A GMAW welding penetration detection method based on dual-band imaging and HOG

Abstract

This paper discloses a GMAW weld penetration test method based on dual-band imaging and HOG. First, a welding temperature field measurement system is designed based on dual-band imaging. Then, a region of interest (ROI) is randomly selected within the weld heat-affected zone (HAZ) surrounding the molten pool, and a HOG algorithm is proposed to extract the distribution characteristics of the ROI temperature field. Finally, an artificial neural network is used to establish a GMAW weld penetration test model. This invention provides a new method for online GMAW weld penetration testing and is of great significance for further improving the quality of arc additive manufacturing.

Description

Technical Field

[0001] The invention relates to a GMAW welding penetration detection method based on dual-band imaging and HOG. Background Art

[0002] In recent years, arc additive manufacturing technology has developed rapidly and has been widely used in aerospace, energy and power, intelligent manufacturing and other fields. Traditional arc additive manufacturing quality diagnosis is through pre-welding process experiments or post-welding inspections, which have problems such as low detection efficiency and low detection accuracy. It is difficult to meet the requirements of modern manufacturing for production and processing efficiency. Therefore, realizing online high-precision monitoring of arc additive manufacturing quality has become a research hotspot in the welding field, which is of great significance to promoting the development of arc additive manufacturing technology.

[0003] In the arc additive manufacturing process, the penetration of the base weld determines the welding quality to a large extent. Therefore, it is particularly important to realize online monitoring of the penetration of the base weld. At present, the main methods for detecting welding penetration include molten pool visual sensing method, arc sound detection method, etc. For the molten pool visual sensing method, since the visual characteristics of the molten pool are closely related to the welding penetration, many scholars at home and abroad currently detect the welding penetration by extracting the two-dimensional visual characteristics of the molten pool. However, the strong arc light generated during the welding process poses a great challenge to the acquisition of high signal-to-noise ratio molten pool images. At present, the general method is to install a neutral density plate and a filter in front of the camera lens to reduce the influence of the arc light on the acquisition of the molten pool image. This method generally requires a lot of debugging experiments, and in some cases, it cannot achieve good results. For arc sound detection, it is relatively easier to collect arc sound than to collect molten pool images. Currently, many scholars have tried to extract various typical features of arc sound for online detection of welding penetration. However, arc sound signals are easily interfered by industrial background noise. Therefore, these research results can only remain in the preliminary experimental stage and are difficult to apply to actual monitoring environments.

[0004] Temperature is also an important information in the welding process. The distribution and changes of the welding temperature field are closely related to the welding forming quality. The distribution of the welding temperature field is measured online by means of radiation temperature measurement. Subsequently, its important features are extracted using image processing algorithms, and then its relationship with the welding forming quality is established. This is also an important means to achieve online monitoring of the welding forming quality. Summary of the Invention

[0005] This paper presents a method for detecting gas metal arc welding (GMAW) penetration based on dual-band imaging and histograms of oriented gradients (HOG). First, a welding temperature field measurement system is designed based on dual-band imaging. A region of interest (ROI) is randomly selected within the weld heat-affected zone (HAZ) surrounding the molten pool, and a HOG algorithm is proposed to extract the distribution characteristics of the ROI temperature field. Finally, an artificial neural network is used to establish a GMAW weld penetration detection model. This invention provides a new method for online GMAW weld penetration detection, which is of great significance for further improving the quality of arc additive manufacturing.

[0006] The technical solutions adopted in the present invention are:

[0007] A GMAW welding penetration detection method based on dual-band imaging and HOG, comprising:

[0008] 1) Set up a welding temperature field measurement system, which includes a computer, a spectroscope mounted on the welding gun, two infrared band filters, and two infrared cameras. The corresponding functions of each component in the system are as follows:

[0009] The beam splitter is used to split the incident light into two identical outputs;

[0010] The infrared band filter is used to select light in a specific infrared band to pass through;

[0011] The infrared camera is used for imaging the weldment in the specific infrared band;

[0012] The computer is used to process the image output by the infrared camera and calculate the welding temperature field;

[0013] 2) After the transient welding temperature field measurement around the molten pool is completed by the welding temperature field measurement system, a region of interest (ROI) is randomly selected in the welding heat-affected zone around the molten pool. The ROI is divided into four small areas. HOG feature extraction is then performed on each of the four small areas to obtain the HOG feature vector of each small area. Finally, the HOG feature vectors of all the small areas are concatenated to obtain the final HOG feature vector of the temperature field of the ROI.

[0014] 3) With the extracted HOG feature vector of the temperature field of the region of interest as input and the GMAW welding penetration as output, a GMAW welding penetration detection model is established based on an artificial neural network. This model is used for online detection of four penetration states: incomplete penetration, penetration, over-penetration, and weld leakage.

[0015] Furthermore, the calculation and derivation process of the welding temperature field is as follows:

[0016] The exposure time and gain of the two infrared cameras are set to the same. The grayscale values ​​of the output images of the two infrared cameras are N(λ1,T) and N(λ2,T), respectively. The ratio is:

[0017]

[0018] Where λ1 and λ2 are the center wavelengths of the two infrared band filters, δλ1 and δλ2 are the bandwidths of the two infrared band filters, η(λ) is the spectral response rate of the two infrared cameras, τ(λ) is the spectral transmittance of the lens in the infrared camera, γ(λ) is the spectral transmittance of the two infrared band filters, and L(λ,T) is the spectral radiance of the weldment at temperature T at wavelength λ.

[0019] According to the integral mean value theorem, formula (1) can be written as:

[0020]

[0021] According to Planck's radiation law, we can get:

[0022]

[0023] Where ε(λ,T) is the spectral emissivity of the weldment, dimensionless, and C2 is the second radiation constant;

[0024] make:

[0025]

[0026] Formula (3) can be written as:

[0027]

[0028] Assuming ε(λ1,T)≈ε(λ2,T), Equation (5) can be written as:

[0029]

[0030] Wherein, the K value is determined by the welding temperature field measurement system itself and is obtained through calibration experiments.

[0031] Furthermore, the specific process of HOG feature extraction in the region of interest is as follows:

[0032] (1) Calculate the gradient of each pixel in the temperature field image of the region of interest. The gradient of the pixel (x, y) in the image is:

[0033] G x(x,y)=H(x+1,y)-H(x-1,y) (7)

[0034] G y (x,y)=H(x,y+1)-H(x,y-1) (8)

[0035] Where G x (x,y),G y (x,y) and H(x,y) are the horizontal gradient, vertical gradient and pixel value at the pixel point (x,y) in the image respectively;

[0036] From this, we can calculate the gradient magnitude and direction at the pixel point (x, y), which are:

[0037]

[0038]

[0039] (2) The interval of the gradient direction [0, 360°) is set to 30°, and then the gradient direction histogram of the temperature field in the region of interest is calculated;

[0040] (3) Use the vector decomposition method to perform statistics on the gradient direction histogram to obtain the HOG feature vector of the temperature field in the area of ​​interest. The specific statistical method is:

[0041] Suppose two adjacent gradient direction angles are θ1 and θ2, θ1<θ2. When performing gradient direction histogram statistics, if θ1<θ<θ2, use the vector decomposition method to vote on the two adjacent gradient direction histograms. Suppose the gradient amplitude of the pixel in the θ direction is G, the amplitude decomposed into the θ1 direction is G1, and the amplitude decomposed into the θ2 direction is G2, then:

[0042]

[0043] It can be solved:

[0044]

[0045] Formula (12) is the improved gradient direction histogram statistical calculation formula.

[0046] Furthermore, the calibration process of the welding temperature field measurement system is as follows:

[0047] Select a standard high-temperature blackbody, align the input end of the spectroscope with the outlet of the blackbody furnace, first increase the temperature of the blackbody furnace to 100°C, extract the grayscale value of the infrared image output by the infrared camera at this time, and then calibrate the grayscale value of the infrared image corresponding to 100°C;

[0048] Subsequently, the temperature of the blackbody furnace is gradually increased by 20°C, and the above steps are repeated for corresponding calibration until the temperature of the blackbody furnace reaches 1500°C;

[0049] Finally, the back propagation (BP) neural network is used to fit all the calibrated data, with lnR as input and T as output, and the welding temperature field measurement model is completed based on BP neural network training.

[0050] Furthermore, after the two infrared cameras acquire the infrared images of the weldment, they perform symmetry and matching processing on the two infrared images.

[0051] Furthermore, the specific infrared bands are two infrared bands: 1 μm to 2 μm and 2 μm to 3 μm.

[0052] The present invention has the following beneficial effects:

[0053] 1) The present invention can perform high-precision online measurement of the temperature field of the welding heat-affected zone around the molten pool;

[0054] 2) The present invention can eliminate the influence of arc light on GMAW welding penetration detection, and realize high-precision online detection of GMAW welding penetration. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0056] The present invention will be further described below with reference to the accompanying drawings.

[0057] like Figure 1 As shown in the figure, the present invention provides a GMAW welding penetration detection method based on dual-band imaging and HOG. First, a welding temperature field measurement system is designed based on dual-band imaging. The system includes a computer, a spectroscope installed on the welding gun, two infrared band filters and two infrared cameras. The corresponding functions of each component in the system are as follows:

[0058] A beamsplitter is used to split the incident light into two identical output paths.

[0059] Infrared band filters are used to select light in specific infrared bands to pass through. The specific infrared bands are 1μm to 2μm and 2μm to 3μm.

[0060] The infrared camera is used to image the weldment in the specific infrared band.

[0061] The computer is used to process the image output by the infrared camera and calculate the welding temperature field.

[0062] The calculation formula of the welding temperature field is derived from the following formula:

[0063] The exposure time and gain of the two infrared cameras are set to the same. The grayscale values ​​of the output images of the two infrared cameras are N(λ1,T) and N(λ2,T), respectively. The ratio is:

[0064]

[0065] Where λ1 and λ2 are the center wavelengths of the two infrared band filters, δλ1 and δλ2 are the bandwidths of the two infrared band filters, η(λ) is the spectral response rate of the two infrared cameras, τ(λ) is the spectral transmittance of the lens in the infrared camera, γ(λ) is the spectral transmittance of the two infrared band filters, and L(λ,T) is the spectral radiance of the weldment at temperature T at wavelength λ.

[0066] According to the integral mean value theorem, formula (1) can be written as:

[0067]

[0068] According to Planck's radiation law, we can get:

[0069]

[0070] Where ε(λ,T) is the spectral emissivity of the weldment, which is dimensionless, and C2 is the second radiation constant.

[0071] make:

[0072]

[0073] Formula (3) can be written as:

[0074]

[0075] If the two infrared bands are selected to be close, ε(λ1,T)≈ε(λ2,T), and Equation (5) can be written as:

[0076]

[0077] Wherein, the K value is determined by the welding temperature field measurement system itself and needs to be obtained through calibration experiments.

[0078] The specific process of welding temperature field measurement system calibration is as follows:

[0079] Select a standard high-temperature blackbody, align the input end of the spectroscope with the outlet of the blackbody furnace, first increase the temperature of the blackbody furnace to 100°C, extract the grayscale value of the infrared image output by the infrared camera at this time, and then calibrate the grayscale value of the infrared image corresponding to 100°C;

[0080] Subsequently, the temperature of the blackbody furnace is gradually increased by 20°C, and the above steps are repeated until the temperature of the blackbody furnace reaches 1500°C;

[0081] Finally, the back propagation (BP) neural network is used to fit all the calibrated data, with lnR as input and T as output, and the welding temperature field measurement model is completed based on BP neural network training.

[0082] After the welding temperature field measurement system acquires infrared images of the weldment in two bands, it needs to perform symmetry and matching processing on the two infrared images.

[0083] After completing the transient welding temperature field measurement, the welding temperature field measurement system randomly selects an ROI in the welding heat affected zone around the molten pool.

[0084] The HOG algorithm is used to extract the distribution features of the ROI temperature field. The specific process of feature extraction is as follows:

[0085] (1) Calculate the gradient of each pixel in the ROI temperature field image. The gradient of the pixel (x, y) in the image is:

[0086] G x (x,y)=H(x+1,y)-H(x-1,y)(7)

[0087] G y (x,y)=H(x,y+1)-H(x,y-1)(8)

[0088] Where G x (x,y),G y (x,y) and H(x,y) are the horizontal gradient, vertical gradient and pixel value at the pixel point (x,y) in the image respectively.

[0089] From this, we can calculate the gradient magnitude and direction at the pixel point (x, y), which are:

[0090]

[0091]

[0092] (2) The spacing of the gradient direction [0, 360°) is set to 30°, and then the gradient direction histogram of the ROI temperature field is counted.

[0093] (3) Use the vector decomposition method to perform statistics on the gradient direction histogram and obtain the HOG feature vector. The specific statistical method is: let the two adjacent gradient direction angles be θ1 and θ2, θ1<θ2. When performing gradient direction histogram statistics, if θ1<θ<θ2, use the vector decomposition method to vote on the two adjacent gradient direction histograms. Let the gradient amplitude of the pixel in the θ direction be G, the amplitude decomposed into the θ1 direction is G1, and the amplitude decomposed into the θ2 direction is G2, then:

[0094]

[0095] It can be solved:

[0096]

[0097] Formula (12) is the improved gradient direction histogram statistical calculation formula.

[0098] The HOG features of the ROI temperature field are extracted based on the idea of ​​block division. First, the welding temperature field is divided into four small areas. Then, HOG features are extracted for each of the four areas to obtain the HOG feature vector of each area. Finally, the HOG feature vectors of all areas are concatenated together to obtain the final HOG feature vector of the ROI temperature field.

[0099] The extracted multi-dimensional ROI temperature field HOG features are used as input and GMAW welding penetration is used as output. A GMAW welding penetration detection model is established based on artificial neural network. It is used for online detection of four penetration states: incomplete penetration, penetration, over-penetration and weld leakage.

[0100] The above description is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principles of the present invention. These improvements should also be regarded as the scope of protection of the present invention.

Claims

1. A GMAW welding penetration detection method based on dual-band imaging and HOG, characterized by: include: 1) Set up a welding temperature field measurement system, which includes a computer, a spectroscope mounted on the welding gun, two infrared band filters, and two infrared cameras. The corresponding functions of each component in the system are as follows: The beam splitter is used to split the incident light into two identical outputs; The infrared band filter is used to select light in a specific infrared band to pass through; The infrared camera is used for imaging the weldment in the specific infrared band; The computer is used to process the image output by the infrared camera and calculate the welding temperature field; 2) After the transient welding temperature field measurement around the molten pool is completed by the welding temperature field measurement system, a region of interest is randomly selected in the welding heat-affected zone around the molten pool. The region of interest is divided into four small areas. Then, HOG feature extraction is performed on each of the four small areas to obtain the HOG feature vector of each small area. Finally, the HOG feature vectors of all small areas are concatenated to obtain the HOG feature vector of the temperature field of the final region of interest. 3) Using the extracted HOG feature vector of the temperature field of the region of interest as input and the GMAW weld penetration as output, a GMAW weld penetration detection model was established based on an artificial neural network. This model can be used for online detection of four weld penetration states: incomplete weld penetration, weld penetration, excessive weld penetration, and weld leakage. The calculation and derivation process of the welding temperature field is: The exposure time and gain of the two infrared cameras are set to be the same, and the grayscale values ​​of the output images of the two infrared cameras are and , the ratio is: （1） Where, are the center wavelengths of the two infrared band filters, are the bandwidths of the two infrared band filters respectively, is the spectral response rate of the two infrared cameras, is the spectral transmittance of the lens in the infrared camera, is the spectral transmittance of the two infrared band filters, The temperature is The weldment at wavelength spectral radiance under ; According to the integral mean value theorem, formula (1) can be written as: （2） According to Planck's radiation law, we can get: （3） Where, is the spectral emissivity of the weldment, dimensionless, is the second radiation constant; make: （4） Formula (3) can be written as: （5） set up , formula (5) can be written as: （6） Where, The value is determined by the welding temperature field measurement system itself and is obtained through calibration experiments.

2. The GMAW welding penetration detection method based on dual-band imaging and HOG according to claim 1, characterized in that: The specific process of HOG feature extraction in the region of interest is: (1) Calculate the gradient of each pixel in the temperature field image of the area of ​​interest. The gradient of is: （7） （8） Where, 、 、 are pixels in the image The horizontal gradient, vertical gradient and pixel value at ; From this we can calculate the pixel The gradient magnitude and direction at are: （9） （10） (2) The gradient direction The spacing is set to 30 0 , and then calculate the gradient direction histogram of the temperature field in the area of ​​interest; (3) Use the vector decomposition method to perform statistics on the gradient direction histogram to obtain the HOG feature vector of the temperature field in the area of ​​interest. The specific statistical method is: Assume that the two adjacent gradient direction angles are and , , when performing gradient direction histogram statistics, if , use the vector decomposition method to vote for two adjacent gradient direction histograms, and set the pixel point in The gradient magnitude in the direction is , decomposed into The magnitude in the direction is , decomposed into The magnitude in the direction is , then: （11） It can be solved: （12） Formula (12) is the improved gradient direction histogram statistical calculation formula.

3. The GMAW welding penetration detection method based on dual-band imaging and HOG according to claim 1, characterized in that: The calibration process of the welding temperature field measurement system is as follows: Select a standard high-temperature blackbody, align the input end of the spectroscope with the outlet of the blackbody furnace, first increase the temperature of the blackbody furnace to 100°C, extract the grayscale value of the infrared image output by the infrared camera at this time, and then calibrate the grayscale value of the infrared image corresponding to 100°C; Subsequently, the temperature of the blackbody furnace is gradually increased by 20°C, and the above steps are repeated for corresponding calibration until the temperature of the blackbody furnace reaches 1500°C; Finally, the back propagation neural network is used to fit all the calibrated data. As input, As output, the welding temperature field measurement model is completed based on back-propagation neural network training.

4. The GMAW welding penetration detection method based on dual-band imaging and HOG according to claim 1, characterized in that: After the two infrared cameras acquire the infrared images of the weldment, they perform symmetry and matching processing on the two infrared images.

5. The GMAW welding penetration detection method based on dual-band imaging and HOG according to claim 1, characterized in that: The specific infrared bands are two infrared bands: 1 μm to 2 μm and 2 μm to 3 μm.

Images