Defect detection method for HDPE pipe production based on machine vision

Abstract

The invention discloses a defect detection method for HDPE (high-density polyethylene) pipe production based on machine vision, which comprises the following steps: before HDPE pipe welding trace defect detection, firstly setting an initial feeding speed to meet defect detection requirements under normal conditions; and high-precision image acquisition equipment is adopted to carry out real-time image capture on the HDPE pipe in transmission. According to the method, through high-precision image acquisition, intelligent feature extraction and convolutional neural network deep analysis, the complex mode of the welding trace is accurately recognized and is divided into a complex type and a conventional type, and a targeted detection strategy is adopted. The feeding speed is dynamically regulated and controlled according to different types of welding traces, the complex type reduces the speed to ensure the detection precision, the conventional type keeps the initial speed to improve the efficiency, the balance problem of the traditional constant feeding speed between the precision and the efficiency is effectively solved, and the quality control of HDPE pipe production and the operation efficiency of a production line are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of HDPE pipe production, and in particular to a defect detection method for HDPE pipe production based on machine vision. Background Art

[0002] Machine vision-based defect detection for HDPE pipe production is a system that uses high-precision image acquisition and processing technology to monitor and identify defects that may occur in the production process of HDPE (high-density polyethylene) pipes in real time. The system usually collects image data on the surface of the pipe through high-resolution cameras and optical sensors, and uses image processing algorithms (such as edge detection, texture analysis, defect classification, etc.) to analyze and determine whether there are defects in the image, such as cracks, bubbles, uneven surfaces, depressions, etc. Through machine learning or deep learning models, the system can automatically distinguish between normal and defective pipes, and locate or alarm defects in a timely manner. This detection technology not only improves the quality control accuracy in the production process, but also effectively reduces the workload of manual inspection, and improves production efficiency and product quality.

[0003] The weld marks of HDPE pipes are usually inconsistent because factors such as temperature, pressure, welding time and material fluidity involved in each welding process may be different, resulting in variability in the morphology and quality of the welded joints. When defect detection is based on machine vision, the existing technology usually uses a constant feed rate to transport the HDPE pipe and perform detection. When the weld marks are relatively simple and regular, the machine vision system can meet the needs of defect detection. However, if the weld marks are intricate, due to their unevenness and complexity, the machine vision system may have difficulty accurately capturing subtle defects, especially microcracks, bubbles or uneven welded joints. This will result in these defects not being identified in time, which in turn affects the final quality of the pipe and may even cause unqualified products to flow into the market. In addition, if the feed rate is too fast, the rapid changes in the weld marks may interfere with the image processing of the visual system, resulting in the system being unable to accurately identify defects, thereby misjudging flawless welded joints as defective, causing unnecessary waste and production delays.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a defect detection method for HDPE pipe production based on machine vision. Through high-precision image acquisition, intelligent feature extraction and deep analysis of convolutional neural networks, the complex patterns of welding marks can be accurately identified and divided into two categories: intricate and conventional types, and targeted detection strategies are adopted. The feed speed is intelligently adjusted according to the type of welding marks: the feed speed is dynamically reduced for intricate welding marks to ensure detection accuracy; the initial feed speed is maintained for conventional welding marks to improve detection efficiency. This dynamic control mechanism effectively solves the shortcomings of the traditional constant feed speed in balancing detection accuracy and efficiency, significantly improves the quality control level of HDPE pipe production and the overall production line operation efficiency, so as to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a defect detection method for HDPE pipe production based on machine vision, comprising the following steps: Before conducting HDPE pipe welding mark defect detection, first set an initial feed speed to meet the defect detection requirements under normal circumstances; High-precision image acquisition equipment is used to capture real-time images of HDPE pipes in transit. After obtaining clear image data, the welding marks in the image are intelligently identified and located. Extract features from the identified welding trace area, and after feature extraction, screen the extracted features to select key features that reflect the complexity of the welding traces, and after analyzing the selected features, preliminarily quantify the complexity of the welding trace area; The selected key features are input into the trained convolutional neural network for further intelligent analysis. The deep learning capability of the convolutional neural network can automatically identify the complex patterns of welding marks and further quantify the complexity of the welding mark area. According to the quantification results of the convolutional neural network, the welding marks are further divided into two types: intricate and conventional. Different types of defect detection measures are taken for different types of welding marks. The feed speed is intelligently controlled based on the analysis results of the convolutional neural network. Specifically, when the welding marks are detected to be complex, the feed rate is dynamically reduced to ensure the accuracy of the detection; when the welding marks are conventional, the initial feed speed is maintained to carry out transmission detection of the HDPE pipe to improve the detection efficiency.

[0007] Preferably, the welding traces in the image are intelligently identified and located, and the specific steps are as follows: Before intelligent recognition of welding marks, pre-process the image to improve image quality, eliminate background noise and enhance visibility of welding mark areas; The processed images are further analyzed by edge detection algorithms to identify the edges and contours of the welding marks, separate the welding marks, and highlight the shape, size and relative position of the welding joints; Once the welding mark area is successfully segmented, the welding mark is then accurately characterized and located. The welding mark is comprehensively analyzed to locate its exact position in the image.

[0008] Preferably, key features reflecting the complexity of the welding mark are selected, wherein the selected features include the local deformation degree of the molten pool and the frequency distribution of the surface texture. After analyzing the local deformation degree of the molten pool and the frequency distribution characteristics of the surface texture of the selected welding mark, a local molten pool deformation reference value and a surface texture high-frequency distribution reference value are generated respectively. The complexity of the welding mark area is preliminarily quantified by the local molten pool deformation reference value and the surface texture high-frequency distribution reference value.

[0009] Preferably, the screened local molten pool deformation reference value and surface texture high-frequency distribution reference value are input into a trained convolutional neural network, and the complex characteristic coefficient is automatically generated through the deep learning ability of the convolutional neural network. The complexity of the welding mark area is further quantified through the complex characteristic coefficient, which is used to further identify the complex pattern of the welding mark.

[0010] Preferably, the complex characteristic coefficient generated when the complexity of the welding mark area is quantified by the trained convolutional neural network is compared and analyzed with a preset complex characteristic coefficient reference threshold, and the welding mark is further divided, and the division steps are as follows: If the complex characteristic coefficient is greater than the complex characteristic coefficient reference threshold, the welding trace is further classified as an intricate and complex type; If the complex characteristic coefficient is less than or equal to the complex characteristic coefficient reference threshold, the welding mark is further classified as conventional.

[0011] Preferably, the specific steps of generating a local molten pool deformation reference value after analyzing the local deformation degree of the welding trace molten pool are as follows: The local molten pool deformation is calculated based on the small geometric changes of the molten pool surface using the local curvature and local deformation gradient. The local curvature reflects the bending degree of the molten pool surface, while the local deformation gradient reflects the rate of change of the molten pool morphology. The calculation expression of the local molten pool deformation is:

[0012] ,in: A is the area of ​​the weld joint, is the surface curvature, the local curvature of the reaction pool surface, is the local shape change gradient, which indicates the rate of change of the molten pool surface morphology. is the second-order gradient of curvature, reflecting the second-order change of local surface deformation, Indicates the local molten pool deformation; The local molten pool deformation degree is normalized by the spatial distribution characteristics of the molten pool deformation, and combined with the welding temperature to generate the final molten pool deformation index. The generation expression of the molten pool deformation index is:

[0013] ,in: represents the melt pool deformation index, is the pixel position of the welding mark area The temperature gradient at represents the spatial distribution change of the temperature field during welding. is the local melt pool deformation, and the geometric changes of the melt pool surface are calculated.

[0014] Preferably, the specific steps of generating a reference value of the high-frequency distribution of the surface texture after analyzing the frequency distribution characteristics of the surface texture of the welding trace are as follows: The texture of the welding mark area is decomposed into different frequency components by two-dimensional wavelet transform; Calculate the high-frequency coefficients of the original welding trace image obtained by two-dimensional wavelet transform. The calculation expression is:

[0015] ,in, is the mother wavelet function based on different scales and directions of pixel positions, is the coefficient of wavelet transform, which represents the frequency characteristics of the corresponding scale and direction in the image. k Indices representing different scales and directions of wavelet transform, N Represents the total number of scales and directions involved in the wavelet transform, is the pixel position The high frequency coefficients at Represents the high-frequency coefficient reference threshold, which is used to distinguish high-frequency coefficients; After extracting the high-frequency features, the distribution intensity of the high-frequency texture is calculated to quantify the complexity of the welding mark area. In order to quantify the distribution of high-frequency texture, the frequency distribution energy weighted model is introduced to calculate the reference value of the high-frequency distribution of the surface texture, which reflects the energy distribution of the frequency components in the welding mark and focuses on the energy concentration in the high-frequency area. The calculation expression of the reference value of the high-frequency distribution of the surface texture is:

[0016] ,in: is the reference value of high-frequency distribution of surface texture, is the weighting function of the frequency component, reflecting the contribution of the frequency component. is the spatial extent of the welding mark area.

[0017] Preferably, the feed speed is intelligently controlled based on the analysis results of the convolutional neural network, and the specific steps are as follows: According to the comparison result of the complex characteristic coefficient and the reference threshold, a feed speed adjustment factor is dynamically generated to adjust the feed rate of the current welding inspection. By mapping the size of the complex characteristic coefficient difference to the speed adjustment ratio, it is ensured that the intricate welding marks obtain a lower feed speed, while the conventional welding marks maintain the initial speed. The generation expression of the feed speed adjustment factor is:

[0018] in: is the difference of complex characteristic coefficient, , is the complex characteristic coefficient, obtained by the convolutional neural network through analyzing the welding traces. The reference threshold of the complex characteristic coefficient is used to define the demarcation standard of the complexity of welding marks. It is the feed speed adjustment factor, which is used to dynamically adjust the current feed speed. To adjust the amplitude coefficient, , used to control the range of feed speed reduction, is a sensitivity parameter used to control the influence of the complex characteristic coefficient difference on the feed speed adjustment factor. The hyperbolic tangent function is used to perform nonlinear mapping on the difference of complex characteristic coefficients to ensure smooth changes in the feed speed adjustment factor; The actual feed speed is dynamically calculated based on the feed speed adjustment factor and the initial feed speed, and the actual feed speed is applied to the welding detection process to achieve speed closed-loop control based on complex characteristic analysis, thereby improving efficiency while ensuring detection accuracy. The dynamic calculation expression of the actual feed speed is:

[0019] in, is the actual feed speed, i.e. the transmission speed applied in real time during welding inspection. is the initial feed speed, used for conventional welding mark transmission detection.

[0020] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention uses high-precision image acquisition, intelligent feature extraction and deep analysis of convolutional neural networks to accurately identify the complex patterns of welding marks, and classify the welding marks into two categories: intricate and conventional types according to their complexity, and then adopts targeted detection strategies.

[0021] The present invention intelligently and dynamically adjusts the feed speed for different types of welding marks. Specifically, for complex welding marks, the feed speed is dynamically reduced to ensure full detection and avoid missed detection or misjudgment; for conventional welding marks, the initial feed speed is maintained to improve detection efficiency. This dynamic control mechanism effectively overcomes the limitations of traditional constant feed speed in balancing detection accuracy and efficiency, and significantly improves the quality control level and overall production line operation efficiency in the HDPE pipe production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0023] Figure 1 The present invention is a flow chart of a defect detection method for HDPE pipe production based on machine vision. DETAILED DESCRIPTION

[0024] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0025] The present invention provides Figure 1 The defect detection method for HDPE pipe production based on machine vision shown includes the following steps: Before conducting HDPE pipe welding mark defect detection, first set an initial feed speed to meet the defect detection requirements under normal circumstances; Before conducting HDPE pipe welding mark defect detection, you first need to set an initial feed speed, which is determined based on the welding mark characteristics and detection requirements under normal circumstances. The purpose of setting this speed is to ensure that the machine vision system has enough time and space to capture images, identify welding marks, and effectively detect potential defects. This initial feed speed is a standard value, which is suitable for situations where welding marks are relatively simple and regular. It can meet the defect detection needs under normal circumstances, will not cause image blur or defect omission due to excessive speed, and can ensure efficient defect detection of relatively simple and regular welding marks.

[0026] High-precision image acquisition equipment is used to capture real-time images of HDPE pipes in transit. After obtaining clear image data, the welding marks in the image are intelligently identified and located. Intelligently identify and locate welding marks in the image. The specific steps are as follows: Before performing intelligent recognition of welding marks, the image needs to be preprocessed first, including denoising, contrast enhancement, and brightness adjustment. The core purpose of this step is to improve image quality, eliminate background noise, and enhance the visibility of the welding mark area to ensure the prominence of the welding mark in the image. Common denoising methods include Gaussian filtering and mean filtering. Through these preprocessing operations, the system can more accurately identify the contours and edges of welding marks and avoid recognition errors caused by noise interference. This step is particularly important when the welding marks are more complex or there are differences in reflection and gloss on the surface.

[0027] The processed image is further analyzed by edge detection algorithms (such as Canny edge detection or Sobel operator) to identify the edges and contours of the welding marks. Edge detection can help the system clearly determine the boundaries of the welding marks and then perform regional segmentation. The purpose of this step is to separate the welding marks from the background and other areas, highlighting the shape, size and relative position of the welding joint. Through precise regional segmentation, the system can locate the position of the welding marks on the entire HDPE pipe surface, providing clear regional information for subsequent defect detection and feature extraction, and avoiding confusion between welding marks and other pipe surface elements.

[0028] Once the weld mark area is successfully segmented, the next step is to accurately analyze and locate the features of the area. In this process, the system first conducts a comprehensive analysis of the weld mark's geometry (such as length, width, area), texture features (such as uniformity, continuity, symmetry), and color changes. Through multi-dimensional feature extraction, the system can accurately determine the overall structure and quality of the weld mark and locate its precise position in the image. This not only helps to clarify the spatial coordinates of the weld mark, but also identifies the morphological changes of the weld joint and further analyzes whether there are defects such as cracks and bubbles.

[0029] Extract features from the identified welding trace area, and after feature extraction, screen the extracted features to select key features that reflect the complexity of the welding traces, and after analyzing the selected features, preliminarily quantify the complexity of the welding trace area; The key features reflecting the complexity of the welding mark are selected, wherein the selected features include the local deformation degree of the molten pool and the frequency distribution of the surface texture. After analyzing the local deformation degree of the molten pool and the frequency distribution characteristics of the surface texture of the selected welding mark, the local molten pool deformation reference value and the surface texture high-frequency distribution reference value are generated respectively. The complexity of the welding mark area is preliminarily quantified by the local molten pool deformation reference value and the surface texture high-frequency distribution reference value.

[0030] The local deformation degree of the weld mark molten pool refers to the degree of local irregular deformation caused by heat and pressure in the molten pool area during the welding process. The molten pool is the area where the weld metal melts during the welding process and is usually directly affected by the welding heat source. Local deformation usually manifests itself as abnormal morphology at the edge of the molten pool, such as expansion, contraction, tilting or uneven flow of the molten pool, etc. These deformations are usually caused by factors such as uneven welding temperature, unstable welding speed, and poor material fluidity. Local deformation not only affects the appearance of the weld mark, but may also lead to a decrease in the strength and quality of the welded joint, especially it may cause defects such as cracks, pores, and incomplete fusion of the welded joint.

[0031] The frequency distribution of the surface texture of the weld mark refers to the spatial distribution law and frequency of texture features of different scales (such as fluctuations, unevenness, changes in grain, etc.) on the surface of the weld joint during the welding process. These texture features are determined by factors such as heat input, cooling rate, material fluidity, and stability of the welding process during welding. During the welding process, temperature changes and the dynamic flow of the molten pool will cause irregular textures on the surface of the material, which may appear uniformly or unevenly distributed within a certain range. The frequency distribution describes the frequency of occurrence of these texture features on the weld surface. When the frequency is high, it means that the surface texture changes are more delicate and frequent, which usually reflects that the fluctuations or instability in the welding process are greater, which may lead to abnormalities or defects in welding quality, such as microcracks, bubbles, or incomplete welding.

[0032] A large degree of local deformation of the weld mark molten pool usually means a high degree of complexity of the weld mark. The degree of deformation of the molten pool reflects the uneven heat input, changes in material fluidity, and fluctuations in the cooling process during welding. When the molten pool is severely deformed in certain areas, it usually indicates uneven temperature distribution or improper pressure control during welding, which may lead to unstable flow of molten material, which directly affects the morphology and quality of the welded joint. This local deformation often causes more complex defects, such as microcracks, pores, and uneven surface, which significantly increase the complexity of the weld mark. Therefore, the large deformation of the local molten pool is not only a manifestation of unstable welding process, but also usually means that there are greater risks and challenges in welding quality.

[0033] The specific steps for generating a local molten pool deformation reference value after analyzing the local deformation degree of the welding trace molten pool are as follows: The local molten pool deformation is used to measure the degree of deformation of the molten pool in the welding trace area. The complexity of the molten pool is quantified by analyzing the surface curvature changes, shape irregularities, and local geometric changes in the weld joint area. The deformation of the molten pool area is usually closely related to temperature non-uniformity, molten pool fluidity, and cooling rate. In this step, the local molten pool deformation is calculated based on the slight geometric changes on the molten pool surface using the local curvature and local deformation gradient, where the local curvature reflects the degree of curvature of the molten pool surface, and the local deformation gradient reflects the rate of change of the molten pool morphology. The calculation expression of the local molten pool deformation is:

[0034] ,in: A is the area of ​​the weld joint, is the surface curvature, the local curvature of the reaction pool surface, is the local shape change gradient, which indicates the rate of change of the molten pool surface morphology. is the second-order gradient of curvature, reflecting the second-order change of local surface deformation, Indicates the local molten pool deformation; Through the comprehensive analysis of local curvature and deformation gradient, the deformation degree of each area of ​​the molten pool during welding is obtained. The deformation characteristics of the molten pool can be quantified through the local molten pool deformation, revealing the complexity and irregularity of the molten pool in the local area, and further providing a quantitative basis for subsequent analysis.

[0035] The local molten pool deformation reference value further quantifies the complexity of the welding trace area by standardizing the local molten pool deformation. In order to adapt to the changes under different welding conditions, the local molten pool deformation is normalized by the spatial distribution characteristics of the molten pool deformation, and combined with its effect on the welding temperature to generate the final molten pool deformation index. This step combines the influence of the local molten pool deformation with the fluctuation of the welding thermal history to enhance the sensitivity to the dynamic changes of the molten pool. The generation expression of the molten pool deformation index is:

[0036] ,in: represents the melt pool deformation index, is the pixel position of the welding mark area The temperature gradient at represents the spatial distribution change of the temperature field during welding. is the local melt pool deformation, which calculates the geometric changes of the melt pool surface; By normalizing the ratio of molten pool deformation to temperature gradient, the non-uniformity of heat input during welding and the response of the molten pool under different temperature conditions are taken into account, and a relatively balanced molten pool deformation index is obtained; Through normalization processing, the local molten pool deformation is combined with the temperature gradient in the welding process to further enhance the reflection of non-uniform heat input and material fluidity in the welding process, thereby obtaining a more accurate local molten pool deformation reference value. The local molten pool deformation reference value can accurately reflect the complexity of the molten pool in the welding process, especially in the case of uneven heat input and inconsistent cooling rate, helping the machine vision system to dynamically adjust the detection strategy and improve the accuracy of defect recognition.

[0037] It can be seen from the local molten pool deformation reference value that the larger the performance value of the local molten pool deformation reference value generated after analyzing the local deformation degree of the weld mark molten pool, the higher the complexity of the weld mark. The local molten pool deformation reference value can accurately reflect the thermal non-uniformity, molten pool flow instability and local geometric deformation of the welding process by analyzing the local deformation degree of the molten pool during welding, taking into account the changes in the curvature and deformation gradient of the molten pool surface. When there is a large deformation in the molten pool area of ​​the weld mark, it usually means that the temperature fluctuation is large, the molten pool fluidity is poor or the cooling process is unstable, resulting in a more irregular and complex morphology of the weld joint. On the contrary, if the local molten pool deformation reference value is small, it means that the deformation of the molten pool during welding is small, the morphology is more regular, and the weld mark is relatively simple. Therefore, the complexity of the weld mark can be effectively judged by the numerical value of the local molten pool deformation reference value.

[0038] A high frequency distribution of surface texture in weld marks usually indicates a high degree of complexity in weld marks. The high frequency distribution of surface texture reflects subtle surface fluctuations and local inhomogeneities in the welding process, such as microcracks, bubbles, local overheating or uneven cooling, which often occur in complex or irregular weld joints. The presence of high frequency texture means that the surface has changed frequently at a smaller scale, which is usually closely related to welding quality problems (such as unstable weld metal flow, heat source fluctuations, etc.). Since these high frequency changes are usually subtle and not easily captured by traditional detection methods, they are a significant sign of the complexity of weld marks, indicating that there may be more serious defects or instabilities in the welding process.

[0039] The specific steps for generating the reference value of the high-frequency distribution of the surface texture after analyzing the frequency distribution characteristics of the welding trace surface texture are as follows: First, the surface texture of the welding mark is subjected to frequency analysis, focusing on the distribution characteristics of the high-frequency area. Through image processing algorithms, such as 2D Wavelet Transform or Fourier Transform, the texture of the welding mark area is decomposed into different frequency components, especially high-frequency components. High-frequency texture reflects the tiny and rapidly changing details of the surface, which is usually related to complex welding processes and defects (such as microcracks, bubbles, etc.). Wavelet transform is used to analyze the local frequency characteristics of the texture, because wavelet transform has stronger localization ability in the time-frequency domain and can accurately capture the subtle structure and local characteristics of the welding mark; Calculate the high-frequency coefficients of the original welding trace image obtained by two-dimensional wavelet transform. The calculation expression is:

[0040] ,in, is the mother wavelet function based on different scales and directions of pixel positions, is the coefficient of wavelet transform, which represents the frequency characteristics of the corresponding scale and direction in the image. k Indices representing different scales and directions of wavelet transform, N Represents the total number of scales and directions involved in the wavelet transform, is the pixel position The high frequency coefficients at Represents the high-frequency coefficient reference threshold, which is used to distinguish high-frequency coefficients; By the high frequency coefficient By performing calculations and threshold comparisons, the high-frequency variation portion in the welding trace area can be extracted, laying the foundation for subsequent frequency distribution analysis.

[0041] After extracting the high-frequency features, the complexity of the welding mark area is quantified by calculating the distribution intensity of the high-frequency texture. The high-frequency distribution reference value of the surface texture can reflect the complexity of the welding mark by integrating the energy distribution of the high-frequency texture at different scales. A higher frequency distribution usually indicates that there are more tiny inhomogeneities or defects in the welding joint area. In order to quantify this distribution, a frequency distribution energy weighting model is introduced to calculate the high-frequency distribution reference value of the surface texture. The frequency distribution energy weighting model calculates the energy distribution of the frequency components in the welding mark, paying special attention to the energy concentration in the high-frequency area. The calculation expression of the high-frequency distribution reference value of the surface texture is:

[0042] ,in: is the reference value of high-frequency distribution of surface texture, It is a weighted function of the frequency component, reflecting the contribution of the frequency component, and is usually calculated through filters of different scales and directions. is the spatial extent of the welding mark area; By weighted integration of high-frequency features, the reference value of high-frequency distribution of surface texture is calculated , a high value indicates that the welding mark has a high complexity. This method can effectively capture the subtle and complex changes in welding marks, especially defects such as microcracks and bubbles, and then quantify the intricate welding features.

[0043] It can be seen from the reference value of the high-frequency distribution of surface texture that the larger the performance value of the reference value of the high-frequency distribution of surface texture generated after analyzing the frequency distribution characteristics of the surface texture of the welding mark, the higher the complexity of the welding mark. The reason is that the reference value of the high-frequency distribution of surface texture reflects the intensity and frequency of subtle changes on the surface of the welding mark. A higher value means that there are more tiny irregularities, texture fluctuations and local unevenness on the surface of the welding joint, such as microcracks, bubbles, overheating or uneven cooling. These high-frequency changes are usually manifestations of unstable factors in the welding process, so their existence indicates that the complexity of the welding mark is high. A lower reference value of the high-frequency distribution of surface texture indicates that the surface of the welding mark is relatively smooth and regular, lacking subtle structural changes, which usually means that the welding process is relatively stable and of good quality. Therefore, the high or low reference value of the high-frequency distribution of surface texture is directly related to the quality and complexity of the welding mark.

[0044] The selected key features are input into the trained convolutional neural network (CNN) for further intelligent analysis. The deep learning ability of the convolutional neural network can automatically identify the complex patterns of welding marks and further quantify the complexity of the welding mark area. The screened local molten pool deformation reference value and surface texture high-frequency distribution reference value are input into the trained convolutional neural network. The complex characteristic coefficient is automatically generated through the deep learning ability of the convolutional neural network. The complexity of the welding mark area is further quantified through the complex characteristic coefficient, which is used to further identify the complex pattern of the welding mark.

[0045] A trained convolutional neural network refers to a deep learning model that has been pre-trained with a large number of welding mark data samples. The goal of this model is to evaluate the complexity of the welding mark area by automatically extracting and analyzing the pattern of input features (such as the reference value of the local molten pool deformation after screening and the reference value of the high-frequency distribution of the surface texture), and to generate complex characteristic coefficients that can quantify the complexity of the welding mark area. Convolutional Neural Network (CNN) is a deep learning architecture commonly used for image and feature analysis. It learns the feature expression in the input data through a multi-layer stack of convolutional layers, pooling layers, and fully connected layers, and outputs an accurate judgment of the target. Here, the trained CNN already has the ability to recognize complex patterns of welding marks.

[0046] Specifically, the trained convolutional neural network is trained offline based on a large-scale, high-quality welding mark dataset. During the training process, the model receives a large amount of annotated data, which includes the real complexity information of different welding mark areas (such as microcracks, bubbles, uneven texture, overmelting, etc.). Through repeated iterations, the CNN model gradually learns the association between the input features (reference values ​​of local molten pool deformation and reference values ​​of high-frequency distribution of surface texture) and the complex welding pattern. In the convolution layer, the model extracts the local spatial relationship of the input features through the convolution kernel, such as the distribution of high-frequency textures in welding marks and the changes in the local molten pool morphology; in the pooling layer, the model performs dimensionality reduction on the extracted local features to highlight the significance of the main pattern; in the fully connected layer, the model conducts a comprehensive analysis of all features and finally predicts the complex characteristic coefficient of the welding mark area. This data-driven training process enables the model to have a strong generalization ability and can be applied to new welding mark data to achieve automatic recognition of complex patterns and quantification of complexity.

[0047] The trained convolutional neural network runs in inference mode in practical applications, which means that the model has frozen all parameters and no longer needs to adjust the weights. By inputting the filtered local molten pool deformation reference value and the surface texture high-frequency distribution reference value, CNN can quickly calculate the complex characteristic coefficient. Specifically, the fully trained model can capture the potential nonlinear associations and complex patterns in the input features, such as how high-frequency distribution features affect the detailed changes in the welding molten pool, and how the local molten pool deformation reflects the uneven distribution of the heat source. This inference process is efficient and suitable for online real-time analysis. The trained CNN can not only improve the accuracy and efficiency of the complexity analysis of the welding trace area, but also avoid the subjectivity of traditional manual analysis, laying a solid foundation for the subsequent intelligent control of welding quality.

[0048] The convolutional neural network is not specifically limited here, and can realize the deformation index of the molten pool and surface texture high frequency distribution reference value Perform comprehensive analysis to generate complex characteristic coefficients In order to realize the technical solution of the present invention, the present invention provides a specific implementation method; Complex characteristic coefficient The resulting calculation formula is:

[0049] , where , Melt pool deformation index and surface texture high frequency distribution reference value The preset scaling factor of , Both are greater than 0.

[0050] Preset scale factor , These are parameters used to adjust the relative importance of two core indicators (i.e., the molten pool deformation index and the reference value of the high-frequency distribution of the surface texture) in calculating the complex characteristic coefficient. Their main function is to assign weights to different indicators to ensure that in the calculation of the complex characteristic coefficient, the influence of a certain indicator can be reasonably highlighted according to the needs of the actual welding scene, rather than simply treating the two indicators equally.

[0051] Specifically, the preset proportionality factor , The size of determines the contribution of the molten pool deformation index and the reference value of the high-frequency distribution of the surface texture to the overall complex characteristic coefficient. For example, if the influence of the molten pool deformation on the complexity during welding is more significant, it can be appropriately increased. The value of can be increased to amplify the weight of the molten pool deformation index; on the contrary, if the high-frequency characteristics of the surface texture can better reflect the complexity of the welding mark area, the The value of is used to enhance the influence of the reference value of the high-frequency distribution of the surface texture. The setting of the preset proportional coefficient is usually based on experimental data or engineering experience, and can be dynamically adjusted according to actual conditions to more accurately reflect the complex characteristics of the welding mark area.

[0052] It can be seen from the complex characteristic coefficient that the greater the expression value of the local molten pool deformation reference value generated after analyzing the local deformation degree of the welding mark molten pool, the greater the expression value of the surface texture high-frequency distribution reference value generated after analyzing the frequency distribution characteristics of the surface texture of the welding mark, that is, the greater the expression value of the complex characteristic coefficient generated when the complexity of the welding mark area is quantified by the trained convolutional neural network, the higher the complexity of the welding mark, and vice versa.

[0053] According to the quantification results of the convolutional neural network, the welding marks are further divided into two types: intricate and conventional. Different types of defect detection measures are taken for different types of welding marks. The complex characteristic coefficient generated when the trained convolutional neural network quantifies the complexity of the welding mark area is compared and analyzed with the pre-set complex characteristic coefficient reference threshold, and the welding marks are further divided. The division steps are as follows: If the complex characteristic coefficient is greater than the complex characteristic coefficient reference threshold, the welding trace is further classified as an intricate and complex type; If the complex characteristic coefficient is less than or equal to the complex characteristic coefficient reference threshold, the welding mark is further classified as conventional.

[0054] Complex welding marks refer to welding surfaces with significant irregularities and high complexity, usually characterized by dramatic changes in details, dense high-frequency distribution of textures, and uneven energy distribution. This type of welding mark is usually accompanied by defects such as microcracks, bubbles, surface fluctuations, and local molten pool deformation, reflecting possible unstable factors in the welding process, such as heat input fluctuations, insufficient material fluidity, or uneven cooling. These characteristics make welding marks difficult to predict and control, and belong to high-risk areas, requiring more accurate detection methods.

[0055] Conventional weld marks refer to weld areas with relatively regular and uniform weld surfaces, simple texture features, and low frequency distribution. Such weld marks usually reflect the stability of the welding process, such as good control of parameters such as heat source, pressure, and material flow. There may only be a small number of minor defects on the surface (such as slight asymmetry or local texture changes), but these defects will not significantly affect the welding quality and structural strength. Conventional weld marks usually do not require overly complex detection strategies to complete quality assessment.

[0056] Based on the analysis results of the convolutional neural network, the feed speed is intelligently controlled. Specifically, when the welding marks are detected to be intricate, the feed rate is dynamically reduced to ensure the accuracy of the detection; when the welding marks are conventional, the initial feed speed is maintained to transmit and detect the HDPE pipe to improve the detection efficiency. Based on the analysis results of the convolutional neural network, the feed speed is intelligently controlled. The specific steps are as follows: According to the comparison result of the complex characteristic coefficient and the reference threshold, a feed speed adjustment factor is dynamically generated to adjust the feed rate of the current welding inspection. By mapping the size of the complex characteristic coefficient difference to the speed adjustment ratio, it is ensured that the intricate welding marks obtain a lower feed speed, while the conventional welding marks maintain the initial speed. The generation expression of the feed speed adjustment factor is:

[0057] ,in: is the difference of complex characteristic coefficient, , is the complex characteristic coefficient, obtained by the convolutional neural network through analyzing the welding traces. It is the reference threshold of the complex characteristic coefficient and defines the demarcation standard of the complexity of welding traces. It is usually set according to the welding process requirements and detection accuracy requirements. It is the feed speed adjustment factor, which is used to dynamically adjust the current feed speed. To adjust the amplitude coefficient, , used to control the range of feed speed reduction, It is a sensitivity parameter, which is used to control the influence of the complex characteristic coefficient difference on the feed speed adjustment factor. The larger the value, the more sensitive the system is to the changes in complex characteristics. The hyperbolic tangent function is used to perform nonlinear mapping on the complex characteristic coefficient difference to ensure smooth changes in the feed speed adjustment factor.

[0058] The actual feed speed is dynamically calculated based on the feed speed adjustment factor and the initial feed speed, and the actual feed speed is applied to the welding detection process to achieve speed closed-loop control based on complex characteristic analysis, thereby improving efficiency while ensuring detection accuracy. The dynamic calculation expression of the actual feed speed is:

[0059] ,in, is the actual feed speed, i.e. the transmission speed applied in real time during welding inspection. It is the initial feed speed, the preset standard speed, used for transmission detection of conventional welding marks.

[0060] Based on the analysis results of the convolutional neural network, the feed speed is intelligently controlled to achieve a dynamic balance between detection accuracy and detection efficiency. When the welding marks are identified as complex, dynamically reducing the feed rate can provide the machine vision system with more time to collect images and conduct in-depth analysis, thereby accurately identifying tiny defects (such as microcracks, bubbles or uneven welding, etc.), effectively avoiding missed detection and misjudgment; and when the welding marks are identified as conventional, maintaining the initial feed speed can ensure the efficient operation of the production line and avoid affecting the overall efficiency due to unnecessary deceleration. Through this dynamic control mechanism, the detection strategy can be flexibly adjusted according to the complexity of the welding marks, which not only ensures high-precision detection in complex areas, but also maximizes production efficiency and resource utilization.

[0061] Through the above-mentioned machine vision-based HDPE pipe welding mark defect detection solution, a dynamic balance between detection accuracy and production efficiency can be achieved. Through high-precision image acquisition, intelligent feature extraction and in-depth analysis of convolutional neural networks, the solution can accurately identify the complex patterns of welding marks, and divide them into two categories: intricate and conventional according to complexity, and take targeted detection measures. At the same time, the feed speed is intelligently controlled based on the type of welding marks. While ensuring that complex welding marks are fully detected and avoiding missed detection or misjudgment, conventional welding marks maintain the initial speed to improve detection efficiency. This dynamic adjustment mechanism effectively solves the problem in the existing technology that the constant feed rate cannot take into account both detection accuracy and efficiency, thereby significantly improving the quality control level of HDPE pipe production and the overall production line operation efficiency.

[0062] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0063] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0064] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A defect detection method for HDPE pipe production based on machine vision, characterized in that: The following steps are involved: Before conducting HDPE pipe welding mark defect detection, first set an initial feed speed to meet the defect detection requirements under normal circumstances; High-precision image acquisition equipment is used to capture real-time images of HDPE pipes in transit. After obtaining clear image data, the welding marks in the image are intelligently identified and located. Extract features from the identified welding trace area, and after feature extraction, screen the extracted features to select key features that reflect the complexity of the welding traces, and after analyzing the selected features, preliminarily quantify the complexity of the welding trace area; The selected key features are input into the trained convolutional neural network for further intelligent analysis. The deep learning capability of the convolutional neural network can automatically identify the complex patterns of welding marks and further quantify the complexity of the welding mark area. According to the quantification results of the convolutional neural network, the welding marks are further divided into two types: intricate and conventional. Different types of defect detection measures are taken for different types of welding marks. The feed speed is intelligently controlled based on the analysis results of the convolutional neural network. Specifically, when the welding marks are detected to be complex, the feed rate is dynamically reduced to ensure the accuracy of the detection; when the welding marks are conventional, the initial feed speed is maintained to carry out transmission detection of the HDPE pipe to improve the detection efficiency.

2. The method for defect detection in HDPE pipe production based on machine vision according to claim 1, characterized in that: Intelligently identify and locate welding marks in the image. The specific steps are as follows: Before intelligent recognition of welding marks, pre-process the image to improve image quality, eliminate background noise and enhance visibility of welding mark areas; The processed images are further analyzed by edge detection algorithms to identify the edges and contours of the welding marks, separate the welding marks, and highlight the shape, size and relative position of the welding joints; Once the welding mark area is successfully segmented, the welding mark is then accurately characterized and located. The welding mark is comprehensively analyzed to locate its exact position in the image.

3. The method for defect detection in HDPE pipe production based on machine vision according to claim 1, characterized in that: The key features reflecting the complexity of the welding mark are selected, wherein the selected features include the local deformation degree of the molten pool and the frequency distribution of the surface texture. After analyzing the local deformation degree of the molten pool and the frequency distribution characteristics of the surface texture of the selected welding mark, the local molten pool deformation reference value and the surface texture high-frequency distribution reference value are generated respectively. The complexity of the welding mark area is preliminarily quantified by the local molten pool deformation reference value and the surface texture high-frequency distribution reference value.

4. The method for defect detection for HDPE pipe production based on machine vision according to claim 3, characterized in that: The screened local molten pool deformation reference value and surface texture high-frequency distribution reference value are input into the trained convolutional neural network. The complex characteristic coefficient is automatically generated through the deep learning ability of the convolutional neural network. The complexity of the welding mark area is further quantified through the complex characteristic coefficient, which is used to further identify the complex pattern of the welding mark.

5. The method for defect detection for HDPE pipe production based on machine vision according to claim 4, characterized in that: The complex characteristic coefficient generated when the trained convolutional neural network quantifies the complexity of the welding mark area is compared and analyzed with the pre-set complex characteristic coefficient reference threshold, and the welding marks are further divided. The division steps are as follows: If the complex characteristic coefficient is greater than the complex characteristic coefficient reference threshold, the welding trace is further classified as an intricate and complex type; If the complex characteristic coefficient is less than or equal to the complex characteristic coefficient reference threshold, the welding mark is further classified as conventional.

6. The method for defect detection for HDPE pipe production based on machine vision according to claim 3, characterized in that: The specific steps for generating a local molten pool deformation reference value after analyzing the local deformation degree of the welding trace molten pool are as follows: The local molten pool deformation is calculated based on the small geometric changes of the molten pool surface using the local curvature and local deformation gradient. The local curvature reflects the bending degree of the molten pool surface, while the local deformation gradient reflects the rate of change of the molten pool morphology. The calculation expression of the local molten pool deformation is: ； in: A is the area of ​​the weld joint, is the surface curvature, the local curvature of the reaction pool surface, is the local shape change gradient, which indicates the rate of change of the molten pool surface morphology. is the second-order gradient of curvature, reflecting the second-order change of local surface deformation, Indicates the local molten pool deformation; The local molten pool deformation degree is normalized by the spatial distribution characteristics of the molten pool deformation, and combined with the welding temperature to generate the final molten pool deformation index. The generation expression of the molten pool deformation index is: ； in: represents the melt pool deformation index, is the pixel position of the welding mark area The temperature gradient at represents the spatial distribution change of the temperature field during welding. is the local melt pool deformation, and the geometric changes of the melt pool surface are calculated.

7. The method for defect detection for HDPE pipe production based on machine vision according to claim 6, characterized in that: The specific steps for generating the reference value of the high-frequency distribution of the surface texture after analyzing the frequency distribution characteristics of the welding trace surface texture are as follows: The texture of the welding mark area is decomposed into different frequency components by two-dimensional wavelet transform; Calculate the high-frequency coefficients of the original welding trace image obtained by two-dimensional wavelet transform. The calculation expression is: ； in, is the mother wavelet function based on different scales and directions of pixel positions, is the coefficient of wavelet transform, which represents the frequency characteristics of the corresponding scale and direction in the image. k Indices representing different scales and directions of wavelet transform, N Represents the total number of scales and directions involved in the wavelet transform, is the pixel position The high frequency coefficients at Represents the high-frequency coefficient reference threshold, which is used to distinguish high-frequency coefficients; After extracting the high-frequency features, the distribution intensity of the high-frequency texture is calculated to quantify the complexity of the welding mark area. In order to quantify the distribution of high-frequency texture, the frequency distribution energy weighted model is introduced to calculate the reference value of the high-frequency distribution of the surface texture, which reflects the energy distribution of the frequency components in the welding mark and focuses on the energy concentration in the high-frequency area. The calculation expression of the reference value of the high-frequency distribution of the surface texture is: ； in: is the reference value of high frequency distribution of surface texture, is the weighting function of the frequency component, reflecting the contribution of the frequency component. is the spatial extent of the welding mark area.

8. The method for defect detection for HDPE pipe production based on machine vision according to claim 5, characterized in that: Based on the analysis results of the convolutional neural network, the feed speed is intelligently controlled. The specific steps are as follows: According to the comparison result of the complex characteristic coefficient and the reference threshold, a feed speed adjustment factor is dynamically generated to adjust the feed rate of the current welding inspection. By mapping the size of the complex characteristic coefficient difference to the speed adjustment ratio, it is ensured that the intricate welding marks obtain a lower feed speed, while the conventional welding marks maintain the initial speed. The generation expression of the feed speed adjustment factor is: ； in: is the difference of complex characteristic coefficient, , is the complex characteristic coefficient, obtained by the convolutional neural network through analyzing the welding traces. The reference threshold of the complex characteristic coefficient is used to define the demarcation standard of the complexity of welding marks. It is the feed speed adjustment factor, which is used to dynamically adjust the current feed speed. To adjust the amplitude coefficient, , used to control the range of feed speed reduction, is a sensitivity parameter used to control the influence of the complex characteristic coefficient difference on the feed speed adjustment factor. The hyperbolic tangent function is used to perform nonlinear mapping on the difference of complex characteristic coefficients to ensure smooth changes in the feed speed adjustment factor; The actual feed speed is dynamically calculated based on the feed speed adjustment factor and the initial feed speed, and the actual feed speed is applied to the welding detection process to achieve speed closed-loop control based on complex characteristic analysis, thereby improving efficiency while ensuring detection accuracy. The dynamic calculation expression of the actual feed speed is: ； in, is the actual feed speed, i.e. the transmission speed applied in real time during welding inspection. is the initial feed speed, used for conventional welding mark transmission detection.

Images