Thread detection method, system and equipment for high-temperature alloy fastener

Abstract

The invention relates to a thread detection method, system and equipment for a high-temperature alloy fastener. The method comprises the following steps: firstly, acquiring images of a thread to be detected from multiple observation angles, constructing a first image set, and registering the first image set to form a second image set; then, constructing a phase difference image based on the second image set, and extracting thread disturbance characteristics in combination with a direction sensitive filter bank to generate a disturbance enhanced image; and finally, through analysis of the disturbance enhancement image, whether micro-damage exists on the surface of the thread is identified, and an identification result is output. Therefore, on the premise that the thread structure is not damaged, high-precision automatic identification of tiny surface defects can be achieved, the technical problems that traditional visual inspection is poor in stability, low in efficiency and weak in micro-damage identification capacity are solved, and the reliability and the intelligent level of thread detection of the high-temperature alloy fastener are effectively improved.

Description

Technical Field

[0001] The present application relates to the technical fields of image processing and quality inspection, and particularly to a thread detection method, system, and device for superalloy fasteners. Background Art

[0002] With the continuous improvement of the performance requirements for components in high-end manufacturing fields such as aerospace and energy equipment, superalloy materials are widely used in the manufacture of fasteners due to their excellent high-temperature strength, oxidation resistance, and corrosion resistance. As key components for connection and load-bearing, the thread quality of superalloy fasteners is directly related to the connection reliability and service safety of the structure.

[0003] Currently, thread detection mainly relies on manual visual inspection, where an inspector uses a magnifying glass to detect the threads. This traditional detection method has obvious deficiencies: First, the detection process highly depends on the experience and subjective judgment of the inspector, easily resulting in unstable detection results. Second, the efficiency is low, making it difficult to meet the production requirements of large quantities and high consistency. Third, the ability to identify small defects and microscopic damages is limited, posing risks of missed and false detections, and it is difficult to ensure the comprehensiveness and reliability of the thread quality detection of fasteners.

[0004] That is, the traditional thread detection method has technical problems such as poor stability of detection results, low detection efficiency, and insufficient accuracy in defect identification. Summary of the Invention

[0005] Based on this, it is necessary to provide a thread detection method for superalloy fasteners, a thread detection system for superalloy fasteners, and a computer device to address the above technical problems.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: On the one hand, a thread detection method for superalloy fasteners is provided, including: Obtaining a first image set of the thread to be detected; the thread to be detected is the thread on a superalloy fastener, and the first image set is a set of images obtained by imaging the thread to be detected from different observation angles; Registering the image regions of the thread to be detected in the first image set to obtain a second image set; Based on the second image set, constructing a phase difference map; the phase difference map is used to characterize the relative phase differences of the same thread region under different observation angles; Based on the second image set and the phase difference map, using a direction-sensitive filter bank to extract the disturbance features of the thread to be detected, and obtaining a disturbance-enhanced image based on the disturbance features; Based on the perturbation-enhanced image, the image recognition model identifies whether there are micro-damages on the surface of the thread to be detected and outputs the recognition result; the image recognition model is constructed based on a convolutional neural network, and micro-damage refers to surface defects that are invisible under visual observation conditions. The damage types of micro-damage include thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0007] In one embodiment, in the above thread detection method for superalloy fasteners, based on the perturbation-enhanced image, the image recognition model identifies whether there are micro-damages on the surface of the thread to be detected and outputs the recognition result, including: Input the perturbation-enhanced image into the image recognition model; the image recognition model includes a feature adaptation module, a candidate box generation module, and a position regression module; The feature adaptation module performs feature mapping processing on the perturbation-enhanced image and outputs a feature map; The candidate box generation module generates candidate detection regions according to the feature map; The position regression module performs regression processing on the candidate detection regions and outputs the position coordinates of the micro-damage; the recognition result includes the position coordinates.

[0008] In one embodiment, in the above thread detection method for superalloy fasteners, the image recognition model further includes a type discrimination module. The type discrimination module is used to identify the damage type of the micro-damage according to the position coordinates and the corresponding candidate detection regions, and outputs the position coordinates and the corresponding damage type; the recognition result includes the position coordinates and the corresponding damage type.

[0009] In one embodiment, in the above thread detection method for superalloy fasteners, in the first image set, each observation angle includes a corresponding visible light image and an infrared image; Perform registration on the image regions of the threads to be detected in the first image set to obtain a second image set, including: For each observation angle, fuse the corresponding visible light image and infrared image to obtain a fused image at each observation angle; Perform registration on the image regions of the threads to be detected in each fused image to obtain a second image set.

[0010] In one embodiment, in the above thread detection method for superalloy fasteners, after outputting the recognition result, it further includes: performing visual measurement on the micro-damage to obtain the geometric dimension parameters of the micro-damage.

[0011] In one embodiment, in the above thread detection method for superalloy fasteners, performing visual measurement on the micro-damage to obtain the geometric dimension parameters of the micro-damage includes: According to the position coordinates, locate the corresponding visual detection region in the second image set; For the visual detection area, extract the first boundary contour of the micro-damage; According to the first boundary contour, the image acquisition resolution, and the calibration model, calculate the geometric dimension parameters of the micro-damage; the calibration model is used to characterize the mapping relationship between the pixel coordinates in the image acquisition module and the actual space coordinates.

[0012] In one embodiment, in the above-mentioned thread detection method for superalloy fasteners, perform visual measurement on the micro-damage to obtain the geometric dimension parameters of the micro-damage, including: According to the position coordinates, extract the second boundary contour of the micro-damage in the perturbation-enhanced image; According to the second image set, the phase difference map, and the perturbation-enhanced image, construct an image transformation model to obtain the pixel position transformation correction factor during the image enhancement process; the image transformation model is used to characterize the pixel position change between the perturbation-enhanced image and the second image set; According to the second boundary contour, the image acquisition resolution, and the correction factor, calculate the geometric dimension parameters of the micro-damage.

[0013] On the other hand, a thread detection system for superalloy fasteners is also provided, including: An image acquisition module for obtaining a first image set of the thread to be detected; the thread to be detected is the thread on a superalloy fastener, and the first image set is an image set obtained by imaging the thread to be detected from different observation angles; A registration module for registering the image area of the thread to be detected in the first image set to obtain a second image set; A construction module for constructing a phase difference map based on the second image set; the phase difference map is used to characterize the relative phase difference of the same thread area under different observation angles; A generation module for using a direction-sensitive filter bank to extract the perturbation features of the thread to be detected based on the second image set and the phase difference map, and obtaining a perturbation-enhanced image based on the perturbation features; An image recognition model for identifying whether there are micro-damages on the surface of the thread to be detected based on the perturbation-enhanced image and outputting an identification result; the image recognition model is constructed based on a convolutional neural network, and micro-damage refers to surface defects that are invisible under visual observation conditions. The damage types of micro-damage include thermal fatigue microcracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0014] In one embodiment, in the above-mentioned thread detection system for superalloy fasteners, it further includes: a measurement module for performing visual measurement on the micro-damage to obtain the geometric dimension parameters of the micro-damage.

[0015] On the other hand, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned thread detection method for superalloy fasteners are implemented.

[0016] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: For the above-mentioned thread detection method, system and device for superalloy fasteners, first, images of the thread to be detected are collected from multiple observation angles to construct a first image set, and the first image set is registered to form a second image set; subsequently, a phase difference map is constructed based on the second image set, and thread perturbation features are extracted in combination with a direction-sensitive filter bank to generate a perturbation-enhanced image; finally, by analyzing the perturbation-enhanced image, it is identified whether there are micro-damages on the thread surface, and the identification result is output. Therefore, without damaging the thread structure, high-precision automatic identification of micro-surface defects can be realized, solving the technical problems of poor stability, low efficiency and weak micro-damage identification ability of traditional visual inspection, and effectively improving the reliability and intelligent level of thread detection of superalloy fasteners. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of a thread detection method for superalloy fasteners in one embodiment.

[0019] Figure 2 It is a schematic flowchart of thread micro-damage identification based on a perturbation-enhanced image in one embodiment.

[0020] Figure 3 It is a schematic module structure diagram of a thread detection system for superalloy fasteners in one embodiment Figure 1 .

[0021] Figure 4 It is a schematic module structure diagram of a thread detection system for superalloy fasteners in one embodiment Figure 2 . DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will describe in detail specific embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the description of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0023] In the description of the present invention, unless otherwise clearly defined and limited, terms such as "arranged", "installed", "connected", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0024] The orientation or positional relationship indicated by terms such as "upper", "lower", "left", "right", "front", "rear", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the present invention is usually placed during use. It is only for the convenience of description and simplification of the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0025] Terms such as "first", "second", "third", etc. are only used to distinguish elements with similar attributes, rather than indicating or implying relative importance or a specific order.

[0026] The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion. In addition to including the listed elements, it may also include other elements not specifically listed.

[0027] The following will describe in detail the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] In one embodiment, as Figure 1 shown, the embodiment of the present application provides a thread detection method for superalloy fasteners, including the following processing steps S11 - S15: S11, obtaining a first image set of the thread to be detected; the thread to be detected is the thread on a superalloy fastener, and the first image set is a set of images obtained by imaging the thread to be detected from different observation angles.

[0029] It can be understood that the thread structure is a three-dimensional spiral structure, and there may be tiny defects on the surface. By using a multi-angle imaging method (such as rotating shooting or arranging multiple cameras), the thread surface is photographed from multiple different perspectives to obtain the first image set.

[0030] In some embodiments, a multi-angle imaging device can be used to collect images of the threads on a superalloy fastener. The multi-angle imaging device can include a rotation control platform and an industrial camera, and can also include a light source system. First, the superalloy fastener to be detected is fixed on the rotation control platform. Through the step rotation function of the platform, the thread is gradually rotated at a preset angular interval (for example, every 20 degrees). At each angle, the industrial camera is used to image the thread surface to obtain a clear image frame.

[0031] Before shooting, the light source can also be controlled for adaptive adjustment to ensure imaging contrast and edge sharpness. The images obtained at multiple angles (for example, 18 viewing angles) are numbered in order of angle to form the first image set of the thread to be detected.

[0032] In other embodiments, a multi-camera arrangement is used to collect images of the threads on a superalloy fastener. For example, multiple (for example, 4 - 8) industrial cameras can be evenly arranged around the thread to be detected, so that each camera synchronously images the thread surface from different directions. Each camera is arranged at a certain angle relative to the thread axis to cover the visible area of the thread surface.

[0033] During the shooting process, all cameras synchronously collect images under unified control to reduce errors caused by workpiece vibration or environmental changes. In addition, independent or annular adjustable light sources can be configured according to the camera positions to provide sufficient and uniform illumination, and combined with polarizers or diffuse reflection devices to reduce surface reflection interference. The multiple perspective images obtained are sorted according to the camera numbers to form the first image set.

[0034] S12, register the image regions of the threads to be detected in the first image set to obtain the second image set.

[0035] It can be understood that registration refers to aligning two or more images taken from different angles to the same reference coordinate system. That is, the regions at the same physical position in the thread images obtained from different perspectives are aligned in the image space so that they can be compared or fused at the pixel level.

[0036] Specifically, during the image registration process, it mainly includes steps such as geometric pre-registration and feature refinement alignment.

[0037] In some embodiments, rotational shooting is adopted for image acquisition. In the geometric pre-registration stage, according to the angle of each rotation of the platform (for example, 20 degrees each time), combined with the known direction of the rotation axis and the pose of the industrial camera, a unified three-dimensional observation reference coordinate system is constructed. Based on the rotation angle information of each frame of the image, the pose transformation relationship between the industrial camera and the fastener during image acquisition can be deduced, so as to map each image in space to the unified reference coordinate system. Specifically, affine transformation or projective transformation can be initially used to perform geometric mapping on the image, so that the thread regions at different angles are roughly aligned in the image plane to a unified reference frame.

[0038] In some other embodiments, a shooting method with multiple cameras arranged is adopted for image acquisition. In the geometric pre-registration stage, according to the position and orientation parameters of each camera in the three-dimensional space, the images collected by each camera are uniformly projected under a common reference coordinate system, and the geometric correspondence between the images is initially established.

[0039] After completing the geometric pre-registration, it enters the feature refinement and alignment stage. It mainly focuses on the thread region to further improve the alignment accuracy between different images. Specifically, first, distinguishable structural features are automatically identified in the image, and the corresponding relationships of these features are searched among the images from multiple perspectives. Then, by introducing a consistency check mechanism, the possible mis-matched points are excluded, and the true and reliable feature pairs are retained. Through the feature alignment process, the consistency of the thread region in the image under different perspectives can be further improved, providing a more accurate image basis for subsequent thread perturbation analysis, defect identification, etc.

[0040] To further improve the registration accuracy and robustness, auxiliary geometric reference objects can be introduced in some implementations. For example, in the rotational shooting scenario, reference marks can be set on the shooting platform or the background board to assist in identifying the rotation center of the platform and compensating for the cumulative rotation error; while in the multi-camera shooting scenario, reference marks can also be set as common features among multiple perspectives to improve the relative pose estimation accuracy between cameras, which is especially suitable for workpieces with repetitive textures or weak patterns on the surface.

[0041] S13, based on the second image set, construct a phase difference map; the phase difference map is used to characterize the relative phase difference of the same thread region under different observation angles.

[0042] It can be understood that on the premise that the image registration has been completed, by analyzing the relative change of the same physical position in the images from different observation angles in terms of image gray level or texture features, a phase difference map for quantifying the local phase change is constructed. The phase difference map can effectively reflect the response consistency of the thread surface under multi-perspective imaging, and its phase difference can reveal potential micro-perturbation regions on the thread surface, such as phase offsets caused by micro-damages such as micro-cracks and corrosion pits.

[0043] Specifically, first, an image sequence covering the same thread region after registration is extracted from the second image set, and sequence modeling processing is performed on the gray scale or texture features at each image position. For example, within a local image window, a sliding window method is used to obtain the gray scale change sequence of each pixel in multiple images, and Fourier analysis or other phase estimation methods are used to calculate the relative phase value of this position at different viewing angles.

[0044] It should be understood that considering that the thread to be detected is a three-dimensional entity with a 360-degree annular structure, while the phase difference map is a two-dimensional image (2D gray scale image or heat map), the registered images can be divided into several viewing angle regions (for example, 4 viewing angle regions). For each viewing angle region, a phase difference map is constructed to characterize the relative phase change of this viewing angle region at different observation angles. Multiple phase difference maps can be combined to represent the phase response distribution of the entire thread surface.

[0045] S14. Based on the second image set and the phase difference map, use a direction-sensitive filter bank to extract the perturbation features of the thread to be detected, and obtain a perturbation-enhanced image based on the perturbation features.

[0046] It can be understood that on the basis of obtaining the second image set and the phase difference map, a direction-sensitive filter bank is used to respond to the image changes in multiple directions of the thread region to extract the possible perturbation features therein. Among them, the direction-sensitive filter can produce a significant response to the features (such as edges, textures, and perturbations) in a specific direction in the image, and the direction-sensitive filter bank includes multiple direction-sensitive filters with different direction response characteristics. The perturbation feature refers to the region that still exhibits local discontinuity, mutation, or asymmetric characteristics after multi-view fusion and phase analysis.

[0047] Specifically, a Gabor filter bank or a directional derivative filter bank can be used to jointly process the second image set and the phase difference map. In each local region window of the image, filter responses are applied along multiple directions to generate a direction response map, and then fusion is performed according to the amplitude, direction consistency, or local perturbation index of the filter response to generate a perturbation-enhanced image. The perturbation-enhanced image strengthens the micro-perturbation region while maintaining the original structural features, providing a more significant micro-damage characterization input for the subsequent image recognition model.

[0048] S15. Based on the perturbation-enhanced image, an image recognition model identifies whether there is micro-damage on the surface of the thread to be detected and outputs the recognition result; the image recognition model is constructed based on a convolutional neural network, and micro-damage refers to surface defects that are invisible under visual observation conditions. The damage types of micro-damage include thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0049] It can be understood that after obtaining the perturbation-enhanced image, an image recognition model based on a convolutional neural network is used to automatically recognize and discriminate the thread surface in the perturbation-enhanced image to determine whether there are micro-damages.

[0050] Micro-damage refers to surface defects that are invisible under visual observation conditions, but micro-damage has a potential impact on the service performance of fasteners. Superalloy fasteners are often used in environments that withstand harsh service conditions such as high temperature, high stress, and strong corrosion. Surface integrity is crucial for the long-term reliability of fasteners. The thread area, as a stress concentration site, is more prone to early damage, and tiny surface perturbations are very likely to develop into fatigue cracks or corrosion initiation points.

[0051] The damage types of micro-damage on the thread surface of superalloy fasteners mainly include thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0052] For superalloy fasteners, the manufacturing process undergoes process steps such as warm forging, grinding and thread rolling, solution treatment, aging heat treatment, and rolling. During warm forging and grinding and thread rolling, due to the action of the material under high temperature and high strain rate conditions, it is extremely easy to introduce residual tensile stress and microscopic plastic deformation in stress concentration areas such as the thread root. If the local deformation is uneven or the surface quality control is not strict, it may induce thermal fatigue micro-cracks or surface erosion phenomena. Solution treatment and aging heat treatment involve a long-term high-temperature treatment process, and the high-temperature oxidation reaction is active. If the surface protection measures are not in place, it is easy to form local oxidation pits on the thread surface, becoming the starting point for later crack propagation. In addition, if there are situations such as excessive residual stress or surface residual corrosion medium in the manufacturing process, it is easy to trigger stress corrosion cracks.

[0053] Specifically, based on the perturbation-enhanced image of the thread surface to be detected, the constructed image recognition model can be used to detect micro-damage on the thread surface of superalloy fasteners. The image recognition model adopts a convolutional neural network architecture, which can include a module for extracting image features and a module for outputting recognition results.

[0054] The image recognition model can adopt different structural configurations: One way is to only include an image feature extraction module and a judgment module, which is used to output the judgment result of whether there is micro-damage on the thread surface. Another way is that the image recognition model further includes a position regression module, which outputs the position coordinates on the basis of recognizing that there is micro-damage. Preferably, the image recognition model can also integrate a type discrimination module to identify the type of detected micro-damage. The output recognition results include position coordinates and damage types, so as to achieve accurate recognition and classification of thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0055] In the above-mentioned thread detection method for superalloy fasteners, by acquiring thread images from multiple observation angles and performing registration to construct a phase difference map, the characterization ability of complex surface micro-damage is effectively enhanced, and the stability of the detection results is improved. Combining with a direction-sensitive filter bank to extract perturbation features and generate a perturbation-enhanced image makes the micro-perturbations easier to identify, significantly improving the defect recognition accuracy. Finally, an image recognition model based on a convolutional neural network is used to automatically recognize the perturbation-enhanced image, which has good generalization performance and high processing efficiency. In summary, the key technical problems such as low detection efficiency, poor stability of detection results, and insufficient recognition accuracy of micro-damage in the prior art are solved, and it is applicable to the high-reliability detection requirements of surface micro-damage of superalloy fasteners.

[0056] In some embodiments, the above-mentioned thread detection method for superalloy fasteners, as Figure 2 shown, step S15: Based on the perturbation-enhanced image, the image recognition model identifies whether there is micro-damage on the surface of the thread to be detected and outputs the recognition result, specifically including: S151, input the perturbation-enhanced image into the image recognition model; the image recognition model includes a feature adaptation module, a candidate box generation module, and a position regression module.

[0057] It can be understood that the image recognition model can be offline trained on a perturbation-enhanced image dataset containing a large number of thread micro-damage samples to enhance the sensitivity and recognition accuracy of the image recognition model for micro-damage. The overall image recognition model adopts a multi-stage structure. The feature adaptation module can be constructed based on a convolutional neural network and is used to perform feature mapping processing on the perturbation-enhanced image. The candidate box generation module can integrate a region proposal network to generate candidate detection areas for micro-damage. The position regression module can adopt a lightweight fully connected layer or a small convolutional sub-network to perform precise position regression on the candidate detection areas.

[0058] S152, the feature adaptation module performs feature mapping processing on the perturbation-enhanced image and outputs a feature map.

[0059] It can be understood that since the input of the image recognition model is an image processed by perturbation enhancement and has more significant surface detail features, there is no need to perform deep semantic feature extraction anymore. The feature adaptation module is mainly used to perform spatial mapping processing on the effective information in the perturbation-enhanced image and maintain the structural features of the image for subsequent modules to use. The feature adaptation module can be constructed based on a shallow network structure such as ResNet (residual network) or MobileNet (lightweight convolutional neural network), and only the first few convolutional layers and normalization layers are retained to achieve effective feature mapping of the perturbation-enhanced image.

[0060] S153, the candidate box generation module generates candidate detection areas according to the feature map.

[0061] It can be understood that the candidate box generation module is used to propose a series of regions on the feature map that may contain micro-damages, which are called candidate detection regions. Specifically, the candidate box generation module can integrate the Region Proposal Network (RPN), slide at different positions on the feature map to generate multiple anchor boxes, and predict the probability of the existence of micro-damages and the rough bounding box coordinates based on these anchor boxes. Subsequently, by sorting the predicted scores and combining the non-maximum suppression (NMS) strategy, the one with the highest score is retained from the overlapping boxes, and the redundant boxes are suppressed, so as to filter out a group of candidate detection regions with higher confidence for subsequent fine regression processing of the position.

[0062] S154, the position regression module performs regression processing on the candidate detection regions and outputs the position coordinates of the micro-damages; the recognition result includes the position coordinates.

[0063] It can be understood that the position regression module is used to perform fine positioning on the candidate detection regions screened in the previous stage. Specifically, the position regression module can perform coordinate regression on each candidate detection region based on a lightweight fully connected layer or a small convolutional sub-network, and output more accurate bounding box coordinates, that is, the position coordinates of the micro-damages. It should be understood that the candidate box generation module outputs rough target positions, while the position regression module further calibrates these positions and outputs the final recognition result. In addition, the position regression module can also learn the prediction function of the coordinate offset according to the damage annotation information during the training process, so as to improve the positioning accuracy of micro-damage recognition.

[0064] If a certain candidate detection region is judged to have no effective micro-damage after regression during the detection process (for example, the classification confidence is low, or the regression result does not have obvious damage characteristics), then this candidate detection region will be excluded and not included in the final recognition result.

[0065] In the above thread detection method for superalloy fasteners, by introducing the combination of the perturbation-enhanced image and the multi-stage image recognition model, the detection accuracy of the micro-damages on the thread surface of superalloy fasteners is significantly improved. The feature adaptation module, based on the lightweight convolution structure, can efficiently extract the significant texture details in the perturbation-enhanced image. The candidate box generation module filters out the candidate detection regions with high confidence. The position regression module further finely calibrates the position of the detection box and learns to predict the offset by combining the damage annotation information, making the finally output position coordinates of the micro-damages more accurate.

[0066] In some embodiments, for the above thread detection method for superalloy fasteners, the image recognition model further includes a type discrimination module. The type discrimination module is used to identify the damage type of the micro-damage according to the position coordinates and the corresponding candidate detection regions, and output the position coordinates and the corresponding damage type; the recognition result includes the position coordinates and the corresponding damage type.

[0067] It can be understood that the type discrimination module can be constructed using a lightweight classification neural network or a convolutional neural network. By performing offline training on an image dataset containing various micro-damage types, the image recognition model is enabled to distinguish multiple damage types such as thermal fatigue microcracks, oxidation pits, surface erosion, and stress corrosion cracks. The type discrimination module receives the position coordinates output by the position regression module, extracts the image sub-region of the corresponding candidate detection area based on the position coordinates, identifies the damage type of the image sub-region, and finally outputs the position coordinates of each micro-damage and the corresponding damage type.

[0068] In the above thread detection method for superalloy fasteners, by introducing a type discrimination module into the image recognition model, fine recognition of micro-damage on the thread surface of superalloy fasteners is achieved. It can not only accurately locate the position of micro-damage but also clarify the specific damage type. This significantly improves the integrity and practicality of the detection results, especially facilitating damage analysis, life assessment, and maintenance decision-making.

[0069] In some embodiments, in the above thread detection method for superalloy fasteners, in the first image set, each observation angle includes a corresponding visible light image and an infrared image. The image regions of the threads to be detected in the first image set are registered to obtain a second image set, including: for each observation angle, fusing the corresponding visible light image and infrared image to obtain a fused image at each observation angle; registering the image regions of the threads to be detected in each fused image to obtain a second image set.

[0070] It can be understood that each observation angle in the first image set includes a visible light image and an infrared image. To improve the usability and information integrity of the detection images, synchronous visible light and infrared images are acquired for each observation angle, and the two are subjected to image fusion processing to obtain a fused image at each observation angle.

[0071] Visible light images can provide clear surface structure and texture information, while infrared images reveal the uneven heat conduction caused by micro-damage by showing the thermal radiation distribution and temperature differences. Especially for deep cracks and microcracks that are difficult to observe with the naked eye, infrared images can provide key information. By fusing the two types of images, the image recognition model can more comprehensively evaluate the micro-damage on the thread surface, significantly improving the accuracy of micro-damage detection.

[0072] During the actual acquisition process, a dual-camera system can be used to synchronously acquire visible light images and infrared images. The cameras can be a color camera and an infrared thermal imager respectively, and through a perspective correction and synchronous triggering mechanism, the consistency of image data at the same observation angle is achieved.

[0073] For each observation angle, the corresponding visible light image and infrared image are fused to obtain a fused image at each observation angle. Specifically, after registering the visible light image and the infrared image, an image fusion algorithm can be used to integrate the feature information of the two images.

[0074] For example, using the weighted average method, by assigning weights to each pixel point, the pixel values of the visible light image and the infrared image are weighted and fused according to certain weights to finally obtain a fused image.

[0075] Another example is that the principal component analysis (PCA) algorithm can also be used. By extracting the main features of the visible light image and the infrared image and combining these features in a new space, a fused image is obtained. Specifically, first, the visible light image and the infrared image are respectively converted into grayscale images and subjected to principal component analysis to obtain their respective principal component matrices. Then, the principal component matrices of the two images are combined, appropriate principal components are selected for synthesis, and finally, the fused principal component image is converted back to a spatial image through inverse transformation.

[0076] Register the image regions of the threads to be detected in each fused image to obtain a second image set. Specifically, first, select a fused image with a clear thread structure as the reference view, and manually or automatically mark the thread region as the template image. The periodic features of the thread are extracted by edge detection combined with Fourier transform to determine the axial direction and pitch of the thread. The thread region is initially located in other fused images through template image matching.

[0077] For the thread region in each fused image, first, perform gray normalization on the thread regions of the template image and the current image to reduce the influence of brightness differences on the registration accuracy. Subsequently, an iterative optimization method based on image gradients is used to estimate the affine transformation parameters, including geometric relationships such as rotation, scaling, translation, and shear. During the optimization process, a residual weighting mechanism can be introduced to suppress or eliminate pixel points with larger errors by weighting, so as to improve the robustness and accuracy of the registration process. Finally, the affine transformation matrix of the thread region of the current image relative to the template image is obtained.

[0078] After obtaining the affine transformation matrix, the bicubic interpolation method is used to transform and align the current thread image region into the coordinate system of the template image to ensure pixel-level continuity and smoothness. After registration, the thread image region after the overall transformation is retained. The registered thread images of all fused images are combined in the order of observation angles to form a second image set. The second image set has a highly consistent spatial structure, significantly improving the training and detection accuracy of the image recognition model.

[0079] In the above thread detection method for superalloy fasteners, by fusing visible light images and infrared images, the surface texture and thermal distribution information are fully combined, improving the detectability and recognition accuracy of thread micro-damage. Further, through the precise registration of the fused images, a second image set with consistent structure is constructed, effectively eliminating the spatial deviation between multi-angle images, improving the training effect and detection stability of the subsequent image recognition model, and thus significantly enhancing the accurate recognition ability of thread micro-damage in superalloy fasteners.

[0080] In some embodiments, in the above thread detection method for superalloy fasteners, after outputting the recognition result, it further includes: visually measuring the micro-damage to obtain the geometric dimension parameters of the micro-damage.

[0081] It can be understood that in the thread detection method for superalloy fasteners, after outputting the recognition result, the micro-damage area is further visually measured to obtain the geometric dimension parameters of the micro-damage. The geometric dimension parameters may include but are not limited to the length, width, and orientation angle of micro-cracks, or the area and depth of micro-pits and surface erosion areas, which can be used for structural integrity assessment and failure risk analysis.

[0082] The embodiments of the present application provide two ways for visual measurement. The first way: identify based on the perturbation-enhanced image and perform visual measurement on the second image set. That is, the recognition result output based on the perturbation-enhanced image can be used to find the corresponding position of the micro-damage on the second image set, and visual measurement is performed on the second image set. The second way: directly perform visual measurement on the perturbation-enhanced image, and then correct the deviation of the measurement result through a correction factor.

[0083] In the above thread detection method for superalloy fasteners, after identifying the micro-damage, further visual measurement can obtain the specific geometric dimension parameters of the micro-damage, which helps to realize the quantitative assessment of the thread damage degree, thereby providing a more accurate basis for subsequent use safety assessment or scrapping decision-making. In addition, introducing visual quantitative measurement can reduce the interference of human subjective factors and make the detection results more consistent, standardized, and objective.

[0084] In some embodiments, in the above thread detection method for superalloy fasteners, visually measuring the micro-damage to obtain the geometric dimension parameters of the micro-damage includes: According to the position coordinates, locate the corresponding visual detection area in the second image set; for the visual detection area, extract the first boundary contour of the micro-damage; according to the first boundary contour, image acquisition resolution, and calibration model, calculate the geometric dimension parameters of the micro-damage; the calibration model is used to represent the mapping relationship between the pixel coordinates in the image acquisition module and the actual space coordinates.

[0085] The second image set, understood as a collection of precisely aligned and multi-angled images without perturbation enhancement, exhibits high geometric and pixel accuracy. Using the position coordinates output from the perturbation-enhanced images and pixel position mapping, the micro-damage is located within the second image set and the visual inspection area is extracted. Boundary contours of the visual inspection area are extracted, and the image resolution and calibration model are combined to achieve high-precision measurement of the micro-damage's geometric dimensions.

[0086] The calibration model is used to describe the mapping relationship between the image pixel coordinate system and the actual physical space coordinate system. By photographing a calibration plate with known size features, a conversion relationship between pixels and actual space coordinates can be established, so that the measurement data in the image can accurately reflect the true physical size of the target.

[0087] In one embodiment, the position coordinates are x=410 and y=285, and these position coordinates are used as the positioning basis and mapped to the corresponding image in the second image set. The area near the position coordinates, for example, 100*100 pixels, is extracted as the visual detection area. In the visual detection area, the first boundary outline of the microcrack is extracted using edge detection or image segmentation. If the image acquisition resolution is 8μm / pixel and the crack length is 92 pixels, the corresponding actual length is: .in, px represents pixels, Indicates micrometers.

[0088] In the above-mentioned thread detection method for high-temperature alloy fasteners, visual measurement is performed on a second image set with higher geometric realism based on the coordinate information of micro-damage output by the disturbance-enhanced image, which has the advantages of balancing recognition accuracy and measurement precision. The disturbance-enhanced image improves the recognizability of micro-damage by introducing texture, contrast or structural enhancement, making it easier for the image recognition model to accurately locate the micro-damaged area. The second image set does not introduce disturbance processing, maintaining the original geometric relationship and high-precision pixel information of the image, making it suitable as a basic image for fine measurement. Through spatial mapping across image sets, while ensuring recognition sensitivity, measurement errors caused by deformation caused by image enhancement are avoided, thereby achieving a coordinated unity of high recognition rate and high measurement accuracy.

[0089] In some embodiments, the thread inspection method for high-temperature alloy fasteners described above performs visual measurement of micro-damage to obtain geometrical dimension parameters of the micro-damage, including: According to the position coordinates, the second boundary contour of the micro-damage is extracted from the disturbance-enhanced image.

[0090] It can be understood that the image recognition model detects and locates the possible micro-damage based on the input disturbance-enhanced image and then outputs the position coordinates.Figure 1 a local area, in which an edge detection algorithm (for example, Canny edge detection algorithm) is used to obtain a clear boundary contour of the microdamage (i.e., the second boundary contour).

[0091] According to the second image set, the phase difference map, and the perturbation-enhanced image, construct an image transformation model to obtain a pixel position transformation correction factor during the image enhancement process; the image transformation model is used to characterize the pixel position change between the perturbation-enhanced image and the second image set.

[0092] It can be understood that since the pixel position transformation occurs in the perturbation-enhanced image relative to the second image set during the generation process, there will be errors in directly measuring the geometric dimensions based on the perturbation-enhanced image. Therefore, it is necessary to establish an image transformation model to correct the geometric dimension parameters of the second boundary contour obtained directly based on the perturbation-enhanced image.

[0093] Specifically, first, based on the second image set and the perturbation-enhanced image, extract structural features such as image key points, edge information, and / or texture gradients. Further, introduce the features characterizing the phase intensity and direction distribution in the phase difference map to enhance the characterization ability of the micro-perturbation region. The above various features can be formed into joint features through methods such as feature map splicing, weighted fusion, or multi-channel parallel connection. Then, based on the joint features, an affine transformation or a non-rigid transformation can be used to establish the pixel position mapping relationship between the perturbation-enhanced image and the second image set, that is, construct an image transformation model.

[0094] Furthermore, extract the pixel position transformation correction factor from the image transformation model, such as a pixel displacement vector field, an affine transformation matrix, or a non-linear deformation function, as the correction basis for the geometric backtracking of the microdamage contour.

[0095] Calculate the geometric dimension parameters of the microdamage according to the second boundary contour, the image acquisition resolution, and the correction factor.

[0096] It can be understood that the pixel position transformation correction factor is used to geometrically correct the second boundary contour of the microdamage in the perturbation-enhanced image. Based on the known image acquisition resolution, convert the corrected boundary coordinate pixel unit into a physical unit, so as to obtain the geometric dimension parameters of the microdamage.

[0097] In some embodiments, in the above-mentioned thread detection method for superalloy fasteners, directly obtaining the second boundary contour based on the perturbation-enhanced image can significantly improve the recognizability and boundary clarity of microdamage, thereby improving the accuracy of boundary extraction. The introduction of the correction factor effectively corrects the spatial position deviation introduced during the enhancement process of the second image set, taking into account both the detection sensitivity and the measurement accuracy, and finally improves the reliability of thread microdamage recognition and quantitative characterization.

[0098] It should be understood that although Figure 1 - Figure 2 each step in the flowchart is shown in sequence according to the indication of the arrow, these steps are not necessarily executed in sequence according to the order indicated by the arrow. Unless otherwise clearly stated in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 - Figure 2 at least a part of the steps in

[0099] may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps. Figure 3 In one embodiment, as shown in an thread detection system 200 for superalloy fasteners is provided, including: an image acquisition module 21, configured to obtain a first image set of the thread to be detected; the thread to be detected is the thread on the superalloy fastener, and the first image set is a set of images obtained by imaging the thread to be detected from different observation angles; a registration module 22, configured to register the image regions of the thread to be detected in the first image set to obtain a second image set; a construction module 23, configured to construct a phase difference map based on the second image set; the phase difference map is used to characterize the relative phase difference of the same thread region under different observation angles; a generation module 24, configured to extract the disturbance features of the thread to be detected by using a direction-sensitive filter bank based on the second image set and the phase difference map, and obtain a disturbance-enhanced image based on the disturbance features;

[0100] In the above-mentioned thread detection system 200 for superalloy fasteners, by acquiring thread images from multiple observation angles and performing registration, a phase difference map is constructed, effectively enhancing the characterization ability of micro-damage on complex surfaces and improving the stability of detection results. Combining with a direction-sensitive filter bank to extract perturbation features and generate a perturbation-enhanced image makes tiny perturbations easier to identify, significantly improving the defect recognition accuracy. Finally, an image recognition model based on a convolutional neural network is used to automatically recognize the perturbation-enhanced image, which has good generalization performance and high processing ability. In summary, the key technical problems such as low detection efficiency, poor stability of detection results, and insufficient accuracy of micro-damage recognition in the prior art are solved, and it is applicable to the high-reliability detection requirements of surface micro-damage of superalloy fasteners.

[0101] In some embodiments, in the above-mentioned thread detection system 200 for superalloy fasteners, the image recognition model 25 includes an input module, a feature adaptation module, a candidate box generation module, and a position regression module. The input module is used to input the perturbation-enhanced image into the image recognition model. The feature adaptation module is used to perform feature mapping processing on the perturbation-enhanced image and output a feature map. The candidate box generation module is used to generate candidate detection regions according to the feature map. The position regression module is used to perform regression processing on the candidate detection regions and output the position coordinates of the micro-damage; the recognition result includes the position coordinates.

[0102] In some embodiments, in the above-mentioned thread detection system 200 for superalloy fasteners, the image recognition model 25 further includes a type discrimination module, which is used to identify the damage type of the micro-damage according to the position coordinates and the corresponding candidate detection regions, and output the position coordinates and the corresponding damage type; the recognition result includes the position coordinates and the corresponding damage type.

[0103] In some embodiments, in the above-mentioned thread detection system 200 for superalloy fasteners, in the first image set collected by the image acquisition module 21, each observation angle includes a corresponding visible light image and an infrared image. The registration module 22 includes: a fusion sub-module, which is used to fuse the corresponding visible light image and infrared image for each observation angle to obtain a fused image at each observation angle; a registration sub-module, which is used to register the image regions of the threads to be detected in each fused image to obtain a second image set.

[0104] In some embodiments, as Figure 4 shown, the above-mentioned thread detection system 200 for superalloy fasteners further includes a measurement module 26, which is used to perform visual measurement on the micro-damage to obtain the geometric size parameters of the micro-damage.

[0105] In some embodiments, for the thread detection system 200 of superalloy fasteners, the measurement module 26 includes: a positioning sub-module for positioning a corresponding visual detection area in the second image set according to position coordinates; a first extraction sub-module for extracting the first boundary contour of micro-damage for the visual detection area; and a first calculation sub-module for calculating the geometric dimension parameters of the micro-damage according to the first boundary contour, the image acquisition resolution, and a calibration model, where the calibration model is used to represent the mapping relationship between the pixel coordinates in the image acquisition module and the actual space coordinates.

[0106] The measurement module 26 may be composed of the sub-modules in the previous embodiment. In other embodiments, the measurement module 26 may include: a second extraction sub-module for extracting the second boundary contour of micro-damage from the disturbance-enhanced image according to position coordinates; a correction sub-module for constructing an image transformation model based on the second image set, the phase difference map, and the disturbance-enhanced image to obtain a pixel position transformation correction factor during the image enhancement process, where the image transformation model is used to represent the pixel position change between the disturbance-enhanced image and the second image set; and a second calculation sub-module for calculating the geometric dimension parameters of the micro-damage according to the second boundary contour, the image acquisition resolution, and the correction factor.

[0107] For the specific limitations of the thread detection system for superalloy fasteners, reference can be made to the limitations of the thread detection method for superalloy fasteners in the above text, which will not be elaborated here. Each module in the above thread detection system for superalloy fasteners can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0108] In one embodiment, a computer device is provided, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Obtain a first image set of the thread to be detected, where the thread to be detected is a thread on a superalloy fastener, and the first image set is an image set obtained by imaging the thread to be detected from different observation angles; Register the image area of the thread to be detected in the first image set to obtain a second image set; Construct a phase difference map based on the second image set, where the phase difference map is used to represent the relative phase difference of the same thread area under different observation angles; Based on the second image set and the phase difference map, use a direction-sensitive filter bank to extract the disturbance features of the thread to be detected, and obtain a disturbance-enhanced image based on the disturbance features; Based on the perturbation-enhanced image, an image recognition model identifies whether there are micro-damages on the surface of the screw thread to be detected and outputs the recognition result; the image recognition model is constructed based on a convolutional neural network, and micro-damage refers to surface defects that are invisible under visual observation conditions. The types of micro-damage include thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

[0109] It can be understood that in addition to the memory and processor involved above, the above computer device also includes other software and hardware components not listed in this specification. Specifically, it can be determined according to the model of the specific computer device in different application scenarios, and will not be listed in detail one by one in this specification.

[0110] In one embodiment, when the processor executes the computer program, it can also implement the additional steps or sub-steps in each of the above embodiments of the thread detection method for superalloy fasteners.

[0111] It should be noted that the embodiments in this specification are all described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts between the embodiments, reference can be made to each other.

[0112] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims

1. A thread detection method for superalloy fasteners, characterized in that, Including: Obtaining a first image set of the thread to be detected; the thread to be detected is a thread on a superalloy fastener, and the first image set is a set of images obtained by imaging the thread to be detected from different observation angles; Registering the image regions of the thread to be detected in the first image set to obtain a second image set; Based on the second image set, constructing a phase difference map; the phase difference map is used to characterize the relative phase differences of the same thread region at different observation angles; Based on the second image set and the phase difference map, using a direction-sensitive filter bank to extract the disturbance features of the thread to be detected, and obtaining a disturbance-enhanced image based on the disturbance features; Based on the disturbance-enhanced image, an image recognition model recognizes whether there are micro-damages on the surface of the thread to be detected and outputs a recognition result; the image recognition model is constructed based on a convolutional neural network, and the micro-damage refers to surface defects that are invisible under visual observation conditions, and the damage types of the micro-damage include thermal fatigue micro-cracks, oxidation pits, surface erosion, and stress corrosion cracks.

2. The thread detection method for superalloy fasteners according to claim 1, wherein The step of, based on the disturbance-enhanced image, the image recognition model recognizing whether there are micro-damages on the surface of the thread to be detected and outputting a recognition result includes: Inputting the disturbance-enhanced image into the image recognition model; the image recognition model includes a feature adaptation module, a candidate box generation module, and a position regression module; The feature adaptation module performs feature mapping processing on the disturbance-enhanced image and outputs a feature map; The candidate box generation module generates candidate detection regions according to the feature map; The position regression module performs regression processing on the candidate detection regions and outputs the position coordinates of the micro-damage; the recognition result includes the position coordinates.

3. The thread detection method for superalloy fasteners according to claim 2, characterized in that The image recognition model further includes a type discrimination module, and the type discrimination module is used to identify the damage type of the micro-damage according to the position coordinates and the corresponding candidate detection regions, and output the position coordinates and the corresponding damage type; the recognition result includes the position coordinates and the corresponding damage type.

4. The thread detection method for superalloy fasteners according to claim 2, characterized in that, In the first image set, each observation angle includes a corresponding visible light image and an infrared image; The step of registering the image regions of the thread to be detected in the first image set to obtain a second image set includes: For each observation angle, fusing the corresponding visible light image and the infrared image to obtain a fused image at each observation angle; Registering the image regions of the thread to be detected in each of the fused images to obtain the second image set.

5. The thread detection method for superalloy fasteners according to any one of claims 2-4, characterized in that After outputting the recognition result, it further includes: Performing visual measurement on the micro-damage to obtain geometric dimension parameters of the micro-damage.

6. The thread detection method for superalloy fasteners according to claim 5, characterized in that, The step of performing visual measurement on the micro-damage to obtain geometric dimension parameters of the micro-damage includes: According to the position coordinates, positioning a corresponding visual detection region in the second image set; For the visual detection region, extracting a first boundary contour of the micro-damage; Calculate the geometric dimension parameters of the microdamage according to the first boundary profile, the image acquisition resolution, and the calibration model; the calibration model is used to characterize the mapping relationship between the pixel coordinates in the image acquisition module and the actual space coordinates.

7. The thread detection method for superalloy fasteners according to claim 5, characterized in that The visual measurement of the microdamage to obtain the geometric dimension parameters of the microdamage includes: Extract the second boundary profile of the microdamage from the perturbation-enhanced image according to the position coordinates. Construct an image transformation model based on the second image set, the phase difference map, and the perturbation-enhanced image to obtain the pixel position transformation correction factor during the image enhancement process; the image transformation model is used to characterize the pixel position change between the perturbation-enhanced image and the second image set. Calculate the geometric dimension parameters of the microdamage according to the second boundary profile, the image acquisition resolution, and the correction factor.

8. A thread detection system for superalloy fasteners, characterized in that, including: An image acquisition module for obtaining a first image set of the thread to be detected; the thread to be detected is a thread on a superalloy fastener, and the first image set is an image set obtained by imaging the thread to be detected from different observation angles. A registration module for registering the image regions of the thread to be detected in the first image set to obtain a second image set. A construction module for constructing a phase difference map based on the second image set; the phase difference map is used to characterize the relative phase difference of the same thread region under different observation angles. A generation module for extracting the perturbation features of the thread to be detected by using a direction-sensitive filter bank based on the second image set and the phase difference map, and obtaining a perturbation-enhanced image based on the perturbation features. An image recognition model for identifying whether there is a microdamage on the surface of the thread to be detected based on the perturbation-enhanced image and outputting an identification result; the image recognition model is constructed based on a convolutional neural network, and the microdamage refers to surface defects that are invisible under visual observation conditions. The damage types of the microdamage include thermal fatigue microcracks, oxidation pits, surface erosion, and stress corrosion cracks.

9. The thread detection system for superalloy fasteners according to claim 8, wherein, It also includes: A measurement module for visually measuring the microdamage to obtain the geometric dimension parameters of the microdamage.

10. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the thread detection method for superalloy fasteners according to any one of claims 1 to 7.

Images