is detected within T consecutive frames, and meets the following conditions: the detection confidence level is higher than the first threshold (area change rate threshold), the centroid displacement of adjacent frames satisfies the trajectory continuity condition, the area change rate satisfies the preset generation feature condition, and the leakage source residual deduced from the trajectory of the consecutive frames is less than the second threshold, it is determined to be a valid leakage bubble. If the candidate bubble appears only in a single frame or a few frames, or its centroid trajectory changes abruptly, its area changes abnormally, or its source point deduced residual is too large, it is determined to be a transient interference bubble or a noise target. This can suppress the influence of liquid surface fluctuations, reflections, water flow disturbances, and random bubbles on the detection results.

[0067] Since the image acquisition module is arranged in a preset orientation with the membrane fiber end face, preferably vertically or approximately vertically, the temporal evolution characteristics of the bubble region can be extracted by acquiring multiple consecutive images during the pressing process. The central control unit tracks the candidate bubble regions in the consecutive images to form a sequence of changes in the spatial position of the bubbles.

[0068] The planar coordinates of the centroid of the detected bubble region are calculated for each frame of the image to characterize the spatial center position of the bubble region. Assume the bubble region contains... Each pixel has its centroid coordinates. The calculation is as follows:

[0069] in, The first in the bubble region The coordinates of each pixel. This represents the total number of pixels. (This is achieved by analyzing consecutive...) By processing the frame image, the time series of the centroid of the bubble region can be obtained. A bubble region is considered a valid leaking bubble when it is detected in N consecutive frames and meets the following conditions: the rate of change of area is less than the first threshold, the centroid displacement is less than the second threshold, and the detection confidence is higher than the third threshold. If it appears only in a single frame or a few frames and does not meet the continuity constraint, it is considered a transient interference.

[0070] Based on the characteristics of bubble escaping from the end of the leaking membrane filament and its floating, diffusion, and lateral drift in the liquid medium, the central control unit constructs a leak source inversion model. This model does not directly use the bubble position in a single frame as the leak location. Instead, it transforms the centroid coordinates of multiple consecutive frames of bubbles to the physical coordinate system of the membrane end face. Combining the geometric relationship between the preset observation area and the membrane end face, bubble drift parameters, and image acquisition time series, it solves for the source point coordinates that minimize the residual bubble trajectory across multiple frames, which is then used as the leak source inversion location. Subsequently, the leak source inversion location is mapped to the preset membrane aperture template coordinate system, and aperture matching is performed within a preset distance tolerance range. If multiple candidate membrane apertures exist, the target leaking membrane filament location is determined by considering the minimum spatial distance, template row and column constraints, and consistency of historical trajectories.

[0071] Let the set of coordinates of the membrane holes in the template be... Then, by calculating the distance between the leak source location and the coordinates of each membrane pore, the membrane pore with the smallest distance is selected as the final leak plane coordinate. The calculation formula is as follows:

[0072] By combining the spatial distribution information of bubbles with their temporal evolution characteristics through centroid calculation, diffusion source inference, and template matching, the reverse inference from bubble diffusion phenomenon to the location of leakage source can be achieved. This enables accurate location of leakage membrane filaments even when bubbles are obscured, thus improving detection accuracy and robustness.

[0073] The central control unit generates the target coordinates of the robotic arm based on the template matching results, the membrane end face height, and the preset posture. The estimated coordinates are then converted into motion control commands that the mechanical actuator can recognize, thereby driving the mechanical actuator to perform automatic blocking actions.

[0074] After obtaining the spatial coordinates of the target leaking membrane filament, the central control unit converts the coordinates into the target pose in the coordinate system of the mechanical plugging execution unit, controls the mechanical plugging execution unit to move along a preset trajectory to the target position, and completes the plugging action according to preset plugging parameters. The plugging action includes inserting a plug, applying adhesive, hot-melt sealing, injecting plugging material, or a combination thereof. Preferably, when using the plug insertion method, the size of the plug matches the inner diameter of the membrane filament to block the internal channel of the target membrane filament; when using the adhesive application method, the central control unit controls the amount of adhesive applied, the application position, and the curing time to achieve reliable sealing of the end of the target membrane filament.

[0075] To prevent damage to the membrane fiber structure during the sealing process, the central control unit limits the sealing depth, sealing pressure, advance speed, dispensing volume, or curing time of the mechanical sealing actuator. Let the membrane fiber end face reference plane be... The actual depth of the sealant nail is Then the depth of the sealing nail should be controlled to meet the following requirements:

[0076] in, The maximum safe insertion depth is preset, and its value is determined based on the membrane fiber length, material properties, and structural strength to avoid the repair nail penetrating or excessively compressing the internal structure of the membrane fiber.

[0077] After sealing is completed, the system continues to maintain wet pressure and the image acquisition module continuously monitors the area where the target leaking membrane filament is located, repeatedly performing bubble detection, time-series consistency analysis, and source point back-tracing processes to determine whether stable leaking bubbles still exist at the target location. If no continuous stable bubbles corresponding to the target leaking membrane filament are detected within the preset re-inspection time window, the sealing is considered successful; if continuous bubble generation is still detected, the sealing is considered a failure, and manual review is initiated. The central control unit also monitors the force feedback or pose deviation of the mechanical sealing execution unit in real time. When the sealing force exceeds a preset safety threshold or the target pose deviation exceeds the allowable range, the sealing action is stopped and an abnormal signal is output.

[0078] The central control unit is used for unified scheduling and data interaction control of the image acquisition module, bubble detection module, time series analysis module, leak location module, membrane pore template location module, and mechanical plugging execution unit. When abnormal bubbles are detected, the location reliability is insufficient, or plugging fails, the central control unit automatically adjusts the acquisition frequency, re-inspection number, and plugging execution parameters to form a closed-loop control mechanism for leak detection, location, repair, and re-inspection. When the cumulative number of plugged membrane fibers of a single membrane module exceeds a preset maintenance ratio threshold, the central control unit determines that the membrane module does not meet the preset quality requirements and records its number, leak location, plugging number, and judgment result. Preferably, the preset maintenance ratio threshold can be set to 5% of the total number of membrane fibers.

[0079] Example 3: As Figure 1 and Figure 2 As shown, a system and method for detecting and automatically sealing defects in column-type hollow fiber membranes under wet pressure is described, based on a column-type hollow fiber membrane production line embodiment using bubble centroid time-series back deduction, membrane pore template matching, and automatic sealing re-inspection.

[0080] Leakage source inversion and membrane fiber localization method based on bubble centroid timing: Step S1: In one embodiment of the present invention, to achieve accurate localization of the leaking membrane filament position during wet pressure testing, a leak source back-calculation algorithm based on bubble centroid timing information is proposed. This algorithm combines the physical laws of bubble buoyancy, image observation constraints, and multi-frame optimization solution mechanisms to analyze the bubble image sequence collected in the preset observation area, thereby deduce the initial spatial location of the underwater leak source.

[0081] Step S2: During the wet pressure test, due to the pressure difference between the inside and outside of the membrane fiber at the defect location, gas continuously escapes from the defect location and forms continuous leakage bubbles. After the bubbles are generated at the leakage source, they are subjected to the combined action of various physical factors in the liquid and float to the surface of the liquid along a certain trajectory.

[0082] Specifically, firstly, the bubble is subject to buoyancy in the liquid and moves upward vertically. Its vertical upward velocity is denoted as and can be approximately followed by a modified Stokes' law. Secondly, due to potential local flow field disturbances, wake effects from nearby bubbles, or micro-circulation in the detected water body, the bubble will also drift horizontally. Its horizontal drift velocity can be expressed as . Secondly, as bubbles rise in the liquid, their depth gradually decreases, the hydrostatic pressure drops, and the bubble volume expands, resulting in a corresponding increase in its equivalent diameter. Furthermore, while the actual bubble trajectory deviates to some extent from the ideal vertical upward path, statistically, its average trajectory can still be considered approximately perpendicular to the liquid surface.

[0083] Step S3: Based on the above physical process, a reverse time mapping model is established, that is: by observing the position of the bubble's center of mass at multiple consecutive moments in the preset observation area, and combining the bubble's vertical upward velocity and lateral drift velocity, the bubble's trajectory is reversed to obtain the initial spatial coordinates of the underwater leakage source.

[0084] Step S4: To facilitate the description of the reverse calculation algorithm of this invention, a spatial rectangular coordinate system is established. - Among them, the origin Located at the geometric center of the end face of the membrane filament; shaft and The axes are all located within the plane of the membrane fiber end face, and are aligned with the horizontal and vertical directions of the image sensor, respectively. The axis is perpendicular to the liquid surface and takes the upward direction as positive; the preset observation plane is defined as... ,in This indicates the depth of immersion in water, expressed in mm.

[0085] Step S5: During the detection process, if multiple membrane filament leakage areas exist simultaneously, the central control unit can independently track and locate multiple leakage targets based on the position of the bubble centroid and the diffusion range. In this coordinate system, let the coordinates of the leak source location be... , No. The coordinates of the centroid of the bubble detected in the frame image are: The corresponding collection time is .

[0086] in, Indicates the density of the liquid. Indicates gas density, Indicates the dynamic viscosity of a liquid. Represents gravitational acceleration. Indicates the equivalent diameter of the bubble. This represents the bubble shape correction factor. This indicates the number of valid image frames used in the back-calculation. Indicates the first Confidence weights for frame bubble observations This indicates the effective area range of the membrane fiber end face.

[0087] Step S6: In this embodiment, the final vertical velocity of a small bubble floating in water can be characterized by a modified Stokes model, specifically expressed as follows:

[0088] in, For the density of the liquid, For gas density, It is the acceleration due to gravity. The equivalent diameter of the bubble. For the dynamic viscosity of the liquid, This is a shape correction factor. For near-spherical bubbles with a diameter of 0.5 mm to 2.0 mm, preferably, the shape correction factor is... The value ranges from 0.85 to 0.95.

[0089] The average upward velocity of the bubble in the vertical direction can be obtained by formula (1), which provides basic parameters for subsequent leakage source inference.

[0090] Step S7: Assume the leak source is located at The time when bubbles are generated is In the first Frame acquisition time The theoretical position of the bubble in space can be expressed as:

[0091] in, and These represent the bubbles in direction and Lateral drift velocity component in the direction, Indicates that the bubbles are in Vertical upward velocity in the direction of ascent.

[0092] In actual testing, the camera primarily captures images of bubbles near the liquid surface. When a bubble reaches the observation plane, i.e., near the liquid surface, it satisfies... At that time, it was captured by the image acquisition device.

[0093] From the third equation in formula (2), we can obtain that the time it takes for the bubble to rise from the leakage source to the observation plane satisfies:

[0094] in, This refers to the depth of immersion in water.

[0095] Step S8: Substituting formula (3) into the first two equations of formula (2), an inverse mapping relationship can be established between the location of the leakage source and the observed location of the bubble centroid, that is:

[0096] Therefore, it can be concluded that, given the known water immersion depth... vertical upward speed of bubbles and lateral drift speed Under these conditions, it is possible to determine the first liquid near the liquid surface. Bubble centroid coordinates observed in frame image By deducing the initial position coordinates of the leakage source at the end face of the membrane filament, .

[0097] Step S9: Considering that there may be image noise, bubble recognition error and random fluctuations in lateral drift during the actual observation process, in order to improve the accuracy and robustness of the leakage source back-inference results, this invention further introduces a multi-frame fusion optimization mechanism to perform weighted solution on the back-inference results of multiple consecutive observation frames.

[0098] Specifically, the objective function for inferring the location of the leakage source is defined as follows:

[0099] in, This indicates the time it takes for the bubble to rise from the leak source to the observation plane; For the first The confidence weight of the frame observation results is determined by the bubble segmentation quality, contour integrity, or image sharpness, and satisfies the following conditions: .

[0100] This invention obtains the optimal estimate of the leakage source location by solving the following optimization problem:

[0101] in, This represents the optimal backward coordinates of the leakage source obtained after integrating multiple observation frames.

[0102] Step S10: To determine the horizontal drift velocity of the bubble, this invention uses a statistical averaging method of the centroid displacement of the bubble in adjacent frames to determine the horizontal drift velocity. An estimate is made. The calculation formula is as follows:

[0103] in, and These represent the positions of the bubble centroids between adjacent frames. direction and Displacement in the direction, This indicates the time interval between adjacent frames.

[0104] The average lateral drift velocity of the bubble during the observation process can be obtained by the above statistical averaging method, thus providing parameter support for the back-calculation in formula (4) and formula (5).

[0105] Step S11: To ensure the physical rationality and engineering feasibility of the back-calculation results, the present invention sets the following constraints on the leakage source location solution process:

[0106] in, This indicates the effective area of ​​the membrane fiber end face, meaning that the location of the leakage source obtained by reverse calculation must fall within the membrane fiber end face area; The reasonable upper limit for the lateral drift speed is preferably 5 mm / s in this embodiment; This indicates that the number of valid observation frames used for back-calculation is no less than 5 frames, to ensure that the multi-frame optimization results have sufficient stability.

[0107] When the back-calculation results do not meet the above constraints, it can be determined that the current bubble trajectory is greatly affected by abnormal disturbances, or that there are false detections in the observation data. In this case, a re-acquisition, anomaly removal, or multi-candidate verification mechanism can be triggered to further improve the positioning accuracy.

[0108] Through the above embodiments, the present invention realizes online, high-precision wet pressure defect detection and automatic sealing of column membranes, which significantly improves product consistency and production efficiency.

[0109] Bubble recognition and validity determination based on wet pressing and deep learning

[0110] The present invention applies controlled gas pressure (preferably within the range of 0.1 to 0.2 MPa) to a fully immersed liquid environment of the membrane module, causing defective membrane fibers to generate continuous leakage bubbles due to the internal and external pressure difference, and continuously acquires multiple frames of images of the membrane fiber end face area using a high frame rate industrial camera.

[0111] In view of the characteristics of bubbles on the end face of membrane fibers, such as small size, high density, and susceptibility to light and water surface fluctuations, this invention uses an improved deep learning target detection model to accurately detect and segment the bubble region, and extract key features such as the bubble's position coordinates, area, contour, and centroid.

[0112] The improved deep learning object detection model introduces a deformable convolution (DCN) module into the backbone feature extraction network and an attention mechanism module into the feature fusion network to adaptively weight the channel dimension and spatial dimension features.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting and automatically sealing defects in a column-type membrane under wet pressure testing, characterized in that, Includes the following steps: The wet pressure test step involves applying a preset pressure fluid to the membrane module while the end face of the membrane module is immersed in the liquid medium, causing air bubbles to be generated at the end of the leaking membrane fibers. The image acquisition step involves continuously acquiring a sequence of bubble images from the end face of the membrane module or a preset observation area above the end face; The central control steps involve bubble detection, temporal consistency judgment of consecutive frames, extraction of bubble centroid trajectory, and deduction of leakage source coordinates based on the multi-frame bubble centroid trajectory. The membrane pore template positioning step maps the coordinates of the leakage source to the physical coordinate system of the membrane end face and matches them with the preset membrane pore template coordinates to determine the target leaking membrane filament. The mechanical plugging procedure involves performing the plugging action based on the location coordinates of the target leaking membrane filament.

2. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that... The wet pressure test step includes: applying a controlled air pressure range of 0.1 to 0.2 MPa in a liquid environment that fully wets the membrane module, causing the defective membrane filaments to generate continuous leakage bubbles due to the internal and external pressure difference, and continuously acquiring multiple frames of images of the membrane filament end face area using a high frame rate industrial camera.

3. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, The membrane pore template positioning steps include: I. Frame number threshold constraint: Bubbles are detected in N consecutive frames (N≥5) at the same location; II. Area change rate constraint: The bubble area change rate between adjacent frames shall not exceed the set threshold K (preferably 15% to 25%). The leakage source inversion based on the bubble centroid timing and precise membrane fiber positioning introduce the reverse inversion of the bubble rising trajectory into the field of membrane fiber leakage positioning, specifically: I. Centroid Time Sequence Construction: For each valid bubble, extract its centroid coordinates in chronological order; II. Establishment of the reverse model: Combining the spatiotemporal characteristics of bubble diffusion and floating in liquid (including buoyancy, water flow disturbance, and lateral drift factors), a mathematical model for the reverse model of the leakage source is established. Optimization Solution: By solving the optimization problem (least squares) that minimizes the sum of distances from the bubble centroids to candidate leak points across multiple frames, the spatial coordinates of the leak source are derived. After obtaining the continuous spatial coordinates of the leakage source, these coordinates are matched with a pre-constructed cylindrical membrane element end face spatial template (containing the calibration coordinates of all membrane pores). The continuous coordinates are then precisely mapped to specific membrane filament numbers, thereby achieving sub-millimeter-level leakage membrane filament positioning accuracy. For boundary situations where the reverse-input position is close to multiple candidate membrane pores, a multi-candidate membrane pore fault-tolerant verification mechanism is adopted, and the final leaking membrane filament is confirmed through secondary directional detection.

4. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, The central control steps include: the central control unit calibrates and converts the positioning results in the image coordinate system into the execution coordinates of the robotic arm, driving the automatic sealing module to perform a precise leak repair operation on the leaking membrane filaments. Its key control features include: I. Precise control of the insertion depth of the plugging component: The insertion depth is preset according to the membrane fiber diameter and membrane module structure to avoid damage to the membrane fiber due to excessive depth and failure of the plugging due to insufficient depth; II. Automatic re-inspection after sealing: After sealing is completed, the system automatically repressurizes and performs image acquisition and bubble detection to determine whether the sealing was successful; III. Closed-loop exception handling logic: a. If the number of consecutive closure failures of the same membrane filament exceeds the set number, an alarm will be triggered and the operation will be suspended; b. When the cumulative number of blocked membrane modules in a single module exceeds the threshold, the entire membrane module is deemed unusable. c. When new bubbles appear in the adjacent area after sealing, the extended detection process is automatically started.

5. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, It also includes the following steps: Direct detection of wet bubbles: The end face of the membrane module is immersed in the liquid medium, and the leak bubble image is directly acquired under pressure. The membrane fiber defect is judged by the bubble phenomenon. Bubble sequential frame temporal recognition steps: Detect, track, and judge the temporal consistency of bubbles in a continuous image sequence; distinguish between real leaking bubbles and random interference bubbles based on conditions such as continuous frame existence, centroid displacement, area change, confidence level, and trajectory stability. Leakage source inference steps based on bubble centroid trajectory: The observed bubble position is not directly equivalent to the leaking membrane filament position. The system inversely derives the bubble source coordinates based on multi-frame bubble centroid trajectory, floating / drift characteristics, and the geometric relationship between the observation area and the membrane end face. Leakage source coordinates and membrane pore template matching steps: Map the derived leakage source coordinates to the membrane end face physical coordinate system, and match them with the preset membrane pore template coordinate set to determine the specific leakage membrane wire or pore location; The detection results drive the automatic sealing process: The central control unit generates mechanical execution coordinates based on the location of the target leaking membrane filament, and controls the mechanical sealing execution unit to perform leak repair nail insertion, glue application, hot melting, glue injection or other sealing actions on the target membrane filament; Closed-loop procedure for pressure testing after sealing: After sealing is completed, continue to maintain pressure, collect images again and determine whether there are still continuous and stable bubbles at the target location, thereby confirming whether the sealing is successful, failed, needs to be resealed, or is scrapped.

6. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, It also includes the following steps: Direct detection of wet bubbles: The end face of the membrane module is immersed in the liquid medium, and the leak bubble image is directly acquired under pressure. The membrane fiber defect is judged by the bubble phenomenon. Bubble sequential frame temporal recognition steps: Detect, track, and judge the temporal consistency of bubbles in a continuous image sequence; distinguish between real leaking bubbles and random interference bubbles based on conditions such as continuous frame existence, centroid displacement, area change, confidence level, and trajectory stability. Leakage source inference steps based on bubble centroid trajectory: The observed bubble position is not directly equivalent to the leaking membrane filament position. The system inversely derives the bubble source coordinates based on multi-frame bubble centroid trajectory, floating / drift characteristics, and the geometric relationship between the observation area and the membrane end face. Leakage source coordinates and membrane pore template matching steps: Map the derived leakage source coordinates to the membrane end face physical coordinate system, and match them with the preset membrane pore template coordinate set to determine the specific leakage membrane wire or pore location; The detection results drive the automatic sealing process: The central control unit generates mechanical execution coordinates based on the location of the target leaking membrane filament, and controls the mechanical sealing execution unit to perform leak repair nail insertion, glue application, hot melting, glue injection or other sealing actions on the target membrane filament; Closed-loop procedure for pressure testing after sealing: After sealing is completed, continue to maintain pressure, collect images again and determine whether there are still continuous and stable bubbles at the target location, thereby confirming whether the sealing is successful, failed, needs to be resealed, or is scrapped.

7. The method for detecting and automatically sealing defects in a column membrane under wet pressure as described in claim 1, characterized in that, The image acquisition step also includes the following steps: continuously acquiring images of the end face of the membrane module immersed in the liquid medium or a preset observation area above the end face at a preset frame rate, in order to monitor in real time whether there is continuous bubble generation caused by membrane fiber leakage; the acquired images are synchronously cached by the central control unit and stored in association with the acquisition time, pressure value, membrane module number and camera calibration parameters, and a bubble detection image sequence is constructed in chronological order; To achieve automatic recognition of bubble regions, the central control unit performs frame-by-frame target detection and segmentation on the bubble detection image sequence, extracting the spatial location, contour boundary, area, centroid, and detection confidence of the bubble region. The bubble detection adopts a small target detection model based on deep learning.

8. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, The central control step tracks candidate bubble regions in continuous images to form a sequence of bubble spatial position changes. The planar coordinates of the centroid of the detected bubble region are calculated for each frame of the image to characterize the spatial center position of the bubble region. Assume the bubble region contains... Each pixel has its centroid coordinates. The calculation is as follows: in, The first in the bubble region The coordinates of each pixel. This represents the total number of pixels. (This is achieved by analyzing consecutive...) By processing the frame image, the time series of the centroid of the bubble region can be obtained. A bubble region is considered a valid leaking bubble when it is detected in N consecutive frames and meets the following conditions: the rate of change of area is less than the first threshold, the centroid displacement is less than the second threshold, and the detection confidence is higher than the third threshold. If it appears only in a single frame or a few frames and does not meet the continuity constraint, it is considered a transient interference.

9. The method for detecting and automatically sealing defects in a column membrane under wet pressure testing according to claim 1, characterized in that, The central control steps also include: constructing a leakage source inversion model based on the characteristics of bubble floating, diffusion, and lateral drift in the liquid medium after escaping from the end of the leaking membrane filament. The leakage source inversion model does not directly take the position of a single bubble as the leakage position, but transforms the centroid coordinates of multiple consecutive frames of bubbles to the physical coordinate system of the membrane end face. Combining the geometric relationship between the preset observation area and the membrane end face, bubble drift parameters, and image acquisition time series, the source point coordinates that minimize the residual of the bubble trajectory in multiple frames are solved as the leakage source inversion position. The leakage source inversion position is mapped to the preset membrane pore template coordinate system, and pore position matching is performed within the preset distance tolerance range. If there are multiple candidate membrane pores, the target leaking membrane filament position is determined by the minimum spatial distance, template row and column constraints, and consistency of historical trajectories.

10. A column-type membrane wet pressure testing defect detection and automatic sealing system, characterized in that, It includes a wet pressure testing module, an image acquisition module, a central control unit, a membrane pore template positioning module, and a mechanical sealing execution unit; The wet pressure module is used to apply a preset pressure fluid to the membrane module while the end face of the membrane module is immersed in the liquid medium, so as to generate bubbles at the end of the leaking membrane filament; The image acquisition module is used to continuously acquire bubble image sequences from the end face of the membrane module or a preset observation area above the end face; The central control unit is used to perform bubble detection, consecutive frame temporal consistency judgment, bubble centroid trajectory extraction on bubble image sequences, and deduce the coordinates of the leakage source based on the multi-frame bubble centroid trajectory. The membrane pore template positioning module is used to map the coordinates of the leakage source to the physical coordinate system of the membrane end face and match them with the preset membrane pore template coordinates to determine the target leaking membrane filament; The mechanical plugging actuator is used to perform plugging actions based on the position coordinates of the target leaking membrane filament; The central control unit is also used to maintain the pressure state after sealing and to acquire image sequences again to determine whether the sealing was successful.

Images