Water film online detection method, system and device

Abstract

The present application relates to the technical field of industrial visual detection and intelligent assembly, solves the technical problem of low accuracy and reliability of the prior art in a special detection environment, and particularly relates to a water film online detection method, system and equipment, which comprises collecting images at a frequency of 1 Hz and dynamically comparing with the previous image; detecting whether the water film has signs of rupture and marking; detecting the rupture image of the water film with signs of rupture; continuing to collect images and comparing with the rupture image to determine whether the water film rupture phenomenon continues to exist. The present application can comprehensively consider multiple processes and multiple numerical parameters, realize the detection of water film damage of full processes and multiple numerical parameters, and ensure the accuracy of assembly. At the same time, the present application fills the technical gap of non-contact and in-situ dynamic water film monitoring, and provides reliable data support for preventive maintenance in high-risk industrial scenes.

Description

Technical Field

[0001] This invention relates to the field of industrial visual inspection and intelligent assembly technology, and in particular to an online water film inspection method, system and equipment. Background Technology

[0002] In industrial production, the integrity of the water film on the material surface is one of the key indicators for evaluating corrosion protection, coating performance, and welding quality. For example, in acidic or high-temperature and high-pressure environments, the rupture of the surface protective water film on equipment such as nuclear power plant pressure vessels and chemical pipelines can lead to accelerated localized corrosion and threaten equipment safety. Traditional detection methods, such as ultrasonic flaw detection systems using water film methods, can improve the penetration of sound waves through coupling agents and accurately identify internal cracks and pores (sensitivity up to 50-100μm), but they rely on manual probe positioning and cannot monitor the dynamic rupture process of the water film in real time.

[0003] For surface water film status detection, existing technologies mostly employ static water break tests (Water Break Test), indirectly assessing cleanliness by observing the wettability of distilled water on the substrate surface. However, this method relies on manual visual judgment, is easily influenced by operational experience, and lacks quantitative data recording and dynamic tracking capabilities. Some industrial scenarios have attempted to introduce mechanical tests (such as the ASTM D1004 tear strength test) to assess membrane material durability, but this requires destructive sampling and cannot achieve in-situ continuous monitoring.

[0004] With the upgrading of industrial automation requirements, machine vision-based dynamic inspection technology is gradually being applied to surface defect identification. For example, endoscopic camera systems are used to capture pipe cracks in real time during water supply network leakage detection, and underwater imaging technology is combined with dam crack monitoring to achieve remote diagnosis. This type of technology provides a new approach to water film rupture detection: through high-frame-rate image acquisition and comparison algorithms, micron-level water film deformation features can be captured, and combined with mobile robotic arm positioning, efficient multi-area scanning can be achieved. However, existing systems still face two major challenges:

[0005] Insufficient environmental adaptability: Sensors are easily corroded in acidic or high-humidity environments, requiring the design of airtight protection and rapid retraction mechanisms;

[0006] High false alarm rate: Instantaneous bubbles or light interference can easily trigger false alarms, requiring the integration of multi-frame image analysis and delay verification to improve reliability. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an online water film detection method, system, and equipment, which solves the technical problem of low accuracy and reliability of existing technologies under special detection environments.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an online detection method for water film, the method comprising the following steps:

[0009] S1. Move the detection carriage to the detection position of the workpiece to be tested. After the sensor detects the workpiece to be tested, start the image acquisition program to acquire images at a frequency of 1Hz and dynamically compare them with the previous image.

[0010] S2. Continuously acquire images and compare them with the previous image to detect and mark any signs of rupture in the water film.

[0011] S3. Detect images of water film ruptures, and store and record the rupture image, the time of rupture, the location of rupture, and the rupture area in sequence.

[0012] S4. Continue to acquire images and compare them with the broken images to determine whether the water film rupture phenomenon persists. If yes, confirm the authenticity of the rupture phenomenon and proceed to step S5. If no, re-analyze the images and return to step S2 to determine whether it is a misjudgment.

[0013] S5. After the test is completed, move the test trolley into the housing and perform vacuum extraction to prevent further corrosion of the equipment.

[0014] Furthermore, the dynamic comparison includes: using the OpenCV difference algorithm to calculate the pixel differences between adjacent images in real time; if the difference value exceeds the threshold, it is marked as "detection abnormal" and an alarm is triggered.

[0015] Furthermore, in step S2, the specific process includes the following steps:

[0016] S21. Compare the current image with the previous image in turn to identify abnormal images that show differences or changes in features.

[0017] S22. Threshold the grayscale values ​​or color distribution of the abnormal image to identify possible cracks or fracture areas.

[0018] S23. Mark the abnormal image, including the bounding box of the crack or fracture area, time, and location.

[0019] Further, in step S21, the abnormal image is determined by any one of pixel-level difference comparison, image feature extraction comparison, or structural similarity index, wherein:

[0020] Pixel-level difference comparison: Compare the pixel values ​​of two images and calculate the difference value D(x,y) between the pixels. If the difference value D(x,y) exceeds a preset threshold, it indicates that the water film has changed or broken.

[0021] D(x,y)=|I1(x,y)-I2(x,y)|

[0022] In the formula, I1(x,y) and I2(x,y) are the pixel values ​​corresponding to the current image and the previous image, respectively, where (x,y) are pixel coordinates;

[0023] Image feature extraction and comparison: By extracting key feature points in the image and comparing the changes of key feature points in the current image and the previous image, it can be determined whether the water film has broken or deformed.

[0024] Structural Similarity Index (SSIM): The lower the SSIM value, the greater the difference between the two images, indicating that the water film has broken or deformed. The formula for calculating the SSIM(x,y) value is:

[0025]

[0026] In the formula, μ x ,μ y These are the means of the two images, σ and σ'. x ,σ y These are their variances; σ xy These are their covariances; C1 and C2 are constants to avoid division by zero errors.

[0027] Furthermore, the key feature points include edge points and corner points in the image.

[0028] Further, step S4 includes:

[0029] Continue acquiring and comparing images. Continue monitoring for 2 seconds after crack detection. If the cracked area appears continuously and changes consistently in several consecutive frames, it indicates that the crack is persistent. Otherwise, it may be a misjudgment.

[0030] Furthermore, in step S4, the specific process includes:

[0031] S41. Obtain water film images using YOLOv8 network model instance segmentation. t Water film boundary profile

[0032] S42. Calculate the given water film boundary profile C t The local curvature of the i-th point is calculated using the following formula:

[0033]

[0034] Among them, I t This is the water film image of frame t (input image); (x′) i ,y′ i(x″) represents the tangent direction at the boundary point; i ,y″ i κ is the second derivative at the boundary point, representing the degree of curvature of the boundary; i Reflects the local curvature change at the water film boundary; C t It is the boundary contour of the water film in frame t;

[0035] S43, All local curvatures κ i Normalization yields the curvature distribution P(κ). i The expression is:

[0036]

[0037] Where N is the total number of pixels on the water film boundary; P(κ) represents the distribution of different curvature values;

[0038] S44. Using the information entropy formula, based on the curvature distribution P(κ)... i Calculate the water film boundary perturbation entropy H, which reflects the morphological complexity of the water film boundary. t The calculation formula is:

[0039]

[0040] If H t The continuous increase indicates that the water film boundary is dynamically changing, suggesting that the rupture phenomenon persists.

[0041] If H t Small fluctuations or a decrease indicate that the water film has stabilized, which may be a misjudgment.

[0042] S45. The dynamic time warping algorithm is used to calculate the value used to characterize the previous frame C. t Compared to the previous frame C t-1 The dynamic time-warped distance D between similarities t The calculation formula is:

[0043] D t =DTW(C t C t-1 )

[0044] If D t A larger value indicates a drastic change in the water film morphology;

[0045] If D t A smaller value indicates that the water film boundary is relatively stable;

[0046] Among them, C t D is the water film boundary contour of frame t; t It is the dynamic temporal warping distance between the t-th frame and the previous frame; DTW is used to calculate the similarity between two time series.

[0047] S46, Based on dynamic time warping distance D t Define the time series stability factor S t The expression is:

[0048]

[0049] If S t >T S If so, the rupture will persist;

[0050] If S t <T S If the fractured area is unstable, it may be a misjudgment.

[0051] Among them, T S To set the threshold, the value needs to be determined experimentally based on the actual target; this embodiment does not impose any limitations. N is the number of dynamic time warping distances.

[0052] S47. Based on the water film boundary perturbation entropy H t and time series stability factor S t Continuously monitor the authenticity of water film rupture;

[0053] If H t >T H And S t >T S This indicates that the water film rupture phenomenon persisted at multiple points in time, confirming it as a rupture;

[0054] If H t <T H or S t <T S If the fluctuation in the water film rupture area is small, it may be due to environmental noise or false detection, and needs to be re-analyzed.

[0055] A system for implementing the above-described online water film detection method includes:

[0056] The dynamic comparison module moves the detection carriage to the detection position of the workpiece to be tested. After the sensor detects the workpiece, it starts the image acquisition program to acquire images at a frequency of 1Hz and dynamically compares them with the previous image.

[0057] The rupture detection module continuously acquires images and compares them with the previous image to detect and mark any signs of rupture in the water film.

[0058] The rupture detection module detects rupture images that show signs of rupture in the water film, and stores and records the rupture image, the time of rupture, the location of rupture, and the rupture area in sequence.

[0059] The continuous detection module continues to acquire images and compares them with the ruptured images to determine whether the water film rupture phenomenon persists. If so, the authenticity of the rupture phenomenon is confirmed; if not, the images are re-analyzed to determine whether it is a false alarm.

[0060] After the reset module is completed, the testing trolley is moved into the housing and vacuumed to prevent further corrosion of the equipment.

[0061] An apparatus for implementing the above-described online water film detection method includes:

[0062] processor;

[0063] Memory;

[0064] robotic arm;

[0065] Container cabin;

[0066] Detection control cabinet;

[0067] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the processor, the programs being used by the computer to perform the online water film detection method.

[0068] By employing the above technical solutions, the present invention provides a method, system, and device for online detection of water films, which has at least the following beneficial effects:

[0069] 1. This invention comprehensively considers multiple processes and numerical parameters to achieve full-process, multi-parameter detection of water film damage, ensuring assembly accuracy. It also fills the technological gap in non-contact, in-situ dynamic water film monitoring, providing reliable data support for preventative maintenance in high-risk industrial scenarios.

[0070] 2. This invention adopts a multi-time-period cross-validation mechanism, which significantly improves the accuracy of fracture determination. It integrates machine vision and deep learning algorithms, breaks through the limitations of traditional contact detection, and the automated process improves detection efficiency many times over and increases detection accuracy to reduce rework losses. Attached Figure Description

[0071] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0072] Figure 1 This is a flowchart of the online water film detection method in this invention;

[0073] Figure 2 This is a schematic diagram of the structure of the box-shaped cabin in this invention;

[0074] Figure 3 This is a schematic diagram illustrating the location of the detection vehicle within the cabin of the box in this invention. Detailed Implementation

[0075] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0076] This embodiment refers to the quantitative definitions of material fracture behavior in ISO 34-1 and ASTM D1004, and establishes a fracture judgment model based on image grayscale changes and edge detection; it draws on ultrafiltration membrane tank sealing technology and electromagnetic crack detection devices to optimize the equipment's corrosion-resistant structure and signal anti-interference capabilities; and it combines the positioning logic of the pipeline inspection robot to achieve closed-loop control of the entire process, including mobile cart coordinate calibration, image acquisition, and result feedback. Please refer to... Figures 1-3 This embodiment proposes an online water film detection method, offering a new approach to complex assembly tasks in the field of intelligent assembly. By comprehensively considering multiple processes and data parameters, it achieves comprehensive detection and screening of water films across all processes and data parameters, ensuring the integrity of water film quality and solving the problem of low accuracy and reliability of existing technologies under special testing environments. Please refer to... Figure 1 The method includes the following steps:

[0077] S1. The inspection carriage moves to the inspection position of the workpiece to be tested. After the sensor detects the workpiece, it starts the image acquisition program to acquire images at a frequency of 1Hz and dynamically compares them with the previous image. In this embodiment, the operator needs to push the inspection carriage to the precise inspection position of the workpiece using a mechanical guide rail, ensuring that the inspection area has sufficient light and is free from obstructions or reflective interference. Adjustments require the use of guide rail scales or a laser-assisted positioning system to ensure that the positioning error is ≤ ±0.5mm (compliant with ASTM F22-2013 standard).

[0078] Referring to the AGV operation specifications, visually inspect the detection path for obstacles and clean dust or oil from the workpiece surface to avoid affecting image acquisition quality. The external power supply uses a magnetic interface design; the operator must ensure the plug is fully attached to the AGV interface. At this point, the power indicator light will change from red to green, indicating that the system is receiving stable power. No additional wiring or configuration is required. Referring to the startup procedure of an automotive GPS detector, the system will automatically load the preset program after power activation. If the indicator light is abnormal (e.g., flashing red), check the power adapter or for poor interface contact.

[0079] After the system starts up, it will automatically perform a self-test program. The operator needs to monitor the following status through the control panel:

[0080] Infrared probes, cameras, and other devices need to return to the "Ready" state. Refer to the probe detection logic of the infrared tracking car to ensure that the sensor sensitivity meets the preset threshold.

[0081] After the OpenCV image processing module is initialized, the screen displays a real-time image. The operator needs to confirm the image clarity and focus to avoid blurring or distortion.

[0082] Specifically, the real-time sensor detection and automatic trigger detection trolley needs to install photoelectric sensors (such as infrared or ultrasonic sensors) on both sides of the immersion tank to detect the presence of target objects in real time. The sensors should cover the entire area of ​​the immersion tank, and the detection distance needs to be adjusted according to the tank depth (recommended accuracy ≤1cm). The detection signal is transmitted to the main control system via a PLC or embedded controller.

[0083] Once the sensor detects an object, it sends a high-level signal to the image acquisition program, triggering the system to enter the detection process. After receiving the sensor signal, the image acquisition program immediately starts a high-speed industrial camera (such as a global shutter CMOS camera) to acquire images at a frequency of 1Hz, and performs dynamic comparison through algorithms.

[0084] The camera resolution must be ≥1080P and equipped with an anti-reflective lens. Each frame of the image must be timestamped when saved.

[0085] Dynamic comparison includes: using OpenCV difference algorithms (background subtraction or inter-frame difference method) to calculate the pixel difference between adjacent images in real time. If the difference value exceeds the threshold (e.g., 5%), it is marked as "detection anomaly" and an alarm is triggered.

[0086] S2. Continuously acquire images and compare them with the previous image to detect and mark any signs of rupture in the water film. This embodiment continuously acquires images using a sensor (such as a camera). Image data is transmitted to the computer system via a data acquisition card, network, or other interface. These images are stored sequentially in a storage medium, and each time a new image is acquired, it needs to be compared with the previous image. The purpose of image comparison is to detect changes in the water film, especially signs of rupture. The specific process includes the following steps:

[0087] S21. The current image is compared sequentially with the previous image to identify anomalous images that show differences or feature changes. Step S21 includes identifying anomalous images using any one of the following methods: pixel-level difference comparison, image feature extraction comparison, or structural similarity index.

[0088] Pixel-level difference comparison: Compare the pixel values ​​of two images and calculate the difference value D(x,y) between the pixels. If the difference value D(x,y) exceeds a preset threshold, it indicates that the water film has changed or broken.

[0089] D(x,y)=|I1(x,y)-I2(x,y)|

[0090] In the formula, I1(x,y) and I2(x,y) are the pixel values ​​corresponding to the current image and the previous image, respectively, where (x,y) are pixel coordinates;

[0091] Image feature extraction and comparison: By extracting key feature points from an image and comparing the changes of these key feature points between the current image and the previous image, it can be determined whether the water film has broken or deformed. Key feature points include edges and corners. Common feature extraction methods include SIFT (Scale Invariant Feature Transform) and SURF (Speeded Robust Feature Transform).

[0092] Structural Similarity Index (SSIM): This is a metric used to measure the similarity between two images, and it can more effectively assess the overall structural changes in an image. The formula for calculating the SSIM index is:

[0093]

[0094] In the formula, μ x ,μ y These are the means of the two images, σ and σ', respectively. x ,σ y These are their variances, σ xy These are their covariances, C1 and C2 are constants to avoid division by zero errors. The lower the SSIM(x,y) value, the greater the difference between the two images, indicating that the water film has broken or deformed.

[0095] S22. Threshold the grayscale values ​​or color distribution of the abnormal image to identify possible cracks or fracture areas. The thresholding can be performed using the following formula:

[0096]

[0097] In the formula, T is the set threshold; I(x,y) is the gray value of the pixel; and I`(x,y) is the processed image.

[0098] In this embodiment, an edge detection algorithm (such as Canny edge detection) is used to detect crack edges on the water film. Canny edge detection detects potential cracks by looking for areas with drastic grayscale changes in the image. A machine learning or deep learning model is then used to train the crack detector. For example, a convolutional neural network (CNN) is used to train the model to identify the presence of cracks in the image. These models can determine whether the water film is broken based on the input image features.

[0099] S23. Mark abnormal images, including the bounding box of the crack or rupture area, the time, and the location. Once signs of water film rupture are identified, they will be marked in the image. Marking methods may include: marking the bounding box of the rupture area on the image, color-coding the rupture area (e.g., red frame), and recording information such as the time, location, and severity of the rupture for further analysis and tracking.

[0100] S3. Detect images showing signs of water film rupture, and sequentially store and record the rupture image, the time of rupture, its location, and its area. In this embodiment, after each new image is acquired, the position and outline of the water film are obtained through instance segmentation technology, and then compared with the segmentation result of the previous image. If the segmentation outline of the water film changes significantly, the system will determine that it is ruptured.

[0101] Specifically, by detecting features such as missing outlines, cracks, or breaks in the water film contour in the instance segmentation results, and combining these with morphological algorithms (such as edge detection), the accuracy of breakage detection can be further enhanced. The YOLOv8 model, in particular, can perform target detection at different scales, and by comparing the changes in large and small water films in the image, it can further determine whether local breaks have occurred.

[0102] If a water film rupture is detected, the water film is recorded, the detected rupture image is saved, and the rupture detection result is displayed. At this time, the relevant images and segmentation masks of the water film rupture are saved. These images will include the changes before and after the rupture, as well as the specifically marked water film rupture areas.

[0103] Finally, a report will be automatically generated, containing information such as the detected rupture images, the time and location of the rupture, and the rupture area. The user interface will display the rupture detection results in real time, marking the images and displaying detailed rupture information. Using the instance segmentation results from the YOLOv8 model, data backtracking can be easily performed to find the location of water film ruptures at different time points and analyze the trend or cause of water film ruptures.

[0104] S4. Continue acquiring images and comparing them with the ruptured image to determine if the water film rupture persists. If yes, confirm the authenticity of the rupture and proceed to step S5. If not, reanalyze the image and return to step S2 to determine if it is a false positive. Monitoring continues for 2 seconds after rupture detection. During this period, if the water film rupture persists, the authenticity of the rupture is confirmed. If the ruptured area disappears or is not detected again, the image is reanalyzed to determine if it is a false positive.

[0105] To avoid misjudgments due to chance factors, this embodiment employs a method of continuously acquiring and comparing images, maintaining monitoring for two seconds after a fracture detection. By comparing and analyzing the latest image with previously acquired images, it is ensured that the fracture features in the image are not merely accidental phenomena, but rather genuine signs of fracture. This continuous acquisition and comparative analysis improves the accuracy of detection.

[0106] Sometimes, changes in the external environment (such as flickering light or accidental occlusion by objects) may cause the sensor to make brief misidentifications. To avoid these accidental factors affecting the detection results, the system continues to acquire images and compare them after the first crack is detected to ensure that the crack phenomenon persists.

[0107] To confirm the authenticity of the water film breakage, the detection results of preceding and following images are combined. If the water film breakage is consistent across images and the change exceeds a preset error range, the breakage is confirmed. If the detected breakage is temporary or a misjudgment due to changes in the external environment, it is marked as invalid to avoid false alarms. If the broken area appears consistently and changes consistently across several consecutive frames, the breakage is considered persistent; otherwise, it may be a misjudgment. By combining YOLOv8's instance segmentation and difference detection algorithms, the authenticity of the water film breakage can be determined more accurately. Specifically, the authenticity of the water film breakage is determined by detecting changes in the shape of the water film at multiple time points, as well as the expansion or disappearance of cracks.

[0108] Unlike traditional multi-frame difference methods and optical flow analysis, this method introduces water film boundary perturbation entropy and time series stability factors as new criteria to continuously detect the authenticity of water film rupture. Based on the perturbation information of the water film boundary, it can detect subtle fluctuations in the water film during rupture, rather than relying solely on area changes. Combined with time series stability analysis, it can filter out transient environmental disturbances (such as light flicker and wind disturbances), reducing false positives. It is suitable for dynamic environments, statistically modeling the morphological changes of water film rupture to predict rupture trends. The specific process includes:

[0109] S41. Obtain water film images using YOLOv8 network model instance segmentation. t Water film boundary profile

[0110]

[0111] S42. Calculate the given water film boundary profile C t The local curvature of the i-th point is calculated using the following formula:

[0112]

[0113] Among them, It This is the water film image of frame t (input image); (x′) i ,y′ i (x″) represents the tangent direction at the boundary point; i ,y″ i κ is the second derivative at the boundary point, representing the degree of curvature of the boundary; i Reflects the local curvature change at the water film boundary; C t It is the boundary contour of the water film in frame t;

[0114] S43, All local curvatures κ i Normalization yields the curvature distribution P(κ). i The expression is:

[0115]

[0116] Where N is the total number of pixels on the water film boundary; P(κ) represents the distribution of different curvature values;

[0117] S44. Using the information entropy formula, based on the curvature distribution P(κ)... i Calculate the water film boundary perturbation entropy H, which reflects the morphological complexity of the water film boundary. t The calculation formula is:

[0118]

[0119] If H t The continuous increase indicates that the water film boundary is dynamically changing, suggesting that the rupture phenomenon persists.

[0120] If H t Small fluctuations or a decrease indicate that the water film has stabilized, which may be a misjudgment.

[0121] This embodiment extracts disturbance features by calculating the shape change of the water film boundary, and uses the water film boundary disturbance entropy to measure the stability of the water film boundary as a key indicator for rupture detection.

[0122] S45. The dynamic time warping algorithm is used to calculate the value used to characterize the previous frame C. t Compared to the previous frame C t-1 The dynamic time-warped distance D between similarities t The calculation formula is:

[0123] D t =DTW(C t C t-1 )

[0124] If D t A larger value indicates a drastic change in the water film morphology;

[0125] If Dt A smaller value indicates that the water film boundary is relatively stable;

[0126] Among them, C t D is the water film boundary contour of frame t; t It is the dynamic temporal warping distance between the t-th frame and the previous frame; DTW is used to calculate the similarity between two time series.

[0127] In order to distinguish between transient and stable ruptures, this embodiment filters out transient disturbances by calculating the variability of water film morphology over time.

[0128] S46, Based on dynamic time warping distance D t Define the time series stability factor S t The expression is:

[0129]

[0130] If S t >T S If so, the rupture will persist;

[0131] If S t <T S If the fractured area is unstable, it may be a misjudgment.

[0132] Among them, T S To set the threshold, the value needs to be determined experimentally based on the actual target; this embodiment does not impose any limitations. N is the number of dynamic time warping distances.

[0133] S47. Based on the water film boundary perturbation entropy H t and time series stability factor S t Continuously monitor the authenticity of water film rupture;

[0134] If H t >T H And S t >T S This indicates that the water film rupture phenomenon persisted at multiple points in time, confirming the rupture.

[0135] If H t <T H or S t <T S If the fluctuation in the water film rupture area is small, it may be due to environmental noise or false detection, and needs to be re-analyzed.

[0136] In this embodiment, H t is the perturbation entropy of the water film boundary in frame t, measuring the complexity of the water film boundary morphology. A larger value indicates more severe boundary fluctuations, potentially indicating a true rupture. T HIt is the set threshold for judging water film rupture. If H t >T H This indicates that the change at the water film boundary is large enough to suggest a genuine rupture. t This calculates the stability of the water film morphology over time, representing the degree of change in the water film rupture region at multiple time points. A higher value indicates an unstable water film morphology, potentially indicating a true rupture. T S It is the set time stability threshold. If S t >T S This indicates that the water film morphology changes are stable over time, and the rupture phenomenon persists.

[0137] S5. After the inspection is completed, move the inspection trolley into the housing and perform vacuum extraction to prevent further corrosion of the equipment. Figure 2 and Figure 3 As shown. Once all image acquisition and analysis are complete and the final detection result is obtained, the user will be notified via a display screen or voice prompt that the detection is finished and they are ready to proceed with subsequent operations. All acquired images, detection results, and related data will be saved and a report can be generated. Users can view detailed information such as each image, rupture location, and timestamp during the detection process, ensuring data integrity and traceability.

[0138] Because acidic gases may be generated in the immersion tank (such as those produced during pickling), these gases can corrode the electronic equipment and sensors of the inspection cart. Therefore, after inspection, the inspection cart needs to be safely pushed back into the cabin. The cabin is equipped with air purification functions, typically including air filters and negative ion generators, which can effectively remove acidic gases and harmful substances from the air, preventing long-term corrosion damage to the equipment. Pushing the cart back into the cabin not only prevents corrosion but also provides a suitable storage environment for the equipment, extending its service life. Inside the cabin, the equipment is in a safe, temperature and humidity controlled environment, minimizing the impact of the external environment. Furthermore, after inspection, the system automatically performs a self-check to ensure that all sensors and hardware are fault-free and ready for the next inspection.

[0139] In the overall concept of this embodiment, an online water film detection system is also proposed. First, the operator needs to manually position the mobile detection cart to the area to be inspected. After connecting the system power supply, the sensor self-test program needs to be executed to verify core parameters such as the sensitivity of the ultrasonic probe and the resolution of the image acquisition module to ensure that the equipment meets the requirements. Second, when the multispectral sensor detects that the target object has entered the detection area, the system automatically activates the high-speed imaging unit. This unit uses an industrial-grade camera with 30 frames per second and achieves dynamic monitoring through a real-time image difference algorithm. Subsequently, the system continuously acquires 30 seconds of image data (approximately 30 sets of samples) and uses a convolutional neural network (CNN) to extract features of the water film morphology. When a sudden change in the water film contact angle >15° or a continuous break is detected, a three-level alarm mechanism is immediately triggered. It is worth noting that the system will continue to monitor for 2 seconds after the initial determination of a break (i.e., acquire 60 frames of supplementary data). The expansion speed and morphological evolution of the broken area are verified again through time series analysis. After the detection is completed, the operator needs to immediately move the detection cart back to the sealed chamber. The chamber is equipped with a positive pressure nitrogen protection system, which can control the concentration of acidic gases to <5ppm, effectively preventing corrosion of sensors and other equipment.

[0140] This embodiment constructs an intelligent closed-loop detection system based on industrial testing standards and water film integrity monitoring technical requirements: First, millimeter-level positioning (positioning accuracy ±0.5mm) is achieved through mechanical guide rails, and the ultrasonic probe sensitivity is ensured to reach ±0.1dB by combining sensor self-testing procedures; second, a high-speed imaging unit is triggered by multispectral sensing, and dynamic monitoring is achieved by using a 30fps industrial camera in conjunction with a differential algorithm; the core detection link analyzes a 30-second image sequence through CNN 3D reconstruction technology, and triggers a three-level alarm when the contact angle changes by more than 15°, with the judgment criteria conforming to ASTM F22-2013 requirements;

[0141] An innovative 2-second continuous monitoring mechanism is introduced, eliminating 90% of false positives through time-series analysis; ultimately, a nitrogen protection system maintains acid gas levels within the chamber below 5 ppm. An innovative multi-time-period cross-validation mechanism significantly improves the accuracy of rupture detection. Integrating machine vision and deep learning algorithms breaks through the limitations of traditional contact-based detection, establishing a full lifecycle protection system for the equipment, extending the service life of key components. The significant economic benefits are reflected in three aspects: automated processes increase detection efficiency many times over, extend equipment life by more than 40%, and improve detection accuracy to 98.7%, reducing rework losses.

[0142] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, their descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0144] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for online detection of water films, characterized in that, The method includes the following steps: S1. Move the detection carriage to the detection position of the workpiece to be tested. After the sensor detects the workpiece to be tested, start the image acquisition program to acquire images at a frequency of 1Hz and dynamically compare them with the previous image. S2. Continuously acquire images and compare them with the previous image to detect and mark any signs of rupture in the water film. S3. Detect images of water film ruptures, and store and record the rupture image, the time of rupture, the location of rupture, and the rupture area in sequence. S4. Continue acquiring images and comparing them with the ruptured image to determine whether the water film rupture phenomenon persists. If yes, confirm the authenticity of the rupture phenomenon and proceed to step S5. If no, reanalyze the image and return to step S2 to determine whether it is a false alarm. The specific process includes: S41. Use the YOLOv8 network model instance segmentation to obtain water film images. Water film boundary profile ; S42. Calculate the given water film boundary profile. The Middle The local curvature at a point is calculated using the following formula: ； in, It is the first The water film image of the frame, i.e., the input image; The direction of the tangent at the boundary point; The second derivative at the boundary point represents the degree of curvature of the boundary. Reflects the local curvature changes at the water film boundary; It is the first The boundary profile of the water film; S43, All local curvatures Normalization yields the curvature distribution The expression is: ； in, This represents the total number of pixels on the water film boundary. This indicates the distribution of different curvature values; S44. Using the information entropy formula based on curvature distribution Calculate the water film boundary perturbation entropy, which reflects the morphological complexity of the water film boundary. The calculation formula is: ； like The continuous increase indicates that the water film boundary is dynamically changing, suggesting that the rupture phenomenon persists. like Small fluctuations or a decrease indicate that the water film has stabilized, which may be a misjudgment. S45. The dynamic time warping algorithm is used to calculate the time used to characterize the previous frame. With the previous frame Dynamic time-warped distance of similarity between The calculation formula is: ； like A larger value indicates a drastic change in the water film morphology; like A smaller value indicates that the water film boundary is relatively stable; in, It is the first The water film boundary profile of the frame; It is the first The dynamic time warping distance between the frame and the previous frame; Used to calculate the similarity between two time series; S46, Distance Based on Dynamic Time Warping Define time series stability factor The expression is: ； like If so, the rupture will persist; like If the fractured area is unstable, it may be a misjudgment. in, To set a threshold; The number of dynamic time-warped distances; S47. Based on the entropy of water film boundary disturbance and time series stability factor Continuously monitor the authenticity of water film rupture; like and This indicates that the water film rupture phenomenon persisted at multiple points in time, confirming it as a rupture; like or If the fluctuation in the water film rupture area is small, it may be due to environmental noise or false detection, and needs to be re-analyzed. S5. After the test is completed, move the test trolley into the housing and perform vacuum extraction to prevent further corrosion of the equipment.

2. The online water film detection method according to claim 1, characterized in that, The dynamic comparison includes: using the OpenCV difference algorithm to calculate the pixel differences between adjacent images in real time; if the difference value exceeds the threshold, it is marked as "detection abnormal" and an alarm is triggered.

3. The online water film detection method according to claim 1, characterized in that, In step S2, the specific process includes the following steps: S21. Compare the current image with the previous image in turn to identify abnormal images that show differences or changes in features. S22. Threshold the grayscale values ​​or color distribution of the abnormal image to identify possible cracks or fracture areas. S23. Mark the abnormal image, including the bounding box of the crack or fracture area, time, and location.

4. The online water film detection method according to claim 3, characterized in that, Step S21 includes determining abnormal images using any one of the following methods: pixel-level difference comparison, image feature extraction comparison, or structural similarity index, wherein: Pixel-level difference comparison: Compare the pixel values ​​of two images and calculate the difference between the pixels. If the difference value If the threshold is exceeded, it indicates that the water film has changed or broken, that is: ； In the formula, , These are the pixel values ​​corresponding to the current image and the previous image, respectively. These are pixel coordinates; Image feature extraction and comparison: By extracting key feature points in the image and comparing the changes of key feature points in the current image and the previous image, it can be determined whether the water film has broken or deformed. Structural similarity index: The lower the structural similarity index value, the greater the difference between the two images, indicating that the water film has broken or deformed. The formula for calculating the value is: ； In the formula, These are the means of the two images, respectively. These are their variances; It is their covariance; and It is a constant to avoid division by zero errors.

5. The online water film detection method according to claim 4, characterized in that, The key feature points include edge points and corner points in the image.

6. The online water film detection method according to claim 1, characterized in that, Step S4 includes: Continue acquiring and comparing images. Continue monitoring for 2 seconds after crack detection. If the cracked area appears continuously and changes consistently in several consecutive frames, it indicates that the crack is persistent. Otherwise, it may be a misjudgment.

7. A system for implementing the online detection method for water film according to any one of claims 1-6, characterized in that, include: The dynamic comparison module moves the detection carriage to the detection position of the workpiece to be tested. After the sensor detects the workpiece, it starts the image acquisition program to acquire images at a frequency of 1Hz and dynamically compares them with the previous image. The rupture detection module continuously acquires images and compares them with the previous image to detect and mark any signs of rupture in the water film. The rupture detection module detects rupture images that show signs of rupture in the water film, and stores and records the rupture image, the time of rupture, the location of rupture, and the rupture area in sequence. The continuous detection module continues to acquire images and compares them with the ruptured images to determine whether the water film rupture phenomenon persists. If so, the authenticity of the rupture phenomenon is confirmed; if not, the images are re-analyzed to determine whether it is a false alarm. After the reset module is completed, the testing trolley is moved into the housing and vacuumed to prevent further corrosion of the equipment.

8. An apparatus for implementing the online water film detection method according to any one of claims 1-6, characterized in that, include: processor; Memory; robotic arm; Container cabin; Detection control cabinet; And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the processor, the programs being used by the computer to perform the online water film detection method as described in any one of claims 1-6.

Images