Welded structural part monitoring method and system based on vibration acoustic method

Abstract

The invention provides a welding structural part monitoring method and system based on a vibration acoustic method, and relates to the technical field of welding monitoring, and the method comprises the steps: constructing a defect classification model and a defect strategy model, employing an ultrasonic phased array to detect a to-be-detected welding line of a welding structural part, obtaining a first image, inputting the first image into the defect classification model, and obtaining a second image; a second image is obtained, whether welding defects exist in the second image or not is judged, and if not, the welding quality evaluation result of the to-be-detected weld joint is set to be optimal and output; and if yes, inputting the second image into the defect strategy model to obtain a defect repairing strategy of the second image. According to the method, the defects of the welding structural part are monitored in real time, the welding defects are classified on the basis of the defect classification model, then the welding defects are evaluated, the targeted repairing strategy is obtained through the defect strategy model on the basis of the evaluation result, and the purpose that the welding technology is adjusted in time for repairing on the basis of the welding defects occurring in the welding process is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of welding monitoring, and in particular to a method and system for monitoring welding structures based on a vibroacoustic method. Background Art

[0002] Welding technology plays a vital role in modern industrial production and is widely used in many fields such as automobiles, aerospace, shipbuilding, construction, and energy. During the welding process, common welding defects include cracks, pores, slag inclusions, lack of fusion, and welds that are too large or too small. Traditional welding quality monitoring methods such as ultrasonic, X-ray, and magnetic particle monitoring often focus on post-inspection of finished welded structural parts. Although these methods can detect defects in welded structural parts, they are based on monitoring of completed welded structural parts, so there is a certain time lag. If the problem is not discovered in time during the welding process, the finished welded structural parts with welding defects produced by welding technology will continue to accumulate, which not only increases production costs but also affects welding quality and welding production efficiency.

[0003] The Chinese invention patent with application publication number CN118533972A provides an ultrasonic-based weld intelligent detection system, which sends an ultrasonic signal to the weld of the welded part to be detected through an ultrasonic sensor, and records the ultrasonic feedback information as an ultrasonic acquisition information group; preprocessing is performed based on the ultrasonic acquisition information group to obtain a weld feature data set and an ultrasonic feedback signal feature data set; the ultrasonic feedback signal is extracted from the ultrasonic acquisition information group, and the extracted ultrasonic feedback signal is converted into a digital signal to form an ultrasonic signal digital set, and the difference information of the ultrasonic signal digital set is synchronously counted to obtain a digital signal feedback coefficient Szxs, and then the weld feature data set and the ultrasonic feedback signal feature data set are normalized to form a first data set and a second data set; a weld feature extraction model is established for the first data set and the second data set, and training and extraction are performed, and then fitting is performed with the digital signal feedback coefficient Szxs to obtain a weld feature adjustment index Tzzs; a preset weld evaluation adjustment threshold T is compared with the weld feature adjustment index Tzzs to obtain a weld detection adjustment evaluation strategy scheme, and specific execution and notification are performed according to the content of the weld detection adjustment evaluation strategy scheme.

[0004] With respect to the above technical solution, a weld adjustment plan is obtained based on the weld characteristics of the ultrasonic signal. However, a linkage adjustment mechanism between the welding process and the welding defects has not been established. The welding process cannot be adjusted in time according to the monitored welding defects for repair, thereby affecting the stability and long-term reliability of the welding quality of welded structural parts. Summary of the invention

[0005] In order to timely adjust the welding process and perform repairs according to the monitored welding defects, the present application provides a welding structure monitoring method and system based on a vibroacoustic method.

[0006] In the first aspect, the present application provides a method for monitoring a welded structure based on a vibroacoustic method, which adopts the following technical solution: A method for monitoring a welded structural part based on a vibroacoustic method comprises the following steps: Data collection: Collect historical welding process parameters, historical welding defect images and corresponding defect category labels, and defect repair strategies; Model building: building defect classification model and defect strategy model; Model training: The defect classification model is trained using historical welding process parameters, historical welding defect images and corresponding defect category labels to obtain a trained defect classification model. The defect strategy model is trained using historical welding process parameters, historical welding defect images and defect repair strategies to obtain a trained defect strategy model. First monitoring: using an ultrasonic phased array to detect the weld to be detected of the welded structural component to obtain a first image; First judgment: obtain the current welding process parameters, input the current welding process parameters and the first image into the trained defect classification model, obtain the first image marked with the defect category label, record it as the second image, and judge whether there is a welding defect in the second image: If not, then execute the first setting step; If yes, then execute the strategy acquisition step; First setting: setting the welding quality assessment result of the weld to be inspected to be excellent, and outputting the welding quality assessment result; Strategy acquisition: The current welding process parameters and the second image are input into the trained defect strategy model to obtain the defect repair strategy of the second image.

[0007] By adopting the above technical solution, ultrasonic phased array technology is used to monitor the defects of welded structural parts in real time, and defects are identified and classified based on the current welding process parameters and the detected images, thereby improving the efficiency and accuracy of welding defect detection and evaluation. In addition, based on the evaluation results of welding defects and the current welding process parameters, a targeted repair strategy is obtained by inputting them into the defect repair strategy model, achieving the technical effect of adjusting the welding process according to the defect type, which helps to quickly identify welding defects that occur during the welding process and adjust the welding process in time according to the situation of the welding defects, achieving the purpose of efficiently and accurately repairing welding defects.

[0008] Optionally, after performing the first monitoring step and before performing the first distinguishing step, the method further includes: First collection: collect the current ambient temperature and ambient humidity; obtain the standard temperature and standard humidity; First calculation: calculate the difference between the ambient temperature and the standard temperature, recorded as the first difference, calculate the difference between the ambient humidity and the standard humidity, recorded as the second difference; Speed ​​adjustment: obtaining the standard ultrasonic propagation speed, calculating the ultrasonic propagation speed adjustment value of the ultrasonic phased array based on the environmental compensation formula, and correcting the ultrasonic propagation speed of the ultrasonic phased array based on the calculated ultrasonic propagation speed adjustment value; The calculation model of the environmental compensation formula is: ; in, is the ultrasonic propagation speed adjustment value, is the standard ultrasonic propagation speed, is the influence coefficient of temperature on ultrasonic propagation velocity, is the first difference, is the influence coefficient of humidity on ultrasonic propagation velocity, is the second difference, is the comprehensive influence coefficient of temperature and humidity; Second monitoring: based on the corrected ultrasonic wave propagation velocity, the first image is corrected to obtain a corrected first image, and the corrected first image is updated as the first image.

[0009] By adopting the above technical solution, the ultrasonic propagation speed is adjusted based on the environmental compensation mechanism, which improves the accuracy of the ultrasonic detection results, making the welding defect detection more stable and reliable and not affected by environmental changes. At the same time, based on the corrected ultrasonic propagation speed and the corrected first image, a more accurate image basis is provided for subsequent defect classification and defect evaluation, thereby optimizing the subsequent defect repair strategy.

[0010] Optionally, after performing the first monitoring step and before performing the first distinguishing step, the method further includes: First construction: build a defect recognition model; Collect samples: collect images containing complete welding defects and corresponding complete labels, which are recorded as the first sample; collect images containing incomplete welding defects and corresponding incomplete labels, which are recorded as the second sample; the integrity labels of welding defects include complete labels and incomplete labels; First training: using the first sample and the second sample to train the defect recognition model to obtain a trained defect recognition model; Image segmentation: extracting welding defects from the first image, and segmenting the first image based on the extracted welding defects to obtain a number of sub-images containing welding defects; Defect monitoring: Input the sub-image containing welding defects into the trained defect recognition model, and output the sub-image annotated with integrity labels; Defect screening: Screen out sub-images with incomplete labels from sub-images with complete labels and record them as incomplete images; Defect filling: Using fractal image generation technology, the incomplete welding defects in the incomplete image are filled based on the preset fractal dimension to obtain the complete welding defects, and the filled incomplete image is recorded as the filled image; Parameter adjustment: extract features of welding defects in the image, adjust parameters of the ultrasonic phased array probe based on the extracted features, and use the ultrasonic phased array with adjusted parameters to perform the first monitoring step.

[0011] By adopting the above technical solution, incomplete defects are identified, and defects in the incomplete image are filled by using fractal image generation technology. The ultrasonic probe parameters are adjusted by extracting the welding defect features after filling and the welding defect features before filling. The accuracy and sensitivity of the ultrasonic wave in the detection process are improved based on the adjusted ultrasonic probe parameters, making the detection of welding defects more accurate, providing a more accurate image basis for subsequent defect classification and defect evaluation, optimizing the subsequent defect repair strategy, and further improving the welding quality of welding defect repair. In addition, adjusting the ultrasonic probe parameters based on filling the defects in the incomplete image helps to reduce the number of adjustments to the ultrasonic probe parameters, more accurately locate the ultrasonic probe parameters, and improve the detection efficiency of welding defects.

[0012] Optionally, after executing the defect repair step and before executing the parameter adjustment step, the method further includes: First verification: The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the incomplete welding defects in the incomplete image, which is recorded as the first matrix. The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the remaining part containing defects in the image, which is recorded as the second matrix. Second verification: calculate the contrast of the first matrix, recorded as the first contrast, calculate the contrast of the second matrix, recorded as the second contrast, calculate the contrast difference between the first contrast and the second contrast, recorded as the third difference, and determine whether the third difference is less than the preset contrast difference threshold: If yes, then execute the parameter adjustment step; If not, the preset fractal dimension is adjusted and the defect filling step is performed.

[0013] By adopting the above technical solution, the grayscale co-occurrence matrix method is used to analyze the spatial relationship between the pixel grayscale values ​​of the welding defect image, identify the structural and texture differences between the filled image and the original image, and calculate the contrast of the grayscale co-occurrence matrix of the image before and after filling, so as to quantify the filling effect and verify the filled image, thereby improving the consistency and accuracy of defect filling. In addition, by dynamically verifying and adjusting the preset fractal dimension, the filled defect image is more consistent with the characteristics of the defect before filling, which helps to improve the accuracy and sensitivity of the ultrasonic wave in the detection process based on the filled defect image, making the detection of welding defects more accurate, providing a more accurate image basis for subsequent defect classification and defect evaluation, optimizing the subsequent defect repair strategy, and further improving the welding quality of welding defect repair.

[0014] Optionally, the parameter adjustment step further includes: First processing: in the filled image, the incomplete welding defect before filling is recorded as the original part, and the remaining part containing the defect in the filled image is recorded as the filled part; Construct coordinates: Take any pixel in the filled image as the origin, the length direction of the filled image as the x-axis, and the width direction of the filled image as the y-axis to construct a coordinate system; Second calculation: using an edge detection method to obtain coordinates of boundary points of the original part, calculating the center coordinates of the original part based on the coordinates of the boundary points of the original part, and marking the calculated center coordinates of the original part as the first center; Third calculation: using edge detection method to obtain the coordinates of the boundary points of the complete welding defect obtained in the filled image, based on the coordinates of the boundary points of the complete welding defect, calculating the center coordinates of the defect, and marking the calculated center coordinates of the defect as the second center; Angle calculation: record the vector between the origin and the first center as the first vector, calculate the angle between the direction of the first vector and the positive direction of the x-axis, record it as the first angle, record the vector between the origin and the second center as the second vector, calculate the angle between the direction of the second vector and the positive direction of the x-axis, record it as the second angle; First adjustment: obtaining an initial angle of the ultrasonic phased array probe, calculating an adjustment angle through proportional operation based on the initial angle, the first angle and the second angle, and using the adjustment angle as the adjusted angle of the ultrasonic phased array probe.

[0015] By adopting the above technical scheme, the center coordinates of the welding defect before filling and the center coordinates of the complete welding defect after filling are calculated, and the corresponding vectors are constructed based on the center coordinates of the welding defects before and after filling. The center angle difference of the defects before and after filling is calculated by the vectors of the welding defects before and after filling, and the angle of the ultrasonic phased array probe is calculated based on the angle difference, so as to improve the accuracy and detection range of ultrasonic waves in the detection process, better adapt to welding defects in different directions or shapes, and improve the reliability of ultrasonic detection.

[0016] Optionally, after performing the first adjustment step, the method further includes: Fourth calculation: based on the second angle, in the filled image, the second center is updated as the origin of the coordinate system, the direction of the second vector is used as the new positive direction of the x-axis, and a new y-axis and a new z-axis are constructed based on the new x-axis; First projection: Project the boundary points of the complete welding defect onto the new x-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new x-axis and the coordinate value of the projection point with the minimum coordinate value on the new x-axis, and record the obtained difference as the first length; Second projection: Project the boundary points of the complete welding defect onto the new y-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new y-axis and the coordinate value of the projection point with the minimum coordinate value on the new y-axis, and record the obtained difference as the first width; Third projection: Project the boundary points of the complete welding defect onto the new z-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new z-axis and the coordinate value of the projection point with the minimum coordinate value on the new z-axis, and record the obtained difference as the first height; Second adjustment: performing a product operation on the first length, the first width, and the first height to obtain a first volume, acquiring an initial frequency and a preset standard volume of the ultrasonic phased array probe, calculating a first frequency according to a frequency formula based on the initial frequency, the preset standard volume, and the first volume, and using the calculated first frequency as the adjusted frequency of the ultrasonic phased array probe; The calculation model of the frequency formula is: ; in, is the first frequency, is the initial frequency of the ultrasonic phased array probe, To preset the standard volume, is the first volume.

[0017] By adopting the above technical solution, a coordinate system is constructed based on the geometric center of the filled welding defect as the origin. Based on the filled welding defect, the maximum value of the projection of the filled welding defect on different coordinate axes is calculated in the constructed coordinate system. The volume of the repaired welding defect is calculated based on the maximum value of the projection on different coordinate axes. The frequency of the ultrasonic phased array probe is adjusted based on the calculated volume, so that the frequency of the ultrasonic probe is better adapted to the actual size of the defect, ensuring the accuracy and efficiency of ultrasonic detection. Especially in the case of complex welding structures and various types of defects, the above dynamic adjustment mechanism improves the adaptability of ultrasonic detection and reduces the possibility of false alarms and missed alarms of welding defects.

[0018] Optionally, after performing the first determination step and before performing the strategy acquisition step, the method further includes: First extraction: extracting the geometric features of the welding defects in the second image, recorded as the first features; Second extraction: extracting the size feature of the welding defect in the second image, recorded as the second feature; The third extraction: obtaining the position data of the boundary points of the key load-bearing area of ​​the weld to be monitored, extracting the position data of the boundary points of the area corresponding to the welding defect in the second image, calculating the Euclidean distance between the boundary points of the area corresponding to the welding defect in the second image and the boundary points of the key load-bearing area, and recording the smallest Euclidean distance as the third feature; Second processing: encoding the first feature, the second feature and the third feature of the welding defect to obtain the features after encoding, and performing weighted summation on the features after encoding to obtain the first data of the welding defect in the second image; Second judgment: judge whether the first data of welding defects is not higher than the preset parameter threshold: If yes, then execute the first setting step; If not, execute the strategy acquisition step.

[0019] By adopting the above technical scheme, by extracting the geometric features, dimensional features and Euclidean distances between defects and key load-bearing areas of welding defects, a comprehensive quantitative analysis of welding defects can be performed. Based on the quantitative analysis, the evaluation steps of welding defects are optimized, and the accuracy of welding quality evaluation is improved, so that welds with welding defects but meeting the corresponding quality standards do not need to perform defect repair steps, thereby achieving the purpose of saving repair costs and time, avoiding unnecessary repair work, and improving the timeliness and efficiency of adjusting the welding process to repair welding defects.

[0020] Optionally, after performing the second determination step and before performing the first setting step, the method further includes: Third judgment: Determine whether the number of defect category labels of the second image is 1: If yes, then execute the first setting step; If not, the third processing step is executed; The third processing: segmenting the second image based on the defect category label to obtain a plurality of welding defect images, and integrating the plurality of welding defect images into a third image set; Fifth calculation: using the box counting method to calculate the fractal dimension of each welding defect image in the third image set, obtaining a number of fractal dimensions, and collecting them to obtain the first dimension set; First judgment: judge whether each fractal dimension in the first dimension set is lower than the preset dimension threshold: If yes, then execute the first setting step; If not, execute the strategy acquisition step.

[0021] By adopting the above technical solution, the welding defect assessment is further optimized, and the welding defects are classified and processed based on the number of welding defects, and the single defect and multiple defects are distinguished, which is helpful for the refined assessment of welding defects based on welds of different complexity. The box counting method is used to calculate the fractal dimension of the welding defect image, which provides more refined feature data for each welding defect, which helps to distinguish defects of different types and forms and identify the nature and complexity of each welding defect, improves the accuracy of welding defect assessment, and makes it unnecessary to perform defect repair steps for welds with welding defects but meeting the corresponding quality standards, thereby achieving the purpose of saving repair costs and time, avoiding unnecessary repair work, and improving the timeliness and efficiency of adjusting the welding process to repair welding defects.

[0022] Optionally, after the step of executing the policy acquisition, the method further includes: The third monitoring: executing the defect repair strategy of the obtained second image, recording the weld after the defect repair strategy is executed as the first weld, using the ultrasonic phased array to detect the first weld of the welded structural part, and obtaining a first verification image; obtaining the welding process parameters after the defect repair strategy is executed, and recording them as the first welding process parameters; Second judgment: Input the first verification image and the first welding process parameter into the trained defect classification model to obtain the first verification image marked with the defect category label, recorded as the second verification image, and determine whether the second verification image has welding defects: If not, then execute the first setting step; If yes, the quality determination step is performed; Quality judgment: Take the second verification image as the new second image, perform the first extraction to the second processing steps in sequence, obtain the first data of the welding defect in the new second image, and judge whether the first data of the welding defect in the new second image is not higher than the preset parameter threshold: If yes, then execute the third judgment step; If not, then execute the model update step; Third judgment: counting the number of defect category labels in the second verification image, recorded as a first number, counting the number of defect category labels in the second image, recorded as a second number, and determining whether the first number is less than the second number; If yes, then execute the first setting step; If not, then execute the model update step; Model update: The second image, the corresponding defect repair strategy and the first welding process parameters are used as negative sample data of the defect strategy model, and the model is trained according to the negative sample data to obtain a new trained defect strategy model, and then the strategy acquisition step is re-executed.

[0023] By adopting the above technical solution, the defects of the repaired welds are re-identified and detected, and the quality of the repaired welds is evaluated to determine whether the acquired repair strategy can repair the welding defects. If the defect repair does not meet the standards, the defect repair strategy and welding process parameters are input as negative sample data into the defect strategy model for model update, so that the defect repair strategy model is continuously adjusted according to the actual situation to provide a more accurate repair strategy, reduce the risk of repair failure, and improve the accuracy and stability of the welding process in repairing welding defects.

[0024] In the second aspect, the present application provides a welding structure monitoring system based on a vibroacoustic method, which adopts the following technical solution: A welding structure monitoring system based on a vibroacoustic method, the system comprising: Processor and memory; The memory stores program code; When the processor calls the program code in the memory, the steps of the method described in the first aspect are executed.

[0025] By adopting the above technical solution, the defects of welded structural parts are monitored in real time and the defects are identified and classified, which improves the efficiency and accuracy of welding defect detection and evaluation. In addition, based on the evaluation results of welding defects and the current welding process parameters, the targeted repair strategy is obtained by inputting them into the defect repair strategy model, which achieves the technical effect of adjusting the welding process according to the defect type, helps to quickly identify welding defects that occur during the welding process and adjust the welding process in time according to the situation of the welding defects, achieving the purpose of efficiently and accurately repairing welding defects.

[0026] In summary, the present application includes at least one of the following beneficial technical effects: 1. Real-time monitoring of defects in welded structural parts, defect identification and classification, improves the efficiency and accuracy of welding defect detection and evaluation. In addition, based on the evaluation results of welding defects and the current welding process parameters, the defect repair strategy model is input to obtain a targeted repair strategy, achieving the technical effect of adjusting the welding process according to the defect type, which helps to quickly identify welding defects that occur during the welding process and adjust the welding process in time according to the situation of the welding defects, achieving the purpose of efficiently and accurately repairing welding defects.

[0027] 2. Identify incomplete defects, use fractal image generation technology to fill the defects in the incomplete image, and verify the filled image. Based on the verified image, extract the welding defect characteristics after filling and the welding defect characteristics before filling to adjust the angle and frequency of the ultrasonic probe to improve the accuracy and detection range of ultrasonic waves during the detection process, which helps the ultrasonic phased array probe to better adapt to welding defects in different directions or shapes, improves the reliability of ultrasonic detection, makes the detection of welding defects more accurate, and provides a more accurate image basis for subsequent defect classification and defect evaluation, optimizes the subsequent defect repair strategy, and further improves the welding quality of welding defect repair.

[0028] 3. By extracting the geometric features, dimensional features and Euclidean distances between defects and key load-bearing areas of welding defects, a comprehensive quantitative analysis of welding defects can be performed. Based on the quantitative analysis, the evaluation steps of welding defects are optimized, and the accuracy of welding quality evaluation is improved. This makes it unnecessary to perform defect repair steps for welds that have welding defects but meet the corresponding quality standards, thereby saving repair costs and time, avoiding unnecessary repair work, and improving the timeliness and efficiency of adjusting the welding process to repair welding defects.

[0029] 4. Re-identify and detect defects on the repaired welds, conduct quality assessment on the repaired welds, and determine whether the acquired repair strategy can repair the welding defects. If the defect repair does not meet the standards, the defect repair strategy and welding process parameters are input as negative sample data into the defect strategy model for model update, so that the defect repair strategy model is continuously adjusted according to actual conditions to provide more accurate repair strategies, reduce the risk of repair failure, and improve the accuracy and stability of welding process in repairing welding defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 is a flowchart of S41 image correction in Example 2 of the present application; Figure 3 is a flow chart of adjusting parameters in S42 of Example 3 of the present application; Figure 4 is a flow chart of Example 4 of the present application; Figure 5 This is a flow chart of Example 5 of the present application. DETAILED DESCRIPTION

[0031] The following combination Figures 1 to 5 This application is described in further detail.

[0032] Embodiment 1: This embodiment discloses a method for monitoring a welded structure based on a vibroacoustic method, such as Figure 1 As shown, the method includes: collecting historical welding process parameters, historical welding defect images and corresponding defect category labels, defect repair strategies, constructing a defect classification model and a defect strategy model, training the defect classification model and the defect strategy model based on the collected data, obtaining a trained defect classification model and a trained defect strategy model, using an ultrasonic phased array to detect the weld to be detected of the welded structural part, obtaining a first image, obtaining current welding process parameters, inputting the current welding process parameters and the first image into the trained defect classification model, obtaining a second image, and determining whether there is a welding defect in the second image: if not, setting the welding quality assessment result of the weld to be detected to excellent, and outputting the welding quality assessment result; if so, inputting the current welding process parameters and the second image into the trained defect strategy model, and obtaining a defect repair strategy for the second image. This embodiment includes the following steps: S1 Data Collection: Collect historical welding process parameters, historical welding defect images, and corresponding defect category labels and defect repair strategies.

[0033] The historical welding process parameters include welding current, welding voltage, welding machine speed, heat input, wire diameter, welding position, shielding gas type, shielding gas flow, welding electrode, welding material, welding preheating temperature, post-heat treatment temperature, welding gap, welding time, ambient temperature and ambient humidity.

[0034] The historical welding defect image is a three-dimensional image of a weld welded under historical welding process parameters collected by ultrasonic phased array technology.

[0035] The welding defects include pores, slag inclusions, cracks, lack of fusion and lack of penetration. Porosity includes single pore, pores in a series, needle-shaped pores and root pores. Slag inclusions include internal slag inclusions, root slag inclusions, interlayer slag inclusions and groove slag inclusions. Cracks include hot cracks, cold cracks and shrinkage cracks. Lack of fusion includes root lack of fusion and side wall lack of fusion. Lack of penetration includes partial lack of penetration and total lack of penetration. Defect category labels include pores-single pores, pores-pores in a series, pores-needle-shaped pores, pores-root pores, slag inclusions-internal slag inclusions, slag inclusions-root slag inclusions, slag inclusions-interlayer slag inclusions, slag inclusions-groove slag inclusions, cracks-hot cracks, cracks-cold cracks, cracks-shrinkage cracks, lack of fusion-root lack of fusion, lack of fusion-side wall lack of fusion, lack of penetration-partial lack of penetration and lack of penetration-total lack of penetration.

[0036] The defect repair strategy is a repair strategy that can solve the above-mentioned various types of welding defects.

[0037] In this embodiment, it is necessary to perform data preprocessing on the collected data, and the steps of data preprocessing include data cleaning and data normalization.

[0038] Data cleaning: De-noising, processing missing values, and removing outliers on the collected data.

[0039] Data normalization: Normalize the collected data.

[0040] S2 Model Construction: Use the deep convolutional neural network model as the basic model of the defect classification model to build the defect classification model. Models such as convolutional neural network, recurrent neural network, and long short-term memory network models can be used as the basic models of the defect strategy model to build the defect strategy model.

[0041] In this embodiment, a deep convolutional neural network model VGGNet is used to construct a defect classification model. The defect strategy model can be trained with historical welding process parameters, historical welding defect images, corresponding defect category labels, and defect repair strategies to output defect repair strategies.

[0042] S3 model training: The defect classification model is trained using historical welding process parameters, historical welding defect images and corresponding defect category labels to obtain a trained defect classification model. The defect strategy model is trained using historical welding process parameters, historical welding defect images, corresponding defect category labels and defect repair strategies to obtain a trained defect strategy model.

[0043] The model training process of the defect classification model includes: marking the defect category label on the corresponding historical welding defect image to obtain the first classification sample, marking the historical welding process parameters as parameter labels on the first classification sample to obtain the second classification sample, collecting the second classification samples to obtain the classification sample set, and using the classification sample set to train the defect classification model to obtain the trained defect classification model. The trained defect classification model has the function of outputting a welding defect image marked with a defect category label according to the input welding defect image and the corresponding welding process parameters.

[0044] The model training process of the defect strategy model includes: marking the defect category label on the corresponding historical welding defect image to obtain the first strategy sample, marking the historical welding process parameters as parameter labels on the first strategy sample to obtain the second strategy sample, collecting the second strategy sample and the corresponding defect repair strategy as a group of samples to obtain a strategy sample set, and using the strategy sample set to train the defect strategy model to obtain a trained defect strategy model. The trained defect strategy model has the function of outputting the defect repair strategy of the welding defect image according to the welding defect image and the corresponding defect category label and welding process parameters.

[0045] S4 First Monitoring: The ultrasonic phased array technology is used to scan the weld to be inspected of the welded structural component from different angles to obtain a three-dimensional inspection image of the weld to be inspected, which is recorded as the first image.

[0046] S5 first judgment: obtain the current welding process parameters, input the current welding process parameters and the first image into the trained defect classification model, obtain the first image marked with the defect category label, record it as the second image, and judge whether there is a welding defect in the second image by judging whether there is a defect category label in the second image.

[0047] If there is no defect category label in the second image, that is, there is no welding defect in the second image, then S6 first setting is performed.

[0048] If there is a defect category label in the second image, that is, there is a welding defect in the second image, then S7 strategy acquisition is executed.

[0049] S6 first setting: setting the welding quality evaluation result of the weld to be inspected to be excellent, and outputting the welding quality evaluation result. The welding quality evaluation result being excellent indicates that the weld to be inspected is qualified, and no welding defect repair operation is required.

[0050] S7 strategy acquisition: input the current welding process parameters and the second image into the trained defect strategy model to obtain the defect repair strategy of the second image, perform the defect repair operation according to the output defect repair strategy, and correct the repair defect.

[0051] In this embodiment, ultrasonic phased array technology is used to monitor the defects of welded structural parts in real time, and defects are identified and classified based on the current welding process parameters and the detected images, thereby improving the efficiency and accuracy of welding defect detection and evaluation. In addition, based on the evaluation results of welding defects and the current welding process parameters, a targeted repair strategy is obtained by inputting them into the defect repair strategy model, achieving the technical effect of adjusting the welding process according to the defect type, helping to quickly identify welding defects occurring during the welding process and timely adjust the welding process according to the situation of the welding defects, thereby achieving the purpose of efficiently and accurately repairing welding defects.

[0052] Embodiment 2: The difference from Embodiment 1 is that: like Figure 2 As shown, after executing S4 first monitoring and before executing S5 first determination, S41 image correction is also included, and S41 image correction includes S411 first acquisition, S412 first calculation, S413 speed adjustment and S414 second monitoring.

[0053] S411 First collection: collect the current ambient temperature and ambient humidity through the temperature sensor and the humidity sensor. Obtain the standard atmospheric conditions under the standard propagation conditions of ultrasonic waves in the air, the standard atmospheric conditions include standard temperature and standard humidity. In this embodiment, the standard temperature is 20° C. and the standard humidity is 50%.

[0054] S412 First calculation: Calculate the difference between the collected ambient temperature and the standard temperature, recorded as the first difference, calculate the difference between the collected ambient humidity and the standard humidity, recorded as the second difference.

[0055] S413 Speed ​​adjustment: Obtain the standard ultrasonic propagation speed under standard atmospheric conditions. In this embodiment, the standard ultrasonic propagation speed is 343.2 m / s. Based on the environmental compensation formula, calculate the ultrasonic propagation speed adjustment value of the ultrasonic phased array, and correct the ultrasonic propagation speed of the ultrasonic phased array based on the obtained ultrasonic propagation speed adjustment value, that is, add the ultrasonic propagation speed adjustment value to the standard ultrasonic propagation speed to obtain the actual ultrasonic propagation speed under the current ambient temperature and ambient humidity conditions.

[0056] The calculation model of the environmental compensation formula is: .

[0057] in, is the ultrasonic propagation speed adjustment value, is the standard ultrasonic propagation speed, is the influence coefficient of temperature on ultrasonic propagation velocity, is the first difference, is the influence coefficient of humidity on ultrasonic propagation velocity, is the second difference, is the comprehensive influence coefficient of temperature and humidity, that is, the comprehensive influence coefficient of temperature and humidity on the propagation velocity of ultrasonic wave. In this embodiment, the influence coefficient of temperature on the ultrasonic propagation velocity is The influence coefficient of humidity on the ultrasonic propagation velocity The comprehensive influence coefficient of temperature and humidity can be obtained through experimental testing and theoretical deduction based on experimental test data. The influence coefficient of temperature on the ultrasonic propagation velocity can be obtained by experimental testing and fitting based on the experimental test data by a multivariate linear regression method. The value is 0.6m / s∙℃, the influence coefficient of humidity on the propagation speed of ultrasonic waves The value is 0.01m / s, the comprehensive influence coefficient of temperature and humidity The value is 0.01m / s∙℃.

[0058] In this embodiment, the unit of humidity is %, and when performing calculations, the humidity expressed in percentage is converted into a decimal form to participate in the calculation of the environmental compensation formula.

[0059] S414 Second monitoring: Based on the corrected ultrasonic propagation velocity, obtain the ultrasonic propagation time before correction, and calculate the distance from each pixel point in the corrected first image to the probe based on the corrected ultrasonic propagation velocity and the ultrasonic propagation time before correction. Apply the distance from each pixel point in the corrected first image to the probe to the first image, update the position information of each point in the first image, use the updated first image as the corrected first image, and update the corrected first image to the first image.

[0060] In this embodiment, changes in temperature and humidity are compensated based on the environmental compensation mechanism, the ultrasonic propagation velocity is corrected, the precision, stability and accuracy of the ultrasonic detection results are improved, and the anti-interference ability of welding defect detection is improved. At the same time, based on the corrected ultrasonic propagation velocity and the corrected first image, a more accurate image basis is provided for subsequent defect classification and defect evaluation, thereby optimizing subsequent defect repair strategies.

[0061] Embodiment 3: The difference from Embodiment 1 is that: like Figure 3As shown, after executing S4 the first monitoring and before executing S5 the first discrimination, it also includes S42 parameter adjustment, and S42 parameter adjustment includes S421 first construction, S422 sample collection, S423 first training, S424 image segmentation, S425 defect monitoring, S426 defect screening, S427 defect filling, S428 filling verification and S429 parameter adjustment.

[0062] S421 First Construction: Use a deep convolutional neural network model as a basic model of a defect recognition model to construct a defect recognition model. In this embodiment, a deep convolutional neural network model VGGNet is used to construct a defect recognition model.

[0063] S422 Collect samples: collect images containing complete welding defects and corresponding complete labels, recorded as the first sample, collect images containing incomplete welding defects and corresponding incomplete labels, recorded as the second sample; the integrity labels of welding defects include complete labels and incomplete labels. For example: collect images containing complete bubbles, the complete label of the complete bubble image is "complete", that is, the first sample is a complete bubble image - complete, collect only The image of the bubble The incomplete label of the bubble image is "incomplete", that is, the second sample is Image of bubbles - incomplete.

[0064] S423 First training: collect the first sample and the second sample collected, record the collected samples as the recognition sample set, use the recognition sample set to train the defect recognition model, and obtain the trained defect recognition model. The trained defect recognition model has the function of outputting a welding defect image marked with an integrity label according to the input welding defect image.

[0065] S424 Image segmentation: extract welding defects in the first image, and segment the first image based on the extracted welding defects to obtain a plurality of sub-images containing welding defects.

[0066] The first image is subjected to image preprocessing, wherein the preprocessing includes removing noise in the first image by using a three-dimensional Gaussian filter; and enhancing the first image by using a histogram equalization method and a Canny edge detection algorithm.

[0067] Each welding defect in the first image is extracted by using a threshold segmentation, edge detection or image segmentation model, and the first image is segmented based on each extracted welding defect to obtain a plurality of sub-images containing only one welding defect. In this embodiment, a Canny edge detection algorithm is used to extract each welding defect in the first image.

[0068] S425 Defect monitoring: All the obtained sub-images containing only one welding defect are input into the trained defect recognition model respectively, and the corresponding sub-images marked with integrity labels are output.

[0069] S426 Defect screening: Screen out sub-images marked with incomplete labels from the obtained sub-images marked with completeness labels, record them as incomplete images, and collect all the screened incomplete images to obtain an incomplete image set.

[0070] S427 defect filling: using fractal image generation technology, based on the preset fractal dimension, the incomplete welding defects in the incomplete image are filled to obtain the complete welding defects, and the filled incomplete image is recorded as the filled image.

[0071] Based on the different defect categories of welding defects in each incomplete image in the incomplete image set, a different fractal dimension range is set for each welding defect. For example, the fractal dimension range set for porosity defects is [1.5, 2.0], the fractal dimension range set for slag inclusion defects is [2.0, 2.5], the fractal dimension range set for crack defects is [1.7, 2.2], the fractal dimension range set for lack of fusion defects is [2.0, 2.5], and the fractal dimension range set for lack of penetration defects is [2.0, 2.5]. Based on the fractal dimensions of different welding defects, the fractal dimension of the welding defect in each incomplete image is compared with the preset filling dimension threshold. For welding defects whose fractal dimension is greater than or equal to the preset filling dimension threshold, a filling area that matches the surrounding area of ​​the welding defect is generated by fractal noise interpolation, fractal model interpolation or multi-scale interpolation, and the welding defect is filled; for welding defects whose fractal dimension is less than the preset filling dimension threshold, a filling area that matches the surrounding area of ​​the welding defect is generated by linear interpolation or mean interpolation, and the welding defect is filled. The filling area is smoothed to eliminate the phenomenon that the boundary of the filling area is too obvious, and a complete welding defect is obtained. The incomplete image after filling is recorded as the filling image.

[0072] In order to more finely fill the welding defects in the incomplete image, in other embodiments, the fractal dimension classification processing can be performed based on the structure of the welding defects. For welding defects with simple structures, the defects are filled according to the preset fractal dimensions. For welding defects with complex structures, the fractal dimensions of the welding defects can be calculated by the box counting method and other methods to fill the defects.

[0073] S428 filling verification: includes S4281 first verification and S4282 second verification.

[0074] S4281 First verification: The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the incomplete welding defects in the incomplete image, which is recorded as the first matrix. The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the remaining part containing defects in the image, which is recorded as the second matrix.

[0075] The gray level of the incomplete welding defect in the incomplete image is set to L discrete values, and 8 directions are set, with the horizontal direction from left to right as 0. o , rotate clockwise, the 8 directions are 0 o , 45 o , 90 o , 135 o , 22.5 o , 67.5 o , 112.5 o and 157.5 o . Set the distance between pixel pairs. In this embodiment, the distance between pixel pairs is set to 1 pixel. In the incomplete welding defect, the gray level combination between each pixel and its adjacent pixels in 8 directions is counted. The probability of occurrence of each gray level combination in the incomplete welding defect is counted based on different directions. Based on the statistical results, fill in the L×L co-occurrence matrix to obtain the co-occurrence matrix in each direction. The co-occurrence matrix in each direction is recorded as the first matrix. According to the above steps, the gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the remaining defect-containing part in the image to obtain the second matrix.

[0076] S4282 Second verification: Calculate the contrast of each co-occurrence matrix in the first matrix according to the contrast calculation formula, calculate the contrast of each co-occurrence matrix in the first matrix by using the arithmetic average method or the weighted average method, and record the calculation result as the first contrast, calculate the contrast of each co-occurrence matrix in the second matrix according to the contrast calculation formula, calculate the contrast of each co-occurrence matrix in the second matrix by using the arithmetic average method or the weighted average method, and record the calculation result as the second contrast, calculate the contrast difference between the first contrast and the second contrast, record it as the third difference, and judge whether the third difference is less than the preset contrast difference threshold: If yes, execute S429 parameter adjustment; If not, then another preset fractal dimension is selected in the fractal dimension range of the corresponding welding defect, and based on the selected preset fractal dimension, the defect filling step S427 is performed again.

[0077] In this embodiment, the first contrast and the second contrast are calculated using an arithmetic mean method.

[0078] The contrast ratio is calculated as: .

[0079] in, represents the first Row gray level, represents the first Column gray level, Indicates grayscale and grayscale The probability of occurrence of a gray level combination.

[0080] S429 parameter adjustment: extract the features of welding defects in the image, adjust the parameters of the ultrasonic phased array probe based on the extracted features, and use the ultrasonic phased array with adjusted parameters to perform S4 first monitoring. S429 parameter adjustment includes S4291 angle adjustment and S4292 frequency adjustment.

[0081] S4291 Angle adjustment: includes S42911 first processing, S42912 building coordinates, S42913 second calculation, S42914 third calculation, S42915 angle calculation and S42916 first adjustment.

[0082] S42911 First processing: In the filled image, the incomplete welding defect before filling is recorded as the original part, and the remaining part containing the defect in the filled image is recorded as the filled part, that is, the filled area in S427 defect filling.

[0083] S42912 constructs coordinates: a coordinate system is constructed with any pixel point in the filled image as the origin, the length direction of the filled image as the x-axis, the width direction of the filled image as the y-axis, and the depth direction of the filled image as the z-axis.

[0084] S42913 Second calculation: use edge detection method to obtain the coordinates of the boundary points of the original part, calculate the center coordinates of the original part based on the coordinates of the boundary points of the original part, and mark the calculated center coordinates of the original part as the first center.

[0085] The coordinates of the boundary points of the original part are obtained by edge detection method, and the coordinate data of the boundary points of the original part are collected and recorded as the first data set. Based on each coordinate data in the first data set, the average coordinates of all boundary points of the original part are calculated, and the calculated average coordinates are used as the center coordinates of the original part, which is recorded as the first center.

[0086] S42914 The third calculation: Use edge detection method to obtain the coordinates of the boundary points of the complete welding defect obtained in the filled image, calculate the center coordinates of the complete welding defect based on the coordinates of the boundary points of the complete welding defect, and mark the calculated center coordinates of the complete welding defect as the second center.

[0087] The edge detection method is used to obtain the coordinates of the boundary points of the complete welding defects in the filled image, and the coordinate data of the boundary points of the complete welding defects are collected and recorded as the second data set. Based on each coordinate data in the second data set, the average coordinates of all boundary points of the complete welding defects are calculated, and the calculated average coordinates are used as the center coordinates of the complete welding defects, which are recorded as the second center.

[0088] S42915 Angle calculation: The vector between the origin and the first center is recorded as the first vector, and the inverse tangent function is used to calculate the angle between the direction of the first vector and the positive direction of the x-axis, which is recorded as the first angle. The vector between the origin and the second center is recorded as the second vector, and the inverse tangent function is used to calculate the angle between the direction of the second vector and the positive direction of the x-axis, which is recorded as the second angle.

[0089] S42916 First adjustment: obtaining the initial angle of the ultrasonic phased array probe, calculating the adjustment angle through proportional operation based on the initial angle, the first angle and the second angle, and using the adjustment angle as the adjusted angle of the ultrasonic phased array probe.

[0090] The beam angle of each ultrasonic phased array probe is obtained, and the azimuth angle of each probe to the first center is calculated based on the first center. The difference between the beam angle of each ultrasonic phased array probe and the azimuth angle of the probe to the first center is calculated and recorded as the angle difference of each probe, and the probe corresponding to the calculated minimum angle difference is used as the ultrasonic phased array probe that needs to be adjusted. The beam angle of the ultrasonic phased array probe that needs to be adjusted is used as the initial angle, and the adjustment angle is calculated by proportional operation based on the initial angle, the first angle and the second angle, and the adjustment angle is used as the angle of the ultrasonic phased array probe after adjustment.

[0091] S4292 Frequency adjustment: includes S42921 fourth calculation, S42922 first projection, S42923 second projection, S42924 third projection and S42925 second adjustment.

[0092] S42921 Fourth calculation: Based on the second angle, in the filled image, the second center is updated as the origin of the coordinate system, the direction of the second vector is used as the new positive direction of the x-axis, and a new y-axis and a new z-axis are constructed based on the new x-axis. A new x-axis unit vector is obtained by normalizing the second vector, a new z-axis unit vector is calculated by a cross product operation based on the second vector and the z-axis vector, and a new y-axis unit vector is calculated by a cross product operation based on the second vector and the new z-axis unit vector. A transformation matrix is ​​constructed based on the new x-axis unit vector, the new y-axis unit vector and the new z-axis unit vector, and the new coordinates of each boundary point in the coordinate system before the update in the updated coordinate system are calculated by matrix multiplication, and the updated new coordinates are marked as the coordinates of the corresponding boundary points.

[0093] S42922 First projection: Project the boundary points of the complete welding defect onto the new x-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new x-axis and the coordinate value of the projection point with the minimum coordinate value on the new x-axis, and record the difference as the first length.

[0094] S42923 Second projection: Project the boundary points of the complete welding defect onto the new y-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new y-axis and the coordinate value of the projection point with the minimum coordinate value on the new y-axis, and record the obtained difference as the first width.

[0095] S42924 Third projection: Project the boundary points of the complete welding defect onto the new z-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new z-axis and the coordinate value of the projection point with the minimum coordinate value on the new z-axis, and record the obtained difference as the first height.

[0096] S42925 Second adjustment: multiply the first length, the first width and the first height to obtain the first volume, obtain the initial frequency of the ultrasonic phased array probe and the preset standard volume, calculate the first frequency according to the frequency formula based on the initial frequency, the preset standard volume and the first volume, and use the calculated first frequency as the adjusted frequency of the ultrasonic phased array probe.

[0097] The calculation model of the frequency formula is: ; in, is the first frequency, is the initial frequency of the ultrasonic phased array probe, The preset standard volume is the volume of the area that the ultrasonic phased array probe can detect under standard working conditions. is the first volume.

[0098] In this embodiment, the ultrasonic phased array probe in the second adjustment at S42925 and the ultrasonic phased array probe in the first adjustment at S42916 are the same probe.

[0099] In this embodiment, incomplete defects are identified, defects in the incomplete image are filled using fractal image generation technology, and the filled image is verified. Based on the verified image, the welding defect characteristics after filling and the welding defect characteristics before filling are extracted to adjust the angle and frequency of the ultrasonic probe to improve the accuracy and detection range of ultrasonic waves during the detection process, which helps the ultrasonic phased array probe to better adapt to welding defects in different directions or shapes, improves the reliability of ultrasonic detection, makes the detection of welding defects more accurate, provides a more accurate image basis for subsequent defect classification and defect evaluation, optimizes subsequent defect repair strategies, and further improves the welding quality of welding defect repair.

[0100] Embodiment 4: The difference from Embodiment 1 is that: like Figure 4 As shown, after executing S5 first determination and before executing S7 strategy acquisition, it also includes: S51 first evaluation and S52 second evaluation.

[0101] S51 first evaluation includes S511 first extraction, S512 second extraction, S513 third extraction, S514 second processing and S515 second discrimination.

[0102] S511 First Extraction: Extract the geometric features of the welding defect in the second image, recorded as the first feature. The geometric features include the geometric shape and position of the welding defect, and the geometric features of the welding defect can be obtained through image processing algorithms (such as edge detection method, contour extraction method, etc.).

[0103] S512 Second extraction: Extract the size features of the welding defects in the second image, recorded as the second features. The size features are mainly the length, width, height and volume of the welding defects, and the length, width and height of the welding defects can be obtained by image processing algorithms (such as edge detection method, contour extraction method, etc.). The volume of the welding defect is calculated according to the shape of the welding defect in the obtained geometric features of the welding defect and the corresponding length, width and height of the welding defect. For example, for welding defects of regular shapes (such as defects such as pores and slag inclusions), that is, the shape of the welding defect can be approximated to a regular geometric shape (such as spherical, elliptical and other geometric shapes), the volume of the welding defect can be calculated by the geometric formula of the corresponding shape. For welding defects of irregular shapes (such as cracks and other defects), the voxelization method is used to calculate the volume of irregular welding defects.

[0104] S513 third extraction: obtain the position data of the boundary points of the key load-bearing area at the weld to be monitored, extract the position data of the boundary points of the corresponding area where the welding defects are located in the second image, calculate the Euclidean distance between the boundary points of the corresponding area where the welding defects are located in the second image and the boundary points of the key load-bearing area, and record the minimum Euclidean distance as the third feature.

[0105] Based on the geometric modeling of the welded structural parts, the material properties of the welded structural parts are defined to obtain the geometric model of the welded structural parts. The material properties include Young's modulus, Poisson's ratio, yield strength, density and thermal expansion coefficient. Based on the preset grid volume, the geometric model of the welded structural parts is divided into several grids, and the divided grids are recorded as finite elements. Gravity load and pressure load are applied to each finite element. The finite element analysis is defined as static analysis, and the finite element equation is constructed. The finite element equation is solved to obtain the key load-bearing area of ​​the welded structural parts. Among the key load-bearing areas of the welded structural parts obtained, the key load-bearing area closest to the weld to be monitored is selected as the key load-bearing area of ​​the weld to be monitored. The position data of all boundary points of the key load-bearing area of ​​the weld to be monitored are obtained, and the position data of the boundary points of the corresponding area where the welding defects are located in the second image are extracted. The Euclidean distance method is used to calculate the Euclidean distance between the boundary points of the corresponding area where the welding defects are located in the second image and the boundary points of the key load-bearing area. All calculated Euclidean distances are collected, and the minimum Euclidean distance is selected as the third feature.

[0106] S514 second processing: normalize the first feature, the second feature and the third feature of the welding defect, then encode the normalized features to obtain the encoded features, perform weighted summation on the encoded features to obtain the first data of the welding defect in the second image.

[0107] In this embodiment, the normalized features are encoded using a hot encoding method.

[0108] In this embodiment, in the weighted summation of the features after encoding processing, the weight of each feature is set according to expert experience, the weight of the first feature of the welding defect is 0.4, the weight of the second feature of the welding defect is 0.3, and the weight of the third feature of the welding defect is 0.3.

[0109] S515 Second judgment: judging whether the first data of welding defects is not higher than the preset parameter threshold: If yes, then execute S52 second evaluation; If not, execute S7 strategy acquisition.

[0110] S52 second evaluation: includes S521 third determination, S522 quantity determination, S523 third processing, S524 fifth calculation and S525 first determination.

[0111] S521 Third determination: Determine whether the number of defect category labels of the second image is 1: If yes, execute S6 first setting; If not, execute S522 quantity determination.

[0112] S522 Quantity determination: determining whether the number of defect category labels of the second image is less than a preset label quantity threshold: If yes, execute the third process in S523; If not, execute S7 strategy acquisition.

[0113] S523 The third processing: Based on the defect category label, the second image is segmented by edge detection method to obtain a plurality of welding defect images containing only one welding defect, and the plurality of welding defect images are integrated into a third image set.

[0114] S524 Fifth calculation: Use the box counting method to calculate the fractal dimension of each welding defect image in the third image set, obtain a plurality of fractal dimensions, and collect them to obtain the first dimension set.

[0115] Set different box sizes, segment the welding defect image in order from large to small based on the box sizes, count the number of boxes that can completely cover the welding defects in the welding defect image at each box scale, calculate the logarithm of the box size based on the box size, calculate the logarithm of the number of boxes based on the number of boxes corresponding to the box size, construct a double logarithmic graph with the logarithm of the box size as the x-axis and the logarithm of the number of boxes as the y-axis, and use a linear fitting method to calculate the slope of the double logarithmic graph based on the constructed double logarithmic graph. Use the calculated slope as the fractal dimension of the welding defect image, collect the fractal dimensions of all welding defect images, and obtain the first dimension set.

[0116] S525 First judgment: judging whether each fractal dimension in the first dimension set is lower than a preset dimension threshold: If yes, execute S6 first setting; If not, execute S7 strategy acquisition.

[0117] In this embodiment, by extracting the geometric features, dimensional features and Euclidean distances between defects and key load-bearing areas of welding defects, a comprehensive quantitative analysis of welding defects can be performed. Based on the quantitative analysis, the evaluation steps of welding defects are optimized, and the accuracy of welding quality evaluation is improved, so that welds with welding defects but meeting the corresponding quality standards do not need to perform defect repair steps, thereby saving repair costs and time, avoiding unnecessary repair work, and improving the timeliness and efficiency of adjusting the welding process to repair welding defects.

[0118] Embodiment 5: The difference from Embodiment 1 is that: like Figure 5 As shown, after executing S7 policy acquisition, S8 policy verification is also included, and S8 policy verification includes S81 third monitoring, S82 second judgment, S83 quality judgment, S84 third judgment and S85 model update.

[0119] S81 Third monitoring: Execute the defect repair strategy of the obtained second image, record the weld after the defect repair strategy is executed as the first weld, use the ultrasonic phased array to detect the first weld of the welded structural part, and obtain a first verification image. Obtain the welding process parameters after the defect repair strategy is executed, and record them as the first welding process parameters.

[0120] S82 Second judgment: Input the first verification image and the first welding process parameters into the trained defect classification model to obtain the first verification image marked with the defect category label, recorded as the second verification image, and determine whether there is a welding defect in the second verification image.

[0121] If not, execute S6 first setting.

[0122] If yes, then execute S83 quality determination.

[0123] S83 quality judgment: take the second verification image as the new second image, perform S511 first extraction to S514 second processing in sequence, obtain the first data of welding defects in the new second image, and judge whether the first data of welding defects in the new second image is not higher than the preset parameter threshold.

[0124] If yes, execute S84 third judgment.

[0125] If not, execute S85 model update.

[0126] S84 Third judgment: Count the number of defect category labels in the second verification image, record it as the first number, count the number of defect category labels in the second image, record it as the second number, and determine whether the first number is less than the second number.

[0127] If yes, execute S6 first setting.

[0128] If not, execute S85 model update.

[0129] S85 model update: the second image and the corresponding defect repair strategy and the first welding process parameters are used as negative sample data of the defect strategy model, and the model is trained according to the negative sample data to obtain a new trained defect strategy model, and then S7 strategy acquisition is re-executed.

[0130] In this embodiment, the defect identification and detection are re-performed on the repaired weld, and the quality of the repaired weld is evaluated to determine whether the acquired repair strategy can repair the welding defect. If the defect repair does not meet the standard, the defect repair strategy and welding process parameters are input into the defect strategy model as negative sample data for model update, so that the defect repair strategy model is continuously adjusted according to the actual situation to provide a more accurate repair strategy, reduce the risk of repair failure, and improve the accuracy and stability of the welding process in repairing welding defects.

[0131] Embodiment 6: This embodiment discloses a welding structure monitoring system based on a vibroacoustic method, the system comprising: processor and memory, The memory stores program code; When the processor calls the program code in the memory, the steps of the method described in the above embodiment are executed.

[0132] In this embodiment, the defects of the welded structural parts are monitored in real time and the defects are identified and classified, which improves the efficiency and accuracy of welding defect detection and evaluation. In addition, based on the evaluation results of welding defects and the current welding process parameters, the defect repair strategy model is input to obtain a targeted repair strategy, which achieves the technical effect of adjusting the welding process according to the defect type, helps to quickly identify welding defects that occur during the welding process and adjust the welding process in time according to the situation of the welding defects, achieving the purpose of efficiently and accurately repairing welding defects.

[0133] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for monitoring welded structural parts based on a vibroacoustic method, characterized in that: include: Data collection: Collect historical welding process parameters, historical welding defect images and corresponding defect category labels, and defect repair strategies; Model building: building defect classification model and defect strategy model; Model training: The defect classification model is trained using historical welding process parameters, historical welding defect images and corresponding defect category labels to obtain a trained defect classification model. The defect strategy model is trained using historical welding process parameters, historical welding defect images and defect repair strategies to obtain a trained defect strategy model. First monitoring: using an ultrasonic phased array to detect the weld to be detected of the welded structural component to obtain a first image; First judgment: obtain the current welding process parameters, input the current welding process parameters and the first image into the trained defect classification model, obtain the first image marked with the defect category label, record it as the second image, and judge whether there is a welding defect in the second image: If not, then execute the first setting step; If yes, then execute the strategy acquisition step; First setting: setting the welding quality assessment result of the weld to be inspected to be excellent, and outputting the welding quality assessment result; Strategy acquisition: The current welding process parameters and the second image are input into the trained defect strategy model to obtain the defect repair strategy of the second image.

2. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 1, characterized in that: After executing the first monitoring step and before executing the first distinguishing step, the method further includes: First collection: collect the current ambient temperature and ambient humidity; obtain the standard temperature and standard humidity; First calculation: calculate the difference between the ambient temperature and the standard temperature, recorded as the first difference, calculate the difference between the ambient humidity and the standard humidity, recorded as the second difference; Speed ​​adjustment: obtaining the standard ultrasonic propagation speed, calculating the ultrasonic propagation speed adjustment value of the ultrasonic phased array based on the environmental compensation formula, and correcting the ultrasonic propagation speed of the ultrasonic phased array based on the calculated ultrasonic propagation speed adjustment value; The calculation model of the environmental compensation formula is: ； in, is the ultrasonic propagation speed adjustment value, is the standard ultrasonic propagation speed, is the influence coefficient of temperature on ultrasonic propagation velocity, is the first difference, is the influence coefficient of humidity on ultrasonic propagation velocity, is the second difference, is the comprehensive influence coefficient of temperature and humidity; Second monitoring: based on the corrected ultrasonic wave propagation velocity, the first image is corrected to obtain a corrected first image, and the corrected first image is updated as the first image.

3. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 1, characterized in that: After executing the first monitoring step and before executing the first distinguishing step, the method further includes: First construction: build a defect recognition model; Collect samples: collect images containing complete welding defects and corresponding complete labels, which are recorded as the first sample; collect images containing incomplete welding defects and corresponding incomplete labels, which are recorded as the second sample; the integrity labels of welding defects include complete labels and incomplete labels; First training: using the first sample and the second sample to train the defect recognition model to obtain a trained defect recognition model; Image segmentation: extracting welding defects from the first image, and segmenting the first image based on the extracted welding defects to obtain a number of sub-images containing welding defects; Defect monitoring: Input the sub-image containing welding defects into the trained defect recognition model, and output the sub-image annotated with integrity labels; Defect screening: Screen out sub-images with incomplete labels from sub-images with complete labels and record them as incomplete images; Defect filling: Using fractal image generation technology, the incomplete welding defects in the incomplete image are filled based on the preset fractal dimension to obtain the complete welding defects, and the filled incomplete image is recorded as the filled image; Parameter adjustment: extract features of welding defects in the image, adjust parameters of the ultrasonic phased array probe based on the extracted features, and use the ultrasonic phased array with adjusted parameters to perform the first monitoring step.

4. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 3, characterized in that: After the step of performing defect repair and before the step of performing parameter adjustment, the method further includes: First verification: The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the incomplete welding defects in the incomplete image, which is recorded as the first matrix. The gray level co-occurrence matrix method is used to construct the gray level co-occurrence matrix of the remaining part containing defects in the image, which is recorded as the second matrix. Second verification: calculate the contrast of the first matrix, recorded as the first contrast, calculate the contrast of the second matrix, recorded as the second contrast, calculate the contrast difference between the first contrast and the second contrast, recorded as the third difference, and determine whether the third difference is less than the preset contrast difference threshold: If yes, then execute the parameter adjustment step; If not, the preset fractal dimension is adjusted and the defect filling step is performed.

5. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 3, characterized in that: The parameter adjustment steps also include: First processing: in the filled image, the incomplete welding defect before filling is recorded as the original part, and the remaining part containing the defect in the filled image is recorded as the filled part; Construct coordinates: Take any pixel in the filled image as the origin, the length direction of the filled image as the x-axis, and the width direction of the filled image as the y-axis to construct a coordinate system; Second calculation: using an edge detection method to obtain coordinates of boundary points of the original part, calculating the center coordinates of the original part based on the coordinates of the boundary points of the original part, and marking the calculated center coordinates of the original part as the first center; Third calculation: using edge detection method to obtain the coordinates of the boundary points of the complete welding defect obtained in the filled image, based on the coordinates of the boundary points of the complete welding defect, calculating the center coordinates of the defect, and marking the calculated center coordinates of the defect as the second center; Angle calculation: record the vector between the origin and the first center as the first vector, calculate the angle between the direction of the first vector and the positive direction of the x-axis, record it as the first angle, record the vector between the origin and the second center as the second vector, calculate the angle between the direction of the second vector and the positive direction of the x-axis, record it as the second angle; First adjustment: obtaining an initial angle of the ultrasonic phased array probe, calculating an adjustment angle through proportional operation based on the initial angle, the first angle and the second angle, and using the adjustment angle as the adjusted angle of the ultrasonic phased array probe.

6. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 4, characterized in that: After performing the first adjustment step, the method further includes: Fourth calculation: based on the second angle, in the filled image, the second center is updated as the origin of the coordinate system, the direction of the second vector is used as the new positive direction of the x-axis, and a new y-axis and a new z-axis are constructed based on the new x-axis; First projection: Project the boundary points of the complete welding defect onto the new x-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new x-axis and the coordinate value of the projection point with the minimum coordinate value on the new x-axis, and record the obtained difference as the first length; Second projection: Project the boundary points of the complete welding defect onto the new y-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new y-axis and the coordinate value of the projection point with the minimum coordinate value on the new y-axis, and record the obtained difference as the first width; Third projection: Project the boundary points of the complete welding defect onto the new z-axis, calculate the difference between the coordinate value of the projection point with the maximum coordinate value on the new z-axis and the coordinate value of the projection point with the minimum coordinate value on the new z-axis, and record the obtained difference as the first height; Second adjustment: performing a product operation on the first length, the first width, and the first height to obtain a first volume, acquiring an initial frequency and a preset standard volume of the ultrasonic phased array probe, calculating a first frequency according to a frequency formula based on the initial frequency, the preset standard volume, and the first volume, and using the calculated first frequency as the adjusted frequency of the ultrasonic phased array probe; The calculation model of the frequency formula is: ； in, is the first frequency, is the initial frequency of the ultrasonic phased array probe, To preset the standard volume, is the first volume.

7. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 1, characterized in that: After executing the first determination step and before executing the strategy acquisition step, the method further includes: First extraction: extracting the geometric features of the welding defects in the second image, recorded as the first features; Second extraction: extracting the size feature of the welding defect in the second image, recorded as the second feature; The third extraction: obtaining the position data of the boundary points of the key load-bearing area of ​​the weld to be monitored, extracting the position data of the boundary points of the area corresponding to the welding defect in the second image, calculating the Euclidean distance between the boundary points of the area corresponding to the welding defect in the second image and the boundary points of the key load-bearing area, and recording the smallest Euclidean distance as the third feature; Second processing: encoding the first feature, the second feature and the third feature of the welding defect to obtain the features after encoding, and performing weighted summation on the features after encoding to obtain the first data of the welding defect in the second image; Second judgment: judge whether the first data of welding defects is not higher than the preset parameter threshold: If yes, then execute the first setting step; If not, execute the strategy acquisition step.

8. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 7, characterized in that: After performing the second determination step and before performing the first setting step, the method further includes: Third judgment: Determine whether the number of defect category labels of the second image is 1: If yes, then execute the first setting step; If not, the third processing step is executed; The third processing: segmenting the second image based on the defect category label to obtain a plurality of welding defect images, and integrating the plurality of welding defect images into a third image set; Fifth calculation: using the box counting method to calculate the fractal dimension of each welding defect image in the third image set, obtaining a number of fractal dimensions, and collecting them to obtain the first dimension set; First judgment: judge whether each fractal dimension in the first dimension set is lower than the preset dimension threshold: If yes, then execute the first setting step; If not, execute the strategy acquisition step.

9. The method for monitoring welded structural parts based on the vibroacoustic method according to claim 7, characterized in that: After executing the strategy acquisition step, it also includes: The third monitoring: executing the defect repair strategy of the obtained second image, recording the weld after the defect repair strategy is executed as the first weld, using the ultrasonic phased array to detect the first weld of the welded structural part, and obtaining a first verification image; obtaining the welding process parameters after the defect repair strategy is executed, and recording them as the first welding process parameters; Second judgment: Input the first verification image and the first welding process parameter into the trained defect classification model to obtain the first verification image marked with the defect category label, recorded as the second verification image, and determine whether the second verification image has welding defects: If not, then execute the first setting step; If yes, then the quality determination step is performed; Quality judgment: Take the second verification image as the new second image, perform the first extraction to the second processing steps in sequence, obtain the first data of the welding defect in the new second image, and judge whether the first data of the welding defect in the new second image is not higher than the preset parameter threshold: If yes, then execute the third judgment step; If not, then execute the model update step; Third judgment: counting the number of defect category labels in the second verification image, recorded as a first number, counting the number of defect category labels in the second image, recorded as a second number, and determining whether the first number is less than the second number; If yes, then execute the first setting step; If not, then execute the model update step; Model update: The second image, the corresponding defect repair strategy and the first welding process parameters are used as negative sample data of the defect strategy model, and the model is trained according to the negative sample data to obtain a new trained defect strategy model, and then the strategy acquisition step is re-executed.

10. A welding structure monitoring system based on a vibroacoustic method, characterized in that: The system comprises: Processor and memory; The memory stores program code; When the processor calls the program code in the memory, the steps of the method according to any one of claims 1 to 9 are executed.

Images