A light alloy casting flaw detection, film evaluation and positioning marking full-process detection method

Abstract

The application belongs to the technical field of casting quality detection, and discloses a light alloy casting flaw detection, film shooting, evaluation and positioning and marking full-process detection method, which comprises the following steps: obtaining a real flaw detection imaging image and a three-dimensional model of a casting; performing single-piece number casting character recognition after pre-processing the real flaw detection imaging image; generating a virtual no-defect flaw detection image, and obtaining the position of a defect by subtracting the virtual no-defect flaw detection image from the real flaw detection imaging image; sequentially performing secondary detection, defect elimination, category correction and hazard quantitative rating on the real flaw detection defect; converting the real flaw detection imaging image from two-dimensional to three-dimensional, and obtaining three-dimensional position and depth information of the real flaw detection defect, and then performing marking on the real defect. Through the application, intelligent detection can be realized in the whole process, and the application has the advantages of high efficiency, high precision, stability and reliability.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of casting quality detection, and more particularly relates to a light alloy casting flaw detection, film evaluation and positioning and marking full-process detection method. BACKGROUND

[0002] With the rapid development of advanced equipment manufacturing industry, the quality control requirements of light alloy metal components are becoming higher and higher. These components often have complex designs, which greatly increase the difficulty of internal defect detection. In the prior art, the commonly used detection process mainly includes: "X-ray flaw detection two-dimensional DR manual partition parameter adjustment film shooting + manual visual experience film evaluation + manual trial and error three-dimensional positioning".

[0003] However, further research shows that the above prior art still has the following defects or deficiencies:

[0004] Firstly, in the film shooting link, the manual method often has problems such as easy omission of key parts of the casting, large imaging quality fluctuation, etc. Secondly, in the film evaluation link, due to the mixing of various defects and the similarity of various defect appearances, defects are easy to be missed or misdetected. Finally, in the positioning link, due to the unknown depth and size of the defects, positioning and repair are low in quality and efficiency, and the casting may be damaged again. In addition, the film shooting, film evaluation and positioning links are often completed by different personnel, and there is a lack of close connection between the links, which leads to disconnection of the process and further reduces the overall detection efficiency and reliability.

[0005] Correspondingly, the many problems existing in the prior art have become the main bottleneck restricting the intelligent, efficient and high-quality batch production of high-end equipment, and the field urgently needs further research and improvement to better meet the quality needs of modern casting. SUMMARY

[0006] In view of one or more of the above defects or needs of the prior art, the present application provides a light alloy casting flaw detection, film evaluation and positioning and marking full-process detection method, wherein the entire detection process including flaw detection, film evaluation and positioning and marking of the light alloy casting is redesigned and the key steps are improved in view of the processing characteristics and specific needs of the light alloy casting, which can realize intelligent detection throughout the process, and has the advantages of high efficiency, high precision and stability, and is especially suitable for quality detection of light alloy castings with complex structures.

[0007] To achieve the above purpose, according to the present application, a light alloy casting flaw detection, film evaluation and positioning and marking full-process detection method is provided, characterized in that the method comprises the following steps:

[0008] S1, for the light alloy casting to be detected, obtaining real flaw detection imaging images under different flaw detection angles relative to a ray source, and simultaneously obtaining a corresponding three-dimensional model of the casting;

[0009] S2, performing image preprocessing on the real flaw detection imaging images, and then performing single-piece number casting character recognition;

[0010] S3, constructing a virtual imaging system to generate a defect-free flaw detection image; performing gray scale fitting difference calculation on the defect-free flaw detection image and the real flaw detection imaging image to locate the position of the real flaw detection defect;

[0011] S4, performing secondary detection on the obtained real flaw detection defects, increasing the number of real flaw detection defects, and classifying the real flaw detection defects;

[0012] S5, performing defect elimination and category correction on the obtained real flaw detection defects;

[0013] S6, performing hazard quantitative rating on the obtained real flaw detection defects;

[0014] S7, converting the real flaw detection imaging image from two-dimensional to three-dimensional to obtain three-dimensional position and depth information of the real flaw detection defects;

[0015] S8, based on the three-dimensional position and depth information of the real flaw detection defects, using a casting defect marking device to perform marking on the defects, thereby completing the entire detection process.

[0016] As a further preferred embodiment of the present application, in step S2, the image preprocessing is preferably achieved by using a homomorphic filtering method and a local adaptive binarization method.

[0017] As a further preferred embodiment of the present application, in step S2, the single-piece number casting character recognition process is preferably designed as follows:

[0018] S21, detecting the character edges in the real flaw detection imaging image and generating a binary image containing edge information; extracting the contour information of the characters in the binary image, i.e. the character contour line, and then finding its minimum circumscribed rectangle according to the contour information, performing a segmentation operation to obtain the smallest unit of character image;

[0019] S22, constructing a deep learning model and pre-recognizing all characters;

[0020] S23, dividing the real flaw detection imaging image into multiple regions, and then for each region, using the trained deep learning model to perform optical character recognition of the single-piece number casting character and identify the specific content of the single-piece number casting character.

[0021] As a further preferred embodiment of the present application, in step S2, preferably further comprising implementing flaw detection photographing, evaluation and positioning marking procedures and automatic mapping operation of casting model and batch based on the identification result of the single part number.

[0022] As a further preferred embodiment of the present application, in step S3, before generating the non-defect flaw detection image using the virtual imaging system, preferably further comprising a photographing path planning operation, which is preferably designed as follows:

[0023] a. Taking the stress concentration area and defect-prone area of the casting as the quality key control points, extracting and searching the features of these quality key control points and similar structures, and then taking them as the must-pass points of the photographing path of virtual imaging;

[0024] b. Classifying the casting three-dimensional model, and sharing the same must-pass points of the photographing path for castings in the same category after classification;

[0025] c. Generating a complete non-defect flaw detection image using the virtual imaging system according to the photographing path containing the must-pass points of the photographing path and the matching photographing parameters.

[0026] As a further preferred embodiment of the present application, in step S3, the process of gray scale fitting difference calculation is preferably designed as follows:

[0027] S31, respectively calculating the gray scale histograms of the non-defect flaw detection image and the real flaw detection imaging image;

[0028] S32, using the above gray scale histograms, respectively calculating the number and proportion of pixels with a gray scale level less than a preset value in the non-defect flaw detection image and the real flaw detection imaging image, and constructing a mapping relationship between the two images;

[0029] S33, based on the mapping relationship, converting the gray scale level of the non-defect flaw detection image to obtain a fitted non-defect flaw detection image;

[0030] S34, performing pixel difference value comparison between the fitted non-defect flaw detection image and the real flaw detection imaging image, and the area with a pixel difference value greater than a threshold value is the location of the real flaw detection defect.

[0031] As a further preferred embodiment of the present application, in step S4, preferably establishing a network model of defect detection deep learning, and increasing the number of defects by casting simulation and virtual flaw detection simulation, and performing secondary positioning and classification of defects in the network model according to the defect generation mechanism and defect topographic features.

[0032] As a further preferred embodiment of the present application, in step S5, preferably, a corresponding defect multi-dimensional feature matrix is constructed based on the profile, internal topography and gray value and other features of the defect, and the feature matrix is used to implement the above-mentioned defect false elimination and category correction operations.

[0033] As a further preferred embodiment of the present application, in step S5, preferably, the automatic association operation of the same category of defects in the multi-site flaw detection image is implemented based on the diversity level features of the defects.

[0034] As a further preferred embodiment of the present application, in step S6, preferably, the gray scale, texture, area and dispersion degree and other information of the defect are extracted, and the above-mentioned hazard quantitative rating operation is implemented based on these information.

[0035] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0036] 1. The present application re-designs the entire detection process including flaw detection photography, evaluation and positioning marking, and makes targeted improvements on the key steps, so that intelligent detection of the whole process can be realized, and the advantages of high efficiency, high precision and stable reliability are achieved.

[0037] 2. The present application further optimizes the design of the recognition process of the single-piece number cast characters, so that the cast characters and content recognition can be more accurately and quickly realized, and the automatic mapping of flaw detection photography, evaluation and positioning marking with the cast type and batch can be realized.

[0038] 3. The present application can realize accurate positioning of defects and detection of all key parts of the casting by establishing a virtual imaging system and comparing the virtual and real flaw detection imaging images. In addition, the present application also makes targeted design on the related algorithms of gray scale fitting difference and the photography path planning in the virtual imaging process, so that the accuracy of defect positioning can be further improved.

[0039] 4. The present application further constructs a defect multi-dimensional feature matrix to realize defect false elimination and category correction, effectively solving the defect missed detection. In combination with the casting knowledge, production requirements and related standards, the defect hazard evaluation index is designed to realize quantitative rating of defect hazard.

[0040] 5. The whole process detection method of the present application is overall easy to operate, which can avoid human errors to the greatest extent, and can realize X-ray flaw detection of light alloy castings with the advantages of "photographing completely, clearly, accurately, stably, finely, quickly and positioning accurately", so it is especially suitable for high-reliability detection application occasions such as large light alloy complex castings of national defense high-end equipment. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1is a flow chart of a full-process detection method for light alloy casting flaw detection, film shooting, evaluation and positioning marking according to the present application;

[0042] Figure 2 is an application scenario diagram for demonstrating the full-process detection method of the present application;

[0043] Figure 3 is an application scenario diagram for demonstrating virtual imaging system and flaw detection equipment linkage film shooting;

[0044] Figure 4 is an application scenario diagram for demonstrating defect three-dimensional topography reconstruction and positioning marking. DETAILED DESCRIPTION

[0045] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0046] Figure 1 is a flow chart of a full-process detection method for light alloy casting flaw detection, film shooting, evaluation and positioning marking according to the present application, which will be explained more specifically below with reference to Figure 1 .

[0047] Step one, for the light alloy casting to be detected, real flaw imaging images under different flaw detection angles are obtained with respect to the ray source, and the corresponding casting three-dimensional model is also obtained.

[0048] In this step, the ray source may, for example, adopt X-ray, and the detection of all key parts of the casting is realized at different flaw detection angles.

[0049] Step two, the real flaw imaging images are preprocessed, and then the identification of the single-piece number casting character is performed.

[0050] In this step, according to one preferred embodiment of the present application, for the obtained real flaw imaging images, the homomorphic filtering method and the local adaptive binarization method are used for image preprocessing, and then the image text, i.e. the single-piece number casting character, is identified as follows:

[0051] Firstly, the edge of the text in the image is detected, for example, using the Canny edge detection method, and a binary image containing edge information is generated. According to the edge image, the contour information of the character, i.e. the character contour line, is extracted using a contour extraction algorithm. These contour lines are the boundaries between the characters and the background. According to the contour line of the character, the minimum circumscribed rectangle frame of the contour is found, and then a segmentation operation is performed along the minimum circumscribed rectangle frame to obtain the minimum unit character image;

[0052] Then, a deep learning model can be constructed to pre-identify all characters. For example, a convolutional neural network model can be constructed, and the pre-processed related data is input into the model for training based on the character contour line to increase the accuracy of letter recognition. The model principle and specific process are well known in the art, and therefore will not be described again.

[0053] Then, the real flaw imaging image is segmented into multiple regions, and then for each region, the optical character recognition of the single-piece number cast character is performed using the above trained deep learning model, and the specific content of the single-piece number cast character is identified. In addition, film shooting, film evaluation, positioning information and cast model, batch, etc. can be automatically mapped.

[0054] Step three, constructing a virtual imaging system and generating a defect-free flaw detection image; performing gray scale fitting difference calculation on the defect-free flaw detection image and the real flaw imaging image to locate the position of the real flaw.

[0055] In this step, a virtual imaging system can be constructed first. Specifically, for example, the physical imaging process and principles can be used as a blueprint to simulate the processes of ray emission, penetration, imaging plate reception and imaging, etc. to realize complete data tracking and film shooting simulation for the abstract film shooting process, and to construct a model capable of generating high-quality defect images. As understood by those skilled in the art, the virtual imaging system can be implemented in various ways, and preferably the technology disclosed in the accepted patent CN202310335552.7, "A virtual radiographic flaw image generation method for internal defect detection of castings" can be used to implement it.

[0056] Then, according to one preferred embodiment of the present application, the film shooting path planning operation can also be included, which is preferably designed as follows:

[0057] a. The stress concentration area and the defect-prone area of the casting are taken as the quality key control points, and feature extraction and search are performed on these quality key control points and similar structures, which are then taken as the must-pass points of the virtual imaging film shooting path; for example, finite element analysis (FEA) can be used to predict potential stress concentration areas and analyze to find areas where defects frequently occur in the past as quality key control points of the casting;

[0058] b. Classify the casting three-dimensional model, and share the same shooting path must-pass point for the castings in the same category after classification; for example, a casting three-dimensional model classification algorithm can be established, the similarity of the casting three-dimensional model and different model categories is calculated, and the classification of the three-dimensional model is completed through operator spectrum analysis;

[0059] c. According to the shooting path containing the above shooting path must-pass point and the matching shooting parameters, a complete non-defect flaw detection image is generated using a virtual imaging system. For example, a virtual system-flaw detection equipment linkage criterion can be formulated, a complete shooting image and a shooting path preview are generated in the virtual imaging system, the shooting equipment is programmed using the recommended path, the flaw detection equipment is controlled to move, and the recommended flaw detection strategy is automatically converted into a CNC flaw detection.

[0060] Then, the non-defect flaw detection image is subjected to gray fitting difference calculation with the real flaw detection imaging image to locate the position of the real flaw detection defect.

[0061] More specifically, according to another preferred embodiment of the present application, the process can be designed as follows:

[0062] The gray level histogram of the virtual flaw detection image and the real flaw detection image is calculated. For example, assuming that the gray level of the image is r, the number of pixels with the gray level r in the image is h(r), and the gray level histogram can be represented as H(r) = h(r);

[0063] Assuming that the gray level of the image is r, the number of pixels with the gray level less than or equal to r in the image is c(r), and the cumulative distribution function is CDF(r) = c(r) / N; where N is the total number of pixels of the image. Assuming that the gray level of the virtual flaw detection image is r and the gray level of the real flaw detection image is z, a mapping function T(r) can be constructed, such that CDF 虚拟 (r) = CDF 真实 (z), thereby realizing the mapping relationship between the two images;

[0064] The gray fitting difference is performed, and the mapping relationship is used to convert the gray level of the virtual flaw detection image, thereby obtaining the fitted image. Assuming that the pixel value of the non-defect flaw detection image is I1(x, y) and the pixel value of the defect flaw detection image is I2(x, y), the pixel difference value can be represented as D(x, y) = |I1(x, y) - I2(x, y)|. The area where the pixel difference value is greater than the threshold T is the defect area.

[0065] Step four, secondary detection is performed on the obtained real flaw detection defect, the number of real flaw detection defects is increased, and the real flaw detection defects are classified.

[0066] In this step, according to another preferred embodiment of the present application, a network model of defect detection deep learning can be established, and the number of defects can be increased by means of casting simulation and virtual flaw detection simulation, and the defects can be positioned and classified again in the network model according to the defect generation mechanism and defect topographic features. Accordingly, the problems of unbalanced defect categories and single defect morphology can be solved, and the problem of defect detection model “partiality” can be avoided, and high-precision positioning and classification of defects can be realized.

[0067] Step five, the real flaw detection defects obtained are corrected in defect and category.

[0068] In this step, according to another preferred embodiment of the present application, a corresponding defect multi-dimensional feature matrix can be constructed based on the features of the defect contour, internal topography and gray value, and the above-mentioned defect pseudo-correction and category correction operations can be realized by using the feature matrix. In addition, image processing and feature extraction methods such as Gaussian filtering, binarization and morphological processing can also be fused to accurately capture the diversity level features of the defects, create a multi-image defect discrimination algorithm, and realize intelligent automatic association of the same defect on multiple part detection images.

[0069] Step six, the real flaw detection defects obtained are quantitatively rated in hazard.

[0070] In this step, according to another preferred embodiment of the present application, the information of gray scale, texture, area and dispersion degree of the defects can be extracted, and the above-mentioned hazard quantitative rating operation can be realized based on these information. More specifically, for example, different hazard evaluation indexes and feature matrices can be designed for each defect in combination with casting knowledge, production requirements and relevant standards. The defect sub-image is preprocessed by denoising, sharpening and binarization, and the information of defect gray scale, texture, area and dispersion degree is extracted to realize quantitative rating of defect hazard.

[0071] The above-mentioned defect multi-dimensional feature matrix and hazard evaluation index matrix can be represented as follows:

[0072]

[0073] Step seven, the real flaw detection imaging image is converted from two-dimensional to three-dimensional, and the three-dimensional position and depth information of the real flaw detection defects are obtained.

[0074] In this step, there can be various implementation methods for constructing a two-dimensional image into a three-dimensional model, and the preferred method can be the technical solution disclosed in the accepted patent CN202310399158.X and the “casting flaw three-dimensional intelligent positioning method and system based on multi-angle flaw detection image”.

[0075] Then, three-dimensional generation of defects is performed. For example, two-dimensional X-ray images of the casting can be converted into three-dimensional point cloud data, the intersection of the ray source and the casting model is calculated, and the precise position of the intersection is determined by ray casting. On this basis, the key boundary points around the intersection are identified by using a boundary detection algorithm. These boundary points form a characteristic polyhedron, representing the preliminary three-dimensional profile of the defect. In order to accurately reconstruct the defect morphology, edge fitting techniques such as B-spline surface fitting are used to smooth and refine the fitting of these boundary points. Finally, the ray angle is adjusted for multiple projections and optimization, and the three-dimensional position and morphology of the defect are refined to obtain the accurate three-dimensional position and depth information of the defect.

[0076] Step eight, based on the three-dimensional position and depth information of the real defect detected, the casting defect marking equipment is used to mark the defect, thereby completing the entire detection process.

[0077] In this step, for example, the casting defect marking equipment can be used through a mechanical arm, high-precision sensors and image processing techniques are used to avoid accumulated errors in the marking process, a vision system is used to capture the fine features of the casting surface in real time, accurately identify the defect position, depth and other information on the casting, accurately match the actual casting position and the defect information on the casting, and automatically mark the defect.

[0078] In summary, the light alloy casting detection, film shooting, evaluation and positioning marking full-process detection method according to the present application can realize intelligent detection throughout the whole process, and can realize X-ray detection of light alloy casting defects "shooting completely, shooting clearly, evaluation accurately, evaluation stably, evaluation finely, evaluation quickly, and positioning accurately", and thus is especially suitable for high-reliability detection application occasions such as large light alloy complex castings of national defense high-end equipment, and has good practical value and application prospect.

[0079] Those skilled in the art will readily understand that the above description is only a preferred embodiment of the present application and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A full-process detection method for light alloy castings, including filming, film evaluation and positioning marking, characterized in that: The method comprises the following steps: S1. For the light alloy casting to be inspected, obtain real flaw detection imaging images at different flaw detection angles relative to the X-ray source, and simultaneously obtain the corresponding three-dimensional model of the casting; S2. performing image preprocessing on the real flaw detection imaging image, and then performing recognition of the single piece number casting character; S3. Construct a virtual imaging system to generate a flawless flaw detection image; perform grayscale fitting and difference calculation on the flawless flaw detection image and the real flaw detection imaging image to locate the position of the real flaw detection defect; wherein the grayscale fitting and difference calculation process is designed as follows: S31, respectively calculating the grayscale histograms of the defect-free flaw detection image and the true flaw detection imaging image; S32. Using the grayscale histogram, respectively calculate the number and proportion of pixels in the defect-free flaw detection image and the true flaw detection imaging image whose grayscale levels are less than a preset value, and simultaneously establish a mapping relationship between the two images; S33, based on the mapping relationship, converting the grayscale level of the defect-free flaw detection image to obtain a fitted defect-free flaw detection image; S34, performing pixel difference comparison between the fitted flawless detection image and the real flaw detection imaging image, and the area where the pixel difference is greater than the threshold is the location of the real flaw detection defect; S4. Perform secondary inspection on the obtained real flaw detection defects to increase the number of real flaw detection defects and classify the real flaw detection defects; S5. Defect false detection and category correction are performed on the real flaws obtained; S6. Quantitatively rate the hazards of the actual flaws obtained; S7, converting the real flaw detection imaging image from two dimensions into three dimensions, and obtaining three-dimensional position and depth information of the real flaw detection defect; S8. Based on the three-dimensional position and depth information of the actual flaw detection defect, the defect is marked using casting defect marking equipment, thereby completing the entire detection process.

2. The method according to claim 1, wherein In step S2, the recognition process of the piece number casting is designed as follows: S21, detecting the edges of the characters in the real flaw detection imaging image and generating a binary image containing edge information; extracting the outline information of the characters from the binary image, then finding its minimum circumscribed rectangular frame based on the outline information, and performing a segmentation operation to obtain a minimum unit character image; S22. Build a deep learning model and pre-train all characters for recognition; S23. Segment the real flaw detection imaging image into multiple regions, and then use the trained deep learning model to perform optical character recognition of the single piece number casting within each region, and identify the specific content of the single piece number casting.

3. The method according to claim 2, wherein In step S2, it also includes realizing the automatic mapping operation of each process of flaw detection filming, film evaluation and positioning marking with the casting model and batch based on the recognition result of the single piece number casting.

4. The method according to any one of claims 1 to 3, wherein: In step S3, before using the virtual imaging system to generate the defect-free flaw detection image, a filming path planning operation is also included. The process is designed as follows: a. Identify stress concentration areas and high-defect areas of the casting as critical quality control points (CQPs). Extract and search for features of these CQPs and similar structures, and then use them as necessary points in the virtual imaging path. b. Classify the three-dimensional models of castings and ensure that castings classified into the same category share the same necessary points on the filming path; c. According to the filming path including the necessary points of the filming path and the corresponding filming parameters, a complete defect-free flaw detection image is generated using the virtual imaging system.

5. The method according to claim 4, wherein In step S4, a network model for deep learning of defect detection is established, and the number of defects is increased by using casting simulation and virtual flaw detection simulation. At the same time, in the network model, the defects are secondary located and classified according to the defect generation mechanism and defect morphology characteristics.

6. The method according to claim 5, wherein In step S5, a corresponding defect multi-dimensional feature matrix is ​​constructed based on the features of the defect contour, internal morphology and grayscale value, and the defect multi-dimensional feature matrix is ​​used to implement the above-mentioned defect de-aliasing and category correction operations.

7. The method according to claim 6, wherein In step S5, it also includes realizing automatic association operation of defects of the same category in multi-position flaw detection images based on the diverse hierarchical features of defects.

8. The method according to claim 7, wherein In step S6, information such as the grayscale, texture, area and discreteness of the defect is extracted, and the above-mentioned quantitative hazard rating operation is implemented based on this information.

Images