A method and system for monitoring the fusion of weld pools

Abstract

This invention relates to the field of weld pool monitoring, specifically to a method and system for monitoring the fusion of weld pools. It includes: acquiring ultrasonic scanning data of the weld pool using a single-point scanning method; determining a first reflection time and a second reflection time based on the dimensional parameters of the weld pool; determining other maxima points from the ultrasonic scanning data based on the first and second reflection times; identifying the actual defect points from these maxima points; inputting the echo signal characteristics of the actual defect points into a trained Long Short-Term Memory (LSTM) network; outputting the defect type and fusion state through the trained LSM network; and adjusting the fusion parameters and optimization algorithm of the weld pool based on the defect type and fusion state. This invention can identify the defect type and the fusion state of the weld pool, facilitating timely adjustment and optimization of the welding process.

Description

Technical Field

[0001] This invention relates to the field of weld pool monitoring, and specifically to a method and system for monitoring the fusion of welding weld pools. Background Technology

[0002] The weld pool is a region of liquid metal formed during the fusion welding process under the influence of a heat source. Under the action of an electric arc, the molten metal in the pool overcomes gravity and surface tension to flow towards the tail, and after cooling, forms the weld. The temperature of the weld pool exhibits a gradient distribution, with the highest temperature at the center of the arc and decreasing towards the edges. The metal at the head of the arc is in the melting stage, while the tail begins to crystallize, undergoing a dynamic cycle of heating-melting-cooling. This region is characterized by its small volume and rapid cooling rate, requiring strict control of chemical components such as carbon, sulfur, and phosphorus.

[0003] To identify typical defects in the weld pool during welding, such as incomplete fusion, porosity, and cracks, existing visual information-based defect identification methods mainly rely on the surface image features of the weld pool acquired by high-speed cameras or infrared thermal imagers. These methods typically include techniques such as grayscale distribution variation analysis of the weld pool image, surface texture feature extraction, and edge contour detection. However, in dynamic hot processing processes such as arc welding and laser welding, the weld pool is actually a transient high-temperature liquid metal region containing complex three-dimensional flow, heat and mass transfer, and phase transformation behaviors. Its surface morphology (such as weld pool width and trailing angle) can only reflect part of the process state, while the key factors determining welding quality, such as penetration depth, internal porosity distribution, and dendrite growth direction, are hidden in the internal structure of the weld pool.

[0004] Existing two-dimensional image processing methods often neglect the correlation between the thermal-fluid coupling characteristics (such as Marangoni convection and keyhole effect) inside the molten pool and the formation of defects because they cannot obtain longitudinal cross-sectional information of the molten pool. This results in insufficient detection accuracy for hidden defects such as incomplete fusion and root cracks, making it difficult to meet the quality control requirements of high-reliability welding. Summary of the Invention

[0005] This invention provides a method and system for monitoring the fusion of weld pools to solve existing problems.

[0006] The welding pool fusion monitoring method of the present invention adopts the following technical solution:

[0007] One embodiment of the present invention provides a method for monitoring the fusion of a weld pool, the method comprising the following steps:

[0008] Ultrasonic scanning data of the weld pool is obtained by a single-point scanning method, wherein the weld pool includes at least one scanning point;

[0009] Based on the size parameters of the weld pool, the first reflection time and the second reflection time are determined, wherein the first reflection time is the time for the ultrasonic echo signal to reflect back to the surface of the weld pool, and the second reflection time is the time for the ultrasonic echo signal to reflect back to the bottom of the weld pool.

[0010] Based on the first reflection time and the second reflection time, other maxima are determined from the ultrasonic scanning data. These other maxima are local maxima located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data.

[0011] Identify the true defect point from other maxima points;

[0012] The echo signal characteristics of the real defect points are input into the trained Long Short-Term Memory (LSTM) network. The trained LSM network outputs the defect type and fusion state, and the fusion parameters of the weld pool and the optimization algorithm are adjusted based on the defect type and fusion state.

[0013] Optionally, other maxima are determined from the ultrasound scan data based on the first and second reflection times, specifically including:

[0014] For each ultrasound scan point, the local maximum point is determined using the derivative method.

[0015] Arrange the local maxima points in ascending order of time to obtain an ascending time sequence;

[0016] The Otsu multi-threshold segmentation method is used to segment the ascending time sequence to obtain extreme point segments;

[0017] The mean value of each extreme point segment is calculated. The peak of the element in the first extreme point segment in the ultrasonic scanning data is determined as the surface echo of the molten pool, and the peak of the element in the second extreme point segment in the ultrasonic scanning data is determined as the bottom echo of the molten pool. The first extreme point segment is the extreme point segment with the smallest difference between the mean value and the first reflection time, and the second extreme point segment is the extreme point segment with the smallest difference between the mean value and the second reflection time.

[0018] The local maxima located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data are identified as other maxima.

[0019] Optionally, the true defect point can be determined from other maxima, specifically including:

[0020] Identify abnormal echo points from other maximum points and define the ultrasound scan data of abnormal echo points as abnormal echoes.

[0021] New ultrasound scan data for each abnormal echo point is obtained. The new ultrasound scan data is obtained by emitting sound waves within a preset sound intensity range for each abnormal echo point, and the sound intensity of the preset sound intensity range increases sequentially.

[0022] Determine the sequence of trend values ​​based on newly added ultrasound scan data;

[0023] The amplitude sequence and time point sequence of abnormal echo points in multiple scan data of each real defect candidate are determined based on the changing trend value sequence;

[0024] The first actual defect point and the suspected defect point are determined based on the amplitude sequence and the time point sequence.

[0025] Identify the second actual defect from the suspected defect points;

[0026] The first and second real defect points are identified as real defect points.

[0027] Optionally, anomalous echo points can be determined from other maxima points, specifically including:

[0028] The upper envelope of other maxima is determined by cubic spline interpolation, and the fitting point of each point in the upper envelope is determined by the least squares method. The absolute value of the difference between each fitting point and the corresponding point in the upper envelope is calculated to obtain the absolute value of the difference. The ratio of the absolute value of the difference to the corresponding point in the upper envelope is calculated to obtain the convexity value.

[0029] The echo anomaly probability of each point in the upper envelope is determined based on the convexity value, and points with an echo anomaly probability greater than a preset anomaly threshold are identified as abnormal echo points.

[0030] Optionally, a sequence of trend values ​​is determined based on the newly added ultrasound scan data, specifically including:

[0031] For each abnormal echo point, the amplitude corresponding to each sound wave intensity of each abnormal echo point in the newly added ultrasound scan data is obtained to obtain the amplitude sequence.

[0032] Determine the trend value sequence based on the amplitude sequence.

[0033] Optionally, a trend value sequence is determined based on the amplitude sequence, specifically including:

[0034] The order of the elements corresponding to the elements of the first n amplitude sequences is used as the x-axis, and the elements of each amplitude sequence are used as the y-axis to determine the principal component analysis coordinates. The projection vector and the projection value corresponding to each projection vector are determined by the principal component analysis method. The arctangent angle of the ratio of the y-axis to the x-axis of the projection vector with the largest projection value is determined as the first element in the trend value sequence, where n is a preset constant.

[0035] By sequentially determining each element in the trend value sequence, the trend value sequence is obtained.

[0036] Optionally, the amplitude sequence and time point sequence of abnormal echo points in multiple scans of each real defect candidate are determined based on the trend value sequence, specifically including:

[0037] The Otsu multi-threshold segmentation method is used to segment the trend value sequence, determine two trend segmentation segments, calculate the variance of each trend segmentation segment, and obtain the sum of the variances of the two trend segmentation segments to obtain the false defect probability value.

[0038] The probability value of a candidate real defect is determined based on the probability value of a false defect, and abnormal echo points whose probability value of a candidate real defect is greater than a preset defect threshold are identified as candidates real defects.

[0039] Acquire multiple scan data for each real defect candidate, identify abnormal echo points in the multiple scan data, and obtain the amplitude sequence and time point sequence of the abnormal echo points in each multiple scan data.

[0040] Optionally, the first true defect point and the suspected defect point are determined based on the amplitude sequence and the time point sequence, specifically including:

[0041] The coefficient of variation of the amplitude sequence and the standard deviation of the time point sequence are calculated. Abnormal echo points with a coefficient of variation less than a preset first abnormal threshold and a standard deviation less than a preset second abnormal threshold are identified as first real defect points. Abnormal echo points in multiple scan data other than the first real abnormal points are identified as suspected defect points.

[0042] Optionally, a second real defect point is determined from the suspected defect points, specifically including:

[0043] Acquire the final scan data of suspected defect points at different emission energies, identify abnormal echo points in the final scan data, calculate the absolute value of the time difference between the time of the abnormal echo point in each final scan data and the first reflection time, and obtain the time difference sequence of the abnormal echo points in each final scan data.

[0044] Abnormal echo points whose time difference sequence is less than a preset variance threshold are identified as the second true defect points.

[0045] This invention proposes a welding pool fusion monitoring system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the welding pool fusion monitoring method described above.

[0046] The beneficial effects of the technical solution of the present invention are:

[0047] In this embodiment of the invention, by analyzing the ultrasonic echo signals under different emission energies, and combining the surface echo and bottom echo of the molten pool, a pseudo-defect signal with noise removed is obtained, and the real defect scanning point is obtained. Then, the signal characteristics of the defect scanning point are input into the neural network to identify the defect type and the fusion state of the weld pool, which is convenient for further analysis and adjustment. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart illustrating a welding pool fusion monitoring method provided in one embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the weld pool;

[0051] Figure 3 This is a schematic diagram showing the maximum value points of the probe's point-by-point scanning data.

[0052] Figure 4 This is a structural diagram of a welding pool fusion monitoring system provided in one embodiment of the present invention. Detailed Implementation

[0053] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a welding pool fusion monitoring method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0055] The following describes in detail, with reference to the accompanying drawings, a specific scheme for a welding pool fusion monitoring method provided by the present invention.

[0056] This invention provides a method and system for monitoring the fusion of weld pools. Please refer to [link / reference]. Figure 1The diagram illustrates a flowchart of a welding pool fusion monitoring method according to an embodiment of the present invention, the method comprising the following steps:

[0057] S101. Ultrasonic scanning data of the weld pool is obtained by a single-point scanning method, wherein the weld pool includes at least one scanning point.

[0058] In a specific embodiment, such as Figure 2 As shown, Figure 2 This is a schematic diagram of a weld pool. Weld pool fusion is a crucial process in welding where the base material and filler material (such as welding wire) fully mix and form a dense bond in a high-temperature molten pool. Its quality directly affects the mechanical properties (such as strength and toughness) and defect rate (such as lack of fusion and porosity) of the weld. Figure 2 As shown in the figure, the white circle represents the weld pool generated during welding. As the welding path moves forward, the weld pool also moves forward.

[0059] By collecting comprehensive information about the weld pool using ultrasonic sensors, and then analyzing the monitored information, the fusion state of the base material and welding wire in the weld pool can be obtained.

[0060] In an ultrasonic sensor, a piezoelectric crystal is installed inside the ultrasonic probe to act as the ultrasonic sensor, used to excite and receive ultrasonic signals. During detection, the ultrasonic sensor first loads parameters and controls the probe to move to the initial scanning point. After the probe comes to rest, a sinusoidal wave excitation causes the crystal to vibrate, generating pulsed ultrasonic waves. The reflected echo signals are received and stored as the scanning data for that scanning point after analog-to-digital conversion. Subsequently, the probe moves point by point, repeating this process to obtain the ultrasonic scanning data for each scanning point.

[0061] S102. Based on the size parameters of the weld pool, determine the first reflection time and the second reflection time, wherein the first reflection time is the time for the ultrasonic echo signal to reflect back to the surface of the weld pool, and the second reflection time is the time for the ultrasonic echo signal to reflect back to the bottom of the weld pool.

[0062] In one specific embodiment, the probe height can be preset, and then the estimated time T1 (i.e., the first reflection time) for the ultrasonic signal to reach the surface of the molten pool and be reflected back to the probe can be calculated based on the distance between the probe and the surface of the molten pool. Similarly, the distance between the probe and the bottom of the molten pool can be obtained by laser ranging, and then the estimated time T2 (i.e., the second reflection time) for the ultrasonic signal to reach the bottom of the molten pool and be reflected back to the probe can be calculated based on the distance between the probe and the bottom of the molten pool.

[0063] S103. Based on the first reflection time and the second reflection time, determine other maxima points from the ultrasonic scanning data, wherein the other maxima points are local maxima points located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data.

[0064] In this embodiment, other maximum points are determined from the ultrasound scan data based on the first reflection time and the second reflection time, specifically including:

[0065] For each ultrasound scan point, the local maximum point is determined using the derivative method.

[0066] Arrange the local maxima points in ascending order of time to obtain an ascending time sequence;

[0067] The Otsu multi-threshold segmentation method is used to segment the ascending time sequence to obtain extreme point segments;

[0068] The mean value of each extreme point segment is calculated. The peak of the element in the first extreme point segment in the ultrasonic scanning data is determined as the surface echo of the molten pool, and the peak of the element in the second extreme point segment in the ultrasonic scanning data is determined as the bottom echo of the molten pool. The first extreme point segment is the extreme point segment with the smallest difference between the mean value and the first reflection time, and the second extreme point segment is the extreme point segment with the smallest difference between the mean value and the second reflection time.

[0069] The local maxima located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data are identified as other maxima.

[0070] For example, if the fusion state between the base material and the welding wire in the molten pool is good, there are only two relatively obvious peaks in the echo signal at different scanning points. One peak represents the surface echo of the molten pool (reflected wave at the interface between the solid base material and the liquid molten pool), and the other peak represents the bottom echo of the molten pool (reflected wave at the interface between the molten pool and the solid weld).

[0071] If the fusion state is poor and defects exist, defect echoes will be formed, namely scattered waves caused by pores / lack fusion (manifested as time-delayed burr signals). Therefore, by comparing multiple scanning points, signals with relatively fixed positions are first obtained, which are denoted as the surface echo signal and the bottom echo signal of the molten pool.

[0072] For each scan point's scan data (a curved acoustic wave), all local maxima on the data are obtained using the derivative method. Local maxima indicate that the acoustic wave was reflected at the corresponding position in the molten pool. This may be caused by defects resulting from incomplete fusion, or by reflections from the surface and bottom of the molten pool.

[0073] Obtain the time point of each local extreme point on the scan data of all scan points, obtain all time points, arrange all time points in ascending order to form an ascending sequence, and obtain multiple extreme point segments (i.e., extreme point segmentation segments) on the ascending sequence using the Otsu multi-threshold segmentation method, with time points in the same extreme point segment being close to each other.

[0074] For each extreme point segment, the mean value of all time points in that segment is calculated; then, the time difference between all the means and T1 is calculated sequentially, and the peak of the corresponding time point in the scan signal of the extreme point segment with the smallest time difference is recorded as the surface echo of the molten pool; similarly, the time difference between each mean and T2 can be calculated, and the peak of the corresponding time point in the scan signal of the extreme point segment with the smallest time difference is recorded as the bottom echo of the molten pool.

[0075] like Figure 3 As shown, Figure 3 This is a schematic diagram showing the maximum points of the probe's point-by-point scanning data. Figure 3 In the diagram, points 1, 2, and 3 represent the time points of the maximum values ​​in the echo signals obtained by the probe through point-by-point scanning. The dashed lines represent the time points, with the straight line corresponding to T1 representing the time point T1 and the straight line corresponding to T2 representing the time point T2.

[0076] The echoes corresponding to other maxima obtained at this time can include: defect echoes and echoes caused by electronic noise from the ultrasonic probe. The amplitude of the echoes caused by electronic noise is generally small. Therefore, the defect echoes are identified by the convexity of each other maxima. The greater the convexity, the greater the probability that the other maxima belongs to the defect echo.

[0077] S104. Determine the true defect point from other maximum points.

[0078] In this embodiment, determining the true defect point from other maxima points specifically includes:

[0079] Identify abnormal echo points from other maximum points and define the ultrasound scan data of abnormal echo points as abnormal echoes.

[0080] New ultrasound scan data for each abnormal echo point is obtained. The new ultrasound scan data is obtained by emitting sound waves within a preset sound intensity range for each abnormal echo point, and the sound intensity of the preset sound intensity range increases sequentially.

[0081] Determine the sequence of trend values ​​based on newly added ultrasound scan data;

[0082] The amplitude sequence and time point sequence of abnormal echo points in multiple scan data of each real defect candidate are determined based on the changing trend value sequence;

[0083] The first actual defect point and the suspected defect point are determined based on the amplitude sequence and the time point sequence.

[0084] Identify the second actual defect from the suspected defect points;

[0085] The first and second real defect points are identified as real defect points.

[0086] Identifying anomalous echo points from other maxima points includes:

[0087] The upper envelope of other maxima is determined by cubic spline interpolation, and the fitting point of each point in the upper envelope is determined by the least squares method. The absolute value of the difference between each fitting point and the corresponding point in the upper envelope is calculated to obtain the absolute value of the difference. The ratio of the absolute value of the difference to the corresponding point in the upper envelope is calculated to obtain the convexity value.

[0088] The echo anomaly probability of each point in the upper envelope is determined based on the convexity value, and points with an echo anomaly probability greater than a preset anomaly threshold are identified as abnormal echo points.

[0089] The trend value sequence was determined based on the newly added ultrasound scan data, specifically including:

[0090] For each abnormal echo point, the amplitude corresponding to each sound wave intensity of each abnormal echo point in the newly added ultrasound scan data is obtained to obtain the amplitude sequence.

[0091] Determine the trend value sequence based on the amplitude sequence.

[0092] Determining the trend value sequence based on the amplitude sequence specifically includes:

[0093] The order of the elements corresponding to the elements of the first n amplitude sequences is used as the x-axis, and the elements of each amplitude sequence are used as the y-axis to determine the principal component analysis coordinates. The projection vector and the projection value corresponding to each projection vector are determined by the principal component analysis method. The arctangent angle of the ratio of the y-axis to the x-axis of the projection vector with the largest projection value is determined as the first element in the trend value sequence, where n is a preset constant.

[0094] By sequentially determining each element in the trend value sequence, the trend value sequence is obtained.

[0095] Based on the trend value sequence, the amplitude sequence and time point sequence of abnormal echo points in the multiple scan data of each real defect candidate are determined, specifically including:

[0096] The Otsu multi-threshold segmentation method is used to segment the trend value sequence, determine two trend segmentation segments, calculate the variance of each trend segmentation segment, and obtain the sum of the variances of the two trend segmentation segments to obtain the false defect probability value.

[0097] The probability value of a candidate real defect is determined based on the probability value of a false defect, and abnormal echo points whose probability value of a candidate real defect is greater than a preset defect threshold are identified as candidates real defects.

[0098] Acquire multiple scan data for each real defect candidate, identify abnormal echo points in the multiple scan data, and obtain the amplitude sequence and time point sequence of the abnormal echo points in each multiple scan data.

[0099] The first true defect point and suspected defect points are determined based on the amplitude sequence and time point sequence, specifically including:

[0100] The coefficient of variation of the amplitude sequence and the standard deviation of the time point sequence are calculated. Abnormal echo points with a coefficient of variation less than a preset first abnormal threshold and a standard deviation less than a preset second abnormal threshold are identified as first real defect points. Abnormal echo points in multiple scan data other than the first real abnormal points are identified as suspected defect points.

[0101] Identifying the second actual defect from suspected defect points specifically includes:

[0102] Acquire the final scan data of suspected defect points at different emission energies, identify abnormal echo points in the final scan data, calculate the absolute value of the time difference between the time of the abnormal echo point in each final scan data and the first reflection time, and obtain the time difference sequence of the abnormal echo points in each final scan data.

[0103] Abnormal echo points whose time difference sequence is less than a preset variance threshold are identified as the second true defect points.

[0104] For example, for other local maxima in the scan data of each scan point, cubic spline interpolation is used to connect adjacent local maxima to form an upper envelope.

[0105] For each point on the upper envelope corresponding to the scan data of each scan point, the data before that point is fitted using the least squares method to obtain the fitting equation. The time point of that point is input to obtain the fitted point n1. The absolute value of the difference between the fitted value n1 and the actual value n2 of that point is calculated. The absolute value of the difference is calculated, and the ratio of the absolute value of the difference to the actual value n2 is calculated to obtain the convexity value. The smaller the convexity value b, the closer the fitted point is to the actual value, the smaller the convexity of that point, and the smaller the probability of anomaly in the echo corresponding to that point.

[0106] The probability of echo anomalies at each point in the upper envelope can be determined based on the convexity value using the following formula:

[0107]

[0108] in, Indicates the convexity value. Represents the natural constant. This indicates the probability of an echo anomaly.

[0109] Echo anomaly probability The larger the value, the greater the probability that the point corresponds to an echo anomaly. Points on the upper envelope with a value greater than 0.5 are denoted as abnormal echoes.

[0110] Optionally, the preset abnormal threshold can be set to 0.5 or other values, which can be set according to the actual situation, and no specific restrictions are imposed here.

[0111] Because the emitted sound waves themselves affect the molten pool: for surface echoes, they cause periodic fluctuations in the amplitude of the surface echo; for bottom echoes, they cause a reduction in the amplitude of the bottom echo, easily leading to misjudgment of the internal flow of the molten pool as dynamic defects, such as false porosity signals. Therefore, the obtained abnormal echoes not only contain defect echoes but also echoes that are normally present but identified as defects. It is necessary to remove such echoes from the abnormal echoes to obtain the defect echoes, analyze the defect echoes to find the causes of the defects, and then adjust the parameters to avoid the generation of defects and obtain a better fusion state between the base material and the welding wire.

[0112] In this embodiment, sound waves of different intensities are emitted to distinguish between real and pseudo-defects. For real defects, such as pores and unfused defects, the physical properties are relatively stable, while ultrasonic disturbance artifacts, such as molten pool surface vibration and acoustic flow effects, change significantly with the ultrasonic energy.

[0113] For each abnormal echo corresponding to a scanning point, new ultrasound scan data are obtained sequentially for the emitted sound waves with progressively increasing intensity. For the same abnormal echo location, the amplitude at that location is obtained in different new ultrasound scan data, and an amplitude sequence is obtained according to the sequentially increasing intensity of the emitted sound waves.

[0114] Starting from the 5th element in the amplitude sequence, the order values ​​of the first 5 elements in the amplitude sequence are used as the x-axis, and the corresponding elements are used as the y-axis to obtain the principal component analysis coordinates. Multiple coordinate points are input into PCA (Principal Component Analysis) to obtain multiple projection vectors and their corresponding projection values. The projection values ​​represent the projection lengths of these coordinate points in the corresponding projection directions. The arctangent angle of the ratio of the y-axis to x-axis of the projection vector corresponding to the maximum projection length is used as the first element in the trend value sequence. The same method is used to obtain the first 6, 7, etc., trend values, forming a trend value sequence. In this embodiment, n is 5, but it can be set according to actual conditions and is not specifically limited here.

[0115] In the trend sequence of real defects, since the amplitude of real defects exhibits a linear or saturation trend with increasing energy, the element values ​​in the trend sequence corresponding to the linear increase stage are similar, while the element values ​​in the trend sequence corresponding to the saturation stage are close to 0. In contrast, the element values ​​in the trend sequence of artifact signals differ significantly.

[0116] For the trend value sequence, the Otsu threshold segmentation method is used to obtain two trend segments. The variances of the two trend segments are calculated separately, and then summed to obtain the false defect probability value. , The larger the value, the more likely it is to be a false defect at the corresponding scan point.

[0117] The formula used to determine the candidate probability of a true defect based on the false defect probability value can be:

[0118]

[0119] in, This represents the probability value of a true defect candidate. This represents the probability value of a false defect.

[0120] Since some artifacts (such as acoustic vortices) may also exhibit amplitude saturation at high energies, therefore... The corresponding scan points are recorded as real defect candidates, but it is necessary to verify whether they may be high-energy artifacts.

[0121] Optionally, the default defect threshold is set to 0.7, but it can also be set according to the actual situation. No specific restrictions are imposed here.

[0122] For the scanning point corresponding to the candidate real defect, this embodiment further identifies the real defect by scanning the same position multiple times: the position and size of the real defect are fixed, and the detection signals of multiple frames should be highly consistent; artifacts are affected by the dynamics of the melt pool, and the signal fluctuates greatly.

[0123] For each real defect candidate, the location of the abnormal echo in each scan data point is calculated from multiple scan data points, and the amplitude sequence and time point sequence of these locations are obtained.

[0124] The coefficient of variation v of the amplitude sequence is calculated, and the standard deviation m of the time point sequence is obtained. If v < 0.15 and m < 0.2, the corresponding scan point is likely to be a real defect point. Points that do not meet the conditions are judged as suspected defect points.

[0125] Optionally, in this embodiment, the preset first abnormal threshold is set to 0.15 and the preset second abnormal threshold is set to 0.2. The preset first abnormal threshold and the preset second abnormal threshold can also be set according to the actual situation, and no specific restrictions are imposed here.

[0126] Among the obtained real defect candidates, since the surface vibration of the molten pool may cause the signal at the real defect point to be unstable, this embodiment further uses the following: the interface position of the real defect is fixed and the echo time does not change with energy; the artifacts will cause the echo time to be delayed due to the dynamics of the molten pool (vibration, acoustic flow), and the real defect point among the suspected defect points can be identified by whether the echo time is delayed.

[0127] For the final scan data obtained at different emission energies for each suspected defect point, the abnormal echo position in the final scan data is obtained, and then the corresponding time point is obtained. The time difference between each time point and the time point corresponding to the molten pool surface echo in the data is calculated to form a time difference sequence.

[0128] If the time difference variation is small, it is likely a real defect; otherwise, it is likely an artifact. The variance of each time difference sequence is calculated, and suspected defect points with a variance less than 0.3 are recorded as real defect points. At this point, all real defect points have been obtained, and the corresponding echo signal for each real defect point can then be obtained.

[0129] Optionally, the preset variance threshold is set to 0.3, which can be set according to actual needs, and no specific restrictions are imposed here.

[0130] S105. Input the echo signal characteristics of the real defect point into the trained long short-term memory network, output the defect type and fusion state through the trained long short-term memory network, and adjust the fusion parameters of the weld pool and the optimization algorithm through the defect type and fusion state.

[0131] In one specific embodiment, echo signal features (such as amplitude, time-domain waveform, and spectral characteristics) corresponding to each real defect point can be extracted. Subsequently, these defect echo signals are input into a trained deep learning neural network (such as a convolutional neural network (CNN) or a long short-term memory network (LSTM)) to perform defect type classification and fusion status evaluation (such as insufficient penetration, porosity, and unfused area).

[0132] Based on the output of the neural network, the system further combines the welding process parameter library (such as current, voltage, welding speed, etc.) and optimization algorithms (such as reinforcement learning or genetic algorithms) to generate a dynamic adjustment scheme.

[0133] This invention is now complete.

[0134] In summary, in this embodiment of the invention, by analyzing ultrasonic echo signals under different emission energies and combining the surface echo and bottom echo of the molten pool, pseudo-defect signals with noise removed are obtained, and real defect scanning points are obtained. Then, the signal features of the defect scanning points are input into the neural network to identify the defect type and the fusion state of the weld pool, which facilitates further analysis and adjustment.

[0135] This invention also proposes a welding pool fusion monitoring system; please refer to [link / reference]. Figure 4 The diagram shows a structural diagram of a welding pool fusion monitoring system provided in an embodiment of the present invention. The system includes: a data acquisition module 101, a data processing module 102, and a defect output module 103.

[0136] The data acquisition module 101 is used to acquire ultrasonic scanning data of the weld pool through a single-point scanning method, wherein the weld pool includes at least one scanning point;

[0137] The data processing module 102 is used to determine the first reflection time and the second reflection time based on the size parameters of the weld pool, wherein the first reflection time is the time it takes for the ultrasonic echo signal to reflect back to the surface of the weld pool, and the second reflection time is the time it takes for the ultrasonic echo signal to reflect back to the bottom of the weld pool; based on the first reflection time and the second reflection time, other maximum points are determined from the ultrasonic scanning data; and the actual defect points are determined from the other maximum points.

[0138] The defect output module 103 is used to input the echo signal characteristics of the real defect point into the trained long short-term memory network, output the defect type and fusion state through the trained long short-term memory network, and adjust the fusion parameters and optimization algorithm of the weld pool through the defect type and fusion state.

[0139] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the welding pool fusion monitoring system and the welding pool fusion monitoring method embodiment provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.

[0140] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0141] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring the fusion of a weld pool, characterized in that, include: Ultrasonic scanning data of the weld pool is obtained by a single-point scanning method, wherein the weld pool includes at least one scanning point; Based on the size parameters of the weld pool, the first reflection time and the second reflection time are determined, wherein the first reflection time is the time it takes for the ultrasonic echo signal to reflect back to the surface of the weld pool, and the second reflection time is the time it takes for the ultrasonic echo signal to reflect back to the bottom of the weld pool. Based on the first reflection time and the second reflection time, other maxima are determined from the ultrasonic scanning data, wherein the other maxima are local maxima located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data; The true defect point is determined from the other maximum points; The echo signal characteristics of the real defect points are input into a trained long short-term memory network. The trained long short-term memory network outputs the defect type and fusion state, and the fusion parameters and optimization algorithm of the weld pool are adjusted based on the defect type and fusion state.

2. The method for monitoring the fusion of a weld pool according to claim 1, characterized in that, The step of determining other maxima points from the ultrasound scan data based on the first reflection time and the second reflection time specifically includes: For each ultrasound scan point, the local maximum point is determined using the derivative method. Arrange the local maxima points in ascending order of time to obtain an ascending time sequence; The time ascending sequence is segmented using the Otsu multi-threshold segmentation method to obtain extreme point segments; The mean of each extreme point segment is calculated, and the peak of the element in the first extreme point segment in the ultrasonic scanning data is determined as the surface echo of the molten pool, and the peak of the element in the second extreme point segment in the ultrasonic scanning data is determined as the bottom echo of the molten pool. The first extreme point segment is the extreme point segment with the smallest difference between the mean and the first reflection time, and the second extreme point segment is the extreme point segment with the smallest difference between the mean and the second reflection time. The local maxima located between the surface echo and the bottom echo of the molten pool in the ultrasonic scanning data are identified as the other maxima.

3. The method for monitoring the fusion of a weld pool according to claim 1, characterized in that, Determining the true defect point from the other maxima points specifically includes: Abnormal echo points are identified from the other maximum points, and the ultrasound scan data of the abnormal echo points are identified as abnormal echoes. New ultrasound scan data for each abnormal echo point is obtained. The new ultrasound scan data is obtained by emitting sound waves of a preset sound intensity range to each abnormal echo point, and the sound intensity of the preset sound intensity range increases sequentially. Determine the sequence of trend values ​​based on the newly added ultrasound scan data; Based on the changing trend value sequence, determine the amplitude sequence and time point sequence of abnormal echo points in the multiple scan data of each real defect candidate; The first actual defect point and the suspected defect point are determined based on the amplitude sequence and the time point sequence; Identify the second actual defect point from the suspected defect points; The first real defect point and the second real defect point are determined as the real defect points.

4. The method for monitoring the fusion of a weld pool according to claim 3, characterized in that, The step of determining the abnormal echo point from the other maxima points specifically includes: The upper envelope of the other maxima points is determined using cubic spline interpolation, and the fitting point of each point in the upper envelope is determined using the least squares method. The absolute value of the difference between each fitting point and the corresponding point in the upper envelope is calculated to obtain the absolute value of the difference. The ratio of the absolute value of the difference to the corresponding point in the upper envelope is calculated to obtain the convexity value. The echo anomaly probability of each point in the upper envelope is determined based on the convexity value, and points whose echo anomaly probability is greater than a preset anomaly threshold are identified as abnormal echo points.

5. The method for monitoring the fusion of a weld pool according to claim 3, characterized in that, The step of determining the trend value sequence based on the newly added ultrasound scan data specifically includes: For each abnormal echo point, the amplitude of each abnormal echo point in the newly added ultrasound scan data corresponding to each sound wave intensity is obtained to obtain an amplitude sequence; The trend value sequence is determined based on the amplitude sequence.

6. The method for monitoring the fusion of a weld pool according to claim 5, characterized in that, The step of determining the trend value sequence based on the amplitude sequence specifically includes: The element order corresponding to the elements of the first n amplitude sequences is used as the x-axis, and the element of each amplitude sequence is used as the y-axis to determine the principal component analysis coordinates. The projection vector and the projection value corresponding to each projection vector are determined by the principal component analysis method. The arctangent angle of the ratio of the y-axis to the x-axis of the projection vector with the largest projection value is determined as the first element in the trend value sequence, where n is a preset constant. The trend value sequence is obtained by sequentially determining each element in the trend value sequence.

7. The method for monitoring the fusion of a weld pool according to claim 3, characterized in that, The step of determining the amplitude sequence and time point sequence of abnormal echo points in multiple scan data of each real defect candidate based on the changing trend value sequence specifically includes: The Otsu multi-threshold segmentation method is used to segment the trend value sequence, determine two trend segmentation segments, calculate the variance of each trend segmentation segment, and obtain the sum of the variances of the two trend segmentation segments to obtain the false defect probability value. Based on the false defect probability value, the real defect candidate probability value is determined, and abnormal echo points whose real defect candidate probability value is greater than the preset defect threshold are determined as real defect candidates. Acquire multiple scan data for each real defect candidate, determine abnormal echo points in the multiple scan data, and obtain the amplitude sequence and time point sequence of each abnormal echo point in the multiple scan data.

8. The method for monitoring the fusion of a weld pool according to claim 3, characterized in that, The step of determining the first actual defect point and the suspected defect point based on the amplitude sequence and the time point sequence specifically includes: Calculate the coefficient of variation of the amplitude sequence and the standard deviation of the time point sequence. Identify the abnormal echo points whose coefficient of variation is less than a preset first abnormal threshold and whose standard deviation is less than a preset second abnormal threshold as the first real defect points. Identify the abnormal echo points in the multiple scan data other than the first real abnormal points as suspected defect points.

9. The method for monitoring the fusion of a weld pool according to claim 3, characterized in that, The step of determining the second actual defect point from the suspected defect points specifically includes: The final scan data of the suspected defect point under different emission energies are obtained, and the abnormal echo points in the final scan data are determined. The absolute value of the time difference between the time of each abnormal echo point in the final scan data and the first reflection time is calculated to obtain the time difference sequence of each abnormal echo point in the final scan data. Abnormal echo points whose time difference sequence is less than a preset variance threshold are identified as the second true defect points.

10. A welding pool fusion monitoring system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of a welding pool fusion monitoring method as described in any one of claims 1-9.

Images