Automobile sheet metal part hole shrinkage detection device

Abstract

The invention discloses a test device for measuring thrust of a small engine, and relates to the technical field of visual inspection in the automobile industry. In order to solve the technical problems in the prior art that the detection precision of the conventional method on the hole shrinkage defect is low and the detection reliability is influenced, the technical scheme provided by the invention is as follows: the device comprises an image reference selection module for selecting a plurality of images with different angles as reference images; the image restoration module is used for carrying out denoising processing on the reference image; the perspective transformation module is used for setting an area where diameter shrinkage may occur according to a design drawing and processing technology information of the sheet metal part and realizing automatic alignment of the detection area; the extraction module is used for carrying out multi-scale extraction and frequency domain analysis on the image; and the defect decision-making module is used for carrying out final decision-making on the hole shrinkage defect based on the gradient intensity, the area and the shape characteristics of the hole shrinkage area. The device is suitable for detecting the hole shrinkage of the sheet metal part.

Description

Technical Field

[0001] The present invention relates to the field of visual inspection technology in the automotive industry, and in particular to a sheet metal shrinkage detection method based on the combination of deep learning and multiple image processing algorithms. Background Art

[0002] In the automotive industry, the manufacturing and quality control of sheet metal parts are crucial. Sheet metal parts are usually processed through processes such as die casting and stretching, which can cause deformation or even defects in the material. Existing sheet metal inspection technologies mainly rely on visual inspection and traditional image processing methods. Common technologies include edge detection, gradient analysis, pattern matching, etc.

[0003] In visual inspection, edge detection methods (such as Sobel operator, Canny operator, etc.) are widely used to detect the edges and defects of workpieces. However, since complex textures may be generated on the surface of sheet metal during the stretching process, and the gradient difference between these textures and defects (such as shrinkage) is small, a single edge detection method has difficulty distinguishing between normal textures and defective areas. In addition, with the development of artificial intelligence and deep learning, deep learning models such as convolutional neural networks (CNNs) have been introduced into the field of industrial inspection for automatic identification and classification of defects. These methods have high learning ability and detection accuracy, but the training of deep learning models depends on a large number of labeled samples. In industrial scenarios, obtaining enough high-quality samples for training is a time-consuming and expensive task.

[0004] For example, many companies currently rely on deep learning algorithms to improve the degree of automation when inspecting sheet metal parts. These deep learning methods often build feature extraction networks to automatically classify defects such as shrinkage, but since shrinkage defects appear as very small gradient differences in images, the model is prone to false detection or missed detection. In addition, although traditional machine vision methods such as Hough transform and shape analysis can detect obvious deformations and defects to a certain extent, they are often unable to cope with defects such as shrinkage that are not obvious.

[0005] The prior art has the following deficiencies:

[0006] 1. Limited detection accuracy: Existing methods have low detection accuracy for shrinkage defects, especially for defects with small gradient differences, which are difficult to detect stably by edge detection and traditional machine vision methods.

[0007] 2. Strong dependence on the number of samples: The effectiveness of deep learning models depends on a large number of high-quality training samples, but obtaining these samples is difficult and costly in an industrial environment.

[0008] 3. High false alarm and missed alarm rates: Existing detection methods are not adaptable enough to complex surface textures, which often leads to high false alarm and missed alarm rates in the detection results, affecting the reliability of detection.

[0009] In summary, there is a lot of room for improvement in the existing technology for sheet metal shrinkage detection, and a detection method is needed that can improve detection accuracy and stability without relying on a large number of training samples. Summary of the invention

[0010] In order to solve the problems existing in the prior art, the existing methods have low detection accuracy for shrinkage defects, especially for defects with small gradient differences, edge detection and traditional machine vision methods are difficult to detect stably, the effectiveness of deep learning models depends on a large number of high-quality training samples, and it is difficult and costly to obtain these samples in an industrial environment, and the existing detection methods are insufficiently adaptable to complex surface textures, which often leads to high false alarm and missed alarm rates in the detection results, affecting the reliability of the detection. The technical solution provided by the present invention is:

[0011] A shrinkage detection module for automobile sheet metal parts, comprising:

[0012] An image benchmark selection module selects multiple images from different angles as benchmark images;

[0013] An image restoration module performs denoising on the reference image;

[0014] The perspective transformation module sets the area where the shrinkage may occur according to the design drawings and processing information of the sheet metal parts, and realizes the automatic alignment of the detection area;

[0015] The extraction module performs multi-scale extraction and frequency domain analysis on the image;

[0016] Dynamic threshold setting module, adaptively adjusts the threshold of fine-tuning screening to ensure that the reduced diameter area can be accurately detected under various keyboard conditions;

[0017] The defect decision module makes the final decision on the shrinkage defect based on the gradient intensity, area and shape characteristics of the shrinkage area.

[0018] Furthermore, a preferred embodiment is provided, in which the image restoration module performs denoising on the reference image through multi-level Gaussian blur processing to retain local blur information on the workpiece surface.

[0019] Furthermore, a preferred implementation is provided, a perspective transformation module, which uses a perspective transformation method to achieve automatic alignment of the detection area.

[0020] Furthermore, a preferred implementation is provided, in which an extraction module performs multi-scale extraction on an image based on an arithmetic operator and performs frequency domain analysis on a slice direction in combination with Fourier transform.

[0021] Furthermore, a preferred implementation is provided, in which a defect decision module is combined with a learning machine algorithm to make a final decision on the shrinkage defect.

[0022] Based on the same inventive concept, the present invention also provides a device for detecting shrinkage of automobile sheet metal parts, comprising:

[0023] Module for collecting image data of sheet metal parts;

[0024] The module described.

[0025] A method for detecting shrinkage of automobile sheet metal parts, implemented based on the device of claim 6, comprising:

[0026] Steps for acquiring image data of sheet metal parts;

[0027] The step of selecting a plurality of reference images at different angles;

[0028] A step of performing multi-level Gaussian blur processing on the reference image;

[0029] Setting the shrinkage detection area and implementing the steps of automatic alignment;

[0030] The steps of reducing the path defect determination are performed.

[0031] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program, and when the computer reads the computer program, the computer executes the described method.

[0032] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method described.

[0033] Based on the same inventive concept, the present invention also provides a computer program product, which is a computer program. When the computer program is executed, the method described above is implemented.

[0034] Compared with the prior art, the technical solution provided by the present invention is beneficial in that:

[0035] The present invention provides a device for detecting shrinkage of automobile sheet metal parts, which has higher detection accuracy: the present invention effectively improves the feature resolution of the shrinkage area through multi-scale gradient feature extraction, multi-level Gaussian blurring and frequency domain analysis, and can achieve stable detection under complex surface textures. Compared with existing edge detection and deep learning methods, the present invention combines multiple feature extraction technologies to make the detection of the shrinkage area more accurate and greatly reduce the missed detection rate.

[0036] The present invention provides a device for detecting shrinkage of automobile sheet metal parts, which has stronger adaptability: the present invention can adapt to the detection of workpieces of different lighting conditions and shapes through dynamic gradient threshold setting and stereoscopic geometric perspective transformation, which significantly improves the adaptability of the system. The fixed threshold in the prior art does not work well when facing complex lighting conditions, while the dynamic threshold setting method of the present invention enables the system to adapt to different detection environments and ensure the consistency and reliability of the detection results.

[0037] The present invention provides a device for detecting shrinkage of automotive sheet metal parts, which has a lower false alarm rate: by combining the comprehensive judgment of connected domain analysis and machine learning models, the present invention can effectively reduce the false alarm rate and reduce the need for human intervention and repeated detection. Different from the traditional threshold judgment method based on a single feature, the present invention uses a machine learning model to integrate multiple features, such as gradient strength, shape characteristics and area size, so as to better distinguish normal textures from defects and significantly reduce the false alarm rate.

[0038] The present invention provides a shrinkage detection device for automobile sheet metal parts, which reduces the dependence on the number of samples: the present invention combines traditional image processing technology with machine learning models, and does not completely rely on a large number of labeled samples for deep learning. In an industrial environment, this greatly reduces the cost of sample collection and labeling, and also avoids the problem of model performance degradation caused by insufficient samples.

[0039] The invention provides a device for detecting shrinkage of automobile sheet metal parts, which is suitable for use in sheet metal part shrinkage detection work. DETAILED DESCRIPTION

[0040] In order to make the advantages and benefits of the technical solution provided by the present invention more clearly reflected, the technical solution provided by the present invention is now further described in detail, specifically:

[0041] Embodiment 1: This embodiment provides a shrinkage detection module for automobile sheet metal parts, including:

[0042] An image benchmark selection module selects multiple images from different angles as benchmark images;

[0043] An image restoration module performs denoising on the reference image;

[0044] The perspective transformation module sets the area where the shrinkage may occur according to the design drawings and processing information of the sheet metal parts, and realizes the automatic alignment of the detection area;

[0045] The extraction module performs multi-scale extraction and frequency domain analysis on the image;

[0046] Dynamic threshold setting module, adaptively adjusts the threshold of fine-tuning screening to ensure that the reduced diameter area can be accurately detected under various keyboard conditions;

[0047] The defect decision module makes the final decision on the shrinkage defect based on the gradient intensity, area and shape characteristics of the shrinkage area.

[0048] Implementation method 2: This implementation method further limits the automobile sheet metal shrinkage detection module provided in implementation method 1. The image restoration module performs denoising on the reference image through multi-level Gaussian blur processing to retain local blur information on the workpiece surface.

[0049] Implementation method three: This implementation method further limits the automobile sheet metal shrinkage detection module provided in implementation method one, and the perspective transformation module uses the perspective transformation method to realize the automatic alignment of the detection area.

[0050] Implementation method 4: This implementation method further limits the automobile sheet metal shrinkage detection module provided in implementation method 1. The extraction module performs multi-scale extraction on the image based on an arithmetic operator and performs frequency domain analysis on the slice direction in combination with Fourier transform.

[0051] Implementation method five: This implementation method further limits the automobile sheet metal shrinkage detection module provided in implementation method one. The defect decision module combines the learning machine algorithm to make the final decision on the shrinkage defect.

[0052] Embodiment 6: This embodiment provides a shrinkage detection device for automobile sheet metal parts, including:

[0053] Module for collecting image data of sheet metal parts;

[0054] Module provided in implementation mode 1.

[0055] Specifically, the functions realized by the device are:

[0056] 1. Image preprocessing:

[0057] Reference image selection and reference mask drawing: Select multiple images from different angles as reference images, and draw reference masks in the reference images to represent the metal area of ​​the sheet metal. Different from the prior art, the present invention uses multi-view reference images combined with stereo geometry methods to improve the adaptability of detection to workpiece surface deformation.

[0058] Multi-level Gaussian blur: In the image preprocessing stage, the image is processed multiple times by Gaussian blur with different parameters to reduce noise and interference in the image. Compared with the traditional single Gaussian blur processing, the present invention better retains the gradient information of the local area of ​​the workpiece surface through multi-level processing, thereby improving the sensitivity of subsequent shrinkage detection.

[0059] 2. Definition of reduction area and angle and perspective transformation:

[0060] Shrinkage detection in fixed areas: According to the design drawings and processing information of sheet metal parts, the area where shrinkage may occur and the corresponding angle range are set in advance. The detection area and angle are mapped to the new detection image using the perspective transformation method to achieve automatic alignment and area locking. Compared with the prior art, the present invention further combines the bending and stress distribution analysis of sheet metal parts, automatically adjusts the candidate area for shrinkage detection, and improves adaptability.

[0061] 3. Feature extraction:

[0062] Multi-scale gradient feature extraction: For a newly input image, multiple operators (such as Sobel operator, Scharr operator, etc.) are used to obtain the gradient map in the x and y directions respectively, and trigonometric functions are performed to obtain the angle map. The present invention combines multiple operators to obtain feature information of different scales to ensure that the features of the reduced diameter area can be fully extracted.

[0063] Gradient direction selection and screening: By using the preset shrinkage angle, the area where shrinkage may occur is screened from the angle map, the corresponding gradient map is screened simultaneously, and the gradient direction is analyzed in the frequency domain in combination with Fourier transform to further remove irrelevant texture interference. Different from the prior art, the present invention proposes to combine the analysis of spatial and frequency domain features to improve the robustness of detection.

[0064] 4. Feature matching and threshold detection:

[0065] Dynamic gradient threshold setting: According to the different ambient lighting and workpiece materials, a dynamic threshold setting method is adopted to adjust the threshold of gradient screening through an adaptive algorithm to ensure accurate detection of the shrinkage area under various lighting conditions. The present invention introduces a local contrast enhancement method so that the threshold can be adaptively adjusted to adapt to different lighting scenes.

[0066] Continuous pixel detection: Find continuous pixels in the filtered gradient map and calculate the length, while considering the similarity of neighboring pixels to improve the stability of detection. Compared with the prior art, the present invention further combines connected domain analysis and effectively reduces the false alarm rate by analyzing the shape and size characteristics of the shrinking area.

[0067] 5. Defect determination and alarm:

[0068] Diameter reduction defect determination: Based on the gradient strength, area and shape characteristics of the diameter reduction area, combined with machine learning algorithms (such as support vector machines, random forests, etc.), the final determination of the diameter reduction defect is made. Different from the existing threshold determination method based on a single feature, the present invention integrates multiple features through a machine learning model, which significantly improves the accuracy of diameter reduction defect determination.

[0069] Alarm system: When a shrinkage defect is detected, the system will issue an alarm through the alarm module and generate a test report, which includes the specific location of the defect, feature description and corresponding treatment suggestions.

[0070] Embodiment 7: This embodiment provides a method for detecting shrinkage of automobile sheet metal parts, which is implemented based on the device provided in Embodiment 6 and includes:

[0071] Steps for acquiring image data of sheet metal parts;

[0072] The step of selecting a plurality of reference images at different angles;

[0073] A step of performing multi-level Gaussian blur processing on the reference image;

[0074] Setting the shrinkage detection area and implementing the steps of automatic alignment;

[0075] The steps of reducing the path defect determination are performed.

[0076] Embodiment 8: This embodiment provides a computer storage medium for storing a computer program. When the computer reads the computer program, the computer executes the method provided in embodiment 7.

[0077] Embodiment 9: This embodiment provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method provided in embodiment 7.

[0078] Embodiment 10: This embodiment provides a computer program product, which is a computer program. When the computer program is executed, the method provided in embodiment 7 is implemented.

[0079] Implementation eleven: This implementation further describes the above technical solution in detail through specific examples, specifically:

[0080] 1. Image data acquisition

[0081] Acquiring image data of sheet metal parts serves as the basis for the entire inspection process.

[0082] Detailed description: Through the input module, the multi-view image data of the sheet metal part is obtained. The image data can be collected using a high-resolution camera to capture the full view of the sheet metal part from multiple different angles for subsequent reference image selection and feature extraction.

[0083] 2. Reference image selection and reference mask drawing

[0084] Select multiple fiducial images from different angles and draw fiducial masks to annotate the metal areas of the sheet metal part.

[0085] Detailed description: In the acquired image data, representative images at different angles are selected as reference images, and reference masks are drawn manually or semi-automatically to cover the metal area of ​​the sheet metal. These reference masks are used to define the detection area and improve the adaptability to complex deformation.

[0086] 3. Image Preprocessing

[0087] Multi-level Gaussian blurring is performed on the image data to remove noise.

[0088] Detailed description: Use multi-level Gaussian blur to process the image. Using Gaussian blur with different parameters multiple times can effectively reduce noise while retaining the local gradient information on the surface of the sheet metal. This step aims to provide a stable image foundation for subsequent gradient feature extraction.

[0089] 4. Perspective transformation and detection area setting

[0090] The shrinkage detection area is set using perspective transformation to achieve automatic alignment.

[0091] Detailed description: Based on the design drawings of sheet metal parts and their processing information, determine the area where shrinkage may occur and the corresponding detection angle range. Perspective transformation is used to automatically align the detection area, and adjust the detection area in combination with the stress distribution information of the sheet metal parts to improve the accuracy of shrinkage detection.

[0092] 5. Feature extraction and frequency domain analysis

[0093] Multi-scale gradient feature extraction and frequency domain analysis are performed on the preprocessed image.

[0094] Detailed description: Sobel operator, Scharr operator and other gradient operators are used to calculate the x and y direction gradient maps of the image respectively. Then, the gradient direction is analyzed in the frequency domain in combination with Fourier transform to screen out the features related to the shrinkage and remove the irrelevant background texture to ensure the accuracy and comprehensiveness of feature extraction.

[0095] 6. Dynamic threshold setting

[0096] Adaptively adjust the gradient threshold according to different ambient lighting conditions.

[0097] Detailed description: The dynamic threshold setting module adaptively adjusts the gradient screening threshold according to the ambient lighting conditions and the workpiece material. To ensure the consistency of detection under different lighting conditions, an adaptive contrast enhancement method is used, combined with local histogram equalization, to enhance the contrast of the reduced diameter area and improve the stability and robustness of the detection.

[0098] 7. Defect determination and alarm

[0099] Based on the feature extraction results, the shrinkage defect is determined and an alarm is issued.

[0100] Detailed description: The defect determination module uses machine learning algorithms (such as support vector machines or random forests) to combine gradient strength, area and shape features to make a comprehensive determination of shrinkage defects. When a defect is determined to exist, the alarm module will issue an alarm signal and generate a test report containing the location and feature description of the defect.

[0101] Example 1: General inspection application of sheet metal parts

[0102] 1. Image data acquisition: Use a high-resolution industrial camera to acquire images of sheet metal parts from five different angles to cover all surface features of the sheet metal parts.

[0103] 2. Reference image selection and reference mask drawing: Three representative angles are selected from the five images as reference images, and reference masks are manually drawn to cover the metal area.

[0104] 3. Image preprocessing: Gaussian blur processing with different parameters is performed three times on each reference image to remove noise but retain edge features.

[0105] 4. Perspective transformation and detection area setting: Based on the design drawing, select the bending area in the sheet metal part as the key detection area, and automatically align it through perspective transformation.

[0106] 5. Feature extraction and frequency domain analysis: Use the Sobel operator to extract the x and y direction gradient images, combine with Fourier transform to remove the background texture and retain the shrinkage feature.

[0107] 6. Dynamic threshold setting: Adaptive histogram equalization is used to adjust the gradient threshold according to the lighting conditions at the detection site.

[0108] 7. Defect determination and alarm: Use the support vector machine algorithm to determine whether there is a shrinkage defect in the inspection area. If so, an alarm will be sounded through the buzzer and a test report will be generated.

[0109] Example 2: Inspection of sheet metal parts with complex surfaces

[0110] 1. Image data acquisition: Use a camera with multi-focus capability to collect image data of complex surface sheet metal parts from seven angles.

[0111] 2. Reference image selection and reference mask drawing: Four images from different angles are selected as reference images, and reference masks are automatically drawn through machine learning to reduce manual intervention.

[0112] 3. Image preprocessing: Four-level Gaussian blur processing is applied to ensure that the noise of complex surfaces is effectively removed and the local features are preserved.

[0113] 4. Perspective transformation and detection area setting: Combined with the stress analysis results of sheet metal parts, key detection areas are dynamically set, and regional alignment is performed through perspective transformation.

[0114] 5. Feature extraction and frequency domain analysis: Use the Scharr operator combined with Fourier transform to extract the shrinkage features and remove non-related textures.

[0115] 6. Dynamic threshold setting: In changing lighting environments, a combination of adaptive contrast enhancement and histogram equalization is used to adjust the gradient threshold in real time.

[0116] 7. Defect judgment and alarm: The random forest model is used to make a comprehensive judgment on the shrinkage defect, and the operator is notified through the alarm system, and a report containing detailed defect information is generated.

[0117] Example 3: Inspection of sheet metal parts of different materials

[0118] 1. Image data acquisition: Use a combination of infrared cameras and ordinary cameras to collect images of sheet metal parts of different materials from multiple angles to adapt to reflective materials.

[0119] 2. Reference image selection and reference mask drawing: Through material property analysis, three images at different angles are selected as reference images, and the reference mask is drawn using automated tools.

[0120] 3. Image preprocessing: Perform multi-level Gaussian blur processing on the reference image, especially filtering noise for reflective materials to ensure image clarity.

[0121] 4. Perspective transformation and detection area setting: Combined with the structural characteristics of sheet metal parts, the perspective transformation is dynamically adjusted to ensure accurate alignment of the detection area under different materials.

[0122] 5. Feature extraction and frequency domain analysis: Use the Sobel operator to extract features and combine frequency domain filtering to remove noise to extract effective diameter reduction features.

[0123] 6. Dynamic threshold setting: According to the reflective characteristics of different materials, adjust the dynamic threshold setting strategy to ensure the robustness of detection.

[0124] 7. Defect determination and alarm: The support vector machine algorithm is used to determine the shrinkage defect. If a defect is detected, the operator is notified through the sound and visual alarm system.

[0125] The technical solution provided by this embodiment is different from the prior art in that:

[0126] 1. Different feature extraction methods: The existing technology mainly relies on a single Sobel operator for gradient extraction, while the present invention combines multiple operators such as Sobel and Scharr to obtain richer feature information.

[0127] 2. Multi-view reference image and perspective transformation: Most existing methods are single-view detection. The present invention adopts multi-view reference images and combines them with stereoscopic geometric perspective transformation methods to improve the adaptability to complex deformations.

[0128] 3. Dynamic threshold and frequency domain analysis: The prior art usually adopts a fixed threshold. The present invention introduces dynamic threshold setting and frequency domain analysis to ensure stability and accuracy in different environments.

[0129] 4. Improvement in defect determination method: The present invention uses a machine learning algorithm to make a comprehensive determination of multiple features, which significantly improves the detection accuracy of shrinkage defects. The existing technologies are mostly simple threshold judgments, which have high false alarm and missed alarm rates.

[0130] The technical solution provided by the present invention is further described in detail above through several specific implementation modes in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific implementation modes described above are not intended to be used as limitations on the present invention. Any reasonable modification and improvement of the present invention, combination of implementation modes and equivalent substitution within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A shrinkage detection module for automobile sheet metal parts, characterized in that: include: An image benchmark selection module selects multiple images from different angles as benchmark images; An image restoration module performs denoising on the reference image; The perspective transformation module sets the area where the shrinkage may occur according to the design drawings and processing information of the sheet metal parts, and realizes the automatic alignment of the detection area; The extraction module performs multi-scale extraction and frequency domain analysis on the image; Dynamic threshold setting module, adaptively adjusts the threshold of fine-tuning screening to ensure that the reduced diameter area can be accurately detected under various keyboard conditions; The defect decision module makes the final decision on the shrinkage defect based on the gradient intensity, area and shape characteristics of the shrinkage area.

2. The automobile sheet metal shrinkage detection module according to claim 1, characterized in that: The image restoration module performs noise removal on the reference image through multi-level Gaussian blur processing to retain local blur information on the workpiece surface.

3. The automobile sheet metal shrinkage detection module according to claim 1, characterized in that: The perspective transformation module uses the perspective transformation method to achieve automatic alignment of the detection area.

4. The automobile sheet metal shrinkage detection module according to claim 1, characterized in that: The extraction module performs multi-scale extraction on the image based on an arithmetic operator and performs frequency domain analysis on the slice direction in combination with Fourier transform.

5. The automobile sheet metal shrinkage detection module according to claim 1, characterized in that: The defect decision module combines the learning machine algorithm to make the final decision on the shrinkage defect.

6. A shrinkage detection device for automobile sheet metal parts, characterized in that: include: Module for collecting image data of sheet metal parts; The module of claim 1.

7. A method for detecting shrinkage of automobile sheet metal parts, characterized in that: The device according to claim 6 is implemented, comprising: Steps for acquiring image data of sheet metal parts; The step of selecting a plurality of reference images at different angles; A step of performing multi-level Gaussian blur processing on the reference image; Setting the shrinkage detection area and implementing the steps of automatic alignment; The steps of reducing the path defect determination are performed.

8. A computer storage medium for storing a computing program, characterized in that: When the computer reads the computer program, the computer executes the method of claim 7.

9. A computer, comprising a processor and a storage medium, characterized in that: When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 7 .

10. A computer program product, being a computer program, characterized in that When the computer program is executed, the method of claim 7 is implemented.