A semiconductor pipeline weld intelligent detection and compensation method based on spatial attention

Abstract

The application provides a kind of semiconductor pipeline weld intelligent detection and compensation method based on spatial attention, wherein, the method includes based on the 2D image and 3D point cloud of the collected welded special gas pipeline, the internal fusion zone image, the external fusion zone image, the internal heat affected zone image and the external heat affected zone image of the welded special gas pipeline are identified;The internal fusion zone image, the external fusion zone image, the internal heat affected zone image and the external heat affected zone image of the welded special gas pipeline are respectively input into the pre-trained welding defect identification model to obtain the defect identification result output by the welding defect identification model, and the defect identification result is used to indicate the welding defect type and defect quantization out-of-tolerance degree value existing in the welded special gas pipeline;Based on the defect identification result, the corresponding process compensation scheme is matched in the process parameter association library to adjust the welding process parameters of the tungsten rail welding equipment for subsequent welding special gas pipeline.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and more specifically, to a method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention. Background Technology

[0002] As a key carrier for high-purity gas transportation, the welding quality of semiconductor ultra-high purity pipeline systems directly affects the sealing of gas transportation and the cleanliness of fluids.

[0003] Industry standards have established strict quantitative criteria for the welding quality of ultra-high purity pipelines. For example, the color difference of the weld surface must not exceed ΔE1.5, the crack length must not exceed 50μm, and the pore diameter must not exceed 100μm. The detection of these defects requires both macroscopic geometric dimension measurement and microscopic surface morphology analysis.

[0004] Current mainstream inspection methods rely on technicians using simple tools such as magnifying glasses and calipers for visual inspection. This is not only inefficient but also highly dependent on the experience of the personnel. When inspecting minute cracks, manual methods struggle to distinguish weld lines from actual defects; and for hidden defects such as internal porosity, there is a lack of effective non-destructive testing methods. This traditional inspection method cannot meet the comprehensive quantitative evaluation needs of modern semiconductor factories for pipeline welding quality, and its inspection efficiency is low. Summary of the Invention

[0005] The purpose of this application is to provide a method for intelligent detection and compensation of semiconductor pipeline welds based on spatial attention, in order to solve the technical problems of existing defects in semiconductor ultra-high purity pipeline welding welds relying on manual labor, resulting in poor detection accuracy and low efficiency.

[0006] In a first aspect, the present invention provides a method for intelligent detection and compensation of semiconductor pipeline welds based on spatial attention. The method includes identifying the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline based on the acquired 2D image and 3D point cloud of the welded special gas pipeline; inputting the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline into a welding defect recognition model based on spatial attention to obtain the defect recognition result output by the welding defect recognition model. The defect recognition result is used to indicate the type of welding defect and the degree of defect quantification deviation of the welded special gas pipeline; based on the defect recognition result, matching the corresponding process compensation scheme in the process parameter association library to adjust the welding process parameters of the subsequent welding of special gas pipelines by the tungsten inert gas welding equipment.

[0007] In an optional implementation, under argon protection, three consecutive surface images of the welded special gas pipeline are acquired by an industrial camera positioned at the center of the weld. Pixel alignment and white balance calibration are performed on the second and third surface images using the first surface image; an enhanced brightness map is determined using the calibrated second and third surface images; a fused brightness map is generated based on the enhanced brightness map and the first surface image; and a 2D surface image is generated based on the fused brightness map.

[0008] In an optional implementation, the second frame surface image / third frame surface image is pixel aligned in the following manner: Fourier transforms are performed on the first, second, and third frame surface images respectively to obtain the corresponding spectra; the cross-power spectrum is calculated based on the spectrum corresponding to the first frame surface image and the spectrum corresponding to the second / third frame surface image; the cross-power spectrum is inversely transformed to determine the peak value, and the integer pixel displacement value is determined; for each pixel in the second / third frame surface image, the pixel value of the pixel is added to the integer pixel displacement value to obtain the pixel-aligned second / third frame surface image.

[0009] In an optional implementation, the enhanced brightness map is determined in the following manner: The absolute difference between the calibrated second and third frame surface images is calculated pixel by pixel to determine the differential intensity map; the differential intensity map is then nonlinearly mapped to determine the enhanced brightness map.

[0010] In an optional implementation, the fused brightness map is generated in the following manner: The first frame surface image is converted to a color space, and the corresponding Y, Cb and Cr channels are determined. The Cb and Cr channels corresponding to the first frame surface image and the Y channel corresponding to the enhanced brightness map are combined to generate a fused brightness map.

[0011] In an optional implementation, three consecutive frames of surface images of the welded special gas pipeline and its exterior are acquired using the following method: The first frame of the surface image was acquired under 0° polarization and white light illumination conditions; The second frame of the surface image was acquired under 0° polarization and 470nm blue light illumination conditions. The third frame of the surface image was acquired under 90° polarization and 470nm blue light illumination.

[0012] In an optional implementation, three consecutive frames of surface images of the interior of the welded special gas pipeline are acquired using the following method: The first frame of the surface image was acquired under 470nm blue light illumination. The second frame of the surface image was acquired under 525nm green light illumination. The third frame of the surface image was acquired under 850nm red light illumination.

[0013] Secondly, the present invention provides an intelligent detection and compensation device for semiconductor pipeline welds based on spatial attention, the device comprising: The segmentation module is used to identify the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline based on the acquired 2D images and 3D point clouds of the welded special gas pipeline. The defect identification module is used to input the images of the internal fusion zone, external fusion zone, internal heat-affected zone and external heat-affected zone of the welded special gas pipeline into the welding defect identification model based on spatial attention, so as to obtain the defect identification results output by the welding defect identification model. The defect identification results are used to indicate the type of welding defect and the degree of defect quantification deviation of the welded special gas pipeline. The execution module is used to match the corresponding process compensation scheme in the process parameter association library based on the defect identification results, so as to adjust the welding process parameters of the special gas pipelines to be welded by the tungsten inert gas welding equipment.

[0014] Thirdly, the present invention provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of any of the spatial attention-based intelligent detection and compensation methods for semiconductor pipeline welds as described in the foregoing embodiments.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of any of the aforementioned embodiments of the intelligent detection and compensation method for semiconductor pipeline weld seams based on spatial attention.

[0016] This application provides a method for intelligent detection and compensation of semiconductor pipeline welds based on spatial attention. The method includes identifying the internal fusion zone, external fusion zone, internal heat-affected zone, and external heat-affected zone images of the welded special gas pipeline based on acquired 2D images and 3D point clouds. These images are then input into a welding defect recognition model based on spatial attention to obtain defect recognition results. The defect recognition results indicate the type of welding defect and the degree of defect quantification in the welded special gas pipeline. Based on the defect recognition results, a corresponding process compensation scheme is matched in a process parameter association library to adjust the welding process parameters for subsequent welding of special gas pipelines using tungsten inert gas (TIG) welding equipment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an intelligent detection and compensation method for semiconductor pipeline welds based on spatial attention, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] Defect detection in the weld seams of ultra-high purity semiconductor piping (SUS316L EP grade) is a core aspect of quality control. The industry has clear and detailed quantitative standards for weld seam defects, covering dimensions such as appearance, dimensions, and internal quality. Accurate identification of defects such as color difference, cracks, porosity, undercut, and weld seam dimensional deviations is required, with detection precision matching the need for micron-level defect identification. The quality of the weld seams in this type of piping directly determines the sealing performance and fluid cleanliness of the high-purity gas delivery system, playing a crucial role in the yield of core processes such as semiconductor etching and deposition. Currently, leading companies still use manual visual inspection combined with simple measuring tools for defect detection, which cannot accurately match the industry's quantitative defect judgment standards. Furthermore, there are no corresponding quantitative process compensation schemes after inspection, resulting in a persistently high rework rate and failing to meet the stringent quality requirements of the semiconductor industry for ultra-high purity piping.

[0020] Specifically, relying on manual visual inspection under 1000±100 Lux illumination lacks a unified quantitative standard for defect judgment, allowing only qualitative acceptance / failure assessments. It cannot identify minute defects at the 0.01mm level (such as micropores and slight undercut), nor can it detect potential quantitative out-of-tolerance defects in critical areas, resulting in a high rate of missed inspections. Furthermore, manual weld inspection is slow, and the results are not recorded quantitatively, only documented in paper logs.

[0021] Based on this, this application provides a method for intelligent detection and process compensation of weld defects in semiconductor ultra-high purity pipelines that integrates spatial attention mechanisms.

[0022] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0023] Example 1 In one embodiment of this application, a method for intelligent detection and process compensation of weld defects in semiconductor ultra-high purity pipeline welding seams that integrates spatial attention mechanisms is provided.

[0024] In one feasible implementation, an intelligent quantitative detection model for weld defects in semiconductor ultra-high purity pipelines is constructed based on a comprehensive quantitative defect judgment standard for welds, enabling accurate identification, location, and quantitative judgment of defects. Simultaneously, a quantitative correlation library between defect types and process parameters is established. Based on the degree of quantitative deviation of the defect, a corresponding quantitative compensation scheme for process parameters is output, achieving real-time data linkage between the detection model and welding equipment. This forms a complete quality control system encompassing "intelligent quantitative detection, quantitative defect judgment, quantitative process compensation, and closed-loop welding optimization."

[0025] For the inspection of butt welds in SUS316L EP grade small-diameter pipes, the core implementation standard is 6.35×0.89mm. The specific implementation steps are as follows: We collected visual images (2D) and three-dimensional morphological data (3D) of welds with different types of defects, covering all defect-sensitive areas such as the weld fusion zone, heat-affected zone, and inner / outer weld edges. Based on industry-standard quantitative defect judgment criteria, all defect samples were accurately labeled, clearly defining label information such as defect type, degree of quantitative deviation, and defect location. The specific quantitative judgment criteria are as follows: The quantitative standards for appearance defects include: slight oxidation color difference is allowed for the outer weld bead; no visible color difference is allowed for the inner weld bead; no visible cracks, pores, or inclusions are allowed; cracks / pores with a diameter greater than or equal to 0.01 mm are considered defects; no undercut, centerline shrinkage, or weld depression are allowed; undercut depth greater than or equal to 0.05 mm is considered a defect.

[0026] The dimensional defect quantification standards include: the convexity of the inner and outer welds being less than or equal to 10% of the nominal wall thickness of the pipe; the width of the outer weld being 2 to 2.5 times the pipe wall thickness, and the width of the inner weld being 1 to 2 times the pipe wall thickness; the weld bending being less than or equal to 35% of the pipe wall thickness; the variation in the width of the inner weld being less than or equal to 0.2 mm; and the axial deviation of the weld being less than or equal to 10% of the pipe wall thickness and the angular deviation being less than or equal to ±0.5°. The internal defect quantification standard includes no internal defects such as poor fusion or slag inclusions in the weld and heat-affected zone, and no visible defects under a 25x magnifying glass; the tensile test shows that the weld breaks at the base material and there are no opening defects exceeding 3mm when bent at 180°.

[0027] Here, when acquiring visual data of semiconductor special gas pipelines, a dual-station rotary table and dual-camera module architecture can be used to achieve imaging of the inner and outer walls in a single clamping operation.

[0028] The first station is used for imaging the outer wall, employing a fixed camera to take a top-down shot while the pipe rotates to complete the image capture. The second station is used for imaging the inner wall, with a periscope-style probe inserted into the pipe; the image is taken after the pipe has finished rotating and stabilized to a stationary state.

[0029] The entire structure can be placed under a Class 10 clean laminar flow hood, and moving parts are lubricated with vacuum-degassed grease.

[0030] For an external surface imaging device, it may include a camera, a lens, a light source, and a motion mechanism. The camera can be a 5-megapixel, 2448×2048 high-resolution camera, such as the Basler acA2440-75gm. The lens can be a high-resolution telecentric lens, such as the Moritex MML2-HR65DVI. The light source can be a high-density coaxial incident polarized light source with a built-in polarizing beam splitter, such as the CCSLDR2-50BL2. The motion mechanism may include a motor, a rotary table, and grippers.

[0031] The internal surface imaging device may include a camera, a light source, a probe, a relay lens, and a motion mechanism. The probe can be a sapphire lens with an integrated 90° side-viewing prism, and the viewing angle is adjustable. The relay lens can be an infinity-conjugate lens group, coupling the endoscopic optical image to the camera sensor. The camera uses a 5-megapixel global shutter CMOS sensor, forming the internal wall imaging optical path with the probe; specifically, it can be a FLIR BFS-U3-51S5M. The light source uses a custom-designed ring-shaped fiber optic beam guide, with the fiber optic cable wrapped around the side-viewing prism of the probe and connected to an external high-power LED, allowing for switching of the light source.

[0032] Here, 3D point cloud data can be acquired using a confocal laser scanning microscope (CLSM) and information such as curvature and slope can be derived.

[0033] Figure 1This is a flowchart illustrating a method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention, provided in an embodiment of this application. Figure 1 As shown, in one embodiment of this application, a method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention is provided. The method includes: S1. Based on the acquired 2D images and 3D point clouds of the welded special gas pipeline, identify the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline.

[0034] In one feasible implementation, after the welding robot completes the welding, the welded special gas pipeline can be cooled to 80°C by air cooling (purging with high-purity N2). The grippers then pick up the welded special gas pipeline and move it to the rotary inspection station.

[0035] The 2D images here require images of the welded areas to be acquired from both the inside and outside of the welded special gas pipeline.

[0036] The welding area here includes at least the fusion zone, the heat-affected zone, and the ordinary zone. The fusion zone is the area formed by the melting, atomic-level mixing, and solidification of the base material and filler material of the special gas pipeline. It has a uniform composition but exhibits slight segregation, and is continuously distributed along the weld centerline without obvious boundary abrupt changes.

[0037] The heat-affected zone is the area in which the microstructure and mechanical properties of the base material change due to the welding thermal cycle (but it does not melt). It is a sensitive area for crack initiation and intergranular corrosion, and in this embodiment, it can be regarded as the primary focus area.

[0038] This allows for the acquisition of multiple frames of external and internal surface images of welded special gas pipelines. Through manual annotation and cropping or AI model recognition, images of the internal fusion zone, external fusion zone, internal heat-affected zone, and external heat-affected zone of the welded special gas pipeline can be identified.

[0039] S2. Input the images of the internal fusion zone, external fusion zone, internal heat-affected zone, and external heat-affected zone of the welded special gas pipeline into the pre-trained welding defect recognition model to obtain the defect recognition results output by the welding defect recognition model.

[0040] The defect identification results are used to indicate the type of welding defects and the degree of defect quantification deviation in welded special gas pipelines.

[0041] In this embodiment, a spatial attention module can be embedded after the convolutional layers of a classic CNN convolutional neural network model. Through multiple iterations and optimization of the weight allocation algorithm, high weights are assigned to high-defect areas such as the edges of inner weld seams, while low weights are assigned to defect-free straight pipe areas. The model output includes the defect type, precise location, and quantified deviation. Specifically, a spatial mask can be constructed based on the "defect occurrence probability density function," assigning high weights to high-probability defect occurrence regions.

[0042] Furthermore, the model can be fully trained for 8,000 rounds based on a pre-built dataset of quantified weld defects, continuously optimizing the weight allocation of the spatial attention module and the feature extraction logic of the CNN model.

[0043] Taking the tungsten inert gas (TIG) rail welding process of UHP-grade SUS316L EP (6.35×0.89mm) pipeline as an example, based on more than 2,000 sets of process test data and actual production experience, the defect types can include internal weld oxidation discoloration, weld convexity deviation, micro-porosity defects, undercut defects, etc.

[0044] S3. Based on the defect identification results, the corresponding process compensation scheme is matched in the process parameter association library to adjust the welding process parameters of the special gas pipelines to be welded by the tungsten inert gas welding equipment.

[0045] By deploying a trained and validated welding defect recognition model to the intelligent vision inspection equipment on the production line, real-time data linkage with the SWS-M200 tungsten inert gas (TIG) rail welding machine is achieved via industrial Ethernet. Taking the production of a batch of UHP-grade SUS316L EP pipe welding as an example, the detection model performs real-time inspection of the completed weld seams and discovers a defect in a certain weld seam with slight oxidation and discoloration of the inner weld seam, which is quantitatively judged as "the color difference of the inner weld seam surface exceeds the standard, with no other dimensional defects." The model automatically matches the corresponding process quantitative compensation scheme from the association library, which is "increasing the argon gas inner diameter purging flow rate from 3L / Min to 4L / Min and extending the delayed gas delivery time from 20s to 25s," and sends this compensation scheme to the SWS-M200 welding machine. After receiving the compensation scheme, the SWS-M200 welding machine automatically adjusts the argon gas inner diameter purging flow rate and the delayed gas delivery time, and welds subsequent seams according to the new process parameters.

[0046] Furthermore, if the defect identification result is that the weld convexity exceeds the tolerance by more than 0.089 mm, the process compensation scheme can be to reduce the welding current from 50A to 45~48A, while keeping the welding speed unchanged at 8.91 mm / min.

[0047] If the defect identification result is a micropore defect, the process compensation scheme can be to increase the argon outer diameter flow rate from 12L / Min to 13~15L / Min.

[0048] If the defect identification result is an undercut defect, the process compensation scheme can be to reduce the peak welding current by 4~6A and calibrate the tungsten electrode angle to 19°.

[0049] This application provides an automated, spatial attention-based intelligent inspection and compensation method for semiconductor pipeline welds. By fusing 2D images and 3D point cloud data, it can accurately identify and distinguish the internal / external fusion zone and heat-affected zone of the welding area in special gas pipelines. It utilizes an AI model to automatically identify welding defect types and outputs a quantitative deviation value for each defect. Based on the quantified defect identification results, it automatically matches the corresponding compensation scheme in a process parameter association library and feeds it back to the welding equipment for parameter adjustment. This allows for dynamic optimization of the subsequent welding process, effectively preventing the recurrence of similar defects and significantly improving the pass rate of special gas pipeline welding, ensuring the consistency and stability of welding quality.

[0050] Example 2 In one embodiment of this application, a pipeline weld area segmentation model based on the fusion of 2D images and 3D point clouds is provided to identify the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of a welded special gas pipeline.

[0051] The segmentation model here may include a 2D image segmentation module (YOLOv8-seg + CBAM), a 3D point cloud preprocessing module, a 2D-3D spatial alignment module, a 2D mask-based coarse point cloud screening module, a mask-based fine screening module, and an output fusion and visualization module.

[0052] The 2D image segmentation module performs pixel-level segmentation on the 2D images of the inner and outer surfaces, identifying four types of regions as masks on the image plane. The input to the 2D image segmentation module is the inner / outer surface image, and the output is four types of regions on the outer / inner surface: 0-external fusion area, 1-external heat-affected area, 2-background (reserved), 3-background, and the corresponding detection boxes.

[0053] The 3D point cloud preprocessing module filters, downsamples, and calculates normal vectors from the original point cloud to prepare for subsequent fusion. The module takes point cloud coordinates and RGB values ​​as input and outputs filtered point cloud coordinates, normal vectors for each point, and curvature for each point.

[0054] The 2D-3D spatial alignment module projects 3D point clouds onto a 2D image plane, establishing a correspondence between pixels and point clouds. The inputs to the 2D-3D spatial alignment module can be the output of the 3D point cloud preprocessing module, as well as camera intrinsic parameters and camera distortion coefficients. The outputs are the 2D image coordinates for each 3D point, the depth value for each pixel location, and the 3D point index (index mapping) for each pixel.

[0055] The 2D mask-based point cloud coarse filtering module utilizes the detection boxes output by YOLOv8 to quickly remove background point clouds, reducing subsequent computation. The input to this module is the output of the first three modules. Its output includes the point cloud coordinates within the detection boxes, the 2D pixel coordinates of the corresponding points within the ROI, and a coarse classification label indicating which detection box the point belongs to.

[0056] The mask-based fine-tuning module utilizes 2D segmentation masks to assign precise four-class region labels to each point cloud. The input to the mask-based fine-tuning module is the output of the first and fourth modules. The output of the mask-based fine-tuning module consists of a dictionary of four keys, labels, and confidence scores. Each key corresponds to a set of point clouds representing a region, the labels indicate the final category of each point, and the confidence scores represent the confidence level for each point's classification.

[0057] The output fusion and visualization module integrates the 2D image segmentation results and 3D point cloud classification results to generate the final four-category region data. The inputs to the output fusion and visualization module are the outputs of the first and fifth modules, as well as the inner / outer surface images of the original input. The outputs of the output fusion and visualization module are 2D mask images of the four regions, 3D point clouds of the four regions, and RGB images of the four regions. These RGB images are overlaid with color-labeled visualization results (e.g., red - external fusion region, green - external heat-affected zone, blue - internal fusion region, yellow - internal heat-affected zone).

[0058] Because the reflective metal in the welded area affects the quality of the acquired 2D images, the defect recognition model performs poorly. In one embodiment of this application, three consecutive surface images of the welded special gas pipeline can be acquired under argon protection using an industrial camera positioned at the center of the weld. The second and third surface images are then pixel-aligned and white-balanced using the first surface image. An enhanced brightness map is determined using the calibrated second and third surface images. A fused brightness map is generated based on the enhanced brightness map and the first surface image. A 2D surface image is then generated based on the fused brightness map.

[0059] The first frame here is primarily used to calculate the color temperature deviation of the light source and generate a real-time correction matrix. The second and third frames are spaced less than 5ms apart to freeze argon gas flow disturbances. This ensures more stable and accurate results for subsequent model inputs and evaluation of other quantitative indicators.

[0060] In one feasible implementation, since polarization devices cannot be placed in the inner wall space, a multispectral ratio method can be used instead of the polarization method to distinguish materials. A 5mm endoscope probe is inserted into the tube by a pneumatic pusher until the lens is directly below the weld (confirmed by laser ranging). Without rotating the probe, a 360° view of the inner wall is obtained in a single image using a conical reflector or prism for ring imaging. Specifically, three consecutive frames of surface images of the welded special gas pipeline's interior can be acquired using the following method: The first surface image was acquired under 470nm blue light illumination. The second surface image was acquired under 525nm green light illumination. The third surface image was acquired under 850nm red light illumination.

[0061] For the outer surface, three consecutive frames of surface images of the welded special gas pipeline can be acquired using the following method: The first surface image was acquired under 0° polarization and white light illumination. The second surface image was acquired under 0° polarization and 470nm blue light illumination. The third surface image was acquired under 90° polarization and 470nm blue light illumination.

[0062] The system synchronously triggers the acquisition of three frames of raw images, all of which are stored in the buffer in RAW12 format (12-bit linear data).

[0063] After acquisition, pixel alignment and bad pixel correction are required to avoid sub-pixel shifts that may be caused by minor vibrations or optical vibrations.

[0064] In a specific embodiment, pixel alignment of the second frame surface image / third frame surface image can be performed in the following manner: Fourier transforms are performed on the third frame surface image, the first frame surface image, and the second frame surface image respectively to obtain the corresponding spectra. The cross-power spectrum is calculated based on the spectrum corresponding to the first frame surface image and the spectrum corresponding to the second / third frame surface image. An inverse transform is performed on the cross-power spectrum to determine the peak value, thereby determining the integer pixel displacement value.

[0065] For example, a Fourier transform can be performed on the first frame and the second frame: ; ; in, This is the first frame of the surface image. This is the surface image of the second frame.

[0066] Calculate the cross power spectrum : ; The inverse Fourier transform yields the correlation peaks: ; in, This is a conjugate operation.

[0067] Find the maximum position This can be determined as an integer pixel displacement.

[0068] For each pixel in the second / third frame surface image, the pixel value is added to an integer pixel displacement value to obtain the pixel-aligned second / third frame surface image. Taking the second frame surface image as an example, it can be represented as: .

[0069] Next, polarization difference calculations are performed to eliminate specular reflections. This includes determining the enhanced brightness map using the following methods: The absolute difference between the calibrated second and third frame surface images is calculated pixel-by-pixel to determine the differential intensity map. A nonlinear mapping is then applied to the differential intensity map to determine the enhanced brightness map.

[0070] Specifically, it can be calculated:

[0071] right Pixels with a value greater than 80 can be retained to determine a 12-bit single-channel differential intensity map.

[0072] Next, the weak signals in the difference map (such as the 10 gray-level differences of the microcrack) are stretched to the full dynamic range to facilitate subsequent observation and algorithm processing.

[0073] ; in, This is used to stretch shadow details. In other methods, here... It can also be based on Dynamic adjustment, for example If less than 20, then , If it is between 20 and 200, then , If it is greater than 200, then .

[0074] Finally, an 8-bit single-channel high-contrast brightness map is output. .

[0075] The next step is to use the first frame for chromaticity fusion to restore the true color information while maintaining high contrast (for ΔE color difference calculation).

[0076] The blended brightness map can be generated in the following ways: The first frame surface image is converted to a color space to determine the corresponding Y, Cb, and Cr channels. The Cb and Cr channels corresponding to the first frame surface image and the Y channel corresponding to the enhanced brightness map are combined to generate a fused brightness map.

[0077] Specifically, the chroma can be extracted from the first frame image, and the first frame image (RAW format) can be debayer interpolated to convert it into an RGB image. The RGB image is then converted to the YCbCr color space. The Y channel (luminance) is discarded, while the Cb (blue difference) and Cr (red difference) channels are retained.

[0078] Next, the obtained single-channel high-contrast brightness map As a new Y channel, it is combined to generate a fused brightness map, which is then converted to RGB format, finally obtaining a 24-bit RGB inner / outer surface image.

[0079] The processed image eliminates the effects of reflections and prevents overexposure (saturation in areas above 100℃) in the high-temperature zone of the weld during long exposures, thus avoiding loss of details. This not only improves the recognition accuracy of the four regions, but also amplifies and enhances the contrast of small defects in the image, which is beneficial for the subsequent welding defect recognition model to identify small defects, such as microcracks. It also significantly improves the recognition effect and reduces the error between the calculated color difference value of the oxide film and the actual measured value by the spectrophotometer.

[0080] The technical solution provided in this application ensures that every step from sensor RAW data to the final analyzed image is traceable and quantifiable, fully meeting the stringent requirements of semiconductor defect detection for image consistency and repeatability.

[0081] Example 3 Based on the same concept, one embodiment of this application provides a smart detection and compensation device for semiconductor pipeline weld seams based on spatial attention, the device comprising: The segmentation module is used to identify the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline based on the acquired 2D images and 3D point clouds of the welded special gas pipeline. The defect identification module is used to input the images of the internal fusion zone, external fusion zone, internal heat-affected zone and external heat-affected zone of the welded special gas pipeline into the pre-trained welding defect identification model to obtain the defect identification results output by the welding defect identification model. The defect identification results are used to indicate the type of welding defect and the degree of defect quantification deviation of the welded special gas pipeline. The execution module is used to match the corresponding process compensation scheme in the process parameter association library based on the defect identification results, so as to adjust the welding process parameters of the special gas pipelines to be welded by the tungsten inert gas welding equipment.

[0082] In a preferred embodiment, under argon protection, three consecutive surface images of the welded special gas pipeline are acquired by an industrial camera positioned at the center of the weld. Pixel alignment and white balance calibration are performed on the second and third surface images using the first surface image; an enhanced brightness map is determined using the calibrated second and third surface images; a fused brightness map is generated based on the enhanced brightness map and the first surface image; and a 2D surface image is generated based on the fused brightness map.

[0083] In a preferred embodiment, the second frame surface image / the third frame surface image are pixel aligned in the following manner: Fourier transforms are performed on the first, second, and third frame surface images respectively to obtain the corresponding spectra; the cross-power spectrum is calculated based on the spectrum corresponding to the first frame surface image and the spectrum corresponding to the second / third frame surface image; the cross-power spectrum is inversely transformed to determine the peak value, and the integer pixel displacement value is determined; for each pixel in the second / third frame surface image, the pixel value of the pixel is added to the integer pixel displacement value to obtain the pixel-aligned second / third frame surface image.

[0084] In a preferred embodiment, the enhanced brightness map is determined by the following method: The absolute difference between the calibrated second and third frame surface images is calculated pixel by pixel to determine the differential intensity map; the differential intensity map is then nonlinearly mapped to determine the enhanced brightness map.

[0085] In a preferred embodiment, the fused brightness map is generated in the following manner: The first frame surface image is converted to a color space, and the corresponding Y, Cb and Cr channels are determined. The Cb and Cr channels corresponding to the first frame surface image and the Y channel corresponding to the enhanced brightness map are combined to generate a fused brightness map.

[0086] In a preferred embodiment, three consecutive frames of surface images of the welded special gas pipeline and its exterior are acquired in the following manner: The first frame of the surface image was acquired under 0° polarization and white light illumination conditions; The second frame of the surface image was acquired under 0° polarization and 470nm blue light illumination conditions. The third frame of the surface image was acquired under 90° polarization and 470nm blue light illumination.

[0087] In a preferred embodiment, three consecutive frames of surface images of the interior of the welded special gas pipeline are acquired in the following manner: The first frame of the surface image was acquired under 470nm blue light illumination. The second frame of the surface image was acquired under 525nm green light illumination. The third frame of the surface image was acquired under 850nm red light illumination.

[0088] Example 4 Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 2 As shown, the electronic device 200 includes a processor 210, a memory 220, and a bus 230.

[0089] The memory 220 stores machine-readable instructions that can be executed by the processor 210. When the electronic device 200 is running, the processor 210 and the memory 220 communicate via the bus 230. When the machine-readable instructions are executed by the processor 210, the steps of a method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention as described in the above method embodiment can be executed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0090] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of a method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0091] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0092] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0093] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0095] It should be noted that if the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0097] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent detection and compensation of semiconductor pipeline weld seams based on spatial attention, characterized in that, The method includes: Based on the acquired 2D surface images and 3D point clouds of the welded special gas pipeline, the internal fusion zone image, external fusion zone image, internal heat-affected zone image and external heat-affected zone image of the welded special gas pipeline are identified. The 2D surface images include 2D inner surface images and 2D outer surface images. The images of the internal fusion zone, external fusion zone, internal heat-affected zone, and external heat-affected zone of the welded special gas pipeline are respectively input into the welding defect recognition model based on spatial attention to obtain the defect recognition results output by the welding defect recognition model. The defect recognition results are used to indicate the type of welding defect and the degree of defect quantification deviation in the welded special gas pipeline. Based on the defect identification results, the corresponding process compensation scheme is matched in the process parameter association library to adjust the welding process parameters of the special gas pipelines to be welded by the tungsten inert gas welding equipment. The system employs a dual-station rotary table and dual-camera module architecture to achieve imaging of the inner and outer walls in a single clamping operation. The industrial camera at the first station is used for imaging the outer wall, employing a fixed overhead shooting method, with the welded special gas pipe rotating to complete the imaging. The industrial camera at the second station is used for imaging the inner wall, with the periscope-style probe of the industrial camera extending into the pipe, and the imaging is performed when the welded special gas pipe has finished rotating and stabilized to a stationary state. Under argon protection, the industrial camera acquires three consecutive surface images of the welded special gas pipeline under three different light sources and polarization conditions. Pixel alignment and white balance calibration are performed on the second and third surface images using the first surface image; The enhanced brightness map was determined using the calibrated second and third frame surface images; A fused brightness map is generated based on the enhanced brightness map and the first frame surface image; The 2D surface image is generated based on the fused brightness map.

2. The method according to claim 1, characterized in that, The second / third frame surface image is pixel-aligned using the following method: Fourier transforms were performed on the first frame surface image, the second frame surface image, and the third frame surface image respectively to obtain the corresponding spectra; The cross power spectrum is calculated based on the spectrum corresponding to the first frame surface image and the spectrum corresponding to the second frame surface image / third frame surface image. Perform an inverse transform on the cross power spectrum to determine the peak value, and then determine the integer pixel displacement value; For each pixel in the second frame surface image / third frame surface image, add the pixel value of that pixel to the integer pixel displacement value to obtain the pixel-aligned second frame surface image / third frame surface image.

3. The method according to claim 2, characterized in that, The enhanced brightness map was determined using the following method: The absolute value of the difference between the calibrated second frame surface image and the third frame surface image is calculated pixel by pixel to determine the difference intensity map; The differential intensity map is nonlinearly mapped to determine the enhanced brightness map.

4. The method according to claim 3, characterized in that, The blended brightness map is generated using the following method: The first frame surface image is converted to a color space to determine the corresponding Y channel, Cb channel, and Cr channel; The Cb and Cr channels corresponding to the first frame surface image and the Y channel corresponding to the enhanced brightness map are combined to generate a fused brightness map.

5. The method according to claim 1, characterized in that, Three consecutive frames of surface images of the welded special gas pipeline and its exterior were acquired using the following method: The first frame of the surface image was acquired under 0° polarization and white light illumination conditions; The second frame of the surface image was acquired under 0° polarization and 470nm blue light illumination conditions. The third frame of the surface image was acquired under 90° polarization and 470nm blue light illumination.

6. The method according to claim 1, characterized in that, Three consecutive frames of surface images of the interior of the welded special gas pipeline were acquired using the following method: The first frame of the surface image was acquired under 470nm blue light illumination. The second frame of the surface image was acquired under 525nm green light illumination. The third frame of the surface image was acquired under 850nm red light illumination.

7. A smart detection and compensation device for semiconductor pipeline welds based on spatial attention, characterized in that, The apparatus is applicable to the method according to any one of claims 1-6, wherein the apparatus comprises: The segmentation module is used to identify the internal fusion zone image, external fusion zone image, internal heat-affected zone image, and external heat-affected zone image of the welded special gas pipeline based on the acquired 2D surface image and 3D point cloud of the welded special gas pipeline. The defect identification module is used to input the images of the internal fusion zone, external fusion zone, internal heat-affected zone and external heat-affected zone of the welded special gas pipeline into the welding defect identification model based on spatial attention, so as to obtain the defect identification results output by the welding defect identification model. The defect identification results are used to indicate the type of welding defect and the degree of defect quantification deviation of the welded special gas pipeline. The execution module is used to match the corresponding process compensation scheme in the process parameter association library based on the defect identification results, so as to adjust the welding process parameters of the special gas pipelines to be welded by the tungsten inert gas welding equipment.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the intelligent detection and compensation method for semiconductor pipeline welds based on spatial attention as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the intelligent detection and compensation method for semiconductor pipeline welds based on spatial attention as described in any one of claims 1 to 6.

Images