A single-frame structured light-based forging workpiece crack detection method

Abstract

The application discloses a kind of based on single frame structured light's forging workpiece crack detection method, belong to high-temperature forging defect detection technical field.The method includes: set up single frame structured light high-temperature detection system and complete calibration;Collect high-temperature forging piece single frame fringe image and carry out thermal radiation noise suppression and fringe enhancement preprocessing;Based on improved Fourier transform profilometry completes three-dimensional point cloud reconstruction, generates multidimensional reconstruction data containing point cloud, phase, texture;To reconstruction point cloud is denoised, registration and target segmentation;Combining point cloud geometric features and lightweight CNN model, complete crack identification, positioning and segmentation;Length, depth, width quantization is carried out to crack, and output visual result and early warning information.The application solves the problems that existing detection method is unsafe, weak against high-temperature interference, low recognition accuracy and cannot be quantified, realizes non-contact, online, fast and accurate detection, adapts to 600-1200 DEG C high-temperature scene, can be embedded into production line, provides traceable data support for forging process optimization.

Description

Technical Field

[0001] This invention belongs to the field of high-temperature forging defect detection, three-dimensional reconstruction and machine vision inspection technology. Specifically, it relates to a method for detecting cracks in forged workpieces based on single-frame structured light. The core is to use single-frame structured light to quickly acquire three-dimensional reconstruction data of high-temperature forged workpieces. Through precise processing and in-depth analysis of the reconstructed data, efficient and accurate detection of crack defects can be achieved, solving the problem of crack detection in forged workpieces under high-temperature conditions. Background Technology

[0002] High-temperature forging is a core process in equipment manufacturing, aerospace, and automotive industries. During the high-temperature, high-pressure, and rapid forming process, forgings are prone to surface and subsurface crack defects due to factors such as material inhomogeneity and fluctuations in forging process parameters. If these cracks are not detected in time, they will further expand during subsequent processing and service, seriously affecting the mechanical properties, load-bearing capacity, and service safety of the workpiece, and may even lead to major safety accidents.

[0003] Existing methods for detecting cracks in high-temperature forging have significant limitations and cannot meet the needs of intelligent, online detection in production lines.

[0004] 1. Manual visual inspection: This method relies on the experience and judgment of the inspectors. The environment at the high-temperature forging site is harsh and the temperature is extremely high. Inspectors are prone to burns and dust hazards. In addition, the inspection efficiency is low and the rate of missed detection is high, making it unsuitable for high-speed forging production lines.

[0005] 2. Traditional non-destructive testing: Methods such as ultrasonic testing and eddy current testing require contact with the workpiece being tested. The workpiece is easily damaged by the test probe under high temperature conditions, and online real-time testing cannot be achieved. They are only suitable for offline sampling testing.

[0006] 3. Multi-frame structured light 3D reconstruction and inspection: It requires the acquisition of multiple stripe images with different phases, which results in slow reconstruction speed, making it impossible to match the rapid forming pace of high-temperature forgings. Furthermore, the workpiece is prone to displacement during multi-frame acquisition, leading to a decrease in reconstruction accuracy.

[0007] 4. Conventional single-frame structured light detection: Thermal radiation and industrial dust in high-temperature environments can severely interfere with the quality of stripe images, resulting in blurred stripes, weakened features, low accuracy and insufficient data integrity in 3D reconstruction data, making it difficult to identify minute cracks. At the same time, existing defect detection algorithms do not fully utilize the multi-dimensional features of 3D reconstruction data, relying only on a single geometric feature for detection, resulting in low crack identification accuracy, high false detection rate, and inability to achieve accurate quantitative assessment of crack geometric parameters.

[0008] While deep learning-based defect detection methods perform well in ambient temperature scenarios, they suffer from insufficient generalization ability due to point cloud noise and false defects in high-temperature forging environments, making it difficult to meet the requirements for accurate detection. Therefore, developing a non-contact, fast, high-temperature interference-resistant method for detecting cracks in high-temperature forging is of significant theoretical and engineering application value for improving forging product quality, achieving intelligent upgrades of production lines, and ensuring equipment safety. Summary of the Invention

[0009] To overcome the shortcomings of the existing technologies, this invention provides a method for detecting cracks in forged workpieces based on single-frame structured light. The core of this method lies in fully exploring the value of single-frame structured light 3D reconstruction data, solving the problems of unsafe contact detection, slow detection speed, weak anti-interference ability, low crack identification accuracy, and inability to quantify and evaluate existing detection methods. This method enables non-contact, online, rapid, and accurate detection and quantitative characterization of cracks in high-temperature forged workpieces, while improving the traceability and practicality of the detection data.

[0010] Technical solution

[0011] A method for detecting cracks in forged workpieces based on single-frame structured light includes the following steps:

[0012] S1. Build a single-frame structured light high-temperature forging three-dimensional detection system and complete the system calibration;

[0013] S2. Acquire a single-frame structured light stripe image and perform preprocessing;

[0014] S3. Reconstruction of 3D point clouds based on an improved single-frame structured light algorithm;

[0015] S4. Point cloud denoising, registration and target segmentation;

[0016] S5. Crack recognition and segmentation based on geometric features and deep learning;

[0017] S6. Crack Quantitative Assessment and Result Output.

[0018] Furthermore, the detection system built in S1 has high temperature protection capability. The calibration process uses multiple sets of calibration images from different angles. The accuracy of the calibration parameters is ensured by the three-dimensional coordinate transformation formula (1), which provides a guarantee for the accuracy of the three-dimensional reconstruction data. S2 specifically addresses the problems of high temperature heat radiation and dust interference. The image quality is improved by noise suppression and stripe enhancement, laying the foundation for generating high-quality three-dimensional reconstruction data. S3 adopts the improved Fourier transform profilometry. The accuracy and speed of single-frame phase calculation are improved by formula (2). The three-dimensional coordinate transformation is completed by formula (3), ensuring the integrity and accuracy of the three-dimensional reconstruction data (point cloud, phase, texture). S4 filters invalid data and corrects data deviations by using dual filtering and fine registration, obtaining pure and effective point cloud data. S5 combines the geometric and texture features of the three-dimensional reconstruction data with a lightweight deep learning model to improve the accuracy and generalization ability of crack identification. S6 realizes crack parameter quantification and real-time early warning based on formulas (4), (5), and (6), adapting to the needs of the production line, while retaining the correlation between the reconstruction data and the defect data, which is convenient for subsequent process traceability and optimization.

[0019] Beneficial effects

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. Employing single-frame structured light 3D reconstruction technology, 3D reconstruction can be completed by acquiring only a single frame of stripe image. The reconstruction speed is fast (single frame reconstruction time < 50ms), which is suitable for dynamic online inspection scenarios of rapid forming of high-temperature forgings. It does not require the accumulation of multiple frames of images, thus avoiding detection errors caused by workpiece displacement.

[0022] 2. Targeted thermal radiation noise suppression and stripe enhancement algorithms are designed to effectively overcome the interference of thermal radiation and dust on stripe images under high temperature environments, improve stripe clarity and 3D point cloud reconstruction accuracy, and control the point cloud registration error within 0.02mm;

[0023] 3. Fully leverage the multi-dimensional advantages (point cloud coordinates, phase, texture) of single-frame structured light 3D reconstruction data, and combine point cloud geometric features with a lightweight deep learning model to achieve accurate identification of micro-cracks (minimum detectable width 0.05mm), effectively distinguishing cracks from pseudo-defects such as surface oxide spots and scratches, with an identification accuracy of ≥98.5%, significantly reducing the false detection rate and missed detection rate, while improving the accuracy of defect localization based on reconstruction data;

[0024] 4. It adopts a non-contact testing method, which eliminates the need to contact the high-temperature forgings, avoids damage to the testing equipment caused by high temperatures, and ensures the personal safety of the testing personnel. It can be directly embedded into the forging production line to achieve real-time testing.

[0025] 5. Based on 3D reconstruction data, the system can accurately quantify and evaluate geometric parameters such as crack length, depth, and width, generate 3D visualization results and quantitative reports related to the 3D reconstruction data, synchronize the data to the production line control system and trigger real-time early warnings, provide accurate and traceable data support for forging process optimization and defect handling, and improve the level of product quality control. Attached Figure Description

[0026] Figure 1 This is a diagram showing the overall architecture of a high-temperature forging crack detection system.

[0027] Figure 2 Comparison of preprocessing effects for single-frame structured light stripe images (a is the original image, b is the preprocessed image);

[0028] Figure 3 A schematic diagram illustrating the improved phase calculation process for Fourier transform contouring.

[0029] Figure 4 Comparison of point cloud preprocessing effects for high-temperature forgings (a is the original point cloud, b is the point cloud after noise reduction and registration);

[0030] Figure 5 This is a schematic diagram of the crack defect identification and quantification results (a is the crack segmentation result, b is the crack parameter annotation). Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0032] This invention provides a method for detecting cracks in forged workpieces based on single-frame structured light, which enables non-contact, online, rapid, and accurate detection and quantitative evaluation of surface and subsurface cracks in high-temperature forged workpieces. It is suitable for high-temperature forging scenarios of 600-1200℃ and can be directly embedded into forging production lines. The core of this method is to quickly acquire high-quality three-dimensional reconstruction data through single-frame structured light and perform in-depth processing and application of the data to achieve accurate detection and quantification of crack defects.

[0033] The method of the present invention is specifically implemented in the following steps:

[0034] 1. Detection system setup and calibration

[0035] A detection system consisting of a 2-megapixel infrared-enhanced industrial camera, a single-frame sinusoidal structured light projector, an industrial computer, and a high-temperature heat-insulating protective cover was constructed. The system was fixed 1.5-2.0 meters behind the stamping station on the forging production line to ensure non-contact detection without affecting production line operation. The high-temperature heat-insulating protective cover is made of high-temperature resistant ceramic material and can withstand temperatures above 1500℃, preventing damage to the camera and projector from the high temperatures of forging.

[0036] A 12×9 checkerboard calibration board (300mm×225mm) was used to acquire 25 sets of calibration images at different angles and distances. These images were then input into a calibration algorithm to calculate the camera distortion coefficients, intrinsic parameter matrix, projector intrinsic parameters, and extrinsic parameter matrix between the camera and projector, thus determining the projection-pixel mapping relationship. A 3D world coordinate system (X,Y,Z) and image pixel coordinate system were then established. The transformation model is used to derive the mapping formula (1), ensuring the accuracy of the three-dimensional coordinate transformation and providing a solid benchmark for the subsequent generation of high-quality three-dimensional reconstruction data, thus avoiding distortion of the reconstruction data due to calibration errors.

[0037] ;

[0038] In the formula, (X,Y,Z) represent the three-dimensional world coordinates of the surface of the high-temperature forged workpiece. Here, M represents the image pixel coordinates, M is the camera intrinsic parameter matrix, and R is the camera extrinsic parameter rotation matrix. For camera focal length, The distance from the reference plane to the camera.

[0039] 2. Single-frame image acquisition and preprocessing

[0040] A single frame of 128Hz sinusoidal structured light stripes is projected onto the surface of an 800℃ high-temperature alloy forging. The stripe fundamental frequency f0=100 / W (W is the stripe image width, set to 1280 pixels). An industrial camera synchronously exposes and acquires the original stripe image, with the image resolution set to 1280×720 pixels.

[0041] An adaptive thresholding method combined with db4 wavelet denoising is used to suppress gray-scale noise generated by high-temperature thermal radiation and random noise caused by industrial dust. The adaptive thresholding method dynamically adjusts the filtering threshold according to the local gray-scale features of the image to preserve the stripe edge features. The wavelet denoising method decomposes the image into three layers, and after thresholding the high-frequency noise components, the image is reconstructed, which effectively removes noise while preserving stripe details.

[0042] By combining histogram equalization and the Sobel edge enhancement algorithm, the contrast of stripes is enhanced: histogram equalization adjusts the gray-level distribution of the image, improving the overall brightness uniformity; the Sobel edge enhancement algorithm strengthens the gradient of stripe edges, improving the clarity of stripe edges, resulting in a high-quality preprocessed stripe image. This provides a high-quality image source for subsequent phase calculation and the generation of 3D reconstruction data, ensuring the accuracy and integrity of the 3D reconstruction data. Figure 2 (b) shown Figure 2 (a) is the original noisy image.

[0043] 3. Single-frame structured light 3D reconstruction

[0044] Phase calculation of the preprocessed stripe image is performed based on an improved Fourier transform contouring technique, as follows: Figure 3 As shown: First, a Fourier transform is performed on the fringe image to extract the fundamental frequency component, and then an inverse Fourier transform is performed to calculate the wrapping phase. It satisfies formula (2):

[0045] ;

[0046] In the formula, The fundamental frequency of the projected stripes. The grayscale values ​​of the preprocessed striped image. These are the pixel coordinates of the image.

[0047] An improved quality-guided phase unwrapping method is adopted, using stripe grayscale contrast and phase gradient as quality evaluation indicators. Phase unwrapping is performed on high-quality pixels first, and then gradually diffused to the entire image to eliminate the 2π periodic blur that wraps the phase, thereby obtaining the absolute phase of the surface of the high-temperature forging. .

[0048] Combining the system calibration parameters obtained from S1 with the triangulation principle, the absolute phase is... The coordinates (X, Y, Z) of the surface of the high-temperature forged workpiece are converted into three-dimensional coordinates. The three-dimensional coordinate calculation satisfies formula (3). The original three-dimensional reconstruction data containing point cloud coordinate data, phase data, and texture feature data is generated. The point cloud density is set to 0.02 mm / point to ensure that the three-dimensional reconstruction data can clearly characterize the surface morphology and subtle features of the forged workpiece, providing high-quality, multi-dimensional data support for subsequent defect detection.

[0049] ;

[0050] In the formula, Let be the coordinates of the camera's principal point, d be the baseline distance between the camera and the projector, L be the distance from the projector to the reference plane, k be the phase-to-depth conversion factor, and f be the camera's focal length. It is an absolute phase.

[0051] 4. Point cloud preprocessing

[0052] A combination of statistical filtering and radius filtering is used to remove point cloud noise: statistical filtering selects 25 neighboring points for each point, calculates the average distance and standard deviation of the neighboring points, and removes outliers whose distance is greater than the average distance plus twice the standard deviation; radius filtering sets the radius to 0.05 mm and removes isolated points with fewer than 25 neighboring points, effectively removing high-temperature artifacts and background interference points, filtering invalid 3D reconstruction data, and ensuring the purity and accuracy of point cloud data.

[0053] The Iterative Closest Point (ICP) algorithm is used to precisely register the reconstructed original 3D point cloud data with the 3D model of the standard forged workpiece. The number of iterations is set to 100, and the registration error is controlled within 0.02mm. The reference position of the workpiece under test is located, eliminating the deviation of the 3D reconstruction data caused by the workpiece placement deviation, and ensuring the accuracy of subsequent defect detection.

[0054] By employing a point cloud semantic segmentation algorithm, combining the geometric contour features of the forged workpiece (such as its dimensions, fillets, and holes) with the texture features in the 3D reconstruction data, background interference areas such as forging fixtures and conveying mechanisms are removed. Only pure and effective point cloud data of the target area of ​​the forged workpiece is retained, ensuring that subsequent defect detection targets only the workpiece itself and avoiding invalid data from interfering with the detection results. Simultaneously, the subtle feature data of the workpiece surface are preserved, providing assurance for the detection of minute cracks. Figure 4 (b) shown Figure 4 (a) is the original point cloud.

[0055] 5. Crack defect detection

[0056] Extracting geometric and texture features from clean and effective point clouds: The point cloud surface is fitted using the least squares method, and the normal vector deviation, curvature abrupt change value, and depth gradient change of each point are calculated. At the same time, the texture features (grayscale mean, texture roughness, and texture contrast) in the 3D reconstruction data are combined. The thresholds for normal vector deviation are set to 0.3 rad, curvature abrupt change to 0.5, depth gradient to 0.03 mm, and texture roughness to 0.02 mm. The suspected crack area is initially located, and the interference of false defects such as normal surface texture and oxide spots is eliminated. The multi-dimensional advantages of 3D reconstruction data are fully utilized to improve the accuracy of suspected area location.

[0057] A lightweight crack recognition CNN model is constructed. The model structure includes 4 convolutional layers (with kernel sizes of 3×3, 3×3, 5×5, and 5×5), 2 max pooling layers (with a kernel size of 2×2), 1 fully connected layer, and 1 output layer. The activation function is ReLU, and the output layer uses the Sigmoid function to achieve binary classification of cracks and non-cracks. 1500 sets of 3D reconstruction data samples (including point cloud coordinates, phase, and texture data) of high-temperature forging cracks of different types (transverse cracks, longitudinal cracks, and network cracks) and different sizes (width 0.05-0.5mm, length 5-50mm) were collected and divided into training set, validation set, and test set in a 7:2:1 ratio. The model was trained using the adaptive moment estimation (Adam) optimizer with 200 iterations and an initial learning rate of 0.001. The cross-entropy loss function was used, and multi-dimensional features of the 3D reconstruction data were introduced to improve the model's generalization ability and recognition accuracy until the model converged. The model's test accuracy reached over 98.5%.

[0058] The point cloud of the initially located suspected crack area is input into a trained lightweight CNN model to achieve accurate crack defect identification, location localization, and pixel-level fine segmentation, obtaining the complete outline of the crack and distinguishing it from pseudo-defects such as cracks, scratches, and oxide spots. Figure 5 As shown in (a).

[0059] 6. Defect Quantification and Output

[0060] Based on the segmented crack point cloud data, combined with the depth information and coordinate data in the 3D reconstruction data, the shortest path algorithm is used to calculate the crack length, the point cloud depth difference is used to calculate the crack depth, and the cross-section fitting algorithm is used to calculate the crack width. The specific formulas are as follows (4), (5), and (6). The crack geometric parameters are quantified, and a defect quantification report containing the association information of the 3D reconstruction data (such as crack position coordinates, corresponding phase data, and texture features) is generated. The positional relationship between the crack and the workpiece surface morphology is clarified, providing traceable data support for subsequent defect analysis and process optimization.

[0061] Crack length formula (4): ;

[0062] Crack depth formula (5): ;

[0063] Crack width formula (6): ;

[0064] In the formula, ( ) represents the coordinates of continuous points along the crack centerline, and n represents the number of points along the centerline; This represents the point cloud depth value on the workpiece surface surrounding the crack. This represents the depth of the point cloud inside the crack. , and , These are the coordinates of the corresponding points on both sides of the crack opening edge.

[0065] The location and geometric parameters of crack defects are labeled on the 3D model of the high-temperature forging, generating a 3D visualization of the defects. This visually displays the distribution, morphology, and size of the cracks. Figure 5 As shown in (b).

[0066] The system synchronizes the detection data, defect quantification report, 3D visualization results, and original 3D reconstruction data to the production line control system. The preset crack width threshold is 0.1mm and the length threshold is 10mm. When the detected crack parameters exceed the preset thresholds, the system triggers a real-time audible and visual warning to remind the staff to stop the machine in time. At the same time, the defect data and the corresponding 3D reconstruction data are stored in the database to provide accurate data support for forging process optimization and defect tracing, and realize full-process traceability of detection data.

[0067] Experimental verification

[0068] Experiments were conducted on high-temperature alloy forgings (containing transverse, longitudinal, and network cracks) at 800℃ for verification. The method of this invention was compared with traditional single-frame structured light detection methods and manual detection methods. The results are as follows:

[0069] 1. Detection speed: The single-frame reconstruction time of the method of this invention is 42ms, and the single-item detection time is <1s; the single-item detection time of the traditional single-frame structured light detection method is >3s; and the single-item detection time of manual inspection is >10s.

[0070] 2. Recognition Accuracy: The recognition accuracy of the method of this invention is 98.7%, with a false negative rate of 0.8% and a false positive rate of 0.5%; the traditional single-frame structured light detection method has an accuracy of 82.3%, a false negative rate of 8.5%, and a false positive rate of 9.2%; manual detection has an accuracy of 85.1%, a false negative rate of 7.8%, and a false positive rate of 7.1%.

[0071] 3. Detection accuracy: The method of this invention can detect a minimum crack width of 0.05mm with a size measurement error of <±0.01mm; the traditional single-frame structured light detection method can detect a minimum crack width of 0.15mm with a size measurement error of <±0.03mm.

[0072] Experimental results show that the method of the present invention, relying on high-quality single-frame structured light 3D reconstruction data, is superior to traditional detection methods and manual detection methods in terms of detection speed, recognition accuracy, and detection precision. It can give full play to the multi-dimensional advantages of 3D reconstruction data, effectively solve the crack detection problem in high-temperature forging scenarios, and meet the online detection requirements of high-temperature forging production lines.

[0073] A crack defect detection system for high-temperature forging workpieces based on single-frame structured light 3D reconstruction data, employing the method described in any one of claims 1-7 above, including a system calibration unit, an image acquisition and preprocessing unit, a 3D reconstruction unit, a point cloud processing unit, a defect detection unit, and an output early warning unit.

[0074] The system calibration unit: completes the calibration of the internal and external parameters of the detection system camera and projector, establishes a conversion model between three-dimensional coordinates and pixel coordinates, and provides calibration parameter support;

[0075] The image acquisition and preprocessing unit projects a single frame of structured light stripes onto the high-temperature forging, acquires the original stripe image, performs thermal radiation noise suppression and stripe feature enhancement preprocessing, and outputs a high-quality stripe image.

[0076] The three-dimensional reconstruction unit: based on the improved single-frame structured light phase calculation algorithm, performs phase calculation and three-dimensional coordinate transformation on the preprocessed stripe image to generate original three-dimensional reconstruction data of the high-temperature forging workpiece containing point cloud coordinates, phase, and texture features.

[0077] The point cloud processing unit performs noise reduction, registration, and target region segmentation on the point cloud data in the original 3D reconstruction data, filters out invalid data, and outputs clean and valid point cloud data to provide high-quality 3D reconstruction data support for defect detection.

[0078] The defect detection unit extracts point cloud geometric and texture features from the 3D reconstruction data and uses a lightweight CNN model to identify, locate, segment, and quantize geometric parameters of crack defects.

[0079] The output early warning unit outputs defect detection results, 3D visualization models, quantitative reports, and associated 3D reconstruction data, and synchronizes the data to the production line control system to complete real-time defect early warning and data traceability.

Claims

1. A method for detecting cracks in forged workpieces based on single-frame structured light, characterized in that, Includes the following steps: S1. Build a single-frame structured light high-temperature forging three-dimensional detection system, complete the system calibration of camera, projector and forging station, and obtain internal and external calibration parameters; S2. Acquire a single-frame structured light stripe image of the surface of the high-temperature forging, and perform thermal radiation noise suppression and stripe feature enhancement preprocessing on the image; S3. Based on the improved single-frame structured light phase calculation algorithm, the preprocessed stripe image is reconstructed into a three-dimensional point cloud to generate the original three-dimensional point cloud data of the high-temperature forging workpiece. The three-dimensional reconstruction data includes point cloud coordinate data, phase data, and texture feature data, providing comprehensive data support for subsequent defect detection. S4. Preprocess the point cloud data in the original 3D reconstruction data, and sequentially complete noise removal, point cloud registration and target area segmentation of the forging workpiece, filter invalid data and correct data deviations to obtain clean and valid 3D reconstruction point cloud data with data accuracy that meets the detection requirements. S5. Based on the preprocessed 3D reconstruction data, extract point cloud geometric features and texture features, and combine them with a lightweight deep learning model to complete the automatic identification, location positioning and pixel-level fine segmentation of crack defects in high-temperature forging workpieces, making full use of the integrity and accuracy of 3D reconstruction data to improve the defect detection accuracy. S6. Perform geometric parameter quantification evaluation on the identified crack defects, and output the defect detection results and three-dimensional visualization model.

2. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S1 are as follows: S11. Construct a detection system consisting of an industrial camera, a single-frame structured light projector, an industrial control computer, and a high-temperature protection component, and fix it at a non-contact station on the forging production line. The high-temperature protection component is used to isolate the camera and projector from damage caused by the high temperature of forging. S12. Use a checkerboard calibration board to calibrate the intrinsic and extrinsic parameters of the camera and projector. Collect 20-30 sets of calibration images at different angles and distances. Use the calibration algorithm to calculate the camera distortion coefficient, intrinsic parameter matrix, projector intrinsic parameters, and extrinsic parameter matrix between the camera and projector, and determine the projection-pixel mapping relationship. S13. Establish a transformation model between the three-dimensional world coordinates of the high-temperature forged workpiece and the pixel coordinates of the image. Combine the calibration parameters to derive the mapping formula from pixel coordinates to three-dimensional coordinates (1), providing a coordinate reference for subsequent three-dimensional reconstruction: ； In the formula, (X,Y,Z) represent the three-dimensional world coordinates of the surface of the high-temperature forged workpiece. Image pixel coordinates, For the camera intrinsic parameter matrix, Let be the camera extrinsic rotation matrix. For camera focal length, The distance from the reference plane to the camera.

3. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S2 are as follows: S21. Project a single frame of sinusoidal structured light stripes onto the surface of a high-temperature forging at 600-1200℃. Set the stripe fundamental frequency to f0. Simultaneously expose and collect the original stripe image containing thermal radiation interference and dust interference using an industrial camera. Set the image resolution to H×W. S22. Adaptive threshold filtering and wavelet denoising are combined to suppress gray noise generated by high temperature thermal radiation and random noise caused by industrial dust, while retaining the effective stripe features. The db4 wavelet basis is selected for wavelet denoising, and the number of decomposition layers is set to 3. S23. A stripe contrast enhancement method combining histogram equalization and edge enhancement algorithms is used to strengthen the structured light stripe features weakened by thermal radiation on high-temperature surfaces, improve the clarity of stripe edges, and provide high-quality image data for subsequent phase calculation.

4. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S3 are as follows: S31. Based on the preprocessed single-frame stripe image, the wrapping phase is calculated using the improved Fourier transform contouring technique, as shown in the following formula (2): ； In the formula, The fundamental frequency of the projected stripes. The grayscale values ​​of the preprocessed striped image. To obtain the wrapped phase, These are the pixel coordinates of the image. S32. An improved quality-guided phase unwrapping method is adopted, using stripe grayscale contrast and phase gradient as quality evaluation indicators to eliminate the 2π periodic ambiguity of the encapsulated phase and obtain the absolute phase of the surface of the high-temperature forging. ; S33. Combining the system calibration parameters obtained from S1 with the triangulation principle, the absolute phase is converted into three-dimensional coordinates of the surface of the high-temperature forged workpiece. The three-dimensional coordinate calculation satisfies formula (3), generating original three-dimensional reconstruction data containing point cloud coordinates, phase information, and texture features. The point cloud density is set to 0.01-0.05 mm / point according to the detection accuracy requirements to ensure that the three-dimensional reconstruction data can clearly characterize the surface morphology and subtle features of the forged workpiece, providing a high-quality data foundation for subsequent defect detection. ； In the formula, Let these be the coordinates of the camera's principal point. The baseline distance between the camera and the projector. The distance from the projector to the reference plane. These are the phase-to-depth conversion coefficients. For camera focal length, It is an absolute phase.

5. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S4 are as follows: S41. A combination of statistical filtering and radius filtering is used to remove outlier noise points, high-temperature artifact points, and background interference points from the point cloud. The number of neighborhood points for statistical filtering is set to 20-30, and the radius of radius filtering is set to 0.03-0.08 mm. S42. Using the Iterative Closest Point (ICP) algorithm, the reconstructed original 3D point cloud is precisely registered with the 3D model of the standard forging, with the registration error controlled within 0.02mm, to locate the reference position of the workpiece to be tested. S43. By using a point cloud semantic segmentation algorithm, combining the geometric contour features of the forging workpiece with the texture features in the 3D reconstruction data, background interference areas such as forging fixtures and transmission mechanisms are removed, and only pure and effective point cloud data of the target area of ​​the forging workpiece is retained, ensuring that subsequent defect detection is only for the workpiece itself and avoiding invalid data from interfering with the detection results.

6. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S5 are as follows: S51. Based on the preprocessed 3D reconstruction data, extract the geometric and texture features of the clean and effective point cloud. The geometric features include normal vector deviation, curvature abrupt change value, and depth gradient change. The texture features include grayscale mean, texture roughness, and texture contrast. Set reasonable thresholds for each feature to initially locate suspected crack areas, effectively eliminate interference from normal surface textures, oxide spots, and other false defects, and give full play to the multi-dimensional advantages of 3D reconstruction data. S52. Construct a lightweight crack recognition CNN model. The model uses 4 convolutional layers, 2 pooling layers, and 1 fully connected layer. Collect 1000-2000 sets of 3D reconstruction data samples (including point cloud data, phase data, and texture data) of different types and sizes of high-temperature forging cracks. Divide them into training set, validation set, and test set in a 7:2:1 ratio to train and optimize the model. Introduce multi-dimensional features of 3D reconstruction data to improve the model's generalization ability until the model converges, ensuring that the model can accurately identify crack defects of different shapes. S53. Input the point cloud of the suspected crack area that has been initially located into the trained lightweight CNN model to complete the accurate identification, location and pixel-level fine segmentation of crack defects, and distinguish cracks from pseudo-defects such as surface oxide spots and scratches.

7. The method for detecting crack defects in high-temperature forging workpieces based on single-frame structured light 3D reconstruction data according to claim 1, characterized in that, The specific steps for S6 are as follows: S61. Based on the segmented crack point cloud data, combined with the depth information and coordinate data in the 3D reconstruction data, the shortest path algorithm is used to calculate the crack length, the point cloud depth difference is used to calculate the crack depth, and the cross-section fitting algorithm is used to calculate the crack width. The specific formulas are as follows (4), (5), and (6). The crack geometric parameters are quantified, and a defect quantification report containing the association information of the 3D reconstruction data is generated to clarify the positional relationship between the crack and the workpiece surface morphology. Crack length formula (4): ; Crack depth formula (5): ; Crack width formula (6): ; In the formula, ( ) represents the coordinates of continuous points along the crack centerline, and n represents the number of points along the centerline; This represents the point cloud depth value on the workpiece surface surrounding the crack. This represents the depth of the point cloud inside the crack. , and , These are the coordinates of the corresponding points on both sides of the crack opening edge. S62. Mark the location and geometric parameters of the crack defects on the three-dimensional model of the high-temperature forging to generate a three-dimensional visualization result of the defects, which intuitively shows the distribution and morphology of the cracks. S63. Synchronize the detection data and defect quantification report to the production line control system. When the crack parameters exceed the preset threshold, a real-time warning is triggered to remind the staff to take action.

8. A crack defect detection system for high-temperature forging workpieces based on single-frame structured light 3D reconstruction data, employing the method described in any one of claims 1-7, characterized in that, It includes a system calibration unit, an image acquisition and preprocessing unit, a 3D reconstruction unit, a point cloud processing unit, a defect detection unit, and an output early warning unit; The system calibration unit: completes the calibration of the internal and external parameters of the detection system camera and projector, establishes a conversion model between three-dimensional coordinates and pixel coordinates, and provides calibration parameter support; The image acquisition and preprocessing unit projects a single frame of structured light stripes onto the high-temperature forging, acquires the original stripe image, performs thermal radiation noise suppression and stripe feature enhancement preprocessing, and outputs a high-quality stripe image. The three-dimensional reconstruction unit: based on the improved single-frame structured light phase calculation algorithm, performs phase calculation and three-dimensional coordinate transformation on the preprocessed stripe image to generate the original three-dimensional point cloud of the high-temperature forging. The point cloud processing unit performs noise reduction, registration, and target region segmentation on the original 3D point cloud, and outputs a clean and effective point cloud to provide high-quality point cloud data for defect detection. The defect detection unit extracts point cloud geometric features and uses a lightweight CNN model to identify, locate, segment, and quantize geometric parameters of crack defects. The output early warning unit outputs defect detection results, a 3D visualization model, and a quantitative report, and synchronizes the data to the production line control system to complete real-time defect early warning.

Images