Welding defect detection method and system fusing magneto-optical imaging and infrared thermal imaging
By fusing magneto-optical imaging and infrared thermal imaging, decision vectors from welding defect images are trained and fused, solving the problems of large errors and poor anti-interference ability in single-image detection, and achieving high-precision welding defect detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF TECH
- Filing Date
- 2022-12-27
- Publication Date
- 2026-05-08
AI Technical Summary
Existing welding defect detection methods rely on single-image detection, which suffers from large errors, poor anti-interference ability, and difficulty in accurately detecting small defects.
By fusing magneto-optical imaging and infrared thermal imaging, and by preprocessing images of welding defect types, a sub-classifier is trained to obtain and fuse decision vectors from the magneto-optical and infrared images to determine the welding defect type.
It improves the accuracy and anti-interference ability of welding defect detection, and can accurately identify minute defects.
Smart Images
Figure CN115880265B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of welding defect detection, and more specifically, to a welding defect detection method and system that integrates magneto-optical imaging and infrared thermal imaging. Background Technology
[0002] Welding technology is one of the most important material forming and processing technologies in modern manufacturing, widely used in shipbuilding, automobile manufacturing, petroleum industry, aerospace and other industrial manufacturing fields. During the welding process, due to improper adjustment of welding parameters and complex environmental uncertainties, various defects such as cracks, pits, porosity, and lack of fusion can occur in the weldment, which will seriously affect the quality of the weld. The quality of welding is a crucial factor for the normal operation of related equipment. Commonly used non-destructive testing methods for detecting welding defects include magnetic particle testing, ultrasonic testing, and radiographic testing. Magnetic particle testing involves uniformly spreading magnetic powder on the surface of the weldment and applying a strong current to both sides of the weldment. The leakage magnetic field at the defect location will change the distribution of the magnetic powder, thus revealing the defect. Ultrasonic testing utilizes the energy changes such as reflection, refraction, and attenuation that occur during the propagation of ultrasound in a medium, and the difference in acoustic and physical properties between the defect area and the original material, thereby detecting internal defects in the weldment. Radiographic testing utilizes the different attenuation characteristics of rays (such as X-rays and gamma rays) on different structures, resulting in images with varying penetration intensity, thus enabling the imaging detection of defects.
[0003] like Figure 1 As shown, magneto-optical detection is based on Faraday's magneto-optical effect and the principle of magnetic leakage. It uses external excitation to magnetize the ferromagnetic sample. Due to the difference in magnetic permeability between the sample and air, defective areas in the sample generate a leakage magnetic field that diffuses into the air. The light emitted from the light source in the magneto-optical sensor is polarized after passing through a polarizer. According to Faraday's magneto-optical effect, the polarized light is deflected in the sample's leakage magnetic field. The deflected polarized light, after passing through an analyzer, exhibits differences in intensity, which are then captured by the CMOS camera in the magneto-optical sensor to form a magneto-optical image containing information about the sample's defects.
[0004] like Figure 2 As shown, eddy current thermal imaging is an active infrared thermal imaging detection method. According to Faraday's law of electromagnetic induction, a pulsed current flowing through a current-carrying coil can induce eddy currents in a nearby metal conductor, thereby inductively heating the sample under test. When defects exist in the sample, these defects affect the distribution of eddy currents and the conduction of induced heat, resulting in uneven heat distribution on the sample surface and uneven radiation of infrared rays. By using an infrared sensor, an infrared radiation sequence or a specific frame of infrared radiation over a period of time during heat propagation can be acquired, thus obtaining an infrared image containing information about sample defects.
[0005] Decision fusion is a type of multi-source information fusion technology that integrates data from different data sources to make a more comprehensive judgment on detection tasks. It has the advantages of strong adaptability, strong anti-interference ability, and elimination of errors from individual sensors.
[0006] Existing technology discloses a workpiece welding defect detection device, method, and computer-readable storage medium. The workpiece welding defect detection device includes: a communicator for receiving a first image from an image acquisition module; a processor coupled to the communicator for extracting welding information based on the first image; transmitting the welding information to a logic processing component to form a welding defect; and forming display information of the welding defect to present the concrete form or feature value of the welding defect based on the display information. This application utilizes a charge-coupled device (CCD) camera to capture images after welding and uses an extraction module to extract welding information from the captured images. However, this method only uses a single image, which has inherent errors and poor anti-interference capabilities. Furthermore, it acquires the entire image of the workpiece after welding, resulting in poor detection performance and low detection accuracy for small defects. Summary of the Invention
[0007] To overcome the shortcomings of the prior art in detecting welding defects, the present invention provides a welding defect detection method and system that integrates magneto-optical imaging and infrared thermal imaging, which can achieve accurate detection of welding defect types.
[0008] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:
[0009] This invention provides a welding defect detection method that integrates magneto-optical imaging and infrared thermal imaging, comprising:
[0010] S1: Acquire magneto-optical and infrared images corresponding to the welding defect types;
[0011] S2: Perform preprocessing operations on the magneto-optical image and infrared image corresponding to the welding defect type to obtain the preprocessed magneto-optical image and infrared image;
[0012] S3: Use the preprocessed magneto-optical image to train and construct a magneto-optical image sub-classifier, and use the preprocessed infrared image to train and construct an infrared image sub-classifier, until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, and obtain the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier.
[0013] S4: Acquire magneto-optical and infrared images of various locations at the weld seam of the workpiece under test;
[0014] S5: Input the magneto-optical images of each location point at the weld seam of the workpiece under test into the trained magneto-optical image sub-classifier, and output the magneto-optical classification decision vector of that location point; input the infrared images of each location point at the weld seam of the workpiece under test into the trained infrared image sub-classifier, and output the infrared classification decision vector of that location point.
[0015] S6: Fusion of the magneto-optical classification decision vector and infrared classification decision vector at each location point of the weld seam of the weldment under test to obtain the fused decision vector at each location point of the weld seam of the weldment under test.
[0016] S7: Determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment to be tested.
[0017] Preferably, the welding defect types include no defects, cracks, pits, porosity, and lack of fusion.
[0018] Preferably, the preprocessing operations include grayscale conversion, cropping, rotation, and flipping.
[0019] Preferably, the network structures of the magneto-optical image sub-classifier and the infrared image sub-classifier are the same, both being built based on existing residual neural networks.
[0020] Preferably, the specific method of step S4 is as follows:
[0021] The locations of the weld seams on the workpiece under test are sequentially numbered 0, 1, ..., N. Location 0 is the initial position of the weld seam, and location N is the final position of the weld seam. The distance between adjacent locations is d. At time t0, the magneto-optical image P of location 0 is acquired. CM，0 At time t1, the magneto-optical image P of location 1 is acquired. CM，1 Infrared image P at position 0 CR，0 , t i At any given time, acquire the magneto-optical image P of position i. CM，i Infrared image P of location i-1 CR，i-1 The data points at various locations along the weld seam of the workpiece to be tested are acquired sequentially until t N+1 The infrared image P of location point N is acquired at any time. CR，N In the formula, i = 0, 1, ..., N+1.
[0022] Preferably, the magneto-optical classification decision vector and the infrared classification decision vector are both vectors including five dimensions and dimension scores; the same dimension of the magneto-optical classification decision vector and the infrared classification decision vector correspond to the same type of welding defect, and the dimension score represents the magneto-optical probability score and infrared probability score of the corresponding welding defect type.
[0023] Preferably, the specific method for obtaining the fusion decision vector of each position point at the weld seam of the weldment to be tested is as follows:
[0024] CA i =k·CM i +(1-k)CR i
[0025] In the formula, CA i CM represents the fusion decision vector at position i of the weld seam of the workpiece under test. i CR represents the magneto-optical classification decision vector at position i of the weld seam of the workpiece under test. i represents the infrared classification decision vector at the i-th position of the weld seam of the weldment to be tested, and k represents the weighting parameter;
[0026] The fusion decision vector is a vector consisting of five dimensions and dimension scores. Each dimension of the fusion decision vector corresponds to the same welding defect type as the corresponding dimension of the magneto-optical classification decision vector or infrared classification decision vector. Each dimension score represents the fusion probability score of the welding defect type.
[0027] Preferably, the specific method for determining the welding defect type corresponding to each location point based on the fusion decision vector of each location point is as follows:
[0028] For any point on the weld seam of the workpiece to be tested, the fusion decision vector is compared with the scores of each dimension of the fusion decision vector. The welding defect type corresponding to the dimension with the largest score is the final welding defect type at that point.
[0029] This invention also provides a welding defect detection system that integrates magneto-optical imaging and infrared thermal imaging, for implementing the above-mentioned welding defect detection method integrating magneto-optical imaging and infrared thermal imaging, comprising:
[0030] The training image acquisition module is used to acquire magneto-optical and infrared images corresponding to welding defect types;
[0031] The training image preprocessing module is used to preprocess the magneto-optical and infrared images corresponding to the welding defect types to obtain preprocessed magneto-optical and infrared images.
[0032] The sub-classifier training module trains a magneto-optical image sub-classifier using preprocessed magneto-optical images and an infrared image sub-classifier using preprocessed infrared images until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, thus obtaining the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier.
[0033] Magneto-optical-infrared detection equipment is used to acquire magneto-optical and infrared images of various locations at the weld seam of the workpiece under test;
[0034] The subclassification decision module is used to input the magneto-optical images of each location point at the weld seam of the weldment under test into the trained magneto-optical image subclassifier and output the magneto-optical classification decision vector of that location point; and to input the infrared images of each location point at the weld seam of the weldment under test into the trained infrared image subclassifier and output the infrared classification decision vector of that location point.
[0035] The fusion decision module is used to fuse the magneto-optical classification decision vector and the infrared classification decision vector corresponding to each location point at the weld seam of the weldment under test, and obtain the fusion decision vector of each location point at the weld seam of the weldment under test.
[0036] The final defect identification module is used to determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment under test.
[0037] Preferably, the magneto-optical infrared detection device includes a housing, a controller, a magneto-optical thin film, a polarizing light source, an electromagnet, a magneto-optical probe, an infrared sensor, an induction coil, and a processor;
[0038] The controller is located on the side of the box body and is electrically connected to the polarization light source, electromagnet and induction coil respectively.
[0039] The magneto-optical thin film is disposed on the bottom surface inside the box, and the electromagnet is disposed above the magneto-optical thin film. The polarization light source and the magneto-optical probe are respectively disposed on both sides of the electromagnet. The polarization light source emits polarized light into the magneto-optical thin film, which is refracted by the magneto-optical thin film and enters the magneto-optical probe. The magneto-optical probe generates a magneto-optical image and transmits it to the processor.
[0040] The induction coil and the magneto-optical film are arranged side by side on the bottom surface inside the box, and the distance between the center point of the induction coil and the center point of the magneto-optical film is d; the infrared sensor is arranged on the top surface inside the box, corresponding to the induction coil, and the infrared sensor collects infrared images and transmits them to the processor.
[0041] The processor is equipped with a trained magneto-optical image sub-classifier and a trained infrared image sub-classifier.
[0042] In use, the weldment to be tested is fixed horizontally with the weld seam aligned with the X direction. The magneto-optical infrared detection device is placed at the initial position of the weld seam on the weldment, with the detection window of the magneto-optical thin film 3 facing point 0. The controller controls the electromagnet to be energized and simultaneously controls the polarized light source to emit polarized light towards the magneto-optical thin film. After refraction by the magneto-optical thin film, the light enters the magneto-optical probe, forming a magneto-optical image P at point 0. CM，0 The image is transmitted to the pre-trained magneto-optical image sub-classifier in the processor; the magneto-optical infrared detection device moves a distance d along the X direction, and the detection window of the magneto-optical thin film moves to position 1 of the weld seam of the workpiece to be tested, and continues to acquire the magneto-optical image P at position 1. CM，1Simultaneously, the detection window of the infrared sensor moves to position 0, the controller controls the induction coil to heat position 0, and the infrared sensor acquires the infrared image P of position 0. CR，0 The image is then transmitted to the trained infrared image sub-classifier in the processor; subsequently, the magneto-optical infrared detection device continues to move a distance d along the X direction to complete the magneto-optical image P at position 2. CM，2 Magneto-optical image P at position 1 CM，1 The data is collected; repeat the above process, moving along the X direction until the entire weld seam is inspected.
[0043] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:
[0044] This invention is based on the principles of magneto-optical imaging and infrared imaging. It acquires magneto-optical and infrared images corresponding to different welding defect types. After preprocessing, it trains magneto-optical and infrared sub-classifiers to obtain trained sub-classifiers for subsequent defect detection. Then, it sequentially acquires magneto-optical and infrared images of various locations along the weld seam of the workpiece under test, inputting these images into the trained sub-classifiers to obtain magneto-optical and infrared classification decision vectors for each location. Finally, it combines these two decision vectors to obtain a fused decision vector, determining the final welding defect type for each location along the weld seam. This invention combines the advantages of magneto-optical and infrared imaging, incorporating their unique feature information to increase anti-interference capabilities. Furthermore, the sequential acquisition and continuous detection of various locations along the weld seam of the workpiece under test facilitates the detection of minute defects, resulting in high detection accuracy. Attached Figure Description
[0045] Figure 1 The diagram below illustrates the principle of the magneto-optical detection method described in the background section.
[0046] Figure 2 This is a schematic diagram illustrating the principle of the active infrared thermal imaging detection method described in the background art.
[0047] Figure 3 This is a flowchart of the welding defect detection method that combines magneto-optical imaging and infrared thermal imaging as described in Example 1.
[0048] Figure 4 This is a schematic diagram of the welding defect detection system that integrates magneto-optical imaging and infrared thermal imaging as described in Example 3.
[0049] Figure 5 This is a schematic diagram of the magneto-optical infrared detection device described in Example 3. Detailed Implementation
[0050] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.
[0051] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions;
[0052] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.
[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0054] Example 1
[0055] This embodiment provides a welding defect detection method that integrates magneto-optical imaging and infrared thermal imaging, such as... Figure 3 As shown, it includes:
[0056] S1: Acquire magneto-optical and infrared images corresponding to the welding defect types;
[0057] S2: Perform preprocessing operations on the magneto-optical image and infrared image corresponding to the welding defect type to obtain the preprocessed magneto-optical image and infrared image;
[0058] S3: Use the preprocessed magneto-optical image to train and construct a magneto-optical image sub-classifier, and use the preprocessed infrared image to train and construct an infrared image sub-classifier, until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, and obtain the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier.
[0059] S4: Acquire magneto-optical and infrared images of various locations at the weld seam of the workpiece under test;
[0060] S5: Input the magneto-optical images of each location point at the weld seam of the workpiece under test into the trained magneto-optical image sub-classifier, and output the magneto-optical classification decision vector of that location point; input the infrared images of each location point at the weld seam of the workpiece under test into the trained infrared image sub-classifier, and output the infrared classification decision vector of that location point.
[0061] S6: Fusion of the magneto-optical classification decision vector and infrared classification decision vector at each location point of the weld seam of the weldment under test to obtain the fused decision vector at each location point of the weld seam of the weldment under test.
[0062] S7: Determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment to be tested.
[0063] In this implementation, based on the principles of magneto-optical imaging and infrared imaging, magneto-optical and infrared images corresponding to welding defect types are acquired. After preprocessing, magneto-optical and infrared image sub-classifiers are trained to obtain trained magneto-optical and infrared image sub-classifiers, which serve as classifiers for subsequent defect detection. Then, magneto-optical and infrared images of various locations at the weld seam of the weldment under test are sequentially acquired and input into the trained magneto-optical and infrared image sub-classifiers, respectively, to obtain magneto-optical and infrared classification decision vectors for each location. Finally, the two decision vectors are combined to obtain a fused decision vector, determining the final welding defect type corresponding to each location at the weld seam of the weldment under test. This embodiment combines the advantages of magneto-optical and infrared imaging, incorporating their unique feature information, increasing anti-interference capabilities, and achieving high detection accuracy.
[0064] Example 2
[0065] This embodiment provides a welding defect detection method that integrates magneto-optical imaging and infrared thermal imaging, including:
[0066] S1: Acquire magneto-optical and infrared images corresponding to the welding defect types;
[0067] In this embodiment, the welding defect types include no defects, cracks, pits, porosity, and lack of fusion.
[0068] S2: Perform preprocessing operations on the magneto-optical image and infrared image corresponding to the welding defect type to obtain the preprocessed magneto-optical image and infrared image;
[0069] The preprocessing operations include grayscale conversion, cropping, rotation, and flipping.
[0070] S3: Use the preprocessed magneto-optical image to train and construct a magneto-optical image sub-classifier, and use the preprocessed infrared image to train and construct an infrared image sub-classifier, until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, and obtain the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier.
[0071] The magneto-optical image subclassifier and the infrared image subclassifier have the same network structure, both of which are built based on existing residual neural networks.
[0072] S4: Acquire magneto-optical and infrared images of various locations at the weld seam of the weldment under test; specifically:
[0073] The locations of the weld seams on the workpiece under test are sequentially numbered 0, 1, ..., N. Location 0 is the initial position of the weld seam, and location N is the final position of the weld seam. The distance between adjacent locations is d. At time t0, the magneto-optical image P of location 0 is acquired. CM.0 At time t1, the magneto-optical image P of location 1 is acquired. CM，1 Infrared image P at position 0 CR，0 , t i At any given time, acquire the magneto-optical image P of position i. CM，i Infrared image P of location i-1 CR，i-1 The data points at various locations along the weld seam of the workpiece to be tested are acquired sequentially until t N+1 The infrared image P of location point N is acquired at any time. CR，N In the formula, i = 0, 1, ..., N+1.
[0074] S5: Input the magneto-optical images of each location point at the weld seam of the workpiece under test into the trained magneto-optical image sub-classifier, and output the magneto-optical classification decision vector of that location point; input the infrared images of each location point at the weld seam of the workpiece under test into the trained infrared image sub-classifier, and output the infrared classification decision vector of that location point.
[0075] Both the magneto-optical classification decision vector and the infrared classification decision vector are vectors that include five dimensions and dimension scores. The same dimension of the magneto-optical classification decision vector and the infrared classification decision vector corresponds to the same type of welding defect, and the score of each dimension represents the magneto-optical probability score and the infrared probability score of the corresponding welding defect type.
[0076] In this embodiment, the five dimensions of the magneto-optical classification decision vector correspond to no defects, cracks, pits, pores, and lack of fusion, respectively, and the scores of the five dimensions of the magneto-optical classification decision vector represent the magneto-optical probability scores of the corresponding defect types; the five dimensions of the infrared classification decision vector correspond to no defects, cracks, pits, pores, and lack of fusion, respectively, and the scores of the five dimensions of the infrared classification decision vector represent the magneto-optical probability scores of the corresponding defect types.
[0077] S6: The magneto-optical classification decision vector and the infrared classification decision vector corresponding to each location point at the weld seam of the weldment under test are fused to obtain the fused decision vector for each location point at the weld seam of the weldment under test; specifically:
[0078] CA i =k·CM i +(1-k)CR i
[0079] In the formula, CA i CM represents the fusion decision vector at position i of the weld seam of the workpiece under test. iCR represents the magneto-optical classification decision vector at position i of the weld seam of the workpiece under test. i represents the infrared classification decision vector at the i-th position of the weld seam of the weldment to be tested, and k represents the weighting parameter;
[0080] The fusion decision vector is a vector comprising five dimensions and their scores. Each dimension of the fusion decision vector corresponds to the welding defect type of the corresponding dimension in the magneto-optical or infrared classification decision vector. Each dimension score represents the fusion probability score for the welding defect type. To add more categories, simply increase the dimensions of each decision component accordingly.
[0081] S7: Determine the final welding defect type corresponding to each location point at the weld seam of the weldment under test based on the fused decision vector. Specifically:
[0082] For any point on the weld seam of the workpiece to be tested, the fusion decision vector is compared with the scores of each dimension of the fusion decision vector. The welding defect type corresponding to the dimension with the largest score is the final welding defect type at that point.
[0083] Example 3
[0084] This embodiment provides a welding defect detection system that integrates magneto-optical imaging and infrared thermal imaging, used to implement the welding defect detection method that integrates magneto-optical imaging and infrared thermal imaging as described in Embodiment 1 or 2. Figure 4 As shown, it includes:
[0085] The training image acquisition module is used to acquire magneto-optical and infrared images corresponding to welding defect types;
[0086] The training image preprocessing module is used to preprocess the magneto-optical and infrared images corresponding to the welding defect types to obtain preprocessed magneto-optical and infrared images.
[0087] The sub-classifier training module trains a magneto-optical image sub-classifier using preprocessed magneto-optical images and an infrared image sub-classifier using preprocessed infrared images until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, thus obtaining the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier.
[0088] Magneto-optical-infrared detection equipment is used to acquire magneto-optical and infrared images of various locations at the weld seam of the workpiece under test;
[0089] The subclassification decision module is used to input the magneto-optical images of each location point at the weld seam of the weldment under test into the trained magneto-optical image subclassifier and output the magneto-optical classification decision vector of that location point; and to input the infrared images of each location point at the weld seam of the weldment under test into the trained infrared image subclassifier and output the infrared classification decision vector of that location point.
[0090] The fusion decision module is used to fuse the magneto-optical classification decision vector and the infrared classification decision vector corresponding to each location point at the weld seam of the weldment under test, and obtain the fusion decision vector of each location point at the weld seam of the weldment under test.
[0091] The final defect identification module is used to determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment under test.
[0092] like Figure 5 As shown, the magneto-optical infrared detection device includes a housing 1, a controller 2, a magneto-optical thin film 3, a polarizing light source 4, an electromagnet 5, a magneto-optical probe 6, an infrared sensor 7, an induction coil 8, and a processor 9.
[0093] The controller 2 is located on the side inside the housing 1 and is electrically connected to the control terminals of the polarization light source 4, the electromagnet 5, and the induction coil 8, respectively.
[0094] The magneto-optical thin film 3 is disposed on the bottom surface inside the box 1, and the electromagnet 5 is disposed above the magneto-optical thin film 3; the polarization light source 4 and the magneto-optical probe 6 are respectively disposed on both sides of the electromagnet 5. The polarization light source 4 emits polarized light into the magneto-optical thin film 3, which enters the magneto-optical probe 6 after being refracted by the magneto-optical thin film 3, generating a magneto-optical image that is transmitted to the processor 9.
[0095] The induction coil 8 and the magneto-optical film 3 are arranged side by side on the bottom surface inside the box 1, and the distance between the center point of the induction coil 8 and the center point of the magneto-optical film 3 is d; the infrared sensor 7 is arranged on the top surface inside the box 1 corresponding to the induction coil 8, and collects infrared images and transmits them to the processor 9.
[0096] The processor 9 is equipped with a trained magneto-optical image sub-classifier and a trained infrared image sub-classifier.
[0097] In use, the workpiece to be tested is fixed horizontally with the weld seam aligned with the X direction. The magneto-optical infrared detection device is placed at the initial position of the weld seam on the workpiece, with the detection window of the magneto-optical thin film 3 facing point 0. The controller 2 controls the electromagnet 5 to be energized, and simultaneously controls the polarized light source 4 to emit polarized light into the magneto-optical thin film 3. After refraction by the magneto-optical thin film 3, the light enters the magneto-optical probe 6, forming a magneto-optical image P at point 0. CM，0The image is transmitted to the trained magneto-optical image sub-classifier in processor 9; the magneto-optical infrared detection device moves a distance d along the X direction, and the detection window of the magneto-optical thin film 3 moves to position 1 of the weld seam of the workpiece to be tested, and continues to acquire the magneto-optical image P at position 1. CM，1 Simultaneously, the detection window of infrared sensor 7 moves to position 0, controller 2 controls induction coil 8 to heat position 0, and infrared sensor 7 acquires infrared image P of position 0. CR，0 The image is transmitted to the trained infrared image sub-classifier in processor 9; then the magneto-optical infrared detection device continues to move a distance d along the X direction to complete the magneto-optical image P at position 2. CM，2 Magneto-optical image P at position 1 CM，1 The data is collected; repeat the above process, moving along the X direction until the entire weld seam is inspected.
[0098] The system provided in this embodiment can continuously and sequentially acquire magneto-optical and infrared images of various locations along the entire weld seam of the weldment under test. It is simple to operate and highly automated. The images are input into the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier in the processor 9 to obtain the magneto-optical classification decision vector and the infrared classification decision vector for each location point. Finally, the two decision vectors are combined to obtain the fused decision vector, which determines the final welding defect type corresponding to each location point at the weld seam of the weldment under test. It has strong anti-interference ability and high detection accuracy.
[0099] The same or similar labels correspond to the same or similar parts;
[0100] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.
[0101] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A method for detecting welding defects by integrating magneto-optical imaging and infrared thermal imaging, characterized in that, include: S1: Obtain magneto-optical and infrared images corresponding to welding defect types, wherein the welding defect types include no defects, cracks, pits, pores, and lack of fusion; S2: Perform preprocessing operations on the magneto-optical image and infrared image corresponding to the welding defect type to obtain the preprocessed magneto-optical image and infrared image; S3: Use the preprocessed magneto-optical image to train and construct a magneto-optical image sub-classifier, and use the preprocessed infrared image to train and construct an infrared image sub-classifier, until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, and obtain the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier. S4: Obtain magneto-optical and infrared images of various points at the weld seam of the weldment to be tested. The specific method is as follows: Sequentially number the locations of the weld joints on the workpiece to be tested. Point 0 represents the initial position of the weld seam in the workpiece to be tested. The location point is the end position of the weld seam of the workpiece to be tested, and the distance between adjacent locations is d; Acquire magneto-optical image of position 0 at any time , Acquire magneto-optical image of location 1 at any time Infrared image of position 0 , Get it in real time Magneto-optical image of location number [number] and Infrared image of location number [number] The data points at various locations along the weld seam of the workpiece to be tested are acquired sequentially until... Get it in real time Infrared image of location number [number] In the formula, ; S5: Input the magneto-optical images of each location point at the weld seam of the workpiece under test into the trained magneto-optical image sub-classifier, and output the magneto-optical classification decision vector of that location point; input the infrared images of each location point at the weld seam of the workpiece under test into the trained infrared image sub-classifier, and output the infrared classification decision vector of that location point. Both the magneto-optical classification decision vector and the infrared classification decision vector are vectors that include five dimensions and dimension scores; the same dimension of the magneto-optical classification decision vector and the infrared classification decision vector correspond to the same type of welding defect, and the score of each dimension represents the magneto-optical probability score and the infrared probability score of the corresponding welding defect type. S6: Fusion of the magneto-optical classification decision vector and infrared classification decision vector at each location point of the weld seam of the weldment under test to obtain the fused decision vector at each location point of the weld seam of the weldment under test. S7: Determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment to be tested.
2. The welding defect detection method integrating magneto-optical imaging and infrared thermal imaging according to claim 1, characterized in that, The preprocessing operations include grayscale conversion, cropping, rotation, and flipping.
3. The welding defect detection method integrating magneto-optical imaging and infrared thermal imaging according to claim 1, characterized in that, The magneto-optical image subclassifier and the infrared image subclassifier have the same network structure, both of which are built based on existing residual neural networks.
4. The welding defect detection method integrating magneto-optical imaging and infrared thermal imaging according to claim 1, characterized in that, The specific method for obtaining the fusion decision vector of each location point at the weld seam of the weldment to be tested is as follows: In the formula, Indicates the first weld seam of the workpiece to be tested. The fusion decision vector of position number 1 Indicates the first weld seam of the workpiece to be tested. The magneto-optical classification decision vector at location number 1 Indicates the first weld seam of the workpiece to be tested. The infrared classification decision vector at location number 1. Indicates the weighted parameters; The fusion decision vector is a vector consisting of five dimensions and dimension scores. Each dimension of the fusion decision vector corresponds to the same welding defect type as the corresponding dimension of the magneto-optical classification decision vector or infrared classification decision vector. Each dimension score represents the fusion probability score of the welding defect type.
5. The welding defect detection method integrating magneto-optical imaging and infrared thermal imaging according to claim 1, characterized in that, The specific method for determining the final welding defect type corresponding to a location point based on the fusion decision vector of each location point at the weld seam of the weldment to be tested is as follows: For any point on the weld seam of the workpiece to be tested, the fusion decision vector is compared with the scores of each dimension of the fusion decision vector. The welding defect type corresponding to the dimension with the largest score is the final welding defect type at that point.
6. A welding defect detection system integrating magneto-optical imaging and infrared thermal imaging, used to implement the welding defect detection method integrating magneto-optical imaging and infrared thermal imaging as described in any one of claims 1-5, characterized in that, include: The training image acquisition module is used to acquire magneto-optical and infrared images corresponding to welding defect types; The training image preprocessing module is used to preprocess the magneto-optical and infrared images corresponding to the welding defect types to obtain preprocessed magneto-optical and infrared images. The sub-classifier training module trains a magneto-optical image sub-classifier using preprocessed magneto-optical images and an infrared image sub-classifier using preprocessed infrared images until the corresponding magneto-optical image classification loss function and infrared image classification loss function converge, thus obtaining the trained magneto-optical image sub-classifier and the trained infrared image sub-classifier. Magneto-optical-infrared detection equipment is used to acquire magneto-optical and infrared images of various locations at the weld seam of the workpiece under test; The subclassification decision module is used to input the magneto-optical images of each location point at the weld seam of the weldment under test into the trained magneto-optical image subclassifier and output the magneto-optical classification decision vector of that location point; and to input the infrared images of each location point at the weld seam of the weldment under test into the trained infrared image subclassifier and output the infrared classification decision vector of that location point. The fusion decision module is used to fuse the magneto-optical classification decision vector and the infrared classification decision vector corresponding to each location point at the weld seam of the weldment under test, and obtain the fusion decision vector of each location point at the weld seam of the weldment under test. The final defect identification module is used to determine the final welding defect type corresponding to each location point based on the fusion decision vector of each location point at the weld seam of the weldment under test.
7. The welding defect detection system integrating magneto-optical imaging and infrared thermal imaging according to claim 6, characterized in that, The magneto-optical infrared detection device includes a housing (1), a controller (2), a magneto-optical thin film (3), a polarization light source (4), an electromagnet (5), a magneto-optical probe (6), an infrared sensor (7), an induction coil (8), and a processor (9). The controller (2) is located on the side inside the box (1) and is electrically connected to the polarization light source (4), the electromagnet (5) and the induction coil (8) respectively. The magneto-optical thin film (3) is disposed on the bottom surface inside the box (1), and the electromagnet (5) is disposed above the magneto-optical thin film (3); the polarized light source (4) and the magneto-optical probe (6) are respectively disposed on both sides of the electromagnet (5). The polarized light source (4) emits polarized light into the magneto-optical thin film (3), which enters the magneto-optical probe (6) after being refracted by the magneto-optical thin film (3). The magneto-optical probe (6) generates a magneto-optical image and transmits it to the processor (9). The induction coil (8) and the magneto-optical film (3) are arranged side by side on the bottom surface inside the housing (1), and the distance between the center point of the induction coil (8) and the center point of the magneto-optical film (3) is... d The infrared sensor (7) is set on the top surface inside the box (1) in relation to the induction coil (8). The infrared sensor (7) collects infrared images and transmits them to the processor (9). The processor (9) is equipped with a trained magneto-optical image subclassifier and a trained infrared image subclassifier.
Citation Information
Patent Citations
Weld defect detection method, device, equipment and system and memory medium
CN108711151A
Visible light image and infrared image fusion ship classification method and device based on deep learning
CN113610180A