Corrugated pipe surface defect detection method and equipment thereof

Abstract

The invention discloses a corrugated pipe surface defect detection method and equipment thereof, and a multi-feature multi-verification closed-loop judgment system is constructed through four core technologies of tracer gas color detection, ultrasonic detection, image form analysis and directional blowing verification. Aiming at the defects of cracks and pores, by combining the color characteristics of the tracer gas and the linear distribution analysis of continuous acquisition points, the accurate distinguishing of defect types is realized, and the problem of traditional detection pain points is solved; for bubble defects, through ultrasonic multi-feature and SVM model preliminary positioning, image shape feature verification and directional blowing deformation feature confirmation, real bubbles and pseudo defects are thoroughly distinguished. Each link sets a threshold interval based on a large number of sample statistics, and the judgment reliability is guaranteed. Meanwhile, a corrugated pipe folding or unfolding state judgment mechanism is introduced, differentiated weights are configured in a targeted mode, the folding state is dominated by shape features to resist interference, the unfolding state is dominated by color and ultrasonic features to adapt to all working conditions, and the detection accuracy and applicability are improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and equipment for detecting defects on the surface of corrugated pipes. Background Technology

[0002] Corrugated pipes, as flexible piping components with a ring-shaped corrugated structure, are widely used in municipal water supply and drainage, gas transmission, chemical fluid transmission, and building heating and ventilation due to their excellent resistance to deformation, corrosion, and expansion and contraction. The integrity of the pipe surface directly determines the operational safety and service life of the piping system. If there are leakage defects such as cracks or pinholes on the pipe surface, it can easily lead to media leakage, sudden pressure drops, or even pipe rupture accidents. If there are bubble-like defects on the inner wall, stress concentration points will form due to bubble rupture during service, accelerating pipe wall aging and damage.

[0003] Given the application scenarios and hazards of defects in corrugated pipes, pipe surface defect detection has become a core aspect of production quality control and operation and maintenance. Currently, the mainstream detection technologies in the industry are mainly divided into three categories: First, leak detection technologies primarily rely on tracer gas detection and hydrostatic airtightness testing. Tracer gas detection determines leaks by introducing a specific gas into the pipe and observing gas leaks outside the pipe using sensors or visual inspection. Hydrostatic airtightness testing, on the other hand, determines the presence of leaks by maintaining pressure inside the pipe and monitoring pressure changes. However, traditional tracer gas detection can only determine if a leak has occurred; it cannot distinguish between cracks and pinholes. Furthermore, the repair solutions and risk levels for these two methods differ significantly, easily leading to a misallocation of maintenance resources. Additionally, the process involves filling with water, maintaining pressure, and draining the pipe, making the inspection of a single corrugated pipe time-consuming and unsuitable for the batch inspection needs of a production line.

[0004] Secondly, internal wall defect detection technologies primarily rely on ultrasonic testing and industrial vision inspection. Ultrasonic testing identifies internal wall anomalies by utilizing the reflection characteristics of ultrasonic signals; industrial vision inspection acquires images of the internal wall and determines the defect type based on image algorithms. However, impurities in the pipe wall, coupling agent bubbles, and environmental electromagnetic interference can generate ultrasonic reflection signals similar to bubbles, which traditional single-threshold judgment methods cannot effectively distinguish, resulting in a high false positive rate. Furthermore, relying solely on time-domain features for judgment cannot adapt to the characteristic variations of bubbles of different sizes and concentrations, and its ability to identify microbubbles is insufficient.

[0005] Third, most comprehensive testing equipment consists of combinations of single-function devices. When the airtightness test bench uses a uniform blowing pressure, excessive pressure can easily damage the wall of soft corrugated pipes, while insufficient pressure may prevent the formation of identifiable deformations in tiny bubbles in hard materials, leading to missed detection of bubbles. Relying solely on visual observation of deformation, without quantitative indicators such as geometric shape differences or deformation coefficients, results in a lack of authority and traceability in the judgment. Summary of the Invention

[0006] To achieve the above objectives, one technical solution adopted by the present invention is as follows: The method includes: stably introducing a tracer gas of a preset color into a bellows; collecting color image datasets of the outer surface of the bellows at preset intervals along the length of the bellows; determining whether the preset color of the tracer gas exists in each color image dataset using a color feature analysis strategy; and extracting color features of the colored areas on the bellows surface; emitting ultrasonic waves into the bellows and collecting the reflected ultrasonic signals; processing the ultrasonic signals using an ultrasonic signal processing strategy to initially determine whether there are suspected air bubbles on the bellows surface; and extracting the ultrasonic features and preliminary location of suspected air bubbles; and collecting data at the preliminary location. The first suspected bubble image data is processed using a suspected bubble image processing strategy to extract shape features and center point location. Based on the color image dataset, the corrugated pipe is determined to be in a folded or unfolded state. The preset state weight configurations of ultrasonic features, shape features, and color features of the colored area in the folded or unfolded state are retrieved. The preset state weight configurations are input into a weighted fusion algorithm to normalize the color features, ultrasonic features, and shape features of the colored area, obtaining a comprehensive defect vector of the suspected defect area on the corrugated pipe surface. If the comprehensive defect vector is greater than a preset threshold, it is determined that there is a defect on the corrugated pipe surface, and the defect type and the location of the suspected bubble are obtained.

[0007] Furthermore, the color feature analysis strategy includes: collecting preset color sample data of tracer gases at different concentrations and under different environments; statistically analyzing the grayscale value distribution range in the preset color region; defining the grayscale value distribution range as the preset color threshold interval; performing binarization processing on each image in the color image dataset; extracting the color mean of the preset color region; and defining the color mean of the preset color region as the color feature of the colored region; if the color mean of the preset color region falls within the preset color threshold interval, it is determined that the region contains the preset color of the tracer gas, and the image is defined as a leak sample image. The color feature analysis strategy also includes: if multiple consecutive sampling points at the same axial position detect color regions that meet the conditions, and the regions are linearly distributed, then it is further confirmed that the bellows has cracks and / or fine pore defects.

[0008] Furthermore, the ultrasonic signal processing strategy includes: collecting ultrasonic signal sample data of bubbles of different sizes and concentrations, as well as ultrasonic signal sample data of pipe wall impurities, coupling agent bubbles, and environmental interference; defining bubble characteristic threshold intervals; and training a binary classification model using a support vector machine with the ultrasonic signal sample data, and outputting the judgment result of bubble signal or non-bubble signal; preprocessing the collected suspected bubble ultrasonic signals, extracting the effective frequency bands from the suspected bubble ultrasonic signals, and performing signal benchmarking and normalization processing; calculating the theoretical propagation time of ultrasound from the probe to the pipe wall and back to the probe based on the corrugated pipe wall thickness and ultrasonic velocity, and extracting the signal segment with a preset propagation time as the effective analysis interval; and extracting the duty cycle, rise time, waveform entropy, and pulse interval mean of the signal peak and peak values ​​within the effective analysis interval. The time-domain features of the standard deviation are obtained; a fast Fourier transform is performed on the effective analysis interval to extract the frequency domain features with the dominant frequency and spectral width within the effective analysis interval; a short-time Fourier transform is used to generate a time-frequency diagram of the signal, and time-frequency domain features with the time-frequency centroid and time-frequency entropy within the effective analysis interval are extracted; a two-stage discrimination algorithm is used to compare the time-domain features, frequency domain features, and time-frequency domain features with the bubble feature threshold interval to determine whether the reflected suspected bubble ultrasonic signal is a bubble signal; the ultrasonic features of a suspected bubble include at least one of the bubble depth, equivalent diameter, and echo attenuation coefficient; the acquisition points of the detected bubble signal are recorded and numbered, and the axial position is obtained according to the acquisition interval; the circumferential angle of the bubble is obtained according to the time difference positioning algorithm; based on the propagation time difference of the ultrasonic signal, the radial distance between the bubble and the outer surface of the tube wall is calculated to form the preliminary three-dimensional positioning coordinates of the bubble signal in the axial, circumferential, and radial directions.

[0009] Furthermore, the secondary discrimination algorithm includes: initially marking the time-domain features within the effective analysis interval that fall within the bubble feature threshold interval and whose frequency domain dominant frequency energy ratio is greater than a preset ratio as suspected bubble signals; classifying the suspected bubble signals using an SVM model, and confirming them as bubble signals if the model confidence is greater than a preset confidence level and the sum of the time-frequency centroid and time-frequency entropy is greater than a preset threshold.

[0010] Furthermore, the suspected bubble image processing strategy includes: based on the preliminary three-dimensional position output by ultrasound detection, performing targeted cropping, contrast enhancement, and image normalization on the acquired first suspected bubble image to obtain the first suspected bubble region; processing the first suspected bubble region to extract the edge contour of the first suspected bubble; extracting at least one shape feature from the first suspected bubble contour, including area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance; calculating the mean value of the center point of the suspected bubble contour using the shape center method and the centroid method to obtain the center point coordinates; if the shape features of the first suspected bubble contour are greater than a preset number and all fall within the bubble feature threshold range, it is determined to be a bubble; if they are less than the preset number, it is directly determined to be no bubble, eliminating the misjudgment of the ultrasound signal; establishing the conversion relationship between the image coordinates and the actual physical coordinates of the first suspected bubble image through a camera calibration algorithm to obtain the axial and circumferential actual physical coordinates of the first suspected bubble; combining the radial preliminary three-dimensional positioning coordinates of ultrasound detection with the axial and circumferential actual physical coordinates obtained from image processing to output the three-dimensional determined position coordinates of the first suspected bubble.

[0011] Furthermore, when the corrugated pipe is in a folded state, the weight of shape features is greater than the sum of the weights of color features and ultrasonic features in the colored area; when the corrugated pipe is in an unfolded state, the weight of shape features is less than the sum of the weights of color features and ultrasonic features in the colored area.

[0012] Furthermore, the method also includes: blowing air directionally onto a suspected bubble at a determined location at a preset pressure, and acquiring image data of a second suspected bubble at the determined location; extracting shape features from the second suspected bubble image data using a bubble image processing strategy, and comparing the shape features with those of a first suspected bubble image data; if the shape features of the first suspected bubble image data are different from those of the second suspected bubble image data, then the suspected bubble is determined to be a real bubble; otherwise, it is not a real bubble.

[0013] Furthermore, the bubble image processing strategy includes: preprocessing, contrast enhancement, and image normalization of the acquired second suspected bubble image to obtain the second suspected bubble region; wherein, when acquiring the second suspected bubble image, the acquisition parameters are kept completely consistent with those of the first suspected bubble image; the second suspected bubble region is processed to extract the edge contour of the second suspected bubble; at least one shape feature of area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance is extracted from the second suspected bubble contour; wherein, the features extracted from the second suspected bubble contour are the same as those extracted from the first suspected bubble contour to ensure homology of comparison; at least one deformation feature of deformation coefficient, center displacement, and contour change rate is extracted from the second suspected bubble contour; the area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance of the second suspected bubble contour are calculated. If the absolute difference values ​​of shape features such as shape, aspect ratio, contour smoothness, or grayscale variance are greater than a preset number and all satisfy a preset threshold, they are initially marked as real bubble signals; if the absolute difference values ​​of shape features less than a preset number satisfy the preset threshold, they are judged as non-real bubbles; if the second suspected bubble contour meets any of the following conditions: deformation coefficient greater than a preset threshold, or center displacement greater than a preset threshold, or contour change rate greater than a preset threshold, it is confirmed as a real bubble; calculate the comprehensive confidence level of the real bubble; when the comprehensive confidence level of the real bubble is within a preset confidence level range, the blowing pressure value meets the preset blowing pressure value range; when the comprehensive confidence level of the real bubble is less than the preset confidence level range, adjust the blowing pressure value downward to avoid misjudgment due to insufficient airflow.

[0014] Another technical solution adopted by the present invention is: a corrugated pipe surface defect detection device, which is used in the corrugated pipe surface defect detection method of the above claims. The device includes: a mobile installation mechanism for supporting all components of the device; at least one set of installation platforms; each set of installation platforms is mounted opposite to each other on the mobile installation mechanism, the first installation platform of the set of installation platforms is mounted at the fixed part of the mobile installation mechanism, and the second installation platform of the set of installation platforms is mounted at the sliding part of the mobile installation mechanism; at least two airbags for inflating and fitting against the inner walls of both ends of the corrugated pipe; the two airbags are respectively mounted on the first installation platform and the second installation platform; an air inlet for slowly and steadily introducing a tracer gas of a preset color into the corrugated pipe; and a telescopic cylinder mounted on the mobile installation mechanism, with one end extending through the first installation platform and out of the airbag. A rotary motor is mounted on the telescopic end of the telescopic cylinder located inside the bellows; a gas nozzle is mounted on the rotating end of the rotary motor; the gas nozzle is used to directionally blow air onto suspected air bubbles when they are found on the bellows wall; a first industrial camera is mounted on the rotating end of the rotary motor and is arranged side by side with the gas nozzle along the bellows axis; a mounting bracket spans across the movable mounting mechanism; a second industrial wide-angle camera is hinged to the mounting bracket; an ultrasonic sensor is mounted on the end face of the air bladder located on the first mounting platform; the ultrasonic sensor is used to emit ultrasonic waves into the bellows and receive ultrasonic waves reflected back from the inner wall of the bellows; wherein, when both ends of the bellows are fixed to the equipment by the inflated air bladder, the movable mounting mechanism drives the second mounting platform to move, thereby stretching the folds of the bellows to facilitate the detection of defects on the bellows surface.

[0015] Furthermore, the mobile installation mechanism includes: a mechanism mounting plate; a rotary drive motor mounted on the mechanism mounting plate; a lead screw, one end of which is mounted on the rotating end of the rotary drive motor; a nut mounted on the lead screw; and an installation slider mounted on the nut; wherein, a first installation platform is mounted on the rotary drive motor, a second installation platform is mounted on the installation slider, and a telescopic cylinder is mounted on the rotary drive motor.

[0016] Compared with existing technologies, this invention has the following advantages: This invention integrates four core technologies—tracer gas color detection, ultrasonic detection, image morphology analysis, and directional blowing physical verification—to construct a closed-loop judgment system with multiple features and verifications. For cracks and pinhole defects, by combining tracer gas color feature extraction with linear distribution analysis of continuous acquisition points, it achieves an upgrade from simply identifying whether there is a leak to accurately distinguishing defect types, solving the pain point of traditional detection methods that cannot differentiate between cracks and pinholes. For bubble defects, it first uses multiple features of ultrasonic signals in the time domain, frequency domain, and time-frequency domain, along with an SVM model for preliminary localization, then verifies the image shape features, and finally confirms the defect through deformation features after directional blowing. This thoroughly distinguishes real bubbles from pseudo-defects such as scratches and stains from both static features and dynamic physical characteristics. Simultaneously, each detection step sets threshold ranges based on a large number of samples and adopts the confidence interval principle to ensure the reliability of feature judgment and effectively avoid missed judgments due to single samples or unreasonable thresholds.

[0017] Meanwhile, this invention introduces a mechanism for judging the folded / unfolded state of corrugated pipes. By preprocessing color images and calculating the regularity of corrugations, the pipe state is accurately identified, and differentiated weights are configured accordingly: in the folded state, the shape features with strong anti-interference ability are dominant, avoiding the obstruction and interference of wrinkles on color and ultrasonic signals; in the unfolded state, the color features and ultrasonic features are dominant, making full use of the highly reliable features under stable working conditions. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the method for detecting defects on the surface of a corrugated pipe according to the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of the corrugated pipe surface defect detection device of the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of the airbag, air inlet, telescopic cylinder, rotary motor, gas nozzle, and first industrial camera of the present invention.

[0021] Figure 4 This is a cross-sectional schematic diagram of the mobile installation mechanism of the present invention.

[0022] The components include: 1. Mobile installation mechanism; 101. Mechanism mounting plate; 102. Rotary drive motor; 103. Lead screw; 104. Nut; 105. Mounting slider; 2. Installation platform assembly; 201. First installation platform; 202. Second installation platform; 3. Airbag; 4. Air inlet; 5. Telescopic cylinder; 6. Rotary motor; 7. Gas nozzle; 8. First industrial camera; 9. Mounting bracket; 10. Second industrial wide-angle camera; 11. Ultrasonic sensor. Detailed Implementation

[0023] The technical solutions of the bellows surface defect detection method and equipment provided by the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1 As shown, a method for detecting defects on the surface of a corrugated pipe includes: stably introducing a tracer gas of a preset color into the corrugated pipe; collecting a dataset of color images of the outer surface of the corrugated pipe at preset intervals along the length of the corrugated pipe; determining whether the preset color of the tracer gas exists in each image dataset using a color feature analysis strategy; and extracting the color features of the colored areas on the surface of the corrugated pipe.

[0025] Specifically, the pre-selected color tracer gas refers to a gaseous medium with a specific visible color, stable chemical properties, non-reactive to the corrugated pipe material, non-toxic and harmless, and forming a clear visual contrast with air. For example, red and blue tracer gases must be clearly captured by the image acquisition device under different environments. When acquiring images along the length of the corrugated pipe, a pre-selected interval distance between two adjacent acquisition positions is required. This distance must be set according to the field of view of the image acquisition device and the minimum detectable defect size to avoid missing defects. The color image dataset is a collection of color images containing complete information about the outer surface of the corrugated pipe, continuously acquired by the image acquisition device at pre-selected intervals. Each image must cover the entire circumference of the outer corrugated pipe, and the image parameters must be uniform.

[0026] Furthermore, the color feature analysis strategy includes: collecting preset color sample data of tracer gases at different concentrations and under different environments, statistically analyzing the grayscale value distribution range in the preset color region, and defining the grayscale value distribution range as the preset color threshold interval.

[0027] Specifically, the different environments and concentrations of tracer gases indicate varying volumetric proportions of tracer gases in the air, ranging from low concentrations corresponding to minor leaks to high concentrations corresponding to severe leaks. External conditions during image acquisition, such as lighting, background color, and ambient temperature, are also considered, including common operating conditions like indoor ambient temperature, outdoor sunny days, outdoor cloudy days, and nighttime lighting. This ensures the threshold range covers all possible leak scenarios. The preset color sample data consists of image data containing the preset color of the tracer gas, collected under different concentrations and environmental conditions, used to establish a benchmark reference for color recognition. The preset color region is the set of pixels in the image that present the preset color of the tracer gas, and is the core object of color feature analysis. The grayscale value distribution range is the range of grayscale values ​​for all pixels within the preset color region of the preset color sample data, reflecting the brightness distribution characteristics of that color in the image. The preset color threshold range is a judgment range determined based on the grayscale value distribution range of the preset color sample data, and is the core standard for identifying the presence of tracer gas in an image.

[0028] The color feature analysis strategy further includes: performing binarization processing on each image in the color image dataset, extracting the color mean of a preset color region, and defining the color mean of the preset color region as the color feature of the colored region.

[0029] Specifically, binarization is the process of converting a color image into an image containing only two gray levels. This involves dividing the image pixels into foreground and background regions by setting a threshold. The foreground region corresponds to the suspected tracer gas area, and the background region corresponds to the corrugated pipe surface or the ambient background. The color mean is the arithmetic mean of the gray values ​​of all pixels in a preset color region. This value is used to quantify the color brightness characteristics of that region and serves as the core parameter for comparison with the threshold range. The color mean of the preset color region is defined as the color characteristic of the colored region.

[0030] The color feature analysis strategy further includes: if the average color value of a preset color region falls within a preset color threshold range, then it is determined that the region contains a preset color of tracer gas, and the image is defined as an image of a leak sample.

[0031] Specifically, the image of the leaked sample is determined to contain a tracer gas of a preset color after color feature analysis, and the location of the bellows corresponding to this image is the suspected defect location.

[0032] In this embodiment, the sealing condition at both ends of the bellows is first checked. The sealing joints are then connected and tightened to both ends of the bellows to ensure there is no leakage at the connection. Tracer gas is then continuously introduced for 5-10 minutes to stabilize the pressure inside the pipe, ensuring that any defects in the tracer gas have resulted in a stable leak. A preset interval distance is set, calculated by multiplying the coverage length of the image acquisition device's field of view along the bellows axis by 0.8. This ensures a 20% overlap between adjacent images to avoid missed detections. The images are then named according to the acquisition order to form a color image dataset.

[0033] Secondly, for the labeled area of ​​each set of preset color sample data, extract the grayscale values ​​of all pixels. Calculate the average grayscale value of each set of sample data. and standard deviation The formula is as follows: ,in, This represents the average grayscale value. For the first grayscale value of each pixel; This represents the total number of pixels in the preset color area. This is the sum of the grayscale values ​​of all pixels. ,in, The standard deviation of the grayscale values; For the first The difference between the grayscale value of each pixel and the average value; The sum of the squares of all differences; The total number of pixels in the preset color area.

[0034] Based on the average gray value of all sample data and standard deviation The preset color threshold range is determined using the 95% confidence interval principle, and the formula is as follows: , ,in, This is the lower limit of the preset color threshold range; Set the upper limit of the preset color threshold range; The arithmetic mean of the grayscale values ​​of all sample data; The arithmetic mean of the standard deviations of all sample data; This represents the error range corresponding to the 95% confidence interval.

[0035] Next, connected component analysis is performed on the binarized image. The minimum number of connected pixels is set to 50. Isolated noise points are removed, and connected regions with a pixel value of 255 are extracted, which are the preset color regions. Calculate the average color value of the preset color area. The formula is as follows: ,in, The average color value of the preset color area; For the first color within the preset color area grayscale value of each pixel; This is the sum of the grayscale values ​​of all pixels within the preset color area; This represents the total number of pixels within the preset color area.

[0036] Finally, the calculated color mean With preset color threshold range to Compare them.

[0037] like and If the preset color of the tracer gas is determined to be present in the area, the corresponding image is defined as the image of the leaked sample.

[0038] like or If no tracer gas is detected in the area, the corresponding image is considered a normal sample image.

[0039] In this embodiment, tracer gas visualization and grayscale value quantification analysis can accurately capture trace amounts of tracer gas leaking from cracks and pinholes. Even tiny pinholes from bellows production will cause the tracer gas to form colored areas, which can be reliably identified by comparing the average color value with a threshold range. Compared to traditional manual visual inspection, this reduces the false negative rate and effectively avoids subsequent pipeline leakage problems caused by undetected minute defects.

[0040] The method also includes: emitting ultrasonic waves into the corrugated pipe and collecting the reflected ultrasonic signals; processing the ultrasonic signals using an ultrasonic signal processing strategy to initially determine whether there are any suspected air bubbles on the surface of the corrugated pipe; and extracting the ultrasonic features and preliminary location of the suspected air bubbles.

[0041] Furthermore, the ultrasonic signal processing strategy includes: collecting ultrasonic signal sample data of bubbles of different sizes and concentrations, as well as ultrasonic signal sample data of tube wall impurities, coupling agent bubbles, and environmental interference; defining bubble feature threshold ranges; and using the ultrasonic signal sample data of bubbles and ultrasonic signal sample data to train a binary classification model using a support vector machine, and outputting the determination result of bubble signal or non-bubble signal.

[0042] Among these, different sized bubbles refer to bubbles with varying diameters, ranging from 0.1 mm to 5 mm in diameter, ensuring the sample data covers the actual possible bubble sizes. Different concentrations of bubbles refer to the varying number of bubbles per unit volume, ranging from 1 to 50 bubbles per cubic centimeter, covering both sparse and dense bubble scenarios. Pipe wall impurities refer to foreign matter such as dust, metal particles, and plastic debris adhering to the inner wall of the corrugated pipe, which can cause reflection interference to the ultrasonic signal. Coupling agent bubbles refer to bubbles mixed into the coupling agent applied to the probe and corrugated pipe surfaces during ultrasonic testing. The coupling agent is used to expel air and allow ultrasonic waves to effectively penetrate the pipe wall; bubbles within it can create false reflection signals. Environmental interference ultrasonic signals refer to ultrasonic frequency interference signals generated by factors such as motor vibration, sound waves, and electromagnetic radiation in the detection environment, which can affect the accuracy of target signal recognition. The bubble characteristic threshold range refers to the range of characteristic parameter values ​​obtained from the statistical analysis of bubble ultrasonic signal sample data, serving as a preliminary benchmark for identifying bubble signals.

[0043] Specifically, firstly, if the color image analysis does not detect the preset color of the tracer gas, the moving platform is activated, and the moving speed is set to 0.05 meters per second. The ultrasonic probe moves at a constant speed along the length of the corrugated pipe, with a sampling interval set to 5 centimeters, meaning an ultrasonic signal is collected every 5 centimeters. At each sampling point, four ultrasonic probes simultaneously transmit and receive ultrasonic signals. Each probe collects three signals and takes the average value to avoid interference from single signals. The collected signals are named and stored according to the sampling point number and probe number.

[0044] Secondly, bubble sample data were collected: bubble samples of different sizes were prepared with diameters of 0.1 mm, 0.5 mm, 1 mm, 2 mm, and 5 mm, generated using a microbubble generator, and the environment inside a corrugated pipe was simulated in a transparent container. Simultaneously, different bubble concentrations were set: 1, 5, 10, 20, and 50 bubbles per cubic centimeter. For each size and concentration, 100 sets of ultrasonic signals were collected, for a total of 2500 sets of bubble ultrasonic signal sample data (5 sizes × 5 concentrations × 100 sets = 2500 sets).

[0045] Finally, interference signal sample data were collected: For pipe wall impurity samples, metal particles, dust, and plastic debris of different sizes were attached to the inner wall of the corrugated pipe. 50 sets of ultrasonic signals were collected for each type of impurity, for a total of 3 types of impurities × 50 sets = 150 sets of signals. For coupling agent bubble samples, bubbles with diameters ranging from 0.1 mm to 1 mm were mixed into the coupling agent, and 50 sets of ultrasonic signals were collected. For environmental interference samples, equipment such as motors and fans were activated in the detection environment, and 50 sets of environmental interference ultrasonic signals were collected, for a total of 150 + 50 + 50 = 250 sets of interference signal samples.

[0046] The ultrasonic signal processing strategy further includes: preprocessing the collected and reflected suspected bubble ultrasonic signals, extracting the effective frequency bands from the suspected bubble ultrasonic signals, and performing signal benchmarking and normalization processing.

[0047] Signal preprocessing refers to the process of performing noise reduction and filtering operations on the acquired raw ultrasonic signals to remove interference components and retain the effective signal. The effective frequency band refers to the frequency range in the ultrasonic signal that contains bubble characteristic information, after excluding low-frequency noise and high-frequency interference. Signal benchmark processing refers to the process of eliminating systematic errors and environmental background interference by using the ultrasonic signal from a bubble-free region as a benchmark. Normalization processing refers to the process of converting the amplitude of the ultrasonic signal to a fixed range, making signals of different intensities comparable and facilitating feature extraction and threshold comparison.

[0048] Specifically, time-domain features, frequency-domain features, and time-frequency features are extracted from 2500 bubble signal samples, and the arithmetic mean of each feature is calculated. and standard deviation Its formula is: ,in, For the first The arithmetic mean of the features; For the first The first group of bubble signal samples The values ​​of each feature; For the 2500 groups of samples, the first The sum of the values ​​of each feature. ,in, For the first The standard deviation of each feature; For the first The first group of samples The difference between each feature value and the average value of that feature; It is the sum of the squares of the differences in the 2500 samples.

[0049] The threshold range for each feature is defined using the 95% confidence interval principle, and the formula is as follows: , ,in, This is the lower limit of the feature threshold; This represents the upper limit of the feature threshold.

[0050] Bubble signal samples and non-bubble signal samples were divided into training and test sets in a 7:3 ratio. The training set contained 1750 bubble signals and 175 non-bubble signals, while the test set contained 750 bubble signals and 75 non-bubble signals. Eleven feature parameters (time-domain, frequency-domain, and total) from all samples were used as the input vector. The radial basis function kernel was selected as the kernel function for the SVM model, and a penalty parameter was set. Kernel function parameters The model is trained using the training set until the model's accuracy on the test set is greater than 98%, at which point training stops.

[0051] In this embodiment, the bubble samples cover the full size range and the full concentration scenario, and multiple sets of samples are collected for each size and concentration combination to ensure that the model training data covers all possible bubble shapes in practical applications and avoids the problem of missed detection due to a single sample.

[0052] By selectively collecting interference samples from pipe wall impurities, coupling agent bubbles, and environmental interference, the SVM model can accurately distinguish bubble signals from non-bubble interference signals, solving the industry pain point of misjudging false reflection signals in traditional ultrasonic testing, and improving the detection accuracy to over 98%.

[0053] The preprocessing stage removes low-frequency noise, high-frequency interference, and system errors through noise reduction filtering, effective frequency band extraction, reference calibration, and normalization, making the bubble characteristic signal more prominent. In particular, the calibration method based on the signal in the bubble-free region effectively offsets the background interference caused by the irregularity of the corrugated pipe wall and the difference in probe coupling, thus improving signal stability.

[0054] Based on the feature threshold definition of 95% confidence interval, and combined with 11 feature parameters in the time domain, frequency domain, and time-frequency domain, a multi-dimensional and highly robust bubble feature discrimination system was constructed. Compared with the single feature threshold method, it is more adaptable to feature variations under different bubble sizes and concentrations, and reduces the false positive rate.

[0055] A 7:3 sample split ratio is adopted to ensure sufficient training data, while the model's generalization ability is verified through the test set. A radial basis function kernel function is selected and the parameters are optimized so that the model can effectively handle high-dimensional feature data, accurately fit the nonlinear boundary between bubble and non-bubble signals, and achieve efficient binary classification in complex scenarios.

[0056] The ultrasonic signal processing strategy also includes: calculating the theoretical propagation time of ultrasound from the probe to the pipe wall and back to the probe based on the corrugated pipe wall thickness and ultrasonic velocity, and extracting the signal segment of the preset propagation time as the effective analysis interval.

[0057] The theoretical propagation time refers to the calculated time for the ultrasonic wave to travel from the probe, be reflected by the pipe wall, and return to the probe. It is derived based on the corrugated pipe wall thickness and the ultrasonic velocity. The effective analysis interval refers to the signal segment that matches the theoretical propagation time, excluding invalid signals outside the propagation path, and focusing on the reflected signal of the target area.

[0058] Specifically, wavelet thresholding denoising was applied to the acquired raw ultrasound signals, using the db4 wavelet basis, with a decomposition layer of 3, and a threshold was set. ,in, To reduce the number of signal sampling points, high-frequency noise in the signal is removed. A bandpass filter is used, with a passband frequency range of 1 MHz to 8 MHz, to filter out low-frequency interference below 1 MHz and high-frequency noise above 8 MHz, retaining the effective frequency band of the bubble signal. Ultrasonic signals from bubble-free regions are collected as reference signals, and the preprocessed signal is subtracted from the reference signal to eliminate system errors. The maximum-minimum normalization method is used to convert the signal amplitude to the range of 0 to 1, using the formula: ,in, The original signal amplitude, This is the minimum value of the signal amplitude. This represents the maximum amplitude of the signal.

[0059] Next, the theoretical propagation time is calculated using the following formula: ,in, The theoretical propagation time is expressed in microseconds. The thickness is the wall thickness of the corrugated pipe, in millimeters. The speed at which ultrasonic waves propagate through the corrugated pipe material is expressed in millimeters per microsecond.

[0060] Based on theoretical propagation time Centered on the signal, 2 microseconds before and after the signal are extracted as the effective analysis interval, i.e., the effective analysis interval is... microseconds Microseconds ensure coverage of the bubble reflection signal.

[0061] The ultrasonic signal processing strategy also includes: extracting time-domain features within the effective analysis interval, including the duty cycle, rise time, waveform entropy, mean pulse interval, and standard deviation of the signal peaks.

[0062] The temporal features refer to the characteristic parameters extracted from the time dimension of the ultrasound signal, including the signal peak value, peak duty cycle, rise time, waveform entropy, pulse interval mean, and standard deviation. The signal peak value refers to the maximum amplitude of the ultrasound signal within the effective analysis interval. The peak duty cycle is the ratio of the time during which the signal amplitude exceeds 70% of the peak value within the effective analysis interval to the total effective analysis interval time. The rise time is the time required for the signal to rise from 10% to 90% of the peak value. Waveform entropy is a parameter used to describe the waveform complexity of the ultrasound signal; the higher the entropy value, the more irregular the waveform. The pulse interval mean is the arithmetic mean of the time interval between two adjacent signal pulses within the effective analysis interval. The pulse interval standard deviation is a parameter indicating the dispersion of the time interval between two adjacent signal pulses.

[0063] A Fast Fourier Transform is performed on the effective analysis interval to extract frequency domain features with dominant frequency and spectral width within the effective analysis interval.

[0064] The Fast Fourier Transform (FFT) is a mathematical transformation method that converts a time-domain signal into a frequency-domain signal, used to extract the frequency characteristics of the signal. Frequency-domain characteristics refer to the feature parameters extracted from the frequency dimension of the ultrasound signal, including the dominant frequency and spectral width. Dominant frequency: refers to the frequency component with the highest energy in the frequency-domain signal. Spectral width: refers to the frequency range in the frequency-domain signal that accounts for 90% of the total energy.

[0065] Short-time Fourier transform is used to generate the time-frequency plot of the signal, and time-frequency domain features with time-frequency centroid and time-frequency entropy within the effective analysis interval are extracted.

[0066] The Short-Time Fourier Transform (SFT) is an analytical method that converts a time-domain signal into a time-frequency graph, simultaneously reflecting the signal's time and frequency characteristics. A time-frequency graph is an image with time on the horizontal axis and frequency on the vertical axis, displaying amplitude as color or grayscale, visually representing the signal's frequency distribution at different times. Time-frequency domain features refer to the characteristic parameters extracted from the time-frequency graph, including the time-frequency centroid and time-frequency entropy. The time-frequency centroid refers to the center coordinates of the energy concentration region in the time-frequency graph, reflecting the concentrated location of signal energy on the time-frequency plane. Time-frequency entropy is a parameter describing the uniformity of energy distribution in the time-frequency graph; the smaller the entropy value, the more concentrated the energy distribution.

[0067] The ultrasonic signal processing strategy further includes: comparing the time-domain features, frequency-domain features, and time-frequency-domain features with the bubble feature threshold range using a two-level discrimination algorithm to determine whether the reflected suspected bubble ultrasonic signal is a bubble signal.

[0068] Specifically, the time-domain features are extracted first. Among these, the signal peak value... This is used to read the maximum signal amplitude within the valid analysis interval, in volts.

[0069] Peak duty cycle To statistically analyze the signal amplitude within the effective analysis interval Total time Duty cycle ,in, The effective analysis interval is measured in microseconds.

[0070] Rise time To calculate the signal from Rise to The required time is measured in microseconds.

[0071] Waveform entropy The calculation formula is: ,in, For the first The proportion of the amplitude of each sampling point to the total amplitude.

[0072] Mean pulse interval This is obtained by identifying pulse signals within the valid analysis interval and calculating the arithmetic mean of the intervals between adjacent pulses.

[0073] Pulse Interval Standard Deviation The formula is: ,in, For the first The pulse interval is measured in microseconds. This represents the number of pulse intervals.

[0074] Secondly, frequency domain features are extracted. The signal within the effective analysis interval is subjected to a Fast Fourier Transform to obtain the frequency domain signal. Among these features, the dominant frequency... The frequency with the largest amplitude in the frequency domain signal, measured in megahertz (MHz). Spectral width. This is used to calculate the frequency range in which 90% of the total energy of a frequency domain signal is generated, with the unit being megahertz (MHz).

[0075] Finally, time-frequency domain features are extracted. A short-time Fourier transform is performed on the signal within the effective analysis interval, using a Hanning window with a length of 256 sampling points and an overlap rate of 50%, to generate a time-frequency plot.

[0076] Time-frequency center of gravity The calculation formula is: ,in, The time unit is microseconds. The frequency unit is megahertz. This represents the amplitude at the corresponding point on the time-frequency graph, in microseconds × megahertz.

[0077] Time-frequency entropy The calculation formula is: ,in, This represents the normalized amplitude values ​​of the time-frequency graph.

[0078] Furthermore, the secondary discrimination algorithm includes: initially marking time-domain features within the effective analysis interval that fall within the bubble feature threshold interval and whose frequency domain dominant frequency energy ratio is greater than a preset ratio as suspected bubble signals.

[0079] The dominant frequency energy ratio refers to the ratio of the energy corresponding to the dominant frequency to the total energy of the signal in the entire frequency domain. The preset ratio refers to the critical value of the dominant frequency energy ratio used for preliminary screening of suspected bubble signals, which is determined based on statistical analysis of sample data.

[0080] The secondary discrimination algorithm further includes: classifying suspected bubble signals using an SVM model; if the model confidence is greater than a preset confidence level, and the sum of the time-frequency centroid and the time-frequency entropy is greater than a preset threshold, then it is confirmed as a bubble signal.

[0081] Here, model confidence refers to the reliability of the SVM model's signal classification results, with a value ranging from 0 to 1; the higher the value, the higher the confidence. Preset confidence refers to the critical model confidence value required to confirm a bubble signal; signals below this value are considered non-bubble signals. Preset threshold refers to the critical value used to finally confirm the sum of the time-frequency centroid and time-frequency entropy of the bubble signal.

[0082] Specifically, the first-level discrimination involves checking whether all time-domain features fall within the corresponding bubble feature threshold range. The formula for calculating the main frequency energy ratio is: in, The energy unit corresponding to the main frequency is the joule. The total energy of the frequency domain signal is measured in joules. A preset ratio is set. If all time-domain features fall within the threshold range and If so, it is initially marked as a suspected bubble signal.

[0083] Secondary discrimination: The 11 feature parameters of the suspected bubble signal are input into the trained SVM model to obtain the model confidence score. .

[0084] Set the default trust level ,like Then it is determined to be a non-bubble signal.

[0085] The formula for calculating the sum of the time-frequency centroid and the time-frequency entropy is: Set a preset threshold ,like and If so, it is confirmed as a bubble signal.

[0086] The ultrasonic signal processing strategy further includes: the ultrasonic features of the suspected bubble include at least one of the bubble depth, equivalent diameter, and echo attenuation coefficient.

[0087] Specifically, after confirming the presence of bubble signals, the ultrasonic characteristic parameters of suspected bubbles are further extracted. These characteristic parameters can comprehensively reflect the physical properties of the bubble, providing key data support for subsequent bubble hazard assessment and bellows condition diagnosis. Among them, the ultrasonic characteristics of suspected bubbles include at least one of the bubble depth, equivalent diameter, and echo attenuation coefficient. When conditions permit, all three characteristic parameters can be extracted simultaneously to achieve a comprehensive characterization of the bubble's properties.

[0088] Specifically, bubble depth refers to the depth of the bubble within the corrugated pipe wall. It is calculated using the time difference of ultrasonic wave propagation; that is, based on the time difference between the bubble's reflected signal and the signal reflected from the pipe wall surface, combined with the propagation speed of the ultrasonic wave in the pipe wall material, the bubble's depth is accurately deduced. The formula for calculating bubble depth is: , ,in, This refers to the depth of the bubble. The speed at which ultrasonic waves propagate in the corrugated pipe material; This represents the actual propagation time of the ultrasonic signal after reflection from the bubble. This is the theoretical propagation time of the ultrasonic signal after reflection from the inner surface of the pipe wall; The thickness is the wall thickness of the corrugated pipe.

[0089] The equivalent diameter of a bubble refers to the diameter of an irregularly shaped bubble that is equivalent to a spherical bubble. It is calculated using a model that correlates the peak value, dominant frequency, and other characteristic parameters of the bubble signal with the bubble diameter. This model, trained on a large number of bubble samples of different sizes, has high computational accuracy. The formula for calculating the equivalent diameter of a bubble is: ,in, The equivalent diameter of the bubble; This is the proportionality coefficient. The fixed values ​​corresponding to different ultrasonic frequencies and probe models were obtained through experimental calibration. The peak amplitude of the suspected bubble echo signal; The peak amplitude of the standard bubble echo signal is 2.5 volts. The standard bubble is a spherical bubble with a diameter of 1 mm. The peak amplitude of the echo is obtained under the same detection conditions (same material, same propagation distance).

[0090] The echo attenuation coefficient refers to the degree of energy attenuation of ultrasonic waves after they pass through a bubble during propagation. It is calculated by comparing the amplitude of the reflected signal from the bubble with the amplitude of the reflected signal from a bubble-free region. This parameter reflects the attenuation characteristics of ultrasonic waves by the bubble, and indirectly reflects the bubble's concentration and material properties. The formula for calculating the echo attenuation coefficient is: ,in, The return attenuation coefficient is... The higher the value, the more severe the attenuation of the ultrasound signal after reflection; The reference echo amplitude is the peak echo amplitude of the ultrasonic signal after reflection from the tube wall in the bubble-free region. It is obtained experimentally and is a fixed value under the same detection scenario. This represents the peak amplitude of the suspected bubble echo signal.

[0091] The ultrasonic signal processing strategy further includes: recording and numbering the acquisition points where bubble signals are detected, obtaining the axial position based on the acquisition interval; obtaining the circumferential angle of the bubble based on the time difference positioning algorithm; and calculating the radial distance between the bubble and the outer surface of the tube wall based on the time difference of ultrasonic signal propagation to form the preliminary three-dimensional positioning coordinates of the bubble signal in the axial, circumferential, and radial directions.

[0092] The acquisition point number refers to the sequential numbering of the ultrasonic probe's acquisition positions, used to associate the signal with the acquisition location. Axial position: refers to the position of the bubble along the length of the bellows, calculated based on the acquisition point number and acquisition interval. Circumferential angle: refers to the position angle of the bubble on the circumference of the bellows' cross-section, with the probe's initial position as the 0-degree reference. Radial distance: refers to the perpendicular distance from the bubble to the outer surface of the bellows, reflecting the bubble's position inside and outside the pipe wall.

[0093] Specifically, the formula for calculating the axial position is: ,in, This refers to the axial position, in millimeters. Number the acquisition point where the bubble signal was detected; The sampling interval is in millimeters.

[0094] The circumferential angle was calculated using a time-difference positioning algorithm. Four probes were numbered 1-4, with adjacent probes spaced 90 degrees apart. The time it took for each probe to receive the bubble signal was recorded. , , and Calculate the time difference ; ; Based on ultrasonic speed The circumferential angle of the bubble is calculated using the triangulation formula. The unit is degrees.

[0095] Radial distance refers to the perpendicular distance from the bubble to the outer surface of the bellows. It is calculated by the difference between the actual propagation time and the theoretical propagation time of the ultrasonic signal. The calculation formula is as follows: in, This is the radial distance between the bubble and the outer surface of the tube wall; The speed at which ultrasonic waves propagate in the corrugated pipe material; This is the actual propagation time of the ultrasonic signal from the probe, after being reflected by the bubble, and back to the probe. This is the theoretical propagation time of the ultrasonic signal from the probe, after being reflected by the inner surface of the tube wall, and back to the probe. , The thickness of the corrugated pipe wall; This is the difference between the actual propagation time and the theoretical propagation time. If the bubble is outside the tube wall, ,at this time A negative value indicates the bubble is outside the tube wall; if the bubble is inside the tube wall, , A positive value indicates that the bubble is inside the tube wall.

[0096] In this embodiment, a wavelet thresholding denoising method using db4 wavelet base 3-level decomposition and adaptive thresholding is employed to specifically remove high-frequency noise while preserving the detailed features of the bubble signal. Compared to traditional filtering, this method is more suitable for the non-stationary characteristics of ultrasonic signals. Through a combination of reference signal subtraction and maximum-minimum normalization, fixed interferences such as probe coupling differences and equipment system errors are eliminated, and signals of different intensities are unified to the 0-1 range. This solves the problems of strong signals masking weak signals and the incomparability of signals from different acquisition points, laying a unified benchmark for subsequent feature extraction. By extracting multiple time-domain features such as signal peak value, duty cycle, rise time, waveform entropy, and the mean or standard deviation of pulse intervals, the amplitude, duration, rate of change, and stability of the bubble signal are comprehensively characterized. The dominant frequency and spectral width extracted using Fast Fourier Transform reflect the size and concentration characteristics of the bubbles: microbubbles have higher dominant frequencies and narrower spectra, while dense bubbles have wider spectral widths. The dominant frequency of coupling agent bubbles is usually concentrated in a specific frequency band, showing a significant difference from that of pipe bubbles. The introduction of the dominant frequency energy ratio further strengthens the distinction between effective signals and clutter, avoiding misjudgments caused by low-frequency interference.

[0097] The time-frequency graph generated by using short-time Fourier transform retains both the time and frequency information of the signal, and the time-frequency centroid... Time-frequency entropy reflects the location of energy concentration. Describes the uniformity of energy distribution. Compared to individual time-domain or frequency-domain features, time-frequency domain features can more accurately capture the instantaneous frequency changes of the bubble reflection signal, effectively distinguishing bubbles from fixed interference sources. Simultaneously, all time-domain features must fall within a threshold range and represent a significant proportion of the dominant frequency energy. By using dual conditions to quickly eliminate obvious non-bubble signals, the analysis scope is narrowed down to suspected bubble signals, avoiding the need for the subsequent SVM model to process a large amount of invalid data and improving overall detection efficiency; at the same time, based on a preset ratio of sample statistics... This ensures the rationality of the initial screening criteria, neither omitting real bubbles nor introducing excessive interference signals.

[0098] And introduce SVM model confidence. Double verification of the sum of time and frequency characteristics. To ensure the reliability of the model's classification results, Further verification using time-frequency domain features overcomes the limitations of single-model classification. This significantly improves the accuracy of bubble identification, especially effectively distinguishing easily confused interference signals such as pipe wall impurities and coupling agent bubbles, solving the core pain point of false signal misjudgment in traditional ultrasonic testing. The axial position is directly related to the bubble position along the length of the bellows, and the circumferential angle... The time-difference positioning algorithm determines the circumferential orientation of the bubble within the cross-section, and the radial distance accurately reflects the distance between the bubble and the outer surface of the tube wall. This high-precision three-dimensional positioning provides a clear basis for subsequent processing: if... If the value is positive, targeted exhaust operations can be performed; if... Negative values ​​can be used to check whether the pipe seal has failed, avoiding blind repairs.

[0099] The method further includes: acquiring first suspected bubble image data at the initial location, and extracting shape features and center point position from the first suspected bubble image data using a suspected bubble image processing strategy. Specifically, the first suspected bubble image data is obtained by targeting the initial location output by the ultrasonic detection using an industrial CCD camera, acquiring image data containing candidate defect regions.

[0100] Specifically, based on the structural parameters of the bellows and the camera mounting parameters, a rough mapping relationship between the physical coordinates of the initial position and the image pixel coordinates is established, as shown in the following formula: , ,in: These are the horizontal pixel coordinates of the image corresponding to the initial position; The axial physical coordinates of the bellows corresponding to the preliminary position output by the ultrasonic detection represent the position of the suspected bubble along the length of the bellows. The image vertical pixel coordinates corresponding to the initial position; The original image width; The vertical distance from the camera to the surface of the bellows; This refers to the horizontal field of view of the camera; The length of the corrugated pipe; The diameter of the corrugated pipe; The initial position output by the ultrasonic test corresponds to the circumferential angle coordinates of the bellows, representing the angular position of the suspected bubble in the circumferential direction of the bellows.

[0101] by Centered on the bubble, the size of the cropping window can be adjusted according to the maximum possible size of the bubble to ensure complete coverage of the suspected area. Select the corresponding area in the original image to complete the targeted cropping, resulting in the cropped image. .

[0102] Furthermore, the suspected bubble image processing strategy includes: based on the preliminary three-dimensional position output by ultrasonic detection, targeted cropping, contrast enhancement, and image normalization processing are performed on the acquired first suspected bubble image to obtain the first suspected bubble region. Specifically, targeted cropping, based on the preliminary three-dimensional position output by ultrasonic detection, selects the smallest rectangular region containing the candidate defect region in the first suspected bubble image. The cropped image size must retain a preset pixel redundancy range outside the candidate region edge to avoid missing defect edge information. Contrast enhancement uses an image enhancement algorithm to increase the grayscale difference between the suspected bubble region and the background of the corrugated pipe surface, making the defect outline clearer. The enhanced image must ensure that the grayscale dynamic range covers a preset interval without overexposure or underexposure areas. Image normalization mapping maps the grayscale values ​​of the contrast-enhanced image to a fixed interval, eliminating the influence of grayscale value fluctuations caused by different shooting environments, ensuring consistent image features. The first suspected bubble region is the image region containing the candidate defect after targeted cropping, contrast enhancement, and image normalization processing. This region has removed most irrelevant background information, retaining only the core image content related to defect determination.

[0103] Specifically, contrast enhancement is achieved through an adaptive histogram equalization algorithm, with specific parameters such as block size, contrast limit, and interpolation method set. For example, the image is divided into 8×8 sub-blocks, and histogram equalization is performed on each sub-block; the gray-level distribution of the histogram of each sub-block is limited to avoid noise amplification; bilinear interpolation is used to smooth the gray-level transition at the boundaries of the sub-blocks.

[0104] Image normalization is achieved using a linear normalization algorithm, as shown in the following formula: ,in: For the normalized image in The grayscale value at that location, and its range. ; To enhance the contrast of the image The grayscale value at that location is (0-255); This represents the minimum grayscale value of the image after contrast enhancement. This represents the maximum grayscale value of the image after contrast enhancement.

[0105] The normalized image is obtained after processing. This refers to the first suspected bubble area.

[0106] The suspected bubble image processing strategy includes: processing the first suspected bubble region to extract the edge contour of the first suspected bubble.

[0107] Among them, the edge contour refers to the curve formed by the continuous pixels with abrupt changes in grayscale value in the first suspected bubble region. This curve can reflect the outer boundary of the suspected bubble and is the basis for extracting shape features.

[0108] Specifically, the edge contour extraction is achieved through the Canny edge detection algorithm combined with morphological processing. First, the first suspected bubble region is... Gaussian filtering was used for noise reduction, with the Gaussian kernel size set to 5×5 and the standard deviation... Its formula is: ,in: For Gaussian kernel in The weight value at that point, and The range of values ​​is , The standard deviation of the Gaussian distribution is used to control the smoothness of the filter.

[0109] Secondly, for Perform Canny edge detection with dual thresholds: low threshold High threshold .

[0110] Edge detection process: First, calculate the image gradient to obtain the gradient magnitude and direction; then, retain pixels with gradient magnitudes greater than T_high as strong edges and discard those with gradient magnitudes less than T_high. The pixels; finally, it will be between and Pixels that are connected to strong edges are retained as weak edges, thus obtaining the initial edge image. .

[0111] Finally, for Morphological closing operations are performed to eliminate edge gaps. Both the expansion kernel and the erosion kernel use 3×3 rectangular structure elements.

[0112] The formula for the expansion operation is as follows: ,in, It is a 3×3 rectangular structural element; This indicates a dilation operation, which means that for each pixel in an image, if there is at least one edge pixel in its neighborhood, then the pixel is marked as an edge pixel.

[0113] The formula for corrosion calculation is: ,in, This represents the erosion operation, which means that for each pixel in the image, if all its neighboring pixels are edge pixels, then the pixel is retained as an edge pixel; otherwise, it is discarded.

[0114] The final image of the suspected bubble edge contour is obtained after processing. The outline is a closed curve with a width of one pixel.

[0115] The suspected bubble image processing strategy further includes: extracting at least one shape feature from the first suspected bubble contour, including area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance. The shape features are quantitative indicators used to describe the shape and gray-level distribution of the suspected bubble contour, including area, perimeter, roundness, aspect ratio, contour smoothness, and gray-level variance. The combination of these indicators can distinguish between bubble defects and false defects.

[0116] Specifically, targeting From the edge contour, six shape features are extracted. The specific calculation method is as follows: area This represents the total number of pixels enclosed by the outline. It is calculated by counting the number of pixels inside and on the boundary of the outline.

[0117] perimeter The length of the contour is calculated by iterating through consecutive pixels on the contour and accumulating the Euclidean distances between adjacent pixels, as shown in the following formula: ,in, For the outline of the first The coordinates of a pixel. The total number of pixels on the outline. = , = .

[0118] Circularity To describe how close the outline is to a circle, the formula is as follows: ,in, For area, For the perimeter, Take 3.1416 for roundness. The range of values ​​is round The more irregular the outline, The smaller the value.

[0119] Aspect Ratio The ratio of the length to the width of the bounding rectangle is given by the following formula: ,in, Let be the length of the longer side of the circumscribed rectangle. Let be the length of the shorter side of the circumscribed rectangle, which is an axis-aligned rectangle. , round or square a long and narrow outline The value is relatively large.

[0120] Contour smoothness S describes the smoothness of the contour curve, and the formula is as follows: , ,in, The perimeter of the actual outline; Let be the perimeter of the circumscribed ellipse. Let be the major semi-axis of the circumscribed ellipse. Let be the minor semi-axis of the circumscribed ellipse. The range of values ​​is , The closer it is to 1, the smoother the contour.

[0121] Gray variance To describe the dispersion of grayscale values ​​within the first suspected bubble region, the formula is as follows: ,in: This represents the total number of pixels in the first suspected bubble region. For the first in the region The normalized grayscale value of each pixel. This is the normalized grayscale mean of all pixels within the region. .

[0122] The suspected bubble image processing strategy further includes: calculating the mean center point coordinates of the suspected bubble contour using the shape center method and the centroid method. Specifically, the shape center method calculates the center coordinates of the bounding rectangle of the suspected bubble contour as the reference point for the contour center; the bounding rectangle is the smallest axis-aligned rectangle that completely encloses the contour. The centroid method calculates the gray-level weighted average coordinates of all pixels within the suspected bubble contour as the reference point for the contour center; this method reflects the centroid position of the gray-level distribution of the contour. The mean center point is obtained by arithmetically averaging the center coordinates calculated by the shape center method and the center coordinates calculated by the centroid method, used to improve the accuracy of center point localization. The center point coordinates refer to the image pixel coordinates corresponding to the mean center point of the suspected bubble contour.

[0123] The suspected bubble image processing strategy further includes: if the shape features of the first suspected bubble contour are greater than a preset number and all fall within the bubble feature threshold range, then it is determined to be a bubble; if they are less than the preset number, it is directly determined to be no bubble, thus eliminating misjudgment of the ultrasound signal. The bubble feature threshold range is a reasonable range of values ​​for each shape feature obtained by extracting features from a large number of standard bubble defect samples. This range needs to be determined through sample training based on the actual application scenario to ensure that it covers the feature range of real bubbles and excludes feature values ​​of false defects.

[0124] Specifically, a preset number of items is first defined; that is, at least a preset number of shape features must fall within the corresponding bubble feature threshold range to be considered a bubble defect. The bubble feature threshold range is determined by collecting a preset number of standard bubble defect samples and a preset number of pseudo-defect samples, extracting shape features, and then statistically calculating the area. Circularity Aspect Ratio Contour smoothness Gray-scale variance and perimeter The threshold range was then determined. Next, six shape features of the suspected bubble outline were extracted and compared with the aforementioned threshold range, and the number of features falling within each range was counted. .like If the number of items is greater than or equal to the preset number, the suspected area is determined to be a bubble defect; if... If the number is less than the preset number, it is determined to be without bubbles, thus eliminating the possibility of misjudgment by ultrasonic testing.

[0125] The suspected bubble image processing strategy further includes: establishing a transformation relationship between the image coordinates and actual physical coordinates of the first suspected bubble image through a camera calibration algorithm, thereby obtaining the axial and circumferential actual physical coordinates of the first suspected bubble. Specifically, the camera calibration algorithm establishes a mapping relationship between image pixel coordinates and actual physical coordinates by capturing an image of a calibration plate of known size and solving for intrinsic camera parameters such as focal length, principal point coordinates, and distortion coefficients, as well as extrinsic parameters such as the relative position and orientation of the camera to the corrugated pipe surface. Image coordinates refer to the index coordinates of image pixels on a two-dimensional plane. Actual physical coordinates refer to the actual position coordinates of the suspected bubble on the corrugated pipe surface, including axial and circumferential physical coordinates, which are calculated based on the actual dimensions of the corrugated pipe and the camera calibration results.

[0126] The suspected bubble image processing strategy further includes: combining the preliminary radial three-dimensional positioning coordinates obtained from ultrasonic testing with the axial and circumferential actual physical coordinates obtained from image processing to output the three-dimensional determined position coordinates of the first suspected bubble. Specifically, the three-dimensional determined position coordinates, obtained by fusing the radial coordinates from ultrasonic testing with the axial and circumferential physical coordinates obtained from image processing, yield the precise three-dimensional position coordinates of the bubble defect. These coordinates can be directly used for defect localization and subsequent processing.

[0127] Specifically, the first camera calibration algorithm is implemented using Zhang Zhengyou's calibration method. First, a checkerboard calibration board is prepared and fixed at different positions and orientations on the corrugated pipe surface, and 15-20 calibration images are taken.

[0128] Secondly, extract the pixel coordinates of the checkerboard corner points in each calibration image. And its corresponding actual physical coordinates are known. The camera intrinsic parameter matrix is ​​solved using a calibration algorithm. and extrinsic parameter matrix The intrinsic parameter matrix for: K= , in: for Axial focal length, for Axial focal length, Main point coordinate, Main point coordinate.

[0129] In the extrinsic parameter matrix It is a 3×3 rotation matrix. Use a 3×1 translation vector to establish image pixel coordinates. The transformation relationship with the world coordinate system is expressed by the following formula: ,in, As a scale factor, For actual physical coordinates, These are the pixel coordinates of the image.

[0130] Next, the actual physical coordinates of the axial and circumferential axes are calculated. First, the coordinates of the center point of the bubble defect are obtained. These are the mean coordinates calculated using the shape center method and the centroid method. The shape center method calculates the coordinates of the top-left corner of the circumscribed rectangle of the contour. and the coordinates of the bottom right corner The formula for calculating the geometric center is: .

[0131] The centroid method calculates the average coordinates of all pixels within the contour. The formula is: ,in, This represents the number of pixels within the outline. These are the pixel coordinates within the outline.

[0132] Secondly, the conversion relationship obtained through camera calibration will be... Convert to physical coordinates in the world coordinate system ,in, Corresponding axial physical coordinates of the bellows , Arc length coordinates corresponding to the circumference of the bellows .

[0133] Finally, the coordinates of the arc length S of the circle are converted into coordinates of the angle of the circle. The formula is as follows: ,in, The nominal outer diameter of the bellows. It is the length of the circumference arc.

[0134] Subsequently, the preliminary radial coordinates of the ultrasound detection were obtained. With high accuracy, it can be directly used as the final radial coordinate. Axial physical coordinates obtained from image processing and circumferential physical coordinates As the final axial and circumferential coordinates, output the three-dimensional location coordinates of the bubble defect. .

[0135] In this embodiment, a combination of ultrasonic preliminary positioning and fine image processing is used. Ultrasonic scanning is first used to achieve rapid full-area scanning, followed by image strategy to extract shape features and verify defect attributes. This effectively eliminates false identification of scratches, stains, and other defects by ultrasound, making bubble identification more accurate. By fusing the radial coordinates of the ultrasound with the axial and circumferential physical coordinates after image calibration, the output three-dimensional location determination accuracy is higher, directly locating the specific position of the bubble on the bellows, facilitating subsequent defect tracing or repair. Targeted cropping based on the preliminary ultrasonic location processes only small areas containing suspected defects, avoiding full image processing and significantly reducing computational power consumption, thus meeting the high-speed inspection requirements of production lines. The extracted shape features, such as area and roundness, are quantifiable indicators that can distinguish between bubbles and false defects, and also statistically analyze the size and shape distribution of bubbles, providing data support for optimizing production processes.

[0136] The method also includes: determining whether the bellows is in a folded or unfolded state based on the color image dataset, and retrieving the preset state weight configuration of the ultrasonic features, shape features, and color features of the colored area in the folded or unfolded state.

[0137] Specifically, the corrugated pipe state determination is based on the pipe surface morphology features in the color image dataset. The core logic is to distinguish the state by identifying the corrugation regularity and fold marks. Each image in the color image dataset undergoes preprocessing, including grayscale conversion, binarization, and connected component analysis, to extract the effective area of ​​the pipe surface. The corrugation regularity is calculated using the grayscale projection method, with the following formula: ,in, The ripple regularity is represented by a value ranging from 0 to 1. The closer the value is to 1, the more regular the ripples are. The standard deviation of the gray values ​​in the pipe surface area reflects the degree of dispersion of the gray values; This represents the average grayscale value of the pipe surface area, reflecting the overall brightness of the pipe surface.

[0138] A threshold for ripple regularity was set through extensive sample calibration. If S≥ The pipe surface has regular ripples and no obvious folding marks, indicating it is in the unfolded state. If S < The pipe surface has disordered corrugations and fold marks, indicating that it is in a folded state.

[0139] Furthermore, when the corrugated pipe is in a folded state, the weight of shape features is greater than the sum of the weights of color features and ultrasonic features in the colored area; when the corrugated pipe is in an unfolded state, the weight of shape features is less than the sum of the weights of color features and ultrasonic features in the colored area.

[0140] Based on the characteristic reliability of the bellows in different states, a differentiated weight configuration is designed, with a total weight of 1, as follows: Weight configuration for folded state. In the folded state, shape features are less affected by interference and have high reliability in judgment. Color features in colored areas are easily obscured by wrinkles, and ultrasonic features are easily interfered with by reflections from the folded interface. Therefore, the following weights are set for shape features: ,For example 0.6. Color feature weights for colored regions. ,For example Ultrasonic feature weights ,For example When the weight of shape features is greater than the weight of color features in colored regions plus the weight of ultrasonic features, the requirements are met.

[0141] Expand state weight configuration In the unfolded state, the color features and ultrasonic features of the colored areas are highly reliable for identification, and the shape features are mostly normal wavy structures. Therefore, the following shape feature weights are set: For example 0.3. Color feature weights for colored regions. For example, W2_color=0.4. Ultrasonic feature weights. ,For example When the shape feature weight is less than the color feature weight of the colored region plus the ultrasonic feature weight, the requirement is met.

[0142] The method also includes: inputting a preset state weight configuration into a weighted fusion algorithm, normalizing the color features, ultrasonic features and shape features of the colored area to obtain a comprehensive defect vector of the suspected defect area on the corrugated pipe surface; if the comprehensive defect vector is greater than a preset threshold, it is determined that there is a defect on the corrugated pipe surface, and the defect type and the location of the suspected bubble are obtained.

[0143] Specifically, the min-max normalization method is used to transform each feature parameter to the interval between 0 and 1, as shown in the formula: ,in: The value is the normalized value of the feature, ranging from 0 to 1; These are the original values ​​of the features; This is the minimum sample value for this feature; This is the maximum sample value for this feature.

[0144] The formula for calculating the comprehensive defect vector is: ;in: This is the comprehensive defect vector, ranging from 0 to 1; Weights for shape features; Normalized values ​​for shape features; For the color feature weights of the colored regions; This represents the normalized value of the color feature of the colored region. Weights for ultrasound features; This represents the normalized value of the ultrasound characteristics.

[0145] By statistically analyzing multiple sets of defective and normal samples, the comprehensive defect vector range of defective and normal samples was determined, and a preset threshold was set. :like : Determined to have a defect; if The result is determined to be defect-free.

[0146] When the color feature normalization value of the colored region And the comprehensive defect vector This indicates a significant leak in the tracer gas, and the presence of cracks / pores on the bellows surface. Normalized value of ultrasonic characteristics. And the comprehensive defect vector This indicates that the ultrasonic characteristics are obvious and that there are air bubbles on the surface of the corrugated pipe.

[0147] Fold-derived defects: normalized values ​​of shape features And the comprehensive defect vector This indicates that a folding-derived defect has occurred in the folded state.

[0148] In this embodiment, traditional defect determination methods employ fixed judgment logic, failing to consider the differences in the folded / unfolded states of the corrugated pipe, leading to a high false positive rate for defects in folded areas. This method accurately determines the pipe state using a color image dataset. Its core advantage lies in achieving adaptive matching of state and weight: In the folded state, to address the issues of colored areas being easily obscured by wrinkles and ultrasonic signals being easily interfered with by the folded interface, the shape feature weight is set to be greater than the sum of the color and ultrasonic feature weights, with the shape feature, which has strong anti-interference capabilities, dominating the determination; in the unfolded state, considering the reliability of color and ultrasonic feature determinations, the shape feature weight is set to be less than the sum of the color and ultrasonic feature weights, with the highly reliable feature dominating the determination. This adaptability design allows the method to cover the pipe state throughout its entire lifecycle, including transportation, construction, and maintenance, solving the pain point of traditional methods' "one-size-fits-all" determinations that cannot adapt to complex working conditions, thus improving the working condition adaptability rate.

[0149] Example 2 Further, as described in Example 1, the method includes: directional blowing of air at a predetermined pressure onto a suspected bubble at a determined location, and acquisition of image data of a second suspected bubble at the determined location; extraction of shape features from the second suspected bubble image data using a bubble image processing strategy, and comparison of shape features with the first suspected bubble image data; if the shape features of the first suspected bubble image data differ from those of the second suspected bubble image data, then the suspected bubble is determined to be a real bubble; otherwise, it is not a real bubble. The predetermined pressure is a pre-set airflow pressure value used to blow air over the suspected bubble area, which needs to be determined based on the corrugated pipe material and bubble size. Directional blowing involves aiming a miniature nozzle at the determined location of the suspected bubble and outputting airflow at the predetermined pressure; the core principle is to act only on the suspected bubble area, avoiding interference with other areas of the pipe surface. The second suspected bubble image data, obtained after directional blowing, maintains acquisition parameters completely consistent with the first suspected bubble image. The image data captured at the determined location is used to compare deformation with the first image.

[0150] Specifically, the nozzle axis of the miniature air nozzle should be perpendicular to the surface of the corrugated pipe, and the distance between the nozzle and the designated position should be maintained at 5-10 mm to avoid airflow dispersion. Simultaneously, the airflow pressure should be adjusted to the preset pressure using a solenoid regulating valve. The blowing time should not be too long to ensure that the airflow acts on the bubble without damaging the pipe wall. After blowing, an image of the second suspected bubble should be acquired immediately to prevent the bubble from resetting.

[0151] Furthermore, the bubble image processing strategy includes: preprocessing, contrast enhancement, and image normalization of the acquired second suspected bubble image to obtain the second suspected bubble region; wherein, when acquiring the second suspected bubble image, the acquisition parameters are kept completely consistent with those of the first suspected bubble image. The second suspected bubble region is an image block containing the suspected region obtained after preprocessing, contrast enhancement, and image normalization of the second suspected bubble image; the processing flow is completely consistent with that of the first suspected bubble region, and will not be repeated here.

[0152] The bubble image processing strategy also includes: processing the second suspected bubble region to extract the edge contour of the second suspected bubble. This process is completely consistent with the extraction process for the first suspected bubble contour, and will not be repeated here.

[0153] The bubble image processing strategy also includes extracting at least one shape feature from the second suspected bubble contour, such as area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance. The features extracted from the second suspected bubble contour are the same as those extracted from the second suspected bubble contour, ensuring homology in the comparison, and will not be elaborated further here.

[0154] The bubble image processing strategy also includes extracting at least one deformation feature from the second suspected bubble contour: a deformation coefficient, a center displacement, and a contour change rate. The deformation features, including the deformation coefficient, center displacement, and contour change rate, are quantitative indicators describing the changes in the suspected bubble contour before and after blowing. These features distinguish real bubbles easily blown by airflow from false defects such as scratches and stains that show no obvious deformation. The deformation coefficient represents the percentage change in area of ​​the suspected bubble contour after blowing, reflecting the degree of morphological change caused by airflow compression. The center displacement is the pixel distance between the center points of the suspected bubble contour before and after blowing, reflecting the positional change of the bubble due to airflow. The contour change rate represents the percentage change in the perimeter of the suspected bubble contour before and after blowing, reflecting the degree of distortion in the bubble's shape.

[0155] Specifically, for the extraction of deformation features, the deformation coefficient... The calculation formula is: ,in, The first suspected bubble outline; This is the outline of the second suspected bubble.

[0156] center displacement The calculation formula is: ,in, , The coordinates of the center point of the first suspected bubble outline; , The coordinates are the center point coordinates of the second suspected bubble profile.

[0157] Contour change rate The calculation formula is: ,in, Let be the perimeter of the first suspected bubble outline; Let be the perimeter of the second suspected bubble profile.

[0158] The bubble image processing strategy also includes: calculating the absolute difference values ​​of shape features such as area, perimeter, roundness, aspect ratio, contour smoothness, or gray-level variance in the outline of the second suspected bubble. If the absolute difference values ​​of shape features greater than a preset number all meet a preset threshold, they are initially marked as real bubble signals; if the absolute difference values ​​of shape features less than a preset number meet the preset threshold, they are determined to be non-real bubbles. The absolute difference value is the absolute value of the numerical difference between the corresponding shape features of the first and second suspected bubbles, used to quantify the magnitude of feature change.

[0159] Specifically, the absolute difference between the first and second shape features is calculated and compared with a preset threshold. The number of features that meet the threshold is then counted. ,like If the number of suspected bubbles is greater than or equal to a preset number, it is initially marked as a real bubble signal. The bubble image processing strategy also includes: if the second suspected bubble contour meets any of the following conditions: the deformation coefficient is greater than a preset threshold, or the center displacement is greater than a preset threshold, or the contour change rate is greater than a preset threshold, then it is confirmed as a real bubble. Among them, the preset threshold is a pre-set feature difference judgment boundary, which needs to be determined through training with a large number of real bubble and fake defect samples to ensure that it can distinguish the deformation of real bubbles from the lack of change of fake defects.

[0160] In this embodiment, preliminary ultrasound and first image detection can only distinguish between bubbles and false defects based on shape features. However, the shape features of some false defects highly overlap with those of bubbles, easily leading to misjudgment. By directional air blowing, real bubbles, as elastic cavities, undergo shape or position changes due to airflow, while false defects show no significant deformation. This achieves a fundamental distinction between the two at the physical level, completely eliminating the misjudgment of stains as bubbles and significantly improving the reliability of the detection results. For extremely small microbubbles, their static shape features are easily interfered with by the surface texture, making accurate identification difficult. However, the center displacement and contour changes of microbubbles after air blowing are amplified, and their presence can be easily captured through deformation features, filling the detection blind spot of static images for microbubbles. This step constructs a dual judgment logic based on shape feature differences and deformation feature changes: first, preliminary screening is completed through the absolute difference value of shape features, and then final confirmation is completed through deformation features, forming a closed loop of preliminary marking and accurate confirmation. Compared to single static feature judgment, dual verification significantly reduces the probability of misjudgment caused by accidental factors, making the bubble judgment results more unique and authoritative. The acquisition parameters, preprocessing procedures, and feature extraction dimensions of the second image are completely consistent with those of the first image, ensuring the homogeneity of features before and after blowing. This standardized comparison benchmark avoids feature deviations caused by differences in image acquisition or processing, allowing the changes in shape and deformation features to truly reflect the physical state of the bubble, thus guaranteeing the accuracy of the comparison results.

[0161] Example 3 In the description of Example 1 or Example 2, the color feature analysis strategy further includes: if multiple consecutive sampling points at the same axial position detect color regions that meet the conditions, and the regions are linearly distributed, then it is further confirmed that the bellows has cracks and / or fine pore defects.

[0162] Among them, the qualifying color regions are tracer gas color regions whose average color value falls within a preset color threshold range, as determined by color feature analysis. The existence of such regions indicates a tracer gas leak at the corresponding sampling point location. For the regions to exhibit a linear distribution, the center positions of the qualifying color regions corresponding to multiple consecutive sampling points must meet the linear fit requirement in a two-dimensional coordinate system formed by the axial and circumferential directions of the bellows. This arrangement represents a typical spatial characteristic of crack defects extending axially. The linear fit is used to quantify the degree of fit between the center positions of consecutive color regions and the fitted straight line; the closer the fit value is to 1, the more significant the linear distribution characteristic of the region positions.

[0163] Specifically, for determining axial continuity, the axial judgment range for continuous sampling points is first set: when the difference in axial coordinates between two leak locations does not exceed a preset distance, they are considered axially adjacent. Then, the axial coordinates in the leak location dataset are statistically analyzed, and sampling point groups with axial coordinates within the same preset distance interval and a number not less than a preset number are selected; this number serves as the preset baseline for continuous sampling point determination. For determining circumferential consistency, a circumferential coordinate deviation threshold for continuous sampling points is first set: the absolute value of the difference between the circumferential coordinates of all leak locations within the same continuous sampling point group and the group's mean does not exceed a preset angle. Then, for the selected axial continuous sampling point groups, the mean of the circumferential coordinates within the group is calculated, and the difference between the circumferential coordinates of each location within the group and the mean is compared one by one. Locations with deviations exceeding a preset angle are eliminated. If the number of remaining locations is still not less than a preset number, the linear distribution judgment stage begins.

[0164] At the same time, using the axial coordinate of the bellows The independent variable and the circumferential coordinates A univariate linear regression model is established for the dependent variable, and the model expression is as follows: , , ,in, The slope of the fitted line, The intercept of the fitted line, This represents the number of locations within a continuous data collection point group. For the first The axial coordinates of each position. For the first The circumferential coordinates of each position.

[0165] And use the coefficient of determination The linear fit is expressed by the following formula: , , , ,in, For the first The circumferential coordinates of each position; The slope of the linear fit; For the first The axial coordinates of each position; The intercept of the linear fit; This represents the number of locations within a continuous data collection point group.

[0166] Set a threshold for determining the linear fit. Equal to the preset threshold, when the calculated value is... When the color regions of the sampling points are linearly distributed, it is determined that the location of the sampling points is determined. Specifically, the determination of crack defects is as follows: if multiple consecutive sampling points at the same axial position meet the following two conditions, a crack defect is confirmed to exist.

[0167] Condition 1: The number of continuous collection points is not less than the preset number, and the positions of the color areas are linearly distributed.

[0168] Condition 2: Linear fit Greater than or equal to the preset threshold.

[0169] The determination of a pore defect is as follows: if the leakage location meets any of the following conditions, it is determined to be a pore defect.

[0170] Condition 1: Only a single sampling point detects a color area that meets the condition, and there are no continuous sampling points.

[0171] Condition 2: There are continuous data collection points, but the number is less than the preset number.

[0172] Condition 3: The number of continuous sampling points is not less than the preset number, but the position of the color area does not meet the linear distribution requirement, i.e. Less than the preset threshold.

[0173] In this embodiment, traditional tracer gas detection can only determine whether there is a leak, but cannot distinguish between cracks and pinholes. This new rule, however, uses a three-layer judgment logic of axial continuity, circumferential consistency, and linear fit to quantify the spatial characteristics of the two types of defects into calculable parameters: cracks, because they extend axially, form linearly distributed color regions at continuous sampling points, with a linear fit approaching 1; pinholes are point defects that appear only at single or discontinuous sampling points and have no linear distribution characteristics. This quantitative differentiation method completely solves the industry pain point of confusing the two types of defects, upgrading defect type determination from "qualitative" to quantitative. For short microcracks, the range of the leaked tracer gas color region is small, and traditional detection is prone to misjudging them as pinholes. This new rule, by presetting the base number of continuous sampling points and the axial judgment range, can accurately capture the extension characteristics of microcracks, and then verify their distribution pattern through linear fit, achieving accurate differentiation between microcracks and pinholes, filling the detection blind spot in micro-defect classification. First, discrete leak points are eliminated by axial continuity. Then, offset areas caused by gas diffusion are eliminated by circumferential consistency. Finally, the extension characteristics of the defect are confirmed by linear fitting. This greatly reduces misjudgments caused by factors such as tracer gas diffusion and pipe surface interference, making the judgment results more authoritative.

[0174] Example 4 In the description of Embodiment 1, Embodiment 2, or Embodiment 3, the bubble image processing strategy further includes: calculating the overall confidence level of the real bubble; when the overall confidence level of the real bubble is within a preset confidence level range, the blowing pressure value meets the preset blowing pressure value range; when the overall confidence level of the real bubble is less than the preset confidence level range, the blowing pressure value is adjusted downward to avoid misjudgment caused by insufficient airflow.

[0175] The overall confidence level is a quantitative index calculated by weighting the differences in shape and deformation characteristics of real bubbles. It is used to evaluate the reliability of bubble identification results under blowing pressure, with a value ranging from 0 to 1. The closer the value is to 1, the higher the reliability of the identification result. The preset confidence range is a reasonable range of overall confidence determined in advance through training with a large number of samples. This range is the core basis for determining whether the current blowing pressure is appropriate. The lower limit of the range must ensure the basic reliability of the identification result, and the upper limit of the range must exclude the interference of pipe wall deformation caused by excessive pressure. The preset blowing pressure range is a range of blowing pressure values ​​pre-set for different materials of corrugated pipes and different sizes of bubbles. The pressure within this range can cause identifiable deformation of real bubbles without damaging the corrugated pipe wall.

[0176] Specifically, the weighting coefficients are determined using the analytic hierarchy process (AHP) to assign weights to each indicator. The overall confidence level is then calculated. The calculation formula is: ,in, For the first The weighting coefficients for each indicator; For the first The normalized values ​​of each indicator.

[0177] For sample data collection, two common materials, PE and PVC, were selected as corrugated pipes, and standard bubble samples of multiple sizes were prepared for each. Characteristic index data of each sample were collected under different blowing pressures. Multiple blowing pressures were set, and each sample was repeatedly tested a predetermined number of times under each pressure gradient to form an association dataset containing the confidence scores of pressure features.

[0178] Next, a pre-set confidence level range was defined. Statistical analysis of the associated datasets was performed to select the comprehensive confidence level intervals corresponding to samples with undisputed judgment results, eliminating abnormal samples due to excessively low confidence levels caused by insufficient pressure or excessive pressure causing pipe wall deformation. Finally, the pre-set confidence level range was determined, within which the bubble judgment results were reliable and there was no risk of pipe wall damage. Then, a range of preset blowing pressure values ​​was defined: corresponding preset blowing pressure value ranges were determined for different corrugated pipe materials, such as the range corresponding to softer PE corrugated pipes and the range corresponding to harder PVC corrugated pipes. The calculated comprehensive confidence level of the actual bubbles was then determined. When the current air pressure value is within the preset confidence range, it is determined that if the current air pressure value is within the preset air pressure range, no pressure adjustment is required.

[0179] When the overall confidence level When the airflow is below the lower limit of the preset confidence range, it is determined that the current airflow effect is insufficient, and the blowing pressure value needs to be adjusted downwards. Here, the correction is to adjust upwards because a low confidence level indicates insufficient airflow effect, requiring an increase in pressure. This leads to an upward adjustment of the step size. After adjustment, directional blowing and image acquisition are recalculated to obtain a new comprehensive confidence level until the confidence level falls within the preset range. When the comprehensive confidence level... If the pressure exceeds the upper limit of the preset confidence range, the current blowing pressure is deemed too high, posing a risk of damaging the pipe wall. The pressure needs to be adjusted downwards until the confidence level falls within the preset range. To ensure the safety of the bellows, absolute upper and lower limits for the blowing pressure are set. When the adjusted pressure reaches the upper or lower limit, the adjustment stops and an alarm is triggered, prompting manual intervention for confirmation.

[0180] In this embodiment, the traditional fixed blowing pressure detection method is prone to misjudgment due to excessively low or high pressure. However, by comprehensively considering confidence level and adaptive pressure adjustment, the optimal pressure required by the bubble can be dynamically matched: increasing the pressure when the confidence level is low and decreasing the pressure when the confidence level is high, ensuring that the bubble deformation is within an identifiable and undamaged range, thus eliminating misjudgment caused by improper pressure at its source. The comprehensive confidence level transforms shape feature differences and deformation characteristics into a 0-1 quantitative index, replacing the traditional qualitative judgment and providing a clear standard for measuring the reliability of the judgment results. At the same time, the pre-defined confidence level range further clarifies the boundaries of reliable judgment and avoids disputes caused by ambiguous judgments.

[0181] Example 5 like Figures 2-4 As shown, a corrugated pipe surface defect detection device is used in the corrugated pipe surface defect detection method described in Examples 1-3 above. The device includes: a movable installation mechanism 1, which is used to support all components of the device.

[0182] At least one mounting platform group 2; each mounting platform group 2 is mounted opposite to the mobile mounting mechanism 1, the first mounting platform 201 of the mounting platform group 2 is mounted at the fixed position of the mobile mounting mechanism 1, and the second mounting platform 202 of the mounting platform group 2 is mounted at the sliding position of the mobile mounting mechanism 1.

[0183] At least two airbags 3 are used to inflate and fit against the inner walls of both ends of the corrugated pipe; the two airbags 3 are respectively mounted on the first mounting platform 201 and the second mounting platform 202. Specifically, after inflation, they fit against the inner walls of both ends of the corrugated pipe to achieve end sealing and fixation of the corrugated pipe, while providing a mounting carrier for the ultrasonic sensor 11.

[0184] The air inlet 4 is used to slowly and steadily introduce a tracer gas of a preset color into the bellows. Specifically, the tracer gas of the preset color is slowly and steadily introduced into the bellows. If there are cracks / pores in the pipe wall, the tracer gas can leak from the defect, which facilitates subsequent visual inspection.

[0185] Telescopic cylinder 5 is installed on the mobile mounting mechanism 1, with one end extending out of the airbag 3 through the first mounting platform 201. Telescopic cylinder 5 drives rotary motor 6, gas nozzle 7, and first industrial camera 8 to extend and retract along the bellows axis, adjusting their axial position within the bellows to adapt to bellows of different lengths.

[0186] A rotary motor 6 is installed on the telescopic end of the telescopic cylinder 5 located inside the bellows. The rotary motor 6 drives the gas nozzle 7 and the first industrial camera 8 to rotate around the axis of the bellows, realizing directional air blowing and image acquisition throughout the inner wall of the bellows.

[0187] A gas nozzle 7 is mounted on the rotating end of the rotary motor 6; the gas nozzle 7 is used to directionally blow air onto the suspected air bubble when it is found on the wall of the bellows. Specifically, when a suspected air bubble is detected, air is directionally blown onto the suspected area, and its authenticity is verified by the deformation of the air bubble.

[0188] A first industrial camera 8 is mounted on the rotating end of a rotary motor 6 and is arranged side-by-side with a gas nozzle 7 along the axis of the bellows. The first industrial camera 8 acquires images of suspected bubble areas on the inner wall of the bellows, including first / second suspected bubble images, for subsequent image processing and defect determination.

[0189] Mounting bracket 9 spans across the movable mounting mechanism 1. Mounting bracket 9 supports the second industrial wide-angle camera 10, ensuring it spans over the corrugated pipe and covers the entire pipe surface. Specifically, the mounting bracket can be equipped with transverse and longitudinal moving devices. These moving devices can be trolleys, linear drive motors, or rack and pinion linear drive structures, among other technical solutions. The specific structure of the moving devices will not be detailed here.

[0190] The second industrial wide-angle camera 10 is hinged to the mounting bracket 9; the second industrial wide-angle camera 10 is used to acquire color images of the outer wall of the bellows, to detect the color area formed by the tracer gas leak, and to determine crack / pore defects.

[0191] An ultrasonic sensor 11 is mounted on the end face of the airbag 3 located on the first mounting platform 201. The ultrasonic sensor 11 is used to emit ultrasonic waves into the bellows and receive ultrasonic waves reflected back from the inner wall of the bellows. The ultrasonic sensor 11 is mounted on the end face of the airbag 3, emits ultrasonic waves into the inner wall of the bellows and receives reflected waves, thereby achieving preliminary location of defects in the inner wall.

[0192] When both ends of the corrugated pipe are fixed to the equipment by the inflated airbags 3, the moving installation mechanism 1 drives the second installation platform 202 to move, thereby stretching the folds of the corrugated pipe and facilitating the detection of defects on the surface of the corrugated pipe.

[0193] Furthermore, the mobile installation mechanism 1 includes: a mechanism mounting plate 101. The mechanism mounting plate 101 serves as the mounting base for all components of the equipment, providing a stable support platform. A rotary drive motor 102 is mounted on the mechanism mounting plate 101. The rotary drive motor 102 drives the lead screw 103 to rotate, providing power for the movement of the second installation platform 202. One end of the lead screw 103 is mounted on the rotating end of the rotary drive motor 102. A nut 104 is mounted on the lead screw 103. The rotary motion of the rotary drive motor 102 is converted into the linear motion of the installation slider 105, driving the second installation platform 202 to move. The installation slider 105 is mounted on the nut 104. The installation slider 105 supports the second installation platform 202 and slides linearly along the lead screw 103 with the nut 104. The first installation platform 201 is mounted on the rotary drive motor 102, the second installation platform 202 is mounted on the installation slider 105, and the telescopic cylinder 5 is mounted on the rotary drive motor 102.

[0194] The working process is as follows: First, the two ends of the bellows are fitted over the air bladders 3 of the first and second mounting platforms 202, and the air bladders 3 are inflated to make them fit against the inner wall of the bellows, thus fixing and sealing the bellows at the ends. The rotary drive motor 102 of the moving mounting mechanism 1 is started, driving the lead screw 103 to rotate, which moves the mounting slider 105 and the second mounting platform 202 away from the first mounting platform 201, stretching the folds of the bellows and flattening the pipe surface for subsequent inspection. Second, blue tracer gas is slowly introduced into the bellows through the air inlet 4, keeping the gas pressure inside the pipe stable within the preset pressure. If there are cracks / pores in the pipe wall, the tracer gas will leak from the defect to the outer wall. The second industrial wide-angle camera 10 acquires the color image of the outer wall of the bellows, extracts the color area of ​​the tracer gas through a color feature analysis strategy, and determines whether there are cracks / pore defects by combining the linear distribution of continuous acquisition points. Third, the ultrasonic sensor 11 emits ultrasonic waves towards the inner wall of the bellows, receives and analyzes the reflected wave signals, locates the preliminary three-dimensional position of the suspected bubble, and synchronizes the position information to the control system. The first industrial camera 8 acquires and analyzes the first image of the suspected area. Fourth, the telescopic cylinder 5 drives the rotary motor 6, gas nozzle 7, and first industrial camera 8 to extend and retract to the axial position of the suspected bubble; the rotary motor 6 drives both to rotate to the corresponding circumferential position; the gas nozzle 7 blows air in a directional manner with a preset pressure; the first industrial camera 8 acquires the second image after blowing air, extracts features and calculates the comprehensive confidence level through image processing strategies to determine whether it is a real bubble. Fifth, after the inspection is completed, the airbag 3 deflates, the rotary drive motor 102 drives the second mounting platform 202 to reset, and the inspected bellows is removed to prepare for the next inspection.

[0195] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method of detecting defects in the tube surface of a corrugated pipe, characterized by, The method comprises: Stable color tracer gas is introduced into the bellows, color image data sets of the outer surface of the bellows are collected at preset intervals along the length direction of the bellows, and the color feature analysis strategy is used to determine whether the preset color of the tracer gas exists in each color image data set and extract the color feature of the colored area of the bellows surface; Ultrasonic waves are emitted into the bellows, and the reflected ultrasonic signals are collected, the ultrasonic signal processing strategy is used to preliminarily determine whether there is a suspected bubble on the surface of the bellows, and the ultrasonic feature and preliminary position of the suspected bubble are extracted; The first suspected bubble image data at the preliminary position is collected, and the suspected bubble image processing strategy is used to extract the shape feature and center point position of the first suspected bubble image data; The color image data set is used to determine whether the bellows is in a folded state or an unfolded state, the preset state weight configuration of the ultrasonic feature, the shape feature and the color feature of the colored area in the folded state or the unfolded state is called, and the preset state weight configuration is input into a weighted fusion algorithm to normalize the color feature of the colored area, the ultrasonic feature and the shape feature, so as to obtain a comprehensive defect vector of the suspected defect area of the bellows surface, and if the comprehensive defect vector is greater than a preset threshold value, it is determined that there is a defect on the bellows surface, and the type of the defect and the determined position of the suspected bubble are obtained. The color feature analysis strategy comprises: collecting preset color sample data of the tracer gas under different concentrations and different environments, counting the gray value distribution range in the preset color area, and defining the gray value distribution range as a preset color threshold interval; 2. The method of claim 1, wherein Each image in the color image data set is binarized, the color mean value of the preset color area is extracted, and the color mean value of the preset color area is defined as the color feature of the colored area; If the color mean value of the preset color area falls within the preset color threshold interval, it is determined that the preset color of the tracer gas exists in the area, and the image is defined as a leakage sample image; The color feature analysis strategy further comprises: if a plurality of continuous collection points at the same axial position all detect a color area meeting the condition, and the area positions are linearly distributed, it is further confirmed that the bellows has a crack and / or a fine hole defect. The ultrasonic signal processing strategy comprises: collecting ultrasonic signal sample data of bubbles of different sizes and different concentrations, and ultrasonic signal sample data of impurities in the pipe wall, coupling agent bubbles and environmental interference, defining a bubble feature threshold interval, inputting the bubble ultrasonic signal sample data and the ultrasonic signal sample data into a support vector machine to train a two-classification model, and outputting a determination result of a bubble signal or a non-bubble signal; 3. The method of claim 1, wherein The collected suspected bubble ultrasonic signal is preprocessed, the effective frequency band in the suspected bubble ultrasonic signal is extracted, and signal reference and normalization processing are performed; According to the thickness of the bellows wall and the ultrasonic wave speed, the theoretical propagation time of the ultrasonic wave from the probe to the pipe wall and then reflected back to the probe is calculated, and the signal segment of the preset propagation time is taken as the effective analysis interval. ​ Extracting time domain features of the signal peak value and the duty cycle of the peak value, the rising edge time, the waveform entropy, the mean and standard deviation of the pulse interval in the effective analysis interval; performing fast Fourier transform on the effective analysis interval to extract frequency domain features of the main frequency and the spectrum width in the effective analysis interval; generating a signal time-frequency diagram by using a short-time Fourier transform to extract time-frequency domain features of the time-frequency barycenter and the time-frequency entropy in the effective analysis interval; Comparing the time domain features, the frequency domain features and the time-frequency domain features with the bubble feature threshold interval by using a two-stage discriminant algorithm to determine whether the suspected bubble ultrasonic signal reflected is a bubble signal; The ultrasonic features of the suspected bubble include at least one of the depth, the equivalent diameter and the echo attenuation coefficient of the bubble; Recording the collection points where the bubble signals are detected and numbering them to obtain the axial positions according to the collection interval; obtaining the circumferential angle of the bubble according to the time difference positioning algorithm; calculating the radial distance between the bubble and the outer surface of the pipe wall based on the propagation time difference of the ultrasonic signal to form the preliminary three-dimensional positioning coordinates of the bubble signal in the axial, circumferential and radial directions.

4. The method of claim 3, wherein The two-stage discriminant algorithm includes: If the time domain features in the effective analysis interval fall within the bubble feature threshold interval and the energy proportion of the frequency domain main frequency is greater than a preset proportion, the suspected bubble signal is preliminarily marked. If the model confidence is greater than a preset confidence and the sum of the time-frequency barycenter and the time-frequency entropy is greater than a preset threshold, the suspected bubble signal is confirmed as a bubble signal.

5. The method of claim 4, wherein The suspected bubble image processing strategy includes: based on the preliminary three-dimensional position output by the ultrasonic detection, performing targeted cropping, contrast enhancement and image normalization processing on the collected first suspected bubble image to obtain a first suspected bubble region; Processing the first suspected bubble region to strip out the edge profile of the first suspected bubble; Extracting shape features of at least one of the area, the perimeter, the circularity, the aspect ratio, the contour smoothness or the gray variance in the first suspected bubble profile; Calculating the mean of the center points of the suspected bubble profile by using the shape center method and the barycenter method to obtain the center point coordinates; If the shape features of the first suspected bubble profile are greater than a preset number and all fall within the bubble feature threshold interval, the suspected bubble is determined as a bubble; if the shape features are less than the preset number, the suspected bubble is directly determined as no bubble, and the false judgment of the ultrasonic signal is excluded. Establishing a conversion relationship between the image coordinates and the actual physical coordinates of the first suspected bubble image by using a camera calibration algorithm to obtain the actual physical coordinates of the axial and circumferential directions of the first suspected bubble; Combining the radial preliminary three-dimensional positioning coordinates of the ultrasonic detection with the actual physical coordinates of the axial and circumferential directions obtained by the image processing to output the three-dimensional determined position coordinates of the first suspected bubble.

6. The method of claim 5, wherein When the corrugated pipe is in a folded state, the weight of the shape feature is greater than the sum of the weights of the color feature of the colored area and the ultrasonic feature; when the corrugated pipe is in an unfolded state, the weight of the shape feature is less than the sum of the weights of the color feature of the colored area and the ultrasonic feature.

7. The method of claim 1, wherein The method further comprises: directing the suspected bubble at the determined position to blow at a preset pressure, collecting second suspected bubble image data at the determined position, extracting shape features of the second suspected bubble image data through a bubble image processing strategy, and comparing the shape features of the first suspected bubble image data with those of the second suspected bubble image data, if the shape features of the first suspected bubble image data are different from those of the second suspected bubble image data, determining that the suspected bubble is a real bubble; otherwise, not a real bubble.

8. The method of claim 7, wherein, The bubble image processing strategy comprises: pre-processing, contrast enhancement and image normalization processing of the collected second suspected bubble image to obtain a second suspected bubble region; wherein the second suspected bubble image is collected with the same acquisition parameters as the first suspected bubble image; The second suspected bubble region is processed to strip out the edge profile of the second suspected bubble; At least one shape feature of area, perimeter, circularity, aspect ratio, contour smoothness or gray scale variance in the second suspected bubble profile is extracted; wherein the features extracted from the second suspected bubble profile are the same as those extracted from the first suspected bubble profile, to ensure the homology of the comparison; At least one deformation feature of deformation coefficient, center displacement and contour change rate in the second suspected bubble profile is extracted; The absolute difference values of the shape features of area, perimeter, circularity, aspect ratio, contour smoothness or gray scale variance in the second suspected bubble profile are calculated, if more than a preset number of shape feature absolute difference values meet a preset threshold, the real bubble signal is preliminarily marked; if less than a preset number of shape feature absolute difference values meet a preset threshold, it is determined as a non-real bubble; If the second suspected bubble profile meets any of the following conditions: the deformation coefficient is greater than a preset threshold, or the center displacement is greater than a preset threshold, or the contour change rate is greater than a preset threshold, it is confirmed as a real bubble; The comprehensive confidence of the real bubble is calculated, when the comprehensive confidence of the real bubble is within a preset confidence range, the blowing pressure value at this time meets the preset blowing pressure value range; When the comprehensive confidence of the real bubble is less than the preset confidence range, the blowing pressure value is adjusted downward to avoid misjudgment caused by insufficient air flow effect.

9. A bellow tube face defect detection apparatus characterized by, The device is used in the corrugated pipe surface defect detection method of any one of claims 1-8, and the device comprises: A mobile mounting mechanism (1) for carrying all components of the device; At least one mounting platform group (2); each mounting platform group (2) is mounted on the mobile mounting mechanism (1), the first mounting platform (201) of the mounting platform group (2) is mounted on the fixed part of the mobile mounting mechanism (1), and the second mounting platform (202) of the mounting platform group (2) is mounted on the sliding part of the mobile mounting mechanism (1); At least two air bags (3) for adhering to the inner walls of both ends of the corrugated pipe after inflation; two air bags (3) are mounted on the first mounting platform (201) and the second mounting platform (202), respectively; An air inlet (4) for slowly and stably introducing a preset color tracer gas into the corrugated pipe; A telescopic cylinder (5) is installed on the mobile mounting mechanism (1) and one end of the telescopic cylinder (5) extends out of the air bag (3) through the first mounting platform (201); A rotary motor (6) is installed on the telescopic end of the telescopic cylinder (5) in the bellows; A gas nozzle (7) is installed on the rotary end of the rotary motor (6); the gas nozzle (7) is used for directional blowing of suspected bubbles when suspected bubbles are found on the wall of the bellows; A first industrial camera (8) is installed on the rotary end of the rotary motor (6) and is arranged side by side with the gas nozzle (7) along the axis of the bellows; A mounting bracket (9) is transversely arranged on the mobile mounting mechanism (1); A second industrial wide-angle camera (10) is hingedly arranged on the mounting bracket (9); An ultrasonic sensor (11) is installed on the end face of the air bag (3) on the first mounting platform (201); the ultrasonic sensor (11) is used for emitting ultrasonic waves in the bellows and receiving ultrasonic waves reflected by the inner wall of the bellows; When the two ends of the bellows are fixed on the device by the air bag (3) after inflation, the mobile mounting mechanism (1) drives the second mounting platform (202) to move, thereby stretching the corrugated part of the bellows and facilitating the detection of the pipe surface defects of the bellows.

10. The method of claim 9, wherein, The mobile mounting mechanism (1) comprises a mechanism mounting plate (101); A rotary drive motor (102) is installed on the mechanism mounting plate (101); A lead screw (103) is installed on the rotary end of the rotary drive motor (102); A nut (104) is installed on the lead screw (103); A mounting sliding block (105) is installed on the nut (104); The first mounting platform (201) is installed on the rotary drive motor (102), the second mounting platform (202) is installed on the mounting sliding block (105), and the telescopic cylinder (5) is installed on the rotary drive motor (102).

Images