Welded joint penetration state detection method based on sequential temperature distribution segmented attention

Abstract

The invention provides a welding joint penetration state detection method based on time sequence temperature distribution segmented attention, which comprises the following steps: constructing a welding joint surface temperature field measurement system by a color CCD, and measuring to obtain high-precision welding joint surface temperature field distribution based on a welding joint surface temperature field measurement algorithm of color CCD colorimetric temperature measurement correction; according to a welding joint surface temperature field distribution feature extraction algorithm with segmented attention of temperature distribution, temperature field distribution features are extracted; and taking the extracted time sequence welding joint surface temperature field distribution characteristics as input, establishing a welding joint penetration state prediction model based on an artificial neural network, and outputting the penetration state of the welding joint at the current moment in real time. According to the method, the relationship between the distribution characteristics of the surface temperature field of the welding joint and the penetration state of the welding joint is established by performing online measurement on the surface temperature field of the welding joint, so that online detection of the penetration state of the welding joint is realized, and a new method is provided for online detection of welding forming quality.

Description

Technical Field

[0001] The invention relates to the field of welding forming quality detection, and in particular to a method for detecting the penetration state of a welding joint based on segmented attention to temporal temperature distribution. Background Art

[0002] The stability of the penetration state of the weld joint is an important criterion for evaluating the quality of welding. During the welding process, key process parameters such as welding current and welding speed will affect the stability of the penetration state of the weld joint. At the same time, the structural dimensions of the welding mother plate are also closely related to the stability of the penetration state of the weld joint. Fluctuations in any of the above factors may cause changes in the penetration state of the weld joint. Therefore, it is very necessary to realize online sensing detection of the penetration state of the weld joint. Accurately identifying the real-time penetration state of the weld joint is the key to dynamic welding process monitoring, which is of great significance to improving the quality of welding.

[0003] Welding is a complex physical and chemical change process. The welding process will generate strong light radiation, thermal radiation and acoustic radiation. The above three types of radiation information can reflect the real-time penetration state of the weld to a certain extent. Compared with the weld penetration state detection technology based on molten pool vision and sound sensing, the weld penetration state detection technology based on temperature sensing has been reported less. The arc light in the welding process is strong, the temperature is high and the distribution range is wide, which brings great challenges to the temperature measurement of the welding process. At present, scholars generally use infrared sensing to measure the surface temperature field of the welding plate. However, the temperature field distribution measured by this method is basically affected by the emissivity of the welding plate material. In addition, infrared temperature measurement technology is generally suitable for temperature measurement of low-temperature targets. For the temperature measurement of the high-temperature surface of the welding plate, its measurement accuracy is difficult to guarantee, and for the temperature field measurement of the molten pool area covered by the arc light, its measurement accuracy will also be low. At present, it is necessary to further explore the anti-interference and high-precision welding plate surface temperature field measurement technology, so as to accurately monitor the thermal cycle of the welding process and promote the development of weld penetration state detection technology based on temperature sensing. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for detecting the penetration state of a welding joint based on segmented attention of temporal temperature distribution, and its specific technical solution is as follows:

[0005] A method for detecting the penetration state of a welding joint based on segmented attention of temporal temperature distribution comprises the following steps:

[0006] Step 1: Use color CCD to build a welding joint surface temperature field measurement system, design a welding joint surface temperature field measurement algorithm based on color CCD colorimetric temperature measurement correction, and measure the high-precision welding joint surface temperature field distribution;

[0007] Step 2: Design an algorithm for extracting the temperature field distribution characteristics of the welding joint surface based on the segmented attention of temperature distribution, and extract the temperature field distribution characteristics;

[0008] Step 3: Taking the extracted time-series welding joint surface temperature field distribution characteristics as input, a welding joint penetration state prediction model is established based on artificial neural network, and the welding joint penetration state at the current moment is output in real time.

[0009] Furthermore, in step 1, the color CCD is fixed on the welding gun, and the red and green channel output images of the color CCD are used for colorimetric temperature measurement. The original colorimetric temperature measurement formula is:

[0010]

[0011] In the formula, the K value is determined by the temperature measurement system parameters and needs to be obtained through a blackbody calibration experiment. 2 is the second radiation constant, λ r , g are the central wavelengths of the red and green channels respectively, and R and G are the grayscale values ​​of the red and green channel output images respectively.

[0012] The temperature range of the blackbody selected in the blackbody calibration experiment is 700℃~1600℃. In the blackbody calibration experiment, the input end of the color CCD is aligned with the opening of the blackbody radiation source, and the aperture of the optical lens is set to F16. First, the temperature of the blackbody radiation source is raised to 1000℃, and the temperature measurement system collects the blackbody radiation image. Then, the red and green channels are separated to output the blackbody radiation image, and the grayscale values ​​R and G of the two channel output images are extracted respectively. These two values ​​can be obtained by calculating the average grayscale value of the middle area of ​​the radiation image. Finally, the corresponding K value is calculated according to formula (1). Subsequently, the temperature rise step of the blackbody radiation source is set to 100℃, and calibration is performed once every time a temperature point is reached according to the above steps until the temperature of the blackbody radiation source reaches 1600℃. In the process of gradual increase in the temperature of the blackbody radiation source, the grayscale value of the color CCD output image is prevented from reaching saturation by changing the integration time.

[0013] In fact, the K value is not constant, so the original colorimetric temperature measurement formula needs to be corrected. The K value is regarded as a function of R / G. For uncalibrated temperature points, the corresponding K value is calculated using linear interpolation. The corrected calculation formula for the surface temperature field of the weld joint is:

[0014]

[0015] Furthermore, in step 2, based on the measured temperature field image of the welding joint surface, the distribution range of the temperature of the welding joint surface is divided into multiple intervals, the proportion of pixels in each temperature interval is counted, and the proportion of pixels in each temperature interval is connected in series to form the distribution characteristics of the temperature field of the welding joint surface. Specifically, in the temperature field image of the welding joint surface, the proportion of pixels with a temperature less than 1200K, the proportion of pixels with a temperature distribution between 1200K and 1400K, the proportion of pixels with a temperature distribution between 1400K and 1600K, the proportion of pixels with a temperature distribution between 1600K and 1800K, the proportion of pixels with a temperature distribution between 1800K and 2000K, the proportion of pixels with a temperature distribution between 2000K and 2200K, and the proportion of pixels with a temperature distribution above 2200K constitute the distribution characteristics of the temperature field of the welding joint surface [f 1 ,f 2 ,f 3 ,f 4 ,f 5 ,f 6 ,f 7 ].

[0016] Furthermore, in step 3, the surface temperature field distribution characteristics of the weld joint corresponding to the current frame and its two historical frames are extracted as input, and the weld joint penetration state corresponding to the current frame is used as output, and a back propagation neural network is used to establish a weld joint penetration state prediction model. The input of the weld joint penetration state prediction model is: [f 1 (ti),f 2 (ti),f 3 (ti),f 4 (ti),f 5 (ti),f 6 (ti),f 7 (ti)], where i = 0, 1, 2, the output of the prediction network model is determined as: P(t). The hidden layer of the prediction network model is set to two layers, each hidden layer is set to 12 neurons, and the structure of the established welding joint penetration state prediction model is 21-12-12-1.

[0017] Beneficial effects of the present invention:

[0018] The welding joint surface temperature field measurement algorithm designed by the present invention based on color CCD colorimetric temperature measurement correction realizes high-precision online measurement of the welding joint surface temperature field. The welding joint penetration state prediction model based on time-series temperature distribution segmented attention established by the present invention realizes high-precision online detection of the welding joint penetration state, and provides a necessary strategy for the online detection of the welding joint penetration state in the metal electrode gas shielded welding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a method for detecting the penetration state of a welding joint based on segmented attention to time-series temperature distribution provided by the present invention.

[0020] Figure 2 It is a visual image of the welding joint of the present invention under different penetration states and a corresponding temperature field image obtained by measurement.

[0021] Figure 3 It is a schematic diagram of the process of establishing a prediction model for the penetration state of a welding joint based on segmented attention to time-series temperature distribution according to the present invention.

[0022] Figure 4 It is the prediction result of the welding joint penetration state prediction model of the present invention.

[0023] Figure 5 It is a structural schematic diagram of a welding base material in a generalization capability verification experiment of the present invention.

[0024] Figure 6 These are the front and back images of the welding joint in the generalization ability verification experiment of the present invention.

[0025] Figure 7 It is the prediction result of the generalization ability verification experiment welding joint penetration state prediction model of the present invention. DETAILED DESCRIPTION

[0026] The present invention will now be described in further detail with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0027] The present invention provides a method for detecting the penetration state of a welding joint based on segmented attention to time-series temperature distribution.

[0028] like Figure 1 The flowchart of the method for detecting the penetration state of a welding joint is shown in FIG.

[0029] First, a color CCD is used to build a welding joint surface temperature field measurement system. The color CCD is fixed on the welding gun, and the red and green channel output images of the color CCD are used for colorimetric temperature measurement. The original colorimetric temperature measurement formula is:

[0030]

[0031] In the formula, the K value is determined by the temperature measurement system parameters and needs to be obtained through a blackbody calibration experiment. 2 is the second radiation constant, λ r , g are the central wavelengths of the red and green channels respectively, and R and G are the grayscale values ​​of the red and green channel output images respectively.

[0032] The temperature range of the blackbody selected in the blackbody calibration experiment is 700℃~1600℃. In the blackbody calibration experiment, the input end of the color CCD is aligned with the opening of the blackbody radiation source, and the aperture of the optical lens is set to F16. First, the temperature of the blackbody radiation source is raised to 1000℃, and the temperature measurement system collects the blackbody radiation image. Then, the red and green channels are separated to output the blackbody radiation image, and the grayscale values ​​R and G of the two channel output images are extracted respectively. These two values ​​can be obtained by calculating the average grayscale value of the middle area of ​​the radiation image. Finally, the corresponding K value is calculated according to formula (1). Subsequently, the temperature rise step of the blackbody radiation source is set to 100℃, and calibration is performed once every time a temperature point is reached according to the above steps until the temperature of the blackbody radiation source reaches 1600℃. In the process of gradual increase in the temperature of the blackbody radiation source, the grayscale value of the color CCD output image is prevented from reaching saturation by changing the integration time.

[0033] In fact, the K value is not constant, so the original colorimetric temperature measurement formula needs to be corrected. The K value is regarded as a function of R / G. For uncalibrated temperature points, the corresponding K value is calculated using linear interpolation. The corrected calculation formula for the surface temperature field of the weld joint is:

[0034]

[0035] C 2 is the second radiation constant, λ r , g are the central wavelengths of the red and green channels respectively, and R and G are the grayscale values ​​of the red and green channel output images respectively.

[0036] Then, the temperature field distribution features are extracted by the welding joint surface temperature field distribution feature extraction algorithm based on the temperature distribution segmentation attention. According to the measured welding joint surface temperature field image, the distribution range of the welding joint surface temperature is divided into multiple intervals, the proportion of pixels in each temperature interval is counted, and the proportion of pixels in each temperature interval is connected in series to form the distribution features of the welding joint surface temperature field. Specifically, in the welding joint surface temperature field image, the proportion of pixels with a temperature less than 1200K, the proportion of pixels with a temperature distribution between 1200K and 1400K, the proportion of pixels with a temperature distribution between 1400K and 1600K, the proportion of pixels with a temperature distribution between 1600K and 1800K, the proportion of pixels with a temperature distribution between 1800K and 2000K, the proportion of pixels with a temperature distribution between 2000K and 2200K, and the proportion of pixels with a temperature distribution above 2200K constitute the distribution features of the welding joint surface temperature field [f 1 ,f 2 ,f 3 ,f 4 ,f 5 ,f6 ,f 7 ].

[0037] Finally, the surface temperature field distribution characteristics of the weld joint corresponding to the current frame and its two historical frames are extracted as input, and the weld penetration state corresponding to the current frame is used as output. The back propagation neural network is used to establish a weld penetration state prediction model. The input of the weld joint penetration state prediction model is: [f 1 (ti),f 2 (ti),f 3 (ti),f 4 (ti),f 5 (ti),f 6 (ti),f 7 (ti)], where i = 0, 1, 2, the output of the prediction network model is determined as: P(t). The hidden layer of the prediction network model is set to two layers, each hidden layer is set to 12 neurons, and the structure of the established welding joint penetration state prediction model is 21-12-12-1.

[0038] The present invention is based on the cold metal transfer (CMT) welding process, and detects the penetration state of the welding joint during the butt welding process. The welding joint surface temperature field measurement system is composed of an optical lens (Computar), a color CCD industrial camera (Basler, acA640-750uc), a signal generator (UNI-T, UTG932), and a computing and processing system. The focal length of the optical lens is 25mm. During the welding process, the working mode of the signal generator is set to a square wave signal with a transmission frequency of 500Hz, that is, the CCD camera collects a frame of butt joint image every 2ms, and the integration time of the CCD camera is set to 300μs. The welding experiment uses a stainless steel welding wire with a diameter of 1.2mm, and the welding parent material uses a 304 stainless steel plate with a size of 12cm×6cm×5mm. Along the welding direction, the groove of the welding parent material is processed into a Y-shaped groove, the groove angle is 45°, and the blunt edge is 1mm. The shielding gas is pure argon gas, and the shielding gas flow rate is 20L / min. The motion track length of the welding gun is set to 8 cm. The transition of the penetration state of the butt joint is achieved by adjusting the welding current and welding speed. Six sets of welding process parameters are used for butt welding experiments. The specific welding process parameters and the corresponding penetration states of the welding joint are as follows:

[0039]

[0040] Before measuring the surface temperature field of the welded joint, the welded joint surface temperature field measurement system needs to be calibrated. The data extracted from the calibration experiment are as follows:

[0041]

[0042] The accuracy of the revised welding joint surface temperature field measurement method was subsequently analyzed. The specific process is as follows: the temperature of the blackbody radiation source is set to 1050℃, 1150℃, 1250℃, 1350℃, 1450℃, and 1550℃, respectively. After extracting the R and G values, the corresponding K value is calculated according to the linear interpolation algorithm, and then substituted into formula (2) to calculate the corresponding temperature value. Finally, the calculated temperature value is compared with the actual temperature value to obtain the temperature measurement error. All temperature measurement error analysis results are as follows:

[0043]

[0044] According to the results of temperature measurement error analysis, the accuracy of the corrected welding joint surface temperature field measurement method is better than 1.80%, and it can be applied to the online measurement of the welding joint surface temperature field in the CMT process.

[0045] Subsequently, a frame of welding joint image is selected in each CMT arc extinction stage, and the image is sent to the computing processing system and processed using the modified welding joint surface temperature field measurement algorithm to obtain the welding joint surface temperature field distribution. Figure 2 As shown in the figure, the visual images of the welded joints at different penetration states and the corresponding temperature field images obtained by measurement, the temperature unit shown in the figure is K. On this basis, the surface temperature field distribution characteristics of the welded joints are extracted and the prediction model of the welded joint penetration state is established, as shown in the figure. Figure 3 As shown, it is a schematic diagram of the process of establishing the welding joint penetration state prediction model based on the time-series temperature distribution segmented attention of the present invention. In the surface temperature field image of the welding joint, the statistically obtained pixel point ratios in different temperature intervals are used as the temperature field distribution features, and the extracted time-series welding joint surface temperature field distribution features are feature fused and input into the welding joint penetration state prediction model, so as to output the welding joint penetration state at the current moment. For 6 groups of welding experiments, 498 groups of data were obtained in each group of experiments for the establishment and testing of the welding joint penetration state prediction model, that is, a total of 996 groups of unpenetrated state data, 996 groups of penetrated state data, and 996 groups of over-penetrated state data were obtained. In each group of experiments, the first 398 groups of data were selected for the training of the welding joint penetration state prediction model, and the subsequent 100 groups of data were used for the testing of the welding joint penetration state prediction model. All data were normalized and used for the training and testing of the welding joint penetration state prediction model. The prediction results of the welding joint penetration state prediction model are shown in the figure. Figure 4 As shown in the figure, the overall prediction accuracy of the three penetration states reaches 98.83%.

[0046] Finally, the generalization ability of the established weld joint penetration state prediction model was verified by changing the size and structure of the welding parent material. The size of the welding parent material was designed to be 18cm×6cm×5mm, the blunt edge was retained to be 1mm, and the groove angle of the parent material was processed to 30°-45°-60° at equal distances along the welding direction. The specific structure of the welding parent material is as follows: Figure 5 As shown in the figure, the arc starting point of welding is set to 1 cm away from the edge of the base material, and the welding length is set to 16 cm. The welding process parameters are set as follows: welding current 135A, welding voltage 14.8V, welding speed 4mm / s. After completing the welding experiment, the front and back images of the welded joint are as follows Figure 6 As shown, it can be seen that the front 5cm long weld corresponding to the groove angle of 30° is in an unpenetrated state, the middle 6cm long weld corresponding to the groove angle of 45° is in a fully penetrated state, and the last 5cm long weld corresponding to the groove angle of 60° is in an over-penetrated state. The generalization ability verification experiment obtained a total of 1500 sets of data, including 500 sets of unpenetrated state data, 500 sets of fully penetrated state data, and 500 sets of over-penetrated state data. The first 400 sets of data in the unpenetrated state, the first 400 sets of data in the fully penetrated state, and the first 400 sets of data in the over-penetrated state were selected for training the welding joint penetration state prediction model, and the subsequent 100 sets of unpenetrated state data, 100 sets of fully penetrated state data, and 100 sets of over-penetrated state data were used to test the welding joint penetration state prediction model. The prediction results of the welding joint penetration state prediction model are shown in the figure. Figure 7 As shown, the overall prediction accuracy of the three penetration states reaches 98.33%. It can be seen that the welding joint penetration state detection method based on time series temperature distribution segmentation attention of the present invention has high detection accuracy and strong generalization ability.

Claims

1. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution, characterized in that: The following steps are involved: Step 1: Use color CCD to build a welding joint surface temperature field measurement system. Based on the welding joint surface temperature field measurement algorithm corrected by color CCD colorimetric temperature measurement, high-precision welding joint surface temperature field distribution is measured. Step 2: Based on the temperature distribution segmentation attention, the temperature field distribution characteristics of the welding joint surface are extracted; Step 3: Taking the extracted time-series welding joint surface temperature field distribution characteristics as input, a welding joint penetration state prediction model is established based on artificial neural network, and the welding joint penetration state at the current moment is output in real time.

2. According to claim 1, a method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution, characterized in that: In step 1, the color CCD is fixed on the welding gun, and the red and green channel output images of the color CCD are used for colorimetric temperature measurement. The original colorimetric temperature measurement formula is: In the formula, the K value is determined by the temperature measurement system parameters and needs to be obtained through a blackbody calibration experiment. C2 is the second radiation constant, and λ r , g are the central wavelengths of the red and green channels respectively, and R and G are the grayscale values ​​of the red and green channel output images respectively.

3. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 2, characterized in that: The temperature range of the blackbody selected in the blackbody calibration experiment is 700℃~1600℃. In the blackbody calibration experiment, the input end of the color CCD is aligned with the opening of the blackbody radiation source, and the aperture of the optical lens is set to F16.

4. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 2, characterized in that: The blackbody calibration experiment process is as follows: first, the temperature of the blackbody radiation source is raised to 1000℃, the temperature measurement system collects the blackbody radiation image, and then separates the red and green channels to output the blackbody radiation image, respectively extracts the grayscale values ​​R and G of the two channel output images, and obtains the average grayscale value of the middle area of ​​the radiation image. Finally, the corresponding K value is calculated according to formula (1); Subsequently, the temperature rise step of the blackbody radiation source is set to 100°C, and calibration is performed according to the above steps every time a temperature point is reached until the temperature of the blackbody radiation source reaches 1600°C. In the process of gradual increase in the temperature of the blackbody radiation source, the integration time is changed to avoid the grayscale value of the color CCD output image from reaching saturation.

5. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 4, characterized in that: The original colorimetric temperature measurement formula was modified, and the K value was regarded as a function of R / G. For the uncalibrated temperature points, the corresponding K value was obtained using the linear interpolation method. The modified calculation formula for the surface temperature field of the welding joint is: C2 is the second radiation constant, λ r , g are the central wavelengths of the red and green channels respectively, and R and G are the grayscale values ​​of the red and green channel output images respectively.

6. The method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 1, characterized in that: In step 2, based on the measured surface temperature field image of the welding joint, the distribution range of the surface temperature of the welding joint is divided into multiple intervals, the proportion of pixel points in each temperature interval is counted, and the proportion of pixel points in each temperature interval is connected in series to form the distribution characteristics of the surface temperature field of the welding joint.

7. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 6, characterized in that: In the surface temperature field image of the welding joint, the proportion of pixels with temperature less than 1200K, the proportion of pixels with temperature between 1200K and 1400K, the proportion of pixels with temperature between 1400K and 1600K, the proportion of pixels with temperature between 1600K and 1800K, the proportion of pixels with temperature between 1800K and 2000K, the proportion of pixels with temperature between 2000K and 2200K, and the proportion of pixels with temperature above 2200K constitute the distribution characteristics of the surface temperature field of the welding joint [f1, f2, f3, f4, f5, f6, f7].

8. The method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 7, characterized in that: In step three, the extracted surface temperature field distribution characteristics of the welding joint corresponding to the current frame and its two historical frames are taken as input, and the welding joint penetration state corresponding to the current frame is taken as output, and a back propagation neural network is used to establish a welding joint penetration state prediction model; the input of the welding joint penetration state prediction model is: [f1(ti), f2(ti), f3(ti), f4(ti), f5(ti), f6(ti), f7(ti)], where i = 0, 1, 2, and the output of the prediction network model is determined as: P(t).

9. A method for detecting penetration state of welding joints based on segmented attention of temporal temperature distribution according to claim 8, characterized in that: The hidden layer of the prediction network model is set to two layers, each hidden layer is set to 12 neurons, and the structure of the established welding joint penetration state prediction model is 21-12-12-1.

Images