Online detection system and method for surface defects of petroleum pipe fitting fused with multi-scale characteristics

Abstract

The invention belongs to the technical field of defect detection, and discloses a petroleum pipe fitting surface defect online detection system and method fused with multi-scale characteristics. Comprising the following steps: collecting multivariate detection data of a petroleum pipe fitting, and performing data cleaning to obtain a multivariate original data set; carrying out oil film boundary identification and oil film influence suppression, and outputting an oil film suppression data set; identifying a roughness area and an electromagnetic anomaly response area; executing space projection correction to generate a multi-channel space alignment data set; conjoint analysis is carried out, boundary response features of the weld joint structure area are identified, and a weld joint feature subset is output; carrying out distribution comparison, and outputting a corrosion boundary area; carrying out physical consistency analysis, and outputting a credible defect fusion block set; establishing a defect judgment rule to execute defect detection, and sending a detection result to a preset online monitoring terminal; and the multi-physical quantity fusion perception capability and the multi-scale feature collaborative analysis precision of the online detection of the surface defects of the petroleum pipe fitting are improved.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and more specifically, to an online detection system and method for surface defects of oil pipe fittings that integrates multi-scale features. Background Technology

[0002] Surface defect detection of oil pipe fittings is an important stage and step in the manufacturing and operation of oil pipelines. Multi-physical quantity fusion sensing is used in this process to accurately distinguish between minute defects and false defects, ensuring pipeline service safety and quality control. However, traditional defect detection methods still face many technical bottlenecks when dealing with more complex surface conditions and defects of oil pipe fittings.

[0003] In real-world scenarios, oil pipe fittings often have uneven oil films remaining on their surfaces due to processing or protective techniques. The presence of these oil films can easily lead to distortions in infrared emissivity and optical reflectivity. For example, a microcrack at the same location, covered by a thick oil film, may appear as a uniform temperature area in infrared detection, but as a low-reflectivity area in optical detection. However, traditional defect detection methods neglect the multi-physics response distortion caused by the oil film, failing to accurately separate oil film interference from the actual defect signal, easily leading to missed defects or misidentification of the oil film as a defect. Furthermore, surface roughness differences in pipe fittings can cause spatial misalignment between optical and electromagnetic detection at the same location. Areas with abrupt roughness changes may have clear boundaries in optical images, but their position may shift in eddy current response due to the lift-off effect. However, traditional defect detection methods do not dynamically compensate for the physical correspondence between roughness levels and the lift-off effect, often resulting in registration errors. Accumulated differences lead to the failure of multi-channel feature fusion, resulting in incorrect location judgment of microcracks. Simultaneously, the gradual change in metallographic structure in the weld area causes exponential decay of magnetic permeability and abnormal changes in thermal diffusivity. Traditional defect detection methods lack the ability to perceive the background of this gradual change when judging anomalies, easily leading to load offset phenomena such as small-scale magnetic flux leakage defects being masked by the background change or large-scale thermal anomaly areas being vaguely judged. Furthermore, the corrosion transition zone exhibits physical response differences between optical color gradation and electrochemical impedance step changes. Traditional defect detection methods lack the ability to coordinate the analysis of both gradation and step boundary models, failing to verify the physical consistency of microcracks within the transition zone. This results in consistently high false alarms or high false negatives in defect detection in such areas, making it difficult to guarantee the reliability of detection results under complex operating conditions and affecting the overall quality inspection efficiency of the oil pipe fitting production line.

[0004] In view of this, the present invention proposes an online detection system and method for surface defects of oil pipe fittings that integrates multi-scale features to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned shortcomings of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an online detection method for surface defects of oil pipe fittings integrating multi-scale features, comprising: S1. Collect multivariate inspection data of oil pipe fittings and perform data cleaning to obtain multivariate raw datasets, wherein the multivariate raw datasets include optical feature image data, surface electrical response data and pipe surface temperature data; S2. Combine optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppress oil film influence in the oil film boundary area, adjust multi-level channel values, and output oil film suppression dataset. S3. Extract the suppression optical image from the oil film suppression dataset, combine it with the surface electrical response data to extract the roughness image features and electromagnetic induction change features in the suppression optical image, and identify the roughness region and electromagnetic anomaly response region; perform spatial projection correction based on the difference in position information between the roughness region and the electromagnetic anomaly response region to generate a multi-channel spatially aligned dataset. S4. Extract data slices containing weld structure regions from the multi-channel spatially aligned dataset. Perform joint analysis on the data slices based on the temperature data and electromagnetic response data of the weld structure regions to identify the boundary response characteristics of the weld structure regions and output the weld feature subset. S5. Identify image blocks containing rusted regions in a multi-channel spatial alignment dataset, compare the distribution of these image blocks, and output the rusted boundary regions; perform physical consistency analysis on the rusted boundary regions and output a set of reliable defect fusion blocks; S6. By combining the weld feature subset, the set of credible defect fusion blocks, and the multi-channel spatial alignment dataset, establish defect judgment rules to perform defect detection, and send the detection results to the preset online monitoring terminal.

[0006] On the other hand, an online detection system for surface defects in oil pipe fittings that integrates multi-scale features, used to realize an online detection method for surface defects in oil pipe fittings that integrates multi-scale features, includes: The data acquisition module collects multivariate inspection data of oil pipe fittings and performs data cleaning to obtain a multivariate raw dataset, which includes optical feature image data, surface electrical response data and pipe surface temperature data. The oil film suppression module combines optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppresses oil film influence within the oil film boundary region, adjusts multi-level channel values, and outputs an oil film suppression dataset. The spatial registration module extracts the suppressed optical image from the oil film suppression dataset, and combines it with the surface electrical response data to extract roughness image features and electromagnetic induction change features from the suppressed optical image, identifying roughness regions and electromagnetic anomaly response regions; based on the difference in positional information between roughness regions and electromagnetic anomaly response regions, spatial projection correction is performed to generate a multi-channel spatial alignment dataset. The weld anomaly decoupling module extracts data slices containing weld structure regions from the multi-channel spatially aligned dataset. Based on the temperature and electromagnetic response data of the weld structure regions, it performs joint analysis on the data slices to identify the boundary response characteristics of the weld structure regions and outputs a subset of weld features. The corrosion recognition module identifies image blocks containing corrosion regions in a multi-channel spatially aligned dataset, compares the distribution of these image blocks, and outputs the corrosion boundary regions. It then performs physical consistency analysis on the corrosion boundary regions and outputs a set of reliable defect fusion blocks. The joint detection module combines a subset of weld features, a set of reliable defect fusion blocks, and a multi-channel spatially aligned dataset to establish defect judgment rules, perform defect detection, and send the detection results to a preset online monitoring terminal. The various modules are connected via wired and / or wireless means. The technical effects and advantages of the online detection system and method for surface defects of oil pipe fittings that integrates multi-scale features as described in this invention are as follows: By acquiring a multi-dimensional inspection dataset of oil pipe fittings in real time, and subsequently performing oil film suppression, spatial registration, weld joint analysis, and physical consistency analysis of rusted areas based on this dataset, an online inspection system and method for surface defects of oil pipe fittings integrating multi-scale features was realized. Compared with existing technologies, the accuracy of oil film coverage area boundary identification was improved by establishing a spatial overlap determination between low reflectivity areas and low temperature areas; the multi-channel response coupling distortion caused by uneven oil film thickness was eliminated by physically compensating for optical and infrared channels; and the multi-channel distortion caused by surface roughness differences was eliminated by extracting the spatial offset between roughness areas and electromagnetic anomaly response areas and constructing a coordinate transformation function in conjunction with lift-off effect compensation. This study addresses the issue of spatial registration failure; it achieves gradient background stripping by fitting the magnetic permeability decay curve and identifying the thermal diffusivity change boundary, thus preventing minor defects in the weld heat-affected zone from being masked by the gradient background; it optimizes the fusion conflict between the optical gradient boundary and the electrochemical step boundary in the corrosion transition zone by fitting the gradient trend curve and the step response curve to the color gradient vector and the impedance vector, respectively; and it achieves accurate positioning of credible defect areas in the transition zone through multiple verifications of response direction determination and dual-channel response ratio screening. This enhances the multi-physical quantity fusion sensing capability and multi-scale feature collaborative analysis accuracy of online detection of surface defects in oil pipe fittings, ensuring cross-channel response consistency and detection reliability under complex working conditions. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the online detection method for surface defects of oil pipe fittings that integrates multi-scale features according to the present invention. Figure 2 This is a schematic diagram of the online detection system for surface defects of oil pipe fittings that integrates multi-scale features according to the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] Example 1 Please see Figure 1 As shown in this embodiment, the online detection method for surface defects of oil pipe fittings that integrates multi-scale features includes: S1. Collect multivariate inspection data of oil pipe fittings and perform data cleaning to obtain multivariate raw datasets, wherein the multivariate raw datasets include optical feature image data, surface electrical response data and pipe surface temperature data; S2. Combine optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppress oil film influence in the oil film boundary area, adjust multi-level channel values, and output oil film suppression dataset. S3. Extract the suppression optical image from the oil film suppression dataset, combine it with the surface electrical response data to extract the roughness image features and electromagnetic induction change features in the suppression optical image, and identify the roughness region and electromagnetic anomaly response region; perform spatial projection correction based on the difference in position information between the roughness region and the electromagnetic anomaly response region to generate a multi-channel spatially aligned dataset. S4. Extract data slices containing weld structure regions from the multi-channel spatially aligned dataset. Perform joint analysis on the data slices based on the temperature data and electromagnetic response data of the weld structure regions to identify the boundary response characteristics of the weld structure regions and output the weld feature subset. S5. Identify image blocks containing rusted regions in a multi-channel spatial alignment dataset, compare the distribution of these image blocks, and output the rusted boundary regions; perform physical consistency analysis on the rusted boundary regions and output a set of reliable defect fusion blocks; S6. By combining the weld feature subset, the set of credible defect fusion blocks, and the multi-channel spatial alignment dataset, establish defect judgment rules to perform defect detection, and send the detection results to the preset online monitoring terminal.

[0010] In this embodiment, the multivariate detection data of oil pipe fittings is acquired in real time using imaging equipment and various sensors. The optical feature image data includes both image-type data acquired using imaging equipment such as high-precision industrial cameras and optical parameters such as brightness values. The surface electrical response data refers to parameters related to the electrical response of the pipe fitting surface, such as conductivity, permeability, and induced impedance values, acquired using eddy current sensors. The pipe surface temperature data refers to the radiation temperature of the pipe fitting surface acquired by temperature sensors. The multivariate detection data is cleaned by noise filtering and missing value imputation to generate a higher quality multivariate raw dataset.

[0011] Methods for identifying oil film boundaries include: The pixel brightness value of each optical image in the optical feature image data is extracted, and the incident intensity parameter of the light source is obtained. The surface reflectivity of the pipe is calculated based on the pixel brightness value and the incident intensity parameter of the light source.

[0012] This process involves extracting the brightness value of each pixel in each optical image and determining the incident intensity of the light source based on the settings of the imaging device. Then, it uses existing optical physics knowledge or formulas to calculate the reflectivity of the pipe surface and uses this value to replace the brightness value as a more physically meaningful data basis.

[0013] Pixels with surface reflectivity below the first reflectivity threshold are identified and aggregated to generate low reflectivity regions.

[0014] Based on existing knowledge in the optical field and pipe attribute information, a first reflection threshold is set to filter pixels with low reflectivity on the pipe surface. In this embodiment, a region connectivity algorithm is used to connect and aggregate these pixels to form several individual or connected regions, which are low reflectivity regions. These regions are used to filter out disturbances on the pipe surface that are not oil film-like. It should be noted that the optical image in this embodiment may include multiple shooting angles, but the corresponding regions will not differ due to images from different shooting angles. The generated regions are the regions at the corresponding positions on the oil pipe.

[0015] The radiation temperature of each coordinate sample point in the pipe surface temperature data is extracted and compared with the second temperature reference. The sample point area where the pipe surface radiation temperature is lower than the second temperature reference is extracted as the low temperature area.

[0016] The pipeline surface temperature data is based on the spatial structure of the oil pipe fittings. Each coordinate sample point is the corresponding position on the surface of the oil pipe fitting. At the same time, a reasonable second temperature benchmark is set based on the oil pipe fitting operation and maintenance specifications. The radiation temperature of each coordinate sample point is compared with the second temperature benchmark. For all coordinate sample points that are lower than the second temperature benchmark, the region connectivity algorithm is used to aggregate them to form several low-temperature regions, which are used to initially locate the areas in the infrared channel where the signal is reduced due to the influence of oil film.

[0017] Establish the spatial coordinate correspondence between low reflectivity regions and low temperature regions, calculate the overlapping area of ​​the regions, and select pipe surface regions with an overlapping area higher than the preset spatial overlap threshold as joint response regions.

[0018] By mapping the low reflectivity region and the low temperature region to a unified spatial coordinate system corresponding to the oil pipe fitting, a spatial coordinate correspondence between the two regions is established to ensure that coordinate confusion does not occur. Based on historical oil film boundary identification experience, a reasonable preset spatial overlap threshold is set. If the overlap area between the low reflectivity region and the low temperature region is higher than the threshold, it indicates that the corresponding region has a composite abnormal signal of optical channel and infrared channel temperature at the same time. Therefore, the corresponding region is regarded as the joint response region.

[0019] A multi-scale sliding window is constructed to jointly calculate the local reflectivity gradient and thermal response gradient of the joint response region, outputting the joint gradient map of the pipe surface. Spatial regions in the joint gradient map whose gradient values ​​reach the preset gradient intensity are selected as the brightness-temperature boundary transition region.

[0020] In this embodiment, sliding windows of various scales are constructed to cover the corresponding joint response region, such as sliding windows of various sizes like 3*3, 5*5, and 7*7, to enhance adaptability to different morphological boundaries in the corresponding region. For each sliding window, the gradient vector field of reflectivity of the optical channel and the thermal diffusion gradient of radiation temperature of the infrared channel are calculated using existing relevant knowledge or formulas. The two values ​​are weighted and combined according to the corresponding channel weights based on historical experience to form a joint gradient value, which is then mapped to the corresponding region to obtain a joint gradient map.

[0021] The preset gradient intensity is set based on the physical properties of the oil film. The spatial region formed by several spatial positions in the joint gradient map where the joint gradient value is higher than the gradient intensity is extracted is the brightness-temperature boundary transition region.

[0022] Based on the gradient direction of the joint gradient in the brightness-temperature boundary transition region, a boundary contour curve is constructed, and the oil film boundary region is output. In this embodiment, the edge tracking method is used to derive the contour path curve based on the brightness-temperature boundary transition region, construct the boundary contour curve of the region, and identify the closed oil film boundary region.

[0023] Methods for suppressing the effects of oil film include: Calculate the surface reflectance of the oil film at each pixel in the oil film boundary region, normalize it by combining it with the ambient brightness reference mean, and output the corrected reflectance of the corresponding pixel.

[0024] The ambient brightness reference mean is obtained by calculating the overall brightness mean of the corresponding optical image. The ambient brightness reference mean is used as a reference value. The ratio of the oil film surface reflectance of each pixel in the oil film boundary region to the ambient brightness reference mean is calculated as the normalization result. This reduces the global brightness shift interference caused by uneven ambient lighting, so that the subsequent reflectance is only related to the surface condition of the pipe, such as the oil film thickness, and the output is a corrected reflectance that better reflects the true reflective ability of the material surface.

[0025] The difference between the corrected reflectance and the standard reflectance of the uncoated surface is calculated. Based on this difference, a reflection compensation factor for the corresponding pixel is constructed. The reflectance value of the optical channel where the corresponding pixel is located is adjusted using this reflection compensation factor, and the compensated reflectance is output.

[0026] The standard reflectance of the film-free surface refers to a value set based on optical knowledge and the properties of the oil film, serving as a benchmark. The reflectance offset per unit pixel is obtained by calculating the difference between the corrected reflectance and this benchmark value. The ratio of this reflectance offset to the standard reflectance of the film-free surface is then calculated to obtain the reflection compensation factor. This reflection compensation factor is applied to the existing linear gain formula to adjust the reflectance of the optical channel corresponding to each pixel. The adjusted reflectance is the compensated reflectance, which enables the physical restoration of abnormally darkened areas in the optical channel image.

[0027] Calculate the thermal diffusion gradient at each coordinate sample point in the oil film boundary region, and calculate the thermal flow response offset of the corresponding coordinate sample point based on the thermal diffusion gradient.

[0028] Based on physical knowledge, the thermal diffusion gradient of each coordinate sample point in the oil film boundary region is calculated using the gradient calculation formula in the temperature field. The thermal diffusion gradient is then multiplied by the difference in thermal conductivity between the metal surface and the oil film to obtain the thermal flow response offset, thus identifying the physical conduction offset caused by the reduction in thermal response under oil film coverage.

[0029] The radiation temperature is inversely corrected using the heat flow response offset, and the temperature correction value is output.

[0030] This method involves summing the thermal flow response offset with the radiation temperature of the corresponding coordinate sample point to output the temperature correction value of that coordinate sample point. This simulates the underestimation of the measured temperature caused by the absorption or reflection of infrared thermal waves by the oil film, thereby achieving temperature compensation of the infrared channel for each coordinate sample point and more accurately reflecting the actual temperature state of the material surface.

[0031] The compensated reflectance and temperature correction values ​​are jointly mapped to the corresponding coordinate positions, and all adjusted data are integrated to obtain the oil film suppression dataset. The adjusted compensated reflectance and temperature correction values ​​are then bound to the corresponding positions in the oil film boundary region to ensure that the output oil film suppression dataset is no longer affected by the synchronous antireflection and cooling response caused by the oil film.

[0032] Methods for identifying roughness regions and electromagnetic anomaly response regions include: Each suppressed optical image is divided into grid regions, the reflectivity variance of each grid region is calculated, and the magnitude of microstructure orientation change between adjacent grid regions is also calculated.

[0033] Each suppressed optical image is divided into regions by constructing a grid of fixed length. The length of the grid needs to be sufficient to completely divide the entire suppressed optical image. The reflectivity variance in each grid region is calculated to reflect the degree of reflectivity fluctuation in that region. In this embodiment, a constructed texture direction statistical model is used to identify the texture direction of adjacent grid regions, and the texture change amplitude is extracted as the microstructure direction change amplitude.

[0034] If the reflectivity variance is higher than the roughness fluctuation threshold and the microstructure direction change amplitude is higher than the roughness direction change threshold, then the region formed by the corresponding grid region is taken as the roughness region.

[0035] Based on historical roughness region identification experience, roughness fluctuation threshold and roughness direction change threshold are set. The two thresholds are used to represent the degree of reflectivity fluctuation and the degree of texture direction change under theoretical conditions, respectively. The grid region where the reflectivity variance and the microstructure direction change amplitude are both higher than the corresponding threshold are identified as roughness region. This ensures that the corresponding grid region is considered to have structural disturbance caused by surface roughness only under the dual judgment of high fluctuation and structural disorder.

[0036] Extract the induced impedance change curve of each eddy current spatial point in the surface electrical response data, calculate the response change slope, and if the response change slope of continuous eddy current spatial points is higher than the electromagnetic response slope threshold, and the number of continuous eddy current spatial points is higher than the preset number threshold, then the region formed by the continuous eddy current spatial points is taken as the electromagnetic abnormal response region.

[0037] Eddy current spatial points refer to a spatial location in the surface electrical response data collected by an eddy current sensor. An induced impedance change curve is constructed based on the change of induced impedance at this location with the trajectory position. At the same time, the slope of each point in the curve is calculated as the response change slope. Based on the relevant theoretical knowledge of electromagnetic response, an electromagnetic response slope threshold and a preset quantity threshold are set. If there are multiple consecutive eddy current spatial points that all satisfy the response change slope being higher than the corresponding electromagnetic response slope threshold, then the consecutive point segment is considered to reflect a rapid change in eddy current response, which is a typical eddy current signal jump caused by a sudden change in material surface properties (such as cracks or roughness transitions).

[0038] Methods for performing spatial projection corrections include: Extract the center coordinates of each grid region in the roughness region, and record the mean reflectance and reflectance gradient direction of each corresponding grid region. The center coordinates of the region are calculated based on the pixel position of each grid region in the roughness region, and the mean reflectance and reflectance gradient direction of the region are also calculated as one of the data bases for subsequent operations.

[0039] The spatial center coordinates of each eddy current response sub-region in the electromagnetic anomaly response region are extracted synchronously, and the conductivity of the eddy current spatial points in the eddy current response sub-region is recorded. The mean conductivity and conductivity change gradient are calculated based on the conductivity.

[0040] In the electromagnetic anomaly response region, the region was also divided into sub-regions based on historical experience to ensure that the spatial structure corresponding to the electromagnetic anomaly response region could be completely divided into several sub-regions of the same size. The centroid of each sub-region was calculated as the spatial center coordinates. At the same time, the mean conductivity and conductivity change gradient of each eddy current response sub-region were statistically analyzed as one of the data bases for subsequent operations.

[0041] Spatially adjacent regions are paired with grid regions and eddy current response sub-regions. The spatial distance offset between the center coordinates of each paired region and the spatial center coordinates is calculated. The channel response offset distance is obtained by projecting the spatial distance offset in the main direction.

[0042] In this embodiment, the grid region and eddy current response sub-region in the roughness region are mapped to a unified spatial coordinate system corresponding to the oil pipe fitting. The grid region and eddy current response sub-region are paired according to the principle of closest distance. The Euclidean distance between the center coordinate position of each paired region and the spatial center coordinate is calculated as the spatial distance offset. At the same time, this value is projected according to the reflectivity gradient direction of the roughness region as the main direction to obtain the channel response offset distance between the two channels, which reflects the misalignment behavior generated by the optical and electromagnetic channels performing boundary detection on the same physical region.

[0043] The surface roughness level of the corresponding region is estimated based on the mean reflectivity of the grid region, and the spatial drift compensation of the electromagnetic response is calculated based on the correspondence between the roughness level and the eddy current lift-off effect.

[0044] The roughness rating table is constructed based on historical roughness judgment experience. The average reflectance of each grid region is matched with the table to output the surface roughness rating of the corresponding grid region. The fitting function is constructed using a fitting algorithm based on the relevant theoretical knowledge of eddy current lift-off effect. The roughness rating is mapped to the drift compensation amount required for the corresponding electromagnetic response as the spatial drift compensation amount.

[0045] The channel response offset distance and spatial drift compensation are superimposed, and a coordinate transformation function is constructed by combining the angle between the reflectivity gradient direction and the conductivity change gradient direction. The optical matching coordinates of each eddy current spatial point in the electromagnetic anomaly response region are calculated using this coordinate transformation function.

[0046] The formula for calculating the coordinate transformation function is as follows: ;in, This represents the optical matching coordinates of a point in the vortex space after processing by the coordinate transformation function. Represents the coordinates of the initial vortex space point; They represent , ,in Indicates the channel response offset distance. This represents the amount of spatial drift compensation. This represents the angle between the direction of the reflectivity gradient and the direction of the conductivity gradient; This represents a rotation matrix; coordinate mapping is performed using coordinate transformation functions, and the output is optical matching coordinates that can match the coordinate standard of the optical channel.

[0047] Reflectivity and conductivity are bound to a unified spatial location based on optical matching coordinates, and the correlation of channel responses is calculated. When the correlation is higher than a preset correlation coefficient threshold, the spatial location is determined to be valid, and the fused multi-channel spatially aligned dataset is output.

[0048] The reflectivity and conductivity at the points corresponding to the optical matching coordinates are read, and the Pearson correlation coefficient between the reflectivity of the optical channel and the conductivity of the electromagnetic channel at the matched optical matching coordinates is calculated as the channel response correlation. Based on the theoretical knowledge of the correlation between optics and electromagnetics, a preset correlation coefficient threshold is set. If the channel response correlation is higher than the corresponding threshold, it means that the response trends of the optical channel and the electromagnetic channel are consistent at the corresponding spatial location, and the spatial registration is effective.

[0049] Methods for conducting joint analysis include: Vertical cross-sectional lines are drawn for the weld structure area, and the radiation temperature and magnetic permeability of each sampling point on each cross-section are extracted.

[0050] For the weld structure area on the oil pipe fitting, multiple sections are divided along the direction perpendicular to the weld centerline. Multiple sampling points are evenly distributed on each section line according to coordinates. At this time, the radiation temperature and magnetic permeability of the data slice corresponding to the sampling points are extracted.

[0051] The permeability of each profile line is sorted based on its distance from the weld center, and the rate of change of permeability between adjacent sampling points is calculated.

[0052] This process involves determining the coordinates of the weld center, calculating the distance between each sampling point on each profile line and the weld center, and sorting them according to the distance to reflect the trend change process from the center outwards. The permeability of the corresponding sampling points is extracted in sequence, and the rate of change of permeability of adjacent sampling points is calculated.

[0053] Based on the permeability decay curve fitted by the rate of change of permeability as the benchmark for permeability change, in this embodiment, a decay curve that conforms to the physical trend of electromagnetic response is fitted by a fitting algorithm to describe the typical change state of permeability under normal transition of the weld area structure. Therefore, this curve is used as the background benchmark for permeability change.

[0054] The radiation temperature along each profile line is sorted according to its distance from the weld center. The temperature gradient of adjacent sampling points is calculated, the inflection point of the temperature gradient is identified, and the location of the inflection point is used as the boundary of the thermal diffusivity change.

[0055] The radiation temperature is processed using the same method as that used for magnetic permeability. After calculating the temperature gradient between adjacent sampling points, the location where the temperature gradient changes is further identified as the inflection point. This location indicates a change in the response mechanism of thermal diffusivity, which is used to define the boundary of the thermal response change and to provide a boundary reference for subsequent thermal anomalies.

[0056] Calculate the difference between the permeability and the corresponding permeability change benchmark, and output the permeability difference value; select sampling points with permeability difference values ​​higher than the preset permeability threshold as fine-scale magnetic anomaly response points.

[0057] The permeability difference value is calculated by subtracting the value of the corresponding sampling point from the permeability change benchmark at each sampling point. This value is used to indicate the degree to which the corresponding position deviates from the normal transition trend. At the same time, a permeability threshold under the theoretical permeability state is set based on the relevant theoretical knowledge of electromagnetic channels. If the permeability difference value is higher than the threshold, it indicates that the permeability change of these sampling points is significant, and the degree of interference with the local structure magnetism is much higher than the degree of normal transition change. It may be an abnormal magnetic response caused by welding defects. Therefore, the corresponding sampling point is identified as a fine-scale magnetic anomaly response point.

[0058] Calculate the mean radiation temperature of all sampling points within the weld structure area, and select the continuous area formed by sampling points on each profile line whose temperature difference between the radiation temperature and the mean radiation temperature is greater than the preset temperature deviation threshold. This area is recorded as the coarse-scale thermal anomaly area.

[0059] First, the average radiation temperature of the entire weld structure area is calculated to reflect the global average level as the background baseline. Then, sampling points with a temperature difference between the radiation temperature and the average radiation temperature that is higher than the preset temperature deviation threshold are selected for each profile line. The preset temperature deviation threshold refers to the reasonable temperature deviation limit under theoretical conditions. Points with temperature differences greater than the corresponding threshold and that can form continuous blocks on the profile line are extracted to form a continuous area, which is the coarse-scale thermal anomaly area. This avoids misidentifying single-point temperature jumps or thermal noise as defects.

[0060] It should be noted that the process of expanding the space to obtain a continuous region is constrained by the boundary of thermal diffusivity change, ensuring that only regions with significant differences in physical thermal diffusivity and located within the corresponding boundary are identified as thermal anomalies.

[0061] The fine-scale magnetic anomaly response point is spatially overlapped with the coarse-scale thermal anomaly region. If the spatial location of the fine-scale magnetic anomaly response point is located within the coarse-scale thermal anomaly region, the defect response intensity is obtained by multiplying the magnetic permeability difference value of the fine-scale magnetic anomaly response point with the temperature difference value at the corresponding location.

[0062] First, the regions corresponding to the fine-scale magnetic anomaly response points and the coarse-scale thermal anomaly regions are spatially overlapped to screen for regions where both magnetic and thermal anomalies exist simultaneously. Furthermore, the fine-scale magnetic anomaly response points need to be located within the coarse-scale thermal anomaly regions, which is likely to be a micro-region of a real weld defect. In this case, the defect response intensity is defined by the product of the magnetic permeability difference and the temperature difference, highlighting regions with abrupt changes across channels.

[0063] Fine-scale magnetic anomaly response points with defect response intensity higher than the preset defect response threshold are selected, and their spatial coordinates, magnetic permeability difference value, temperature difference value, and defect response intensity are integrated to output a subset of weld features.

[0064] The process involves setting a preset defect response threshold based on historical defect response intensity calculation experience, which is suitable for the current working conditions. Fine-scale magnetic anomaly response points with defect response intensity higher than the corresponding threshold are selected, and their corresponding spatial coordinates, magnetic permeability difference, temperature difference, and defect response intensity are encapsulated as data items. The output is a subset of weld features obtained by integrating the data items.

[0065] Methods for comparing distributions include: The spatial gradient distribution of color channels in an image block and the spatial difference distribution of impedance amplitude of electrochemical channels in the corresponding spatial region are calculated. Based on the difference in the changing trends of the spatial gradient distribution and the spatial difference distribution, regions with inconsistent responses are extracted, and overlapping regions are selected as corrosion boundary regions.

[0066] For image blocks containing rusted areas, the spatial gradient values ​​of color components are first calculated pixel by pixel in the color channels to output the spatial gradient distribution. The color components refer to values ​​such as R, G, and B channels, which are used to describe the rate of change of color channel component values. In the electrochemical channels, the impedance amplitude of each sampling point in the corresponding area of ​​the image block is calculated to calculate the difference between adjacent points, and the spatial difference distribution is output. Since the optical color in the rust transition zone exhibits a continuous and gradual change, while the electrochemical impedance exhibits a step-like abrupt change, the change trends of the two channels at the same spatial location often differ.

[0067] By comparing the direction and magnitude of change of spatial gradient distribution and spatial difference distribution at various spatial locations, and combining relevant knowledge in this field with historical experience, spatial regions with significantly inconsistent response trends between the two channels are extracted. All inconsistent spatial regions are spatially overlapped, and the overlapping regions are used as corrosion boundary regions to identify the transition region between corrosion products and healthy regions.

[0068] Methods for performing physical consistency analysis include: The color components and electrochemical impedance of each rust sampling point in the rust boundary region are extracted, and the color gradient vector and impedance vector are constructed by combining the spatial gradient distribution and spatial difference distribution, respectively.

[0069] For each rust sampling point, the color component is calculated, and the horizontal and vertical gradients at the corresponding positions are used to form a two-dimensional color gradient vector. The direction represents the spatial direction of the drastic color change, and the magnitude represents the intensity of the color change. For each impedance vector, the horizontal and vertical differences at the corresponding positions of the rust sampling point are calculated to form a two-dimensional impedance vector. The direction represents the spatial direction of the significant impedance response change, and the magnitude represents the intensity of the impedance change.

[0070] Identify the main extension direction of the corrosion boundary region and set a multi-scale analysis window along this main extension direction.

[0071] The main extension direction of the rust boundary region is determined by calculating the principal component directions of all rust sampling points. Multiple analysis sliding windows of various scales are set along the main extension direction. Small scales are used to capture subtle local changes, while large scales are used to reflect the overall trend, taking into account both local anomaly detection and overall trend identification.

[0072] The set of color gradient vectors in each analysis window is fitted to generate a gradient trend curve, and the set of impedance vectors is fitted to generate a step response curve.

[0073] Within each analysis window, all color gradient vectors are sorted according to their spatial location. A suitable fitting algorithm is then used to generate a gradient trend curve describing the overall trend of color change, reflecting the physical law of the continuous transition of color from the healthy area to the rusted area within the corrosion transition zone. Similarly, a suitable fitting algorithm is used to fit the sorted impedance vectors to generate a step response curve describing the characteristics of impedance response change, reflecting the abrupt change characteristics of electrochemical impedance within the corrosion transition zone.

[0074] The deviation between the color component of each rust sampling point and the corresponding position of the gradient trend curve is calculated as the color residual value. At the same time, the deviation between the electrochemical impedance of the rust sampling point and the corresponding position of the step response curve is calculated as the impedance residual value.

[0075] The deviations of the color component values ​​from the corresponding gradient trend curve values ​​and the electrochemical impedance values ​​from the corresponding step response curve values ​​are calculated to obtain color residuals and impedance residuals. The color residuals reflect the degree of deviation of the color response of the rust sampling point from the normal gradient background, and the impedance residuals reflect the degree of deviation of the impedance response of the rust sampling point from the normal step background.

[0076] Determine whether the signs of the color residual value and the impedance residual value of each rust sampling point are consistent, and mark the rust sampling points with the same sign as points with the same response direction.

[0077] The method involves determining whether the signs of the color residual value and the impedance residual value are consistent. If both values ​​are positive, it indicates that the rust sampling point exhibits a higher-than-expected response enhancement in both the optical and electrochemical channels. If both values ​​are negative, it indicates that the rust sampling point exhibits a lower-than-expected response reduction in both channels. Rust sampling points with the same sign are marked as points with the same response direction, indicating that the abnormal deviation direction in the two physical channels is consistent.

[0078] The absolute ratio of the color residual value to the impedance residual value at each point of response in the same direction is calculated as the dual-channel response ratio. Points of response in the same direction with dual-channel response ratios within a preset ratio range are selected as points of physical response consistency.

[0079] The dual-channel response ratio is calculated to reflect the relative proportion of the abnormal response intensity of the same sampling point in two physical channels. A reasonable proportion range is set based on historical experience and specific working conditions as the criterion for judging the consistency of physical response. If the dual-channel response ratio of a certain rust sampling point falls within the corresponding preset proportion range, it is considered that the abnormal response amplitude of the rust sampling point in the two physical channels is coordinated and matched. The response intensity of real defects in different physical channels usually has a stable proportional relationship, which is determined by the physical properties of the defect. False anomalies often show extremely unbalanced response ratios. Therefore, the dual-channel response ratio can be used to screen rust sampling points that are truly physically correlated and mark them as points with consistent physical response.

[0080] The region formed by integrating all physical response consistency points is used as a set of trusted defect fusion blocks.

[0081] This involves spatially aggregating all physical response consistency points that are selected through response direction judgment and dual-channel response ratio screening, organizing them into several continuous blocks based on the spatial connectivity between adjacent points, and integrating all blocks to obtain a set of credible defect fusion blocks.

[0082] Methods for performing defect detection include: The subset of weld features, the set of credible defect fusion blocks, and the multi-channel spatial alignment dataset are integrated into the dataset to be detected. The defect judgment rules are used as the basis to match the dataset to be detected with the corresponding data dimensions in the defect judgment rules and output the detection results.

[0083] In this embodiment, a defect detection model is trained using known oil pipe defect data and historical defect detection data as corpus. The defect detection model outputs judgment conditions such as defect image features and corresponding parameter thresholds for various dimensions. These judgment conditions are integrated into defect judgment rules. The image type data and numerical type data in the dataset to be detected are matched with the corresponding defect image features and various threshold judgment conditions in the defect judgment rules to output the discrimination and detection results of oil pipe defects.

[0084] This embodiment acquires a multi-dimensional inspection dataset of oil pipe fittings in real time. Based on this dataset, it performs oil film suppression, spatial registration, weld joint analysis, and physical consistency analysis of rusted areas, realizing an online detection system and method for surface defects in oil pipe fittings that integrates multi-scale features. Compared with existing technologies, it improves the accuracy of oil film coverage area boundary identification by establishing a spatial overlap determination between low-reflectivity and low-temperature regions. By physically compensating the optical and infrared channels, it eliminates the multi-channel response coupling distortion caused by uneven oil film thickness. By extracting the spatial offset between roughness regions and electromagnetic anomaly response regions and constructing a coordinate transformation function in conjunction with lift-off effect compensation, it eliminates the distortion caused by surface roughness differences. The problem of multi-channel spatial registration failure was addressed; by fitting the permeability decay curve and identifying the boundary of thermal diffusivity change, gradient background stripping was achieved, avoiding the masking of minor defects in the weld heat-affected zone by the gradient background of the microstructure; by fitting the gradient trend curve and step response curve of the color gradient vector and impedance vector respectively, the fusion conflict between the optical gradient boundary and the electrochemical step boundary in the corrosion transition zone was optimized; through multiple verifications of response in the same direction judgment and dual-channel response ratio screening, the accurate positioning of credible defect areas in the transition zone was achieved; the multi-physical quantity fusion perception capability and multi-scale feature collaborative analysis accuracy of online detection of surface defects of oil pipe fittings were improved, ensuring the cross-channel response consistency of defect detection and the detection reliability under complex working conditions.

[0085] Example 2 Please see Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. An online detection system for surface defects in oil pipe fittings integrating multi-scale features is provided, including: The data acquisition module collects multivariate inspection data of oil pipe fittings and performs data cleaning to obtain a multivariate raw dataset, which includes optical feature image data, surface electrical response data and pipe surface temperature data. The oil film suppression module combines optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppresses oil film influence within the oil film boundary region, adjusts multi-level channel values, and outputs an oil film suppression dataset. The spatial registration module extracts the suppressed optical image from the oil film suppression dataset, and combines it with the surface electrical response data to extract roughness image features and electromagnetic induction change features from the suppressed optical image, identifying roughness regions and electromagnetic anomaly response regions; based on the difference in positional information between roughness regions and electromagnetic anomaly response regions, spatial projection correction is performed to generate a multi-channel spatial alignment dataset. The weld anomaly decoupling module extracts data slices containing weld structure regions from the multi-channel spatially aligned dataset. Based on the temperature and electromagnetic response data of the weld structure regions, it performs joint analysis on the data slices to identify the boundary response characteristics of the weld structure regions and outputs a subset of weld features. The corrosion recognition module identifies image blocks containing corrosion regions in a multi-channel spatially aligned dataset, compares the distribution of these image blocks, and outputs the corrosion boundary regions. It then performs physical consistency analysis on the corrosion boundary regions and outputs a set of reliable defect fusion blocks. The joint detection module combines a subset of weld features, a set of credible defect fusion blocks, and a multi-channel spatially aligned dataset to establish defect judgment rules, perform defect detection, and send the detection results to a preset online monitoring terminal; the modules are connected to each other via wired and / or wireless means.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0087] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0088] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An online detection method for surface defects in oil pipe fittings integrating multi-scale features, characterized in that, include: S1. Collect multivariate inspection data of oil pipe fittings and perform data cleaning to obtain multivariate raw datasets, wherein the multivariate raw datasets include optical feature image data, surface electrical response data and pipe surface temperature data; S2. Combine optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppress oil film influence in the oil film boundary area, adjust multi-level channel values, and output oil film suppression dataset. S3. Extract the suppression optical image from the oil film suppression dataset, combine it with the surface electrical response data to extract the roughness image features and electromagnetic induction change features in the suppression optical image, and identify the roughness region and electromagnetic anomaly response region; Spatial projection correction is performed based on the difference in location information between the roughness region and the electromagnetic anomaly response region to generate a multi-channel spatially aligned dataset; S4. Extract data slices containing weld structure regions from the multi-channel spatially aligned dataset. Perform joint analysis on the data slices based on the temperature data and electromagnetic response data of the weld structure regions to identify the boundary response characteristics of the weld structure regions and output the weld feature subset. S5. Identify image blocks containing rusted regions in a multi-channel spatial alignment dataset, compare the distribution of these image blocks, and output the rusted boundary regions; perform physical consistency analysis on the rusted boundary regions and output a set of reliable defect fusion blocks; S6. By combining the weld feature subset, the set of credible defect fusion blocks, and the multi-channel spatial alignment dataset, establish defect judgment rules to perform defect detection, and send the detection results to the preset online monitoring terminal.

2. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 1, characterized in that, The methods for identifying oil film boundaries include: The pixel brightness value of each optical image in the optical feature image data is extracted, and the incident intensity parameter of the light source is obtained. The surface reflectivity of the pipe is calculated based on the pixel brightness value and the incident intensity parameter of the light source. Pixels with surface reflectivity of the pipe that are lower than the first reflection threshold are identified and aggregated to generate low reflectivity regions. Extract the radiation temperature of each coordinate sample point from the pipe surface temperature data, compare the radiation temperature with the second temperature reference, and extract the sample point area where the pipe surface radiation temperature is lower than the second temperature reference as the low temperature area. Establish the spatial coordinate correspondence between low reflectivity region and low temperature region, calculate the region overlap area, and select the pipe surface region with region overlap area higher than the preset spatial overlap threshold as the joint response region. A multi-scale sliding window is constructed to jointly calculate the local reflectivity gradient and thermal response gradient of the joint response region, outputting a joint gradient map of the pipe surface. Spatial regions in the joint gradient map whose gradient values ​​reach the preset gradient intensity are selected as the brightness-temperature boundary transition region. Based on the gradient direction of the combined gradient in the brightness-temperature boundary transition region, a boundary profile curve is constructed, and the oil film boundary region is output.

3. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 2, characterized in that, The methods for suppressing the effects of oil film include: Calculate the oil film surface reflectance of each pixel in the oil film boundary region, normalize it by combining it with the ambient brightness reference mean, and output the corrected reflectance of the corresponding pixel. The difference between the corrected reflectance and the standard reflectance of the uncoated surface is calculated. Based on this difference, a reflectance compensation factor for the corresponding pixel is constructed. The reflectance compensation factor is used to adjust the reflectance value of the optical channel where the corresponding pixel is located, and the compensated reflectance is output. Calculate the thermal diffusion gradient at each coordinate sample point in the oil film boundary region, and calculate the thermal flow response offset at the corresponding coordinate sample point based on the thermal diffusion gradient; use the thermal flow response offset to perform inverse correction on the radiation temperature, and output the temperature correction value. The compensated reflectivity and temperature correction values ​​are jointly mapped to the corresponding coordinate positions, and all adjusted data are integrated to obtain the oil film suppression dataset.

4. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 3, characterized in that, The methods for identifying roughness regions and electromagnetic anomaly response regions include: Each suppressed optical image is divided into grid regions, and the reflectance variance of each grid region is calculated. At the same time, the microstructure orientation change amplitude of adjacent grid regions is calculated. If the reflectance variance is higher than the roughness fluctuation threshold and the microstructure orientation change amplitude is higher than the roughness orientation change threshold, then the region formed by the corresponding grid region is taken as the roughness region. Extract the induced impedance change curve of each eddy current spatial point in the surface electrical response data, calculate the response change slope, and if the response change slope of continuous eddy current spatial points is higher than the electromagnetic response slope threshold, and the number of continuous eddy current spatial points is higher than the preset number threshold, then the region formed by the continuous eddy current spatial points is taken as the electromagnetic abnormal response region.

5. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 4, characterized in that, The methods for performing spatial projection correction include: Extract the center coordinates of each grid region in the roughness region and record the mean reflectivity and reflectivity gradient direction of each grid region; simultaneously extract the spatial center coordinates of each eddy current response sub-region in the electromagnetic anomaly response region and record the conductivity of the eddy current spatial points in the eddy current response sub-region, and calculate the mean conductivity and conductivity change gradient based on the conductivity. Spatially adjacent pairs are formed between the grid region and the eddy current response sub-region. The spatial distance offset between the center coordinate position and the spatial center coordinate of each paired region is calculated. The channel response offset distance is obtained by projecting the spatial distance offset in the main direction. The surface roughness level of the corresponding region is estimated based on the mean reflectivity of the grid region, and the spatial drift compensation of the electromagnetic response is calculated based on the correspondence between the roughness level and the eddy current lift-off effect. The channel response offset distance and spatial drift compensation are superimposed, and a coordinate transformation function is constructed by combining the angle between the reflectivity gradient direction and the conductivity change gradient direction. The optical matching coordinates of each eddy current spatial point in the electromagnetic anomaly response region are calculated using this coordinate transformation function. Reflectivity and conductivity are bound to a unified spatial location based on optical matching coordinates, and the correlation of channel responses is calculated. When the correlation is higher than a preset correlation coefficient threshold, the spatial location is determined to be valid, and the fused multi-channel spatially aligned dataset is output.

6. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 5, characterized in that, The methods for conducting joint analysis include: Vertical profile lines are divided for the weld structure area, and the radiation temperature and magnetic permeability of each sampling point on each profile line are extracted; The permeability of each profile line is sorted based on its distance from the weld center, and the rate of change of permeability between adjacent sampling points is calculated; a permeability decay curve is fitted based on the rate of change of permeability as the benchmark for permeability change; The radiation temperature along each profile line is sorted according to its distance from the weld center. The temperature gradient of adjacent sampling points is calculated, the inflection point of the temperature gradient is identified, and the location of the inflection point is used as the boundary of the thermal diffusivity change. Calculate the difference between the permeability and the corresponding permeability change benchmark, and output the permeability difference value; select sampling points with permeability difference values ​​higher than the preset permeability threshold as fine-scale magnetic anomaly response points; Calculate the mean radiation temperature of all sampling points within the weld structure area, and select the continuous area formed by sampling points on each profile line whose temperature difference between radiation temperature and mean radiation temperature is greater than the preset temperature deviation threshold. This area is recorded as the coarse-scale thermal anomaly area. The fine-scale magnetic anomaly response point is spatially overlapped with the coarse-scale thermal anomaly region. If the spatial location of the fine-scale magnetic anomaly response point is located within the coarse-scale thermal anomaly region, the magnetic permeability difference value of the fine-scale magnetic anomaly response point is multiplied by the temperature difference value at the corresponding location to obtain the defect response intensity. Fine-scale magnetic anomaly response points with defect response intensity higher than the preset defect response threshold are selected, and their spatial coordinates, magnetic permeability difference value, temperature difference value, and defect response intensity are integrated to output a subset of weld features.

7. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 6, characterized in that, The methods for performing distribution comparisons include: The spatial gradient distribution of color channels in an image block and the spatial difference distribution of impedance amplitude of electrochemical channels in the corresponding spatial region are calculated. Based on the difference in the changing trends of the spatial gradient distribution and the spatial difference distribution, regions with inconsistent responses are extracted, and overlapping regions are selected as corrosion boundary regions.

8. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 7, characterized in that, The methods for performing physical consistency analysis include: Extract the color components and electrochemical impedance of each rust sampling point in the rust boundary region, and construct color gradient vector and impedance vector by combining spatial gradient distribution and spatial difference distribution, respectively; Identify the main extension direction of the corrosion boundary region and set a multi-scale analysis window along this main extension direction; fit the set of color gradient vectors in each analysis window to generate a gradient trend curve, and fit the set of impedance vectors to generate a step response curve. The deviation between the color component of each rust sampling point and the corresponding position of the gradient trend curve is calculated as the color residual value. At the same time, the deviation between the electrochemical impedance of the rust sampling point and the corresponding position of the step response curve is calculated as the impedance residual value. Determine whether the signs of the color residual value and the impedance residual value of each corrosion sampling point are consistent, and mark the corrosion sampling points with the same sign as points with the same response direction; The absolute ratio of the color residual value to the impedance residual value of each response point is calculated as the dual-channel response ratio. Response points with dual-channel response ratios within a preset ratio range are selected as physical response consistency points. The region formed by integrating all physical response consistency points is used as a set of reliable defect fusion blocks.

9. The online detection method for surface defects of oil pipe fittings integrating multi-scale features according to claim 8, characterized in that, The defect detection methods include: The subset of weld features, the set of credible defect fusion blocks, and the multi-channel spatial alignment dataset are integrated into the dataset to be detected. The defect judgment rules are used as the basis to match the dataset to be detected with the corresponding data dimensions in the defect judgment rules and output the detection results.

10. An online detection system for surface defects of oil pipe fittings integrating multi-scale features, used to implement the online detection method for surface defects of oil pipe fittings integrating multi-scale features as described in any one of claims 1-9, characterized in that, include: The data acquisition module collects multivariate inspection data of oil pipe fittings and performs data cleaning to obtain a multivariate raw dataset, which includes optical feature image data, surface electrical response data and pipe surface temperature data. The oil film suppression module combines optical feature image data and pipeline surface temperature data to identify oil film boundaries, suppresses oil film influence within the oil film boundary region, adjusts multi-level channel values, and outputs an oil film suppression dataset. The spatial registration module extracts the suppressed optical image from the oil film suppression dataset, and combines it with the surface electrical response data to extract roughness image features and electromagnetic induction change features from the suppressed optical image, thereby identifying roughness regions and electromagnetic anomaly response regions. Spatial projection correction is performed based on the difference in location information between the roughness region and the electromagnetic anomaly response region to generate a multi-channel spatially aligned dataset; The weld anomaly decoupling module extracts data slices containing weld structure regions from the multi-channel spatially aligned dataset. Based on the temperature and electromagnetic response data of the weld structure regions, it performs joint analysis on the data slices to identify the boundary response characteristics of the weld structure regions and outputs a subset of weld features. The corrosion recognition module identifies image blocks containing corrosion regions in a multi-channel spatially aligned dataset, compares the distribution of these image blocks, and outputs the corrosion boundary regions. It then performs physical consistency analysis on the corrosion boundary regions and outputs a set of reliable defect fusion blocks. The joint detection module combines a subset of weld features, a set of credible defect fusion blocks, and a multi-channel spatially aligned dataset to establish defect judgment rules, perform defect detection, and send the detection results to a preset online monitoring terminal; the modules are connected to each other via wired and / or wireless means.

Images