Weld inspection method, system and application based on the fusion of machine vision and vacuum inspection

Abstract

This invention discloses a weld inspection method, system, and application based on the fusion of machine vision and vacuum inspection. The method includes: cleaning the weld surface and coating it with a bubble-forming leak detection medium; covering the weld inspection section with a vacuum chamber and performing a sealing self-test; creating a pressure differential by evacuating and simultaneously acquiring video and pressure sequences; establishing a local coordinate system for the weld through geometric calibration, performing time-series bubble event detection and clustering to locate leak points within the ROI; performing isolation pressure rise measurement to calculate leakage characterization parameters, and fusing them with bubble event characteristics to determine true leaks / interference, outputting leak point coordinates and severity classification; roughly estimating the equivalent defect size for more severe leak points based on calibration mapping; and generating an inspection report containing coordinates, classification, rough estimation results, and evidence. This solution reduces human subjectivity, improves anti-interference and traceability, and is suitable for on-site airtightness inspection of long welds.

Description

Technical Field

[0001] This invention relates to the field of welding inspection technology, and in particular to a weld inspection method, system and application based on the integration of machine vision and vacuum inspection. Background Technology

[0002] As critical connection points in equipment such as pressure vessels, storage tanks, enclosures, and pipelines, welds directly impact equipment safety and service life due to their airtightness and structural integrity. In existing weld airtightness testing, the vacuum chamber bubbling method is widely used due to its simplicity and strong field applicability: by coating the weld surface with a foaming medium and applying a pressure differential, bubbles are generated in the leakage channels to locate defects. However, this method typically relies on manual visual interpretation, resulting in strong subjectivity and poor repeatability. Especially under conditions of interference such as reflection, atomization, foam accumulation, and leakage at the sealing edge, misjudgments or missed detections are prone to occur. Furthermore, the vacuum chamber bubbling method primarily provides "presence and location" information, lacking a unified quantitative basis for classifying the severity of leaks, making it difficult to form auditable detection conclusions. Simultaneously, the lack of relevant video recordings leads to limitations in retrospective verification; moreover, it is difficult to make rough estimates of equivalent dimensions or leakage capacity for more severe leaks to support maintenance prioritization and rework effectiveness evaluation. While some existing quantitative leak detection methods (such as helium mass spectrometry leak detection) can provide the leakage rate, their high equipment cost and stringent operating requirements make them difficult to rapidly implement in large-area, long weld seams.

[0003] Therefore, there is an urgent need for a weld inspection method and system that can balance field applicability and result reliability, achieve defect location, severity classification, and make rough estimation of equivalent dimensions for more serious defects. Summary of the Invention

[0004] In view of this, the purpose of this invention is to propose a weld inspection method, system and application based on the fusion of machine vision and vacuum inspection that is convenient to operate, flexible in application and capable of defect location, severity classification and equivalent size estimation of more serious defects.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: A weld inspection method based on the fusion of machine vision and vacuum inspection, comprising: S1. Clean the surface of the weld to be inspected and apply a foaming leak detection medium to the weld surface; S2. Cover the inspection section of the weld with a vacuum box with a transparent observation window, so that the sealing gasket of the vacuum box is in contact with the surface of the weld, and perform a sealing self-inspection on the vacuum box. S3. Evacuate the vacuum chamber to form a preset pressure difference. During the vacuum holding phase, simultaneously acquire video sequences of the weld area and pressure sequences inside the vacuum chamber. S4. Based on the video sequence, establish the region of interest for the weld detection area, and establish the local coordinate system of the weld through pixel-physical scale calibration; S5. Perform temporal bubble event detection on the video sequence to obtain the position coordinates of at least one bubble event source point in the local coordinate system of the weld, and determine whether the bubble event source point is a suspected leak point based on the temporal stability of the bubble event.

[0006] As a possible implementation method, this solution further includes, in addition to the above: S6. Perform isolation pressure rise measurement during or after the vacuum holding phase, and calculate the pressure rise slope based on the pressure sequence to obtain leakage characterization parameters; S7. The location coordinates of the suspected leak point, bubble event characteristics and leakage volume characterization parameters are fused and judged to output the leak point location result. The severity of the leak is classified according to the preset classification rules to generate a severity classification of weld defects.

[0007] As a possible implementation method, this solution further includes, in addition to the above: S8. For leak points that reach the preset severity threshold, output a rough estimate of the equivalent defect size based on the calibration mapping relationship; S9. Generate an inspection report that includes the coordinates of the leak point, severity rating, and a rough estimate of the equivalent defect size.

[0008] As a preferred implementation option, preferably, in this scheme S2, the sealing self-test includes: after the vacuum chamber is covered but before an effective leak detection is performed on the weld area, testing the pressure stability after the vacuum chamber is evacuated to the target vacuum level. When pressure fluctuations exceed a preset threshold or bubbling occurs concentrated at the sealing edge of the vacuum chamber, it is determined that the chamber is leaking air, and a prompt will be output indicating that the seal should be re-attached or replaced. When the pressure fluctuation does not exceed the preset threshold or there is no concentrated bubbling at the sealing edge of the vacuum chamber, the seal is deemed qualified.

[0009] As a preferred implementation option, preferably, in this scheme S4, the establishment of the local coordinate system of the weld includes: using calibration marks and reference scales set on the inner side of the observation window of the vacuum chamber or the surface of the weld to perform geometric calibration on the video image to obtain the pixel-to-millimeter mapping relationship, and defining the direction along the weld as the x-axis and the direction perpendicular to the weld as the y-axis based on the direction of the weld centerline.

[0010] As a preferred implementation option, in S5 of this scheme, the temporal bubble event detection includes: performing differential or background modeling on adjacent frames in the video sequence to extract newly generated bubble event pixel clusters, performing connected component analysis and morphological filtering on the pixel clusters, and recognizing the pixel clusters as valid bubble events only when they appear within a consecutive preset number of frames.

[0011] As a preferred implementation option, preferably, in this scheme S6, the isolation pressure rise measurement includes: shutting off the vacuum pump passage after the vacuum holding phase to isolate the vacuum chamber from the vacuum pump, collecting the pressure sequence within a preset time window, and obtaining the pressure rise slope through linear fitting.

[0012] As a preferred implementation option, preferably, in scheme S6, the leakage characteristic parameter is calculated from the effective volume of the vacuum chamber and the pressure rise slope. The following definition applies: in, The effective volume of the vacuum chamber The slope of the pressure rise.

[0013] As a preferred implementation option, preferably, in this scheme S7, the bubble event characteristics include at least one or more of the following: bubble delay time, bubble event generation rate per unit time, bubble event continuity, and the growth rate of bubble event coverage area.

[0014] As a preferred implementation option, preferably, in this solution S7, the fusion determination includes determining a valid leak point when the bubble event source point is located in the region of interest of the weld and its leakage characterization parameter exceeds a preset threshold during the vacuum stabilization stage. When bubble events are mainly distributed at the sealing edge of the vacuum chamber or are related to pressure instability characteristics, they are identified as interfering bubble events and are removed.

[0015] As a preferred implementation option, preferably, in this scheme S7, the grading rule is constructed based on bubble event characteristics and leakage quantity characterization parameters, and the leakage severity is divided into at least three levels, with higher levels indicating more severe leakage.

[0016] As a preferred implementation option, preferably, in this scheme S8, the calibration mapping relationship is established by a standard reference leakage channel or a standard leak sample. The calibration mapping relationship is used to map the leakage quantity characterization parameter or its combination with bubble event characteristics into an interval rough estimate of the equivalent defect size.

[0017] Based on the above, this solution also proposes a weld inspection system based on the fusion of machine vision and vacuum inspection, which includes: Vacuum chamber assembly, which includes a transparent observation window, a sealing gasket, and an air extraction interface; A vacuum generating component, connected to the vacuum chamber assembly, is used to evacuate the vacuum chamber assembly and form a preset pressure difference, and includes a valve for isolating pressure rise measurement; A pressure acquisition component is connected to the vacuum chamber assembly and is used to acquire the pressure inside the vacuum chamber to form a pressure sequence; A visual acquisition component is connected to the vacuum chamber assembly and is used to acquire video sequences of the weld area through the transparent observation window; The processor is configured to execute the weld inspection method based on the fusion of machine vision and vacuum inspection described above, and to output inspection results. The memory is configured to store the inspection results, as well as the working parameters generated during the inspection process and the process data collected during the inspection.

[0018] Based on the above, this solution also proposes a weld inspection device based on vacuum detection, which includes the weld inspection system based on the fusion of machine vision and vacuum detection mentioned above. The system also includes a communication module connected to a cloud network. The device also includes a server deployed in the cloud and a management platform for accessing the server. The server actively acquires or passively receives the stored detection results, as well as the corresponding working parameters and process data, transmitted by the system through the communication module at a preset time frequency.

[0019] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: Compared with existing methods for weld airtightness detection such as vacuum chamber bubble leak detection, this solution can significantly improve the objectivity, repeatability, and auditability of the detection results while maintaining the advantages of field applicability and low cost. First, by introducing machine vision into the monitoring of the bubble process in the vacuum chamber observation window, and using ROI limitation, dynamic bubble event extraction, and temporal stability screening, "manual visual bubble finding" is transformed into the output of leak point coordinates in the local coordinate system of the weld, and key frames / video evidence are automatically associated to achieve standardization and traceability of defect location, reducing misjudgments caused by differences in operator experience. Second, during implementation, this solution can effectively distinguish between real weld leaks and false signals such as chamber self-leakage and seal edge leaks through interference suppression strategies such as seal self-inspection, edge leak identification, and reflection / foaming / foam accumulation, thereby improving robustness and reliability under complex field conditions. Furthermore, this solution, based on visual bubble phase characteristics (such as bubbling delay, bubble event frequency, continuity, area growth rate, etc.), introduces valve-controlled isolation pressure rise measurement and derives leakage characterization parameters such as throughput from the pressure curve, achieving an upgrade from "qualitative presence or absence" to "quantifiable grading," making the grading of leakage severity consistent, verifiable, and interpretable. For more severe leaks, this solution can combine standard leak / reference channel calibration mapping to provide a range of rough estimates for equivalent defect sizes (e.g., equivalent aperture), used for maintenance priority ranking, rework strategy selection, and quantitative comparison of effects before and after rework, thus forming a closed-loop quality control process of detection—location—grading—rough estimation—rework—re-inspection; finally, it outputs a structured report and data storage, facilitating quality management, accountability, and project acceptance. Attached Figure Description

[0020] 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.

[0021] Figure 1 This is one example illustrating the implementation process of the detection method in this scheme; Figure 2 This is a brief schematic diagram of part of the implementation process of step S4 of the detection method in this scheme; Figure 3 This is the second example illustrating the implementation process of the detection method in this scheme; Figure 4 This is the third example illustrating the implementation process of the detection method in this scheme; Figure 5 This is a schematic diagram of the unit module connections of the detection system in this solution; Figure 6 This is a schematic diagram of the unit module connections of the testing equipment in this solution. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown in the figure, this embodiment provides a weld inspection method based on the fusion of machine vision and vacuum inspection, which includes: S1. Clean the surface of the weld to be inspected and apply a foaming leak detection medium to the weld surface; S2. Cover the inspection section of the weld with a vacuum box with a transparent observation window, so that the sealing gasket of the vacuum box is in contact with the surface of the weld, and perform a sealing self-inspection on the vacuum box. S3. Evacuate the vacuum chamber to form a preset pressure difference. During the vacuum holding phase, simultaneously acquire video sequences of the weld area and pressure sequences inside the vacuum chamber. S4. Based on the video sequence, establish the region of interest for the weld detection area, and establish the local coordinate system of the weld through pixel-physical scale calibration; S5. Perform temporal bubble event detection on the video sequence to obtain the position coordinates of at least one bubble event source point in the local coordinate system of the weld, and determine whether the bubble event source point is a suspected leak point based on the temporal stability of the bubble event.

[0024] In S1 of this implementation scheme, information recording can be pre-initialized for the workpiece or weld segment to improve the corresponding work details, realize the establishment and traceability binding of weld segment information, and provide data support for subsequent management personnel to conduct inspection, traceability, maintenance, and verification. The corresponding work details may include the following: a) Obtain the weld number, weld type, and inspection length of the weld to be inspected, and divide the inspection length into several inspection segments; b) Establish a segment number for each inspection segment and record the start and end positions of that segment on the global mileage of the weld; c) Bind the device number used in this test, such as: vacuum chamber number, camera number, pressure sensor number, data processor number, etc., or one or more of these.

[0025] In step S1, the cleaning process for the surface of the weld to be inspected may include the following: a) The weld and the preset width range on both sides are treated to remove oil, water, dust and rust; if necessary, the weld spatter and burrs can also be lightly ground. b) Confirm that the surface where the weld is located is dry or meets the surface condition suitable for the leak detection medium; In addition to the above, the surface condition parameters of the weld (such as wet / dry state, contamination level, roughness level, etc.) can also be recorded.

[0026] Cleaning the surface of the weld to be inspected can reduce the problem of abnormal foam accumulation caused by oil or other debris, and can also reduce the abnormal situation of foaming inhibition caused by oil film clogging of defect microchannels.

[0027] The application of a foaming leak detection medium to the weld surface can include the following: a) Select the foaming leak detection medium (foaming liquid / leak detection liquid) and record the model and batch number; b) Apply a pre-defined coating method (spraying, brushing, or scraping) evenly to the centerline and both sides of the weld, and control the coverage width; c) By limiting the amount of spraying, the number of sprayings, the size of the scraper, or the amount used per unit area, the thickness of the medium film is kept within a repeatable range; d) Record information on the coating method and the film thickness formed by the bubbling leak detection medium.

[0028] The above information can be used as the basic data for testing. By associating it with the process data, the test results can be packaged together and output later to facilitate the reliability traceability of the testing process.

[0029] To improve the reliability of testing and reduce testing anomalies caused by problems with the vacuum chamber itself, As a preferred implementation option, preferably, in this scheme S2, the sealing self-test includes: after the vacuum chamber is covered but before an effective leak detection is performed on the weld area, testing the pressure stability after the vacuum chamber is evacuated to the target vacuum level. When pressure fluctuations exceed a preset threshold or bubbling occurs concentrated at the sealing edge of the vacuum chamber, it is determined that the chamber is leaking air, and a prompt will be output indicating that the seal should be re-attached or replaced. When the pressure fluctuation does not exceed the preset threshold or there is no concentrated bubbling at the sealing edge of the vacuum chamber, the seal is deemed qualified.

[0030] As an example of leak detection and anomaly assessment, if the pressure fluctuation amplitude of the vacuum chamber exceeds the preset threshold after it has been evacuated to the target vacuum level, the vacuum chamber is considered to be abnormal. Operators need to perform a re-attachment operation, or further determine whether the gasket in contact with the weld surface of the vacuum chamber is abnormal. If the gasket is abnormal, it should be cleaned, repaired, or replaced directly.

[0031] For step S3, this scheme evacuates the vacuum chamber to form a preset pressure difference, and simultaneously acquires video sequences of the weld area and pressure sequences inside the vacuum chamber during the vacuum holding phase.

[0032] Specifically, the vacuum chamber can be evacuated by controlling a vacuum pump connected to it, so that it reaches the preset target pressure. or target pressure difference .

[0033] During the vacuum holding phase, the method for determining whether the vacuum chamber has entered a stable state can be as follows: The vacuum level inside the vacuum chamber is monitored by sensors for a continuous period of time. If the pressure fluctuation amplitude is less than the preset threshold This allows us to define the vacuum chamber as entering a stable state, i.e., the vacuum holding phase, and from this, we can obtain or determine the starting moment when the stable state is reached. .

[0034] To ensure greater consistency in time among the data, this solution performs time synchronization and sequence alignment when acquiring video sequences from the weld area and pressure sequences from the vacuum chamber. Specifically, this can include: when acquiring data from the weld area, both the camera frames of the video acquisition device (visual acquisition component) and the pressure samples from the pressure acquisition component are timestamped, and both components use the same clock source from the same processor system.

[0035] In step S3 of this scheme, the video sequence of the weld area being synchronously acquired during the vacuum holding stage is defined as... For each sampling time point, the pressure sequence inside the vacuum chamber is defined as: ;in, The first Sampling time and sampling pressure.

[0036] In step S4 of this solution, a region of interest (ROI) for the weld inspection area is established based on the video sequence acquired above. This ROI may include the following: a) Extract image frames from the video sequence, and then perform preliminary ROI cropping based on prompt information (such as user annotations, prior knowledge, coarse localization results, etc.) to obtain a coarse ROI; b) Detect the centerline corresponding to the weld within the coarse ROI. Weld line segments can be obtained by combining edge detection with line fitting or texture enhancement, and then using line feature extraction. c) Using the weld seam segment as the center, expand the preset width to form a precise ROI mask. Then it is applied to subsequent bubble event detection.

[0037] In this step, the cropping and extraction methods used are based on existing image technologies, and their underlying principles will not be elaborated upon here.

[0038] In step S4 of this scheme, the local coordinate system of the weld is established through pixel-physical scale calibration. The establishment of the local coordinate system of the weld includes: using calibration marks and reference scales set on the inner side of the observation window of the vacuum chamber or the surface of the weld to perform geometric calibration on the video image to obtain the mapping relationship from pixels to millimeters, and defining the direction along the weld as the x-axis and the direction perpendicular to the weld as the y-axis based on the direction of the weld centerline.

[0039] As one example, it includes the following: Assuming the surface where the weld is located and the calibration mark are on the same approximate plane, the pixel coordinates and physical plane coordinates satisfy the following homography transformation: in, These are pixel coordinates. These are physical plane coordinates, As a scale factor, This is the homography matrix, used for mapping between the pixel plane and the physical plane. In this scheme, For a 3×3 matrix, at least 4 sets of non-collinear corresponding points are required. The solution can be obtained by establishing a system of linear equations using Direct Linear Transformation (DLT) and then solving it using the least squares method. Alternatively, in engineering implementation, the solution and distortion correction can be accomplished by calling a computer vision library, which is an existing technology and will not be elaborated on here.

[0040] In step S4, it is assumed that the weld direction is local. The axis, perpendicular to the weld direction, is local. The axis, physical plane coordinates are The direction angle of the weld centerline is The local origin is Then the transformation of local coordinates can be defined as follows:

[0041] The 2×2 matrix in the above formula is a rotation matrix, which is used to rotate and align the physical coordinates with the weld direction.

[0042] Through step S4, this solution establishes a coordinate system for the area where the weld is located, providing a foundation for subsequent detection and output of information on the location of leak points and defects.

[0043] As a preferred implementation option, in S5 of this scheme, the temporal bubble event detection includes: performing differential or background modeling on adjacent frames in the video sequence to extract newly generated bubble event pixel clusters, performing connected component analysis and morphological filtering on the pixel clusters, and recognizing the pixel clusters as valid bubble events only when they appear within a consecutive preset number of frames.

[0044] As an example of implementation, in terms of bubble detection, step S5 of this scheme performs temporal bubble event detection on the video sequence through inter-frame differential inference to obtain the position coordinates of at least one bubble event source point in the local coordinate system of the weld.

[0045] As an example implementation, to enhance sensitivity to "newly formed bubbles" rather than static bubble accumulation, this scheme employs inter-frame differential inference to detect dynamically changing regions, with the function defined as follows:

[0046] in, This is the inter-frame difference value. These are the times in the video sequence. The image brightness or grayscale value of the image frame; Thresholding is applied to the pixels within the ROI to obtain a binary event map, which is defined as follows:

[0047] in, This is a binary graph of candidate bubble events. The difference threshold, For indicator functions, This is the fine ROI mask corresponding to the image frame.

[0048] Binary graph of candidate bubble events Perform morphological opening and closing operations to remove noise and fill in interference from small holes to obtain the time step. The set of connected components of candidate bubble events .

[0049] Based on the above, for each connected component Calculate its pixel centroid and centroid coordinates. The definition is as follows:

[0050] in, For connected components centroid pixel coordinates, For the set of pixels in the connected component, These are pixel coordinates. This represents the number of pixels in the connected components. Will Transformed into physical coordinates through homography mapping Then, based on the aforementioned transformation of local coordinates, it is converted into weld local coordinates. This forms an event point flow, which is defined as follows:

[0051] in, This refers to a set of event points or an event stream; the area or number of pixels of the connected components can be used as the event intensity surrogate. .

[0052] The method for determining whether a bubble event source is a suspected leak point based on the temporal stability of bubble events includes the following: For event points Spatial clustering is performed, grouping events whose distance does not exceed the spatial clustering radius threshold into the same candidate source point cluster. .

[0053] In the statistics window Within, statistical clusters Number of events With the number of frames And extract the following bubble phase characteristics: (1) Continuity; (2) Event frequency; (3) Delayed bubbling.

[0054] Among them, the continuity index The function is defined as follows:

[0055] Event frequency The function is defined as follows:

[0056] Delayed foaming The function is defined as follows:

[0057] in, This represents the total number of frames within the window. The time window span, i.e., the duration of the statistical window. For clusters The moment when the condition for consecutive occurrences is first met.

[0058] For continuity, set relevant judgment thresholds (event count thresholds). Continuity index threshold As an example, the stability judgment criteria are as follows: and When the continuity indicator is met, it is identified as a suspected leak point, and then the results are aggregated to form a set of suspected leak points. .

[0059] After step S5 of this solution, a set of suspected leak points can be output. and the characteristics of each suspected leak point. The suspected leak point The local coordinates of the weld, blistering delay, continuity index, and event frequency.

[0060] To facilitate tracing, this solution can also extract or export the corresponding image frames from the video sequences that identify suspected leak points, then package them and associate them with the detection results for tracing and verification.

[0061] This solution significantly improves the objectivity, repeatability, and auditability of inspection results while maintaining on-site applicability and low cost. Firstly, by introducing machine vision into the bubbling process monitoring within the vacuum chamber's observation window, and utilizing ROI limitation, dynamic bubble event extraction, and temporal stability screening, it transforms "manual visual bubble finding" into outputting the leak point coordinates in the weld's local coordinate system. It also automatically links keyframes / video evidence, achieving standardized and traceable defect localization and reducing misjudgments caused by differences in operator experience. Based on the above, in order to further improve the accuracy of identifying different suspected leak points, combined with Figure 3 As shown, as one possible implementation, this embodiment further includes: S6. Perform isolation pressure rise measurement during or after the vacuum holding phase, and calculate the pressure rise slope based on the pressure sequence to obtain leakage characterization parameters; S7. The location coordinates of the suspected leak point, bubble event characteristics and leakage volume characterization parameters are fused and judged to output the leak point location result. The severity of the leak is classified according to the preset classification rules to generate a severity classification of weld defects.

[0062] In S6 of this scheme, the isolation pressure rise measurement includes: after the vacuum holding phase ends or the preset observation time is met, closing the vacuum pump passage to isolate the vacuum chamber from the vacuum pump, so that the vacuum chamber forms an approximately closed volume, and then selecting a pressure rise measurement time window. To avoid valve switching transients, pressure sequences were collected within a preset time window, and the pressure rise slope was obtained through linear fitting. .

[0063] As one method for deriving the pressure rise slope, it can be obtained through least squares derivation.

[0064] The pressure change within the time window can be considered approximately linear, and the following model is established:

[0065] in, The pressure rise slope is calculated using least squares minimization, and its definition is as follows:

[0066] right The solution yields the following:

[0067] in, For the first Secondary sampling pressure For the first Next sampling time As the starting point of the window, For the fitting intercept, It is the pressure rise slope. For time windows The average internal pressure.

[0068] In this scheme, within a closed volume V, the temperature T is approximately constant, and the ideal gas equation can be defined as follows:

[0069] Treating V as a constant, taking the derivative with respect to time yields the following definition:

[0070] Define throughput (Or leakage characterization parameters) are as follows:

[0071] in, The pressure inside the vacuum chamber, The effective volume of the vacuum chamber This refers to the amount of gaseous substance inside the vacuum chamber. The gas constant is... Absolute temperature The slope of the pressure rise.

[0072] In summary, in this scheme S6, the leakage characteristic parameter is calculated from the effective volume of the vacuum chamber and the pressure rise slope. The following definition applies:

[0073] in, The effective volume of the vacuum chamber The slope of the pressure rise.

[0074] Regarding the classification of weld defects, in this scheme S7, the bubble event characteristics include at least one or more of the following: bubble initiation delay time, bubble event generation rate per unit time, bubble event continuity, and the growth rate of the bubble event coverage area.

[0075] In addition, preferably, in S7 of this solution, the fusion determination includes determining a valid leak point when the bubble event source point is located in the region of interest of the weld and its leakage characterization parameter exceeds a preset threshold during the vacuum stabilization stage. When bubble events are mainly distributed at the sealing edge of the vacuum chamber or are related to pressure instability characteristics, they are identified as interfering bubble events and are removed.

[0076] As an example of an implementation scheme, this scheme can output the leak point location result by fusing the location coordinates of the suspected leak point, bubble event characteristics, and leakage quantity characterization parameters. During the fusing determination, a fusing score function is constructed, defined as follows:

[0077] The fusion score is obtained through the Sigmoid function. Compression into confidence level, which is defined as follows:

[0078] in, This is the distance from the leak point to the sealing edge of the vacuum chamber, which is used to suppress edge leakage interference; It is a normalization function (which can be a linear normalization function, a piecewise function, or a logarithmic compression function). ; For fusion score The corresponding confidence level, These are the suspected leak points. Bubble formation delay, continuity indicators, and event frequency. The leakage rate is a parameter that characterizes the throughput. These are the weighting coefficients. It is a natural constant.

[0079] The above solution can solve the problems of "visual false leakage" caused by edge leakage and reflection, as well as "pressure false leakage" caused by self-leakage of the enclosure, and improve reliability through multi-source consistency.

[0080] As a preferred implementation option, preferably, in this scheme S7, the grading rule is constructed based on bubble event characteristics and leakage quantity characterization parameters, and the leakage severity is divided into at least three levels, with higher levels indicating more severe leakage.

[0081] As an example of an implementation plan, this plan can output a set of suspected leak points. In the middle, satisfy The leak points were further classified and roughly estimated to determine the extent of the leak.

[0082] Specifically, a comprehensive severity index can be pre-built during the process of constructing and mapping hierarchical indicators. Its definition is as follows:

[0083] in, These are the suspected leak points. Bubble formation delay, continuity indicators, and event frequency. The leakage rate is a parameter that characterizes the throughput. These are the weighting coefficients.

[0084] By combining severity indicators Corresponding to different threshold segments, they are mapped to levels. Taking level four as an example, it ranges from G1 to G4. The higher the level, the more serious the welding leakage. The corresponding evaluation method is as follows: G1 (minor leak): ; G2 (minor leak): ; G3 (Middle Leak): ; G4 (big leak): .

[0085] Additionally, when suspected leak points are gathered It does not exist in If the leak point is not detected, it is classified as G0, meaning no leak was detected.

[0086] This step can solve the problems of high subjectivity and poor repeatability of relying solely on bubble phase classification; by introducing semi-quantitative pressure and interpretable rules, an auditable classification result can be formed.

[0087] Based on visual bubble phase characteristics (such as bubbling delay, bubble event frequency, continuity, area growth rate, etc.), this solution introduces valve-controlled isolation pressure rise measurement and derives leakage characterization parameters such as throughput from the pressure curve, realizing an upgrade from "qualitative presence or absence" to "quantifiable classification", making the classification of leakage severity consistent, verifiable and interpretable.

[0088] To assist in determining the maintenance plan, for some welds with more serious leaks, this plan further estimates the equivalent dimensions of the more serious leak points.

[0089] Combination Figure 4 As shown, as one possible implementation, this solution further includes, in addition to the above: S8. For leak points that reach the preset severity threshold, output a rough estimate of the equivalent defect size based on the calibration mapping relationship; S9. Generate an inspection report that includes the coordinates of the leak point, severity rating, and a rough estimate of the equivalent defect size.

[0090] Preferably, in this scheme S8, the calibration mapping relationship is established by a standard reference leakage channel or a standard leak sample. The calibration mapping relationship is used to map the leakage quantity characterization parameter or its combination with bubble event characteristics into an interval rough estimate of the equivalent defect size.

[0091] In step S8, the coarse estimate of the equivalent defect size based on the calibration mapping relationship can include the following: The relationship between throughput and volumetric flow rate is established as follows: In the aforementioned throughput Based on the function definition, a reference pressure can be selected in the rough estimation of engineering. (For example, atmospheric pressure) approximates the throughput as a volumetric flow rate. Its definition is as follows:

[0092] By obtaining volumetric flow rate This can provide volumetric flow rate input for subsequent equivalent aperture models, ensuring dimensional consistency in the rough estimation results.

[0093] As an example, this scheme can derive the equivalent orifice diameter of leakage points at the G3 or G4 level using the orifice approximation, with the orifice approximation defined as follows:

[0094] in, Substitute and solve for the equivalent aperture The definition is as follows:

[0095] If based on throughput If indicated, then it can be Substituting the values, we get the following formula:

[0096] in, For equivalent aperture, This is the equivalent aperture area. For flow coefficient, For volumetric flow rate, atmospheric pressure and the internal pressure of the vacuum chamber The difference, air density, For reference pressure.

[0097] This solution provides a rough estimate of the "size magnitude" for severe leaks, facilitating repair prioritization and re-inspection comparison. It is important to note the equivalent pore size. The estimated value is an equivalent quantity, which is mainly used to characterize the leakage channel capacity and is not equivalent to the actual defect geometry. The error may increase for micro-leakage or non-orifice flow conditions, so it is preferred to apply it to more severe levels.

[0098] As an example, to reduce the impact of uncertainties in dielectric film thickness, surface condition, and flow coefficient on the rough estimation, a calibration library can be established in advance: select several standard leak / reference leak channel samples to obtain... The data pairs are used to build a regression model in the logarithmic domain, which is defined as follows:

[0099] This yields the equivalent aperture. Its definition is as follows:

[0100] A rough estimate of the interval is given based on residual statistics. .

[0101] in, The first The throughput and equivalent aperture of a calibration sample, where a and b are regression coefficients. It is an exponential function. The equivalent aperture after calibration correction. This is a rough estimate of the boundary of the interval.

[0102] For more serious leaks, this solution combines standard leak / reference channel calibration mapping to provide a range of rough estimates for equivalent defect sizes (e.g., equivalent aperture). This can be used for maintenance priority ranking, rework strategy selection, and quantitative comparison of the effects before and after rework, thus forming a closed-loop quality control process of detection—location—grading—rough estimation—rework—re-inspection. Finally, it outputs a structured report and data evidence, which facilitates quality management, accountability, and project acceptance.

[0103] Combination Figure 5 As shown, based on the above, this solution also proposes a weld inspection system based on the fusion of machine vision and vacuum inspection, which includes: Vacuum chamber assembly, which includes a transparent observation window, a sealing gasket, and an air extraction interface; A vacuum generating component, connected to the vacuum chamber assembly, is used to evacuate the vacuum chamber assembly and form a preset pressure difference, and includes a valve for isolating pressure rise measurement; A pressure acquisition component is connected to the vacuum chamber assembly and is used to acquire the pressure inside the vacuum chamber to form a pressure sequence; A visual acquisition component is connected to the vacuum chamber assembly and is used to acquire video sequences of the weld area through the transparent observation window; The processor is configured to execute the weld inspection method based on the fusion of machine vision and vacuum inspection described above, and to output inspection results. The memory is configured to store the inspection results, as well as the working parameters generated during the inspection process and the process data collected during the inspection.

[0104] In this solution, the visual acquisition component may include a ring-shaped diffused light source, and further include a polarizer to reduce reflection interference from the observation window, thereby improving the video quality during visual acquisition.

[0105] To facilitate calibration, the vacuum chamber assembly may be equipped with calibration marks or detachable reference scales to establish pixel-to-millimeter mapping relationships and local coordinate systems for weld seams.

[0106] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] Combination Figure 6As shown, based on the above, this solution also proposes a weld inspection device based on vacuum detection, which includes the weld inspection system based on the fusion of machine vision and vacuum detection as described above. The system also includes a communication module connected to a cloud network. The device also includes a server deployed in the cloud and a management platform for accessing the server. The server actively acquires or passively receives the stored detection results, as well as the corresponding working parameters and process data, transmitted by the system through the communication module at a preset time frequency.

[0109] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A weld inspection method based on the fusion of machine vision and vacuum inspection, characterized in that, It includes: S1. Clean the surface of the weld to be inspected and apply a foaming leak detection medium to the weld surface; S2. Cover the inspection section of the weld with a vacuum box with a transparent observation window, so that the sealing gasket of the vacuum box is in contact with the surface of the weld, and perform a sealing self-inspection on the vacuum box. S3. Evacuate the vacuum chamber to form a preset pressure difference. During the vacuum holding phase, simultaneously acquire video sequences of the weld area and pressure sequences inside the vacuum chamber. S4. Based on the video sequence, establish the region of interest for the weld detection area, and establish the local coordinate system of the weld through pixel-physical scale calibration; S5. Perform temporal bubble event detection on the video sequence to obtain the position coordinates of at least one bubble event source point in the local coordinate system of the weld, and determine whether the bubble event source point is a suspected leak point based on the temporal stability of the bubble event.

2. The method for weld detection based on fusion of machine vision and vacuum detection according to claim 1, wherein, It also includes: S6. Perform isolation pressure rise measurement during or after the vacuum holding phase, and calculate the pressure rise slope based on the pressure sequence to obtain leakage characterization parameters; S7. The location coordinates of the suspected leak point, bubble event characteristics and leakage volume characterization parameters are fused and judged to output the leak point location result. The severity of the leak is classified according to the preset classification rules to generate a severity classification of weld defects. 3.The method of claim 2, wherein, It also includes: S8. For leak points that reach the preset severity threshold, output a rough estimate of the equivalent defect size based on the calibration mapping relationship; S9. Generate an inspection report that includes the coordinates of the leak point, severity rating, and a rough estimate of the equivalent defect size.

4. The method of claim 1 to 3, wherein In S2, the self-test for sealing includes: after the vacuum chamber is covered but before an effective leak detection is performed on the weld area, testing the pressure stability after the vacuum chamber is evacuated to the target vacuum level. When pressure fluctuations exceed a preset threshold or bubbling occurs concentrated at the sealing edge of the vacuum chamber, it is determined that the chamber is leaking air, and a prompt will be output indicating that the seal should be re-attached or replaced. When the pressure fluctuation does not exceed the preset threshold or there is no concentrated bubbling at the sealing edge of the vacuum chamber, the seal is deemed qualified.

5. The weld inspection method based on the fusion of machine vision and vacuum inspection as described in claim 3, characterized in that, In S4, the establishment of the local coordinate system of the weld includes: using the calibration marks and reference scales set on the inner side of the observation window of the vacuum chamber or the surface of the weld to perform geometric calibration on the video image to obtain the pixel-to-millimeter mapping relationship, and defining the direction along the weld as the x-axis and the direction perpendicular to the weld as the y-axis based on the direction of the weld centerline. In S5, the temporal bubble event detection includes: performing differential or background modeling on adjacent frames in the video sequence to extract newly generated bubble event pixel clusters, performing connected component analysis and morphological filtering on the pixel clusters, and recognizing the pixel clusters as valid bubble events only when they appear within a consecutive preset number of frames.

6. The method for weld detection based on fusion of machine vision and vacuum detection according to claim 5, wherein, In S6, the isolation pressure rise measurement includes: shutting off the vacuum pump passage after the vacuum holding phase ends to isolate the vacuum chamber from the vacuum pump, collecting the pressure sequence within a preset time window, and obtaining the pressure rise slope through linear fitting; In S6, the leakage quantity characterization parameter is calculated from the effective volume of the vacuum tank and the pressure rise slope, and the leakage quantity characterization parameter satisfies the following definition: wherein is the effective volume of the vacuum chamber, is the pressure rise slope.

7. The method for weld detection based on fusion of machine vision and vacuum detection according to claim 6, wherein, In S7, the bubble event characteristics include at least one or more of the following: bubble delay time, bubble event generation rate per unit time, bubble event continuity, and growth rate of bubble event coverage area. In S7, the fusion determination includes determining a valid leak point when the bubble event source point is located in the region of interest of the weld and its leakage characterization parameter exceeds a preset threshold during the vacuum stabilization stage. When bubble events are mainly distributed at the sealing edge of the vacuum chamber or are related to pressure instability characteristics, they are identified as interfering bubble events and are removed. In S7, the grading rule is constructed based on bubble event characteristics and leakage quantity characterization parameters, and the severity of leakage is divided into at least three levels, with higher levels indicating more severe leakage.

8. The method for weld detection based on fusion of machine vision and vacuum detection according to claim 7, wherein, In S8, the calibration mapping relationship is established by a standard reference leakage channel or a standard leak sample. The calibration mapping relationship is used to map the leakage quantity characterization parameter or its combination with bubble event characteristics into an interval rough estimate of the equivalent defect size.

9. A welding seam detection system based on fusion of machine vision and vacuum detection, characterized in that, It includes: Vacuum chamber assembly, which includes a transparent observation window, a sealing gasket, and an air extraction interface; A vacuum generating component, connected to the vacuum chamber assembly, is used to evacuate the vacuum chamber assembly and form a preset pressure difference, and includes a valve for isolating pressure rise measurement; A pressure acquisition component is connected to the vacuum chamber assembly and is used to acquire the pressure inside the vacuum chamber to form a pressure sequence; A visual acquisition component is connected to the vacuum chamber assembly and is used to acquire video sequences of the weld area through the transparent observation window; The processor is configured to execute the weld inspection method based on the fusion of machine vision and vacuum inspection as described in any one of claims 1 to 8, to output inspection results, and the memory is configured to store the inspection results, as well as the working parameters generated during the inspection process and the process data collected during the inspection.

10. A weld inspection device based on vacuum inspection, characterized in that, It includes the weld inspection system based on the fusion of machine vision and vacuum inspection as described in claim 9, the system further includes a communication module connected to a cloud network; the device further includes a server deployed in the cloud and a management platform for accessing the server, the server actively acquires or passively receives the stored inspection results, as well as the working parameters and process data corresponding to the stored inspection results, transmitted by the system through the communication module at a preset time frequency.

Images