Intelligent welding current adaptive regulation method and system

Abstract

The application provides a kind of intelligent welding current adaptive regulation method and system, belongs to industrial control system technical field.The method realizes the online accurate identification and classification of hidden welding defects such as internal incomplete penetration of weld, porosity, etc.by adopting deep convolutional neural network;At the same time, the current waveform data corresponding to the time interval of the welding defect is accurately matched by timestamp, the inverse mapping model of historical defect characteristics and current parameters is established and the current deviation prediction value is output, and the optimal current compensation and waveform correction function are iterated out by gradient descent optimization algorithm, which solves the problem of fuzzy mapping of defect space characteristics and current time parameters and lack of targetedness of parameter adjustment;Finally, the accurate current regulation instruction of the un-welded area is generated, and the adaptive adjustment of welding current according to different welding defect types is realized, so as to improve the welding quality.

Description

Technical Field

[0001] This application belongs to the field of industrial control system technology, specifically a method and system for intelligent adaptive adjustment of welding current. Background Technology

[0002] Currently, in the industrial manufacturing field, welding quality directly determines the reliability and lifespan of key components. To improve the stability of the welding process, many methods rely on preset process parameters or real-time adjustments based on external signals such as molten pool vision. However, these methods cannot directly perceive and respond to defects already formed inside the weld. Since welding is a dynamic thermal process, the generation of internal defects is closely related to complex states such as heat input and molten pool flow. Relying solely on external observation makes it difficult to accurately judge the true evolution of internal quality, resulting in a lack of targeted adjustments to welding parameters and thus reducing the stability of welding quality.

[0003] The above information in the background section of the invention is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the above problems, this application provides an intelligent welding current adaptive adjustment method and system to solve the problem of being unable to adaptively adjust the welding current to improve welding quality.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows: In a first aspect, embodiments of this application provide an intelligent welding current adaptive adjustment method, the method comprising: acquiring acoustic emission signals and first welding current waveform data during the welding process; determining the energy entropy values ​​of each frequency band sub-signal using a multi-scale decomposition method based on the acoustic emission signals, and determining a high-dimensional feature set based on each energy entropy value; determining one type of welding defect in the currently completed welding area using a deep convolutional neural network based on the high-dimensional feature set, the welding defect type including incomplete penetration and porosity; acquiring the time interval in which the current welding defect type occurs, and extracting second welding current waveform data corresponding to the time interval from the first welding current waveform data. The process involves: acquiring historical defect feature vectors and historical welding current waveform data, and establishing a welding current deviation prediction model based on these data; acquiring the defect feature vector corresponding to the current welding defect type, and inputting the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value; determining the welding current compensation amount and welding current waveform correction function using a gradient descent optimization algorithm based on the current welding current deviation prediction value; and determining the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function.

[0006] Secondly, embodiments of this application provide an intelligent welding current adaptive adjustment system, which includes: a first acquisition module, a first determination module, a second determination module, a second acquisition module, a construction module, a third determination module, a fourth determination module, and a fifth determination module. The first acquisition module is used to acquire acoustic emission signals and first welding current waveform data during the welding process; the first determination module is used to determine the energy entropy values ​​of each frequency band sub-signal based on the acoustic emission signals using a multi-scale decomposition method, and to determine a high-dimensional feature set based on the energy entropy values; the second determination module is used to determine one type of welding defect in the currently completed welding area based on the high-dimensional feature set using a deep convolutional neural network, wherein the welding defect type includes incomplete penetration and porosity; the second acquisition module is used to acquire the time interval in which the current welding defect type occurs, and to extract second welding current waveform data corresponding to the time interval from the first welding current waveform data; the construction module is used to acquire historical defect feature vectors. The system comprises five modules: a first module for determining the welding current deviation prediction value; a second module for determining the welding current deviation prediction value; a third module for determining the welding current deviation prediction value; a fourth module for determining the welding current compensation amount and the welding current waveform correction function based on the current welding current deviation prediction value; and a fifth module for determining the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function.

[0007] This application provides an intelligent welding current adaptive adjustment method and system. By employing a deep convolutional neural network, it achieves online and accurate identification and classification of hidden welding defects such as incomplete penetration and porosity within the weld. Simultaneously, by accurately matching welding defects with current waveform data of corresponding time intervals using timestamps, it establishes an inverse mapping model between historical defect characteristics and current parameters, outputting predicted current deviation values. Then, through gradient descent optimization algorithms, iteratively derives the optimal current compensation amount and waveform correction function, solving the problems of fuzzy mapping between defect spatial characteristics and current time parameters, and lack of targeted parameter adjustment. Finally, it generates precise current adjustment commands for unwelded areas, enabling adaptive adjustment of the welding current according to different welding defect types, thereby improving welding quality. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating an exemplary embodiment of the intelligent welding current adaptive adjustment method provided in this application.

[0010] Figure 2 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0011] Figure 3 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0012] Figure 4 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0013] Figure 5 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0014] Figure 6 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0015] Figure 7 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application.

[0016] Figure 8 This is a flowchart illustrating an intelligent welding current adaptive adjustment method provided in another exemplary embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0018] Currently, in the industrial manufacturing field, welding quality directly determines the reliability and lifespan of key components. To improve the stability of the welding process, many methods rely on preset process parameters or real-time adjustments based on external signals such as molten pool vision. However, these methods cannot directly perceive and respond to defects already formed inside the weld. Since welding is a dynamic thermal process, the generation of internal defects is closely related to complex states such as heat input and molten pool flow. Relying solely on external observation makes it difficult to accurately judge the true evolution of internal quality, resulting in a lack of targeted adjustments to welding parameters and thus reducing the stability of welding quality.

[0019] Specifically, internal weld defects such as incomplete penetration or porosity are characterized by delayed and concealed formation. By the time they are detected by traditional offline inspection, the defect location has already cooled, the process window has closed, and in-situ repair is impossible. Furthermore, because defect signals are spatially distributed, while parameters such as welding current act continuously over time, there is a lack of ability to precisely map the spatial characteristics of defects back to the specific time and instantaneous process parameters that caused them. Therefore, the correlation between the cause of the defect and the current welding parameters is ambiguous. This ambiguous correlation makes it difficult to determine, even if a welding defect is detected, which stage of the welding current parameter setting was improper, and what the specific form of the improper setting is.

[0020] Therefore, how to identify internal defects in welds online while simultaneously performing precise current compensation and current waveform correction for areas that have not yet been welded has become a technical problem that needs to be solved to shift welding quality from passive inspection to active closed-loop control.

[0021] This application provides an intelligent welding current adaptive adjustment method, such as... Figure 1 The illustrated intelligent welding current adaptive adjustment method may include the following steps: Step S110: Acquire acoustic emission signals and first welding current waveform data during the welding process; Step S120: Based on the acoustic emission signal, the energy entropy value of each frequency band sub-signal is determined by the multi-scale decomposition method, and the high-dimensional feature set is determined according to the energy entropy value; Step S130: Based on the high-dimensional feature set, a deep convolutional neural network is used to determine one of the welding defect types in the currently completed welding area. The welding defect types include incomplete penetration and porosity. Step S140: Obtain the time interval in which the current welding defect type occurs, and extract the second welding current waveform data corresponding to the time interval from the first welding current waveform data; Step S150: Obtain historical defect feature vectors and historical welding current waveform data, and establish a welding current deviation prediction model based on historical defect feature vectors and historical welding current waveform data; Step S160: Obtain the defect feature vector corresponding to the current welding defect type, and input the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value. Step S170: Based on the current predicted value of welding current deviation, the gradient descent optimization algorithm is used to determine the welding current compensation amount and the welding current waveform correction function; Step S180: Determine the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function.

[0022] According to the intelligent welding current adaptive adjustment method provided in this application, this method can achieve online accurate identification and classification of hidden welding defects such as incomplete penetration and porosity inside the weld by using a deep convolutional neural network. At the same time, it accurately matches welding defects with current waveform data of the corresponding time interval by timestamp, establishes an inverse mapping model between historical defect features and current parameters, and outputs the predicted value of current deviation. Then, it iteratively obtains the optimal current compensation amount and waveform correction function through gradient descent optimization algorithm, which solves the problems of fuzzy mapping between defect spatial features and current time parameters and lack of targeted parameter adjustment. Finally, it generates accurate current adjustment instructions for unwelded areas, realizing the adaptive adjustment of welding current according to different welding defect types, thereby improving welding quality.

[0023] The steps of the intelligent welding current adaptive adjustment method provided in this application are described in detail below: In one embodiment of this application, in step S110, acoustic emission signals and first welding current waveform data during the welding process are acquired.

[0024] Specifically, the acoustic emission signal during welding refers to the elastic stress wave signal generated by the dynamic behaviors of the weld material, such as micro-fractures, molten pool flow, and bubble escape. This signal can penetrate the base material and directly reflect the microscopic changes inside the weld, serving as a basis for identifying internal defects such as incomplete penetration and porosity. Acoustic emission sensors can be installed on the base material surface on both sides of the weld to ensure effective transmission of elastic waves, thereby continuously acquiring the acoustic emission signal sequence during the welding process. Simultaneously, the original signal undergoes preliminary filtering to remove background noise such as electromagnetic interference from the welding arc and mechanical vibration. The first welding current waveform data refers to the real-time current value sequence output by the welding power source during the welding process. This data can characterize the magnitude and variation of the heat input during welding. A Hall current sensor can be connected in series to the output circuit of the welding power source to ensure interference-free acquisition of the true value of the welding current. The acquired current data is marked and stored according to the same timestamp format as the acoustic emission signal, forming the first welding current waveform dataset. Simultaneously, a timestamp matching algorithm is used to align the time axis of the acoustic emission signal dataset with that of the first welding current waveform dataset.

[0025] For example, in a real-world welding scenario involving butt welding of low-carbon steel flat plates, the welding current is 200A and the welding speed is 5mm / s. An acoustic emission sensor with a frequency range of 20kHz to 1MHz is installed on the base material surface 5mm from the weld. The sampling rate is set to 2MS / s and the sampling accuracy to 16 bits. After starting the acquisition, the acoustic emission signal is continuously acquired for 10 seconds, resulting in an acoustic emission signal sequence containing 10 million sampling points. Subsequently, a Hall current sensor with a range of 0A to 500A is connected to the welding power supply output circuit. The sampling rate is set to 10kS / s, and synchronous triggering technology is used to start it simultaneously with the acoustic emission acquisition system, continuously acquiring welding current data for 10 seconds, resulting in the first welding current waveform data containing 100,000 sampling points.

[0026] In one embodiment of this application, step S120, which determines the energy entropy value of each frequency band sub-signal based on the acoustic emission signal using a multi-scale decomposition method, further includes the following steps: Figure 2 As shown, the specific content is as follows: Step S210: Based on the acoustic emission signal, use the multi-scale decomposition method to determine multiple frequency band sub-signals with the same bandwidth; Step S220: Calculate the frequency band energy ratio of each frequency band sub-signal, and based on the frequency band energy ratio, use the Shannon energy entropy algorithm to determine the energy entropy value of each frequency band sub-signal.

[0027] Specifically, in the multi-scale decomposition method, this application employs wavelet packet transform as its implementation algorithm. This algorithm can uniformly decompose the original acoustic emission signal across the entire frequency range, obtaining multiple frequency band sub-signals with identical bandwidths. This overcomes the shortcomings of traditional wavelet transform, which can only subdivide low-frequency signals and has insufficient resolution for high-frequency signals. It can completely capture the full-band acoustic emission characteristics generated by different defects inside the weld. The frequency band energy ratio refers to the proportion of the energy of a single frequency band sub-signal to the total energy of all frequency band sub-signals, used to quantify the distribution of acoustic emission signal energy in different frequency intervals. The Shannon energy entropy algorithm is a mathematical method used to quantify the degree of disorder in the signal energy distribution. A larger entropy value indicates a more dispersed signal energy distribution and a more complex signal, corresponding to a higher probability of defects inside the weld. A smaller entropy value indicates a more concentrated signal energy distribution and a more uniform signal, corresponding to a normal state inside the weld.

[0028] For example, the acoustic emission signal can be decomposed into 3-level wavelet packets using the db4 wavelet basis function to obtain 8 frequency band sub-signals, each with a bandwidth of 250kHz; then, the energy of each of the 8 frequency band sub-signals can be calculated to obtain the energy of each frequency band. , , , , , , , Total energy Calculate the energy percentage of each frequency band to obtain... , , , , , , , Based on Shannon's energy entropy formula, the energy entropy values ​​of each frequency band sub-signal are: H1=1.24, H2=1.57, H3=1.82, H4=2.05, H5=2.18, H6=2.31, H7=2.45, H8=2.52.

[0029] In one embodiment of this application, step S120, determining the high-dimensional feature set based on each energy entropy value, further includes the following steps: Figure 3 As shown, the specific content is as follows: Step S310: Based on the acoustic emission signal, determine multiple intrinsic mode functions using the empirical mode decomposition method; Step S320: Extract the dominant frequency and amplitude of each intrinsic mode function, and determine the multidimensional feature vector based on the dominant frequency, amplitude and each energy entropy value; Step S330: Normalize the multidimensional feature vectors to generate a high-dimensional feature set.

[0030] Specifically, Empirical Mode Decomposition (EMD) is an adaptive signal decomposition method for processing nonlinear and non-stationary signals. It can decompose complex acoustic emission signals into multiple intrinsic mode functions (EMFs) with different time scales. Each EMF represents the oscillation component of the acoustic emission signal at a specific time scale, effectively capturing the transient non-stationary characteristics generated by dynamic behaviors such as molten pool flow and micro-fractures during welding. This overcomes the shortcomings of wavelet packet transform, which is based on fixed basis function decomposition and cannot adaptively match local signal characteristics. Next, a Hilbert transform is performed on each EMF to obtain its instantaneous frequency and amplitude sequence. The energy spectrum of each EMF is calculated, and the frequency with the highest energy proportion is determined as the dominant frequency of that EMF. Simultaneously, the maximum value of the instantaneous amplitude sequence is extracted as the amplitude of that EMF. The energy entropy values ​​of each frequency band sub-signal obtained in step S220 are sequentially concatenated with the dominant frequency and amplitude of each EMF to form a multidimensional feature vector. For example, if wavelet packet decomposition yields the energy entropy values ​​of 8 frequency band sub-signals, and empirical mode decomposition yields 5 intrinsic mode functions, then each intrinsic mode function corresponds to 2 features (dominant frequency, amplitude), and the multidimensional feature vector has a dimension of 8 + 5 × 2 = 18 dimensions. Furthermore, based on a large number of normal and defective samples collected during historical welding processes, the minimum and maximum values ​​of each feature dimension are calculated as benchmark parameters for normalization. For each feature value in the multidimensional feature vector, the minimum-maximum normalization formula is used to map the feature value to the [0,1] interval. All normalized feature values ​​are recombine in their original order to form a standardized high-dimensional feature vector; the high-dimensional feature vectors from multiple consecutively collected time windows are collected and stored to generate the final high-dimensional feature set, which is used for subsequent online defect identification by a deep convolutional neural network.

[0031] For example, for an acoustic emission signal with a sampling rate of 2 MS / s, a 1 ms segment is extracted for analysis. The energy entropy values ​​of the eight frequency band sub-signals are 1.24, 1.57, 1.82, 2.05, 2.18, 2.31, 2.45, and 2.52, respectively. The empirical mode decomposition method is used to adaptively decompose this 1 ms acoustic emission signal. After seven iterations, five intrinsic mode functions (IMFs) satisfying the conditions are obtained, labeled IMF1, IMF2, IMF3, IMF4, and IMF5 in descending order of frequency. IMF1 corresponds to the highest frequency transient micro-fracture signal, and IMF5 corresponds to the lowest frequency molten pool flow steady-state signal. Subsequently, a Hilbert transform was performed on each intrinsic mode function, yielding the following results: IMF1 has a dominant frequency of 620 kHz and an amplitude of 0.9 V; IMF2 has a dominant frequency of 380 kHz and an amplitude of 0.7 V; IMF3 has a dominant frequency of 210 kHz and an amplitude of 0.5 V; IMF4 has a dominant frequency of 120 kHz and an amplitude of 0.3 V; and IMF5 has a dominant frequency of 50 kHz and an amplitude of 0.2 V. Therefore, the energy entropy values ​​of the eight frequency band sub-signals can be sequentially concatenated with the dominant frequency and amplitude of the five intrinsic mode functions to obtain an 18-dimensional multidimensional feature vector: [1.24,1.57,1.82,2.05,2.18,2.31,2.45,2.52,620,0.9,380,0.7,210,0.5,120,0.3,50,0.2]. Finally, the normalized baseline parameters are called, where the minimum value of energy entropy is 1.0 and the maximum value is 3.5, the minimum value of the main frequency is 0kHz and the maximum value is 1000kHz, and the minimum value of the amplitude is 0V and the maximum value is 1V. The multidimensional feature vector is subjected to feature-wise minimum-maximum normalization, mapping all feature values ​​to the interval [0,1]. The final normalized 18-dimensional high-dimensional feature vector is: [0.096,0.228,0.328,0.42,0.472,0.524,0.58,0.608,0.62,0.9,0.38,0.7,0.21,0.5,0.12,0.3,0.05,0.2].

[0032] In the above method, the acoustic emission signal is adaptively decomposed by empirical mode decomposition, which can effectively extract the nonlinear and non-stationary transient features in the welding process, make up for the limitations of wavelet packet transform based on fixed basis function decomposition, and fully capture the oscillation characteristics of different defects such as incomplete penetration and porosity at different time scales. By fusing the energy entropy values ​​of each frequency band sub-signal with the dominant frequency and amplitude features of each intrinsic mode function, the internal state of the welding can be comprehensively characterized from multiple dimensions such as frequency domain energy distribution and time-frequency domain oscillation characteristics, which greatly improves the discriminability and anti-interference ability of the feature set and avoids the misjudgment problem caused by the susceptibility of a single feature to noise.

[0033] In one embodiment of this application, after determining the energy entropy value of each frequency band sub-signal using the multi-scale decomposition method in step S120, the following step is further included: if the energy entropy value of the current frequency band sub-signal exceeds a preset energy entropy threshold, the frequency band sub-signal is decomposed again using the multi-scale decomposition method to generate multiple frequency band signal units. Specifically, when the energy entropy value of a single frequency band sub-signal exceeds its corresponding preset energy entropy threshold, it indicates that the signal energy distribution within the frequency band is extremely dispersed, and may contain multiple characteristic signals or strong noise components with different defects. The initial coarse-grained decomposition alone cannot effectively separate these features, and a secondary multi-scale decomposition is required. After the secondary decomposition, each frequency band sub-signal will be divided into two frequency band signal units with identical bandwidths, and the bandwidth of each frequency band signal unit is half the bandwidth of the original frequency band sub-signal. The preset energy entropy threshold can be set based on the energy entropy statistics of more than 1,000 historical normal welding samples. The threshold is set using the 3σ criterion, that is, the preset energy entropy threshold for each frequency band is equal to the average value of the energy entropy of that frequency band under normal welding conditions plus 3 times the standard deviation. This setting can cover 99.7% of normal welding conditions, ensuring that only significantly abnormal high entropy frequency bands will trigger secondary decomposition.

[0034] In one embodiment of this application, step S130 involves using a deep convolutional neural network based on a high-dimensional feature set to determine one type of welding defect in the currently completed welding area. The welding defect types include incomplete penetration and porosity. The process also includes the following steps: Figure 4 As shown, the specific content is as follows: Step S410: Based on the high-dimensional feature set, a deep convolutional neural network is used to determine the confidence level corresponding to each welding defect type in the completed welding area; Step S420: Sort the confidence levels in order from the maximum to the minimum, and take the welding defect type corresponding to the maximum confidence level as the welding defect type of the currently completed welding area.

[0035] Specifically, high-dimensional feature vectors are input into the input layer of a deep convolutional neural network, with the input layer's dimension matching that of the high-dimensional feature set. The deep convolutional neural network model then extracts abstract semantic information from the features layer by layer through three convolutional blocks and two pooling layers, strengthening the expression of defect-related features. Finally, after classification mapping through two fully connected layers, it outputs confidence values ​​corresponding to three states: incomplete penetration, porosity, and normal. Simultaneously, each confidence value is associated with and stored with its corresponding type label. Next, the three confidence values ​​are sorted in descending order to obtain a sorted confidence sequence. Then, the maximum value in the sorted confidence sequence is extracted and compared with a preset confidence threshold. If the maximum value is greater than or equal to the preset threshold, the type corresponding to that maximum value is taken as the welding defect type of the currently completed welding area. If the maximum value is less than the preset threshold, the currently completed welding area is determined to be in a normal, defect-free state, and no subsequent parameter compensation process needs to be initiated. The preset confidence threshold is set to ensure that all defective areas can be accurately identified, preventing substandard welds from flowing into subsequent processes and ensuring the reliability of the welded structure.

[0036] For example, the 18-dimensional high-dimensional feature vector is [0.096, 0.228, 0.328, 0.42, 0.472, 0.524, 0.58, 0.608, 0.62, 0.9, 0.38, 0.7, 0.21, 0.5, 0.12, 0.3, 0.05, 0.2]. The deep convolutional neural network adopts the ResNet-50 (Residual Network) architecture, has been trained based on more than 100,000 historical samples, and has been set up in an industrial control system with a preset confidence threshold of 0.6. The 18-dimensional high-dimensional feature vector is input into a deep convolutional neural network. After multiple layers of convolution, pooling, and fully connected layers, the model outputs a confidence score of 0.12 for incomplete penetration, 0.85 for porosity, and 0.03 for normality. The sum of the three confidence scores is 1.0. Then, the three confidence scores are sorted in descending order, resulting in a sorted sequence of 0.85 (porosity), 0.12 (incomplete penetration), and 0.03 (normality). The maximum value of 0.85 is extracted and compared with a preset confidence threshold of 0.6. Since 0.85 is greater than 0.6, the welding defect type of the currently completed welding area is determined to be porosity.

[0037] In the above method, a deep convolutional neural network trained with a large number of historical normal samples and samples with incomplete penetration and porosity defects is used to perform deep feature extraction and classification on the high-dimensional feature set. This method can automatically learn the complex feature patterns corresponding to different welding defects, effectively capture the implicit correlation between multi-dimensional features, and significantly improve the accuracy of defect identification and the ability to resist industrial noise interference.

[0038] In one embodiment of this application, in step S140, the time interval in which the current welding defect type occurs is obtained, and the second welding current waveform data corresponding to the time interval is extracted from the first welding current waveform data.

[0039] Specifically, the time interval for the occurrence of the current welding defect type refers to the complete time range from the initial formation of the welding defect to its complete solidification. This time interval is earlier than the timestamp when the defect is identified by the deep convolutional neural network, which is determined by the physical characteristics of the welding process. Welding defects form inside the molten pool and only generate micro-fracture signals that can be captured by acoustic emission sensors after the molten pool solidifies. Therefore, there is an inherent time lag in defect identification.

[0040] Furthermore, the precise timestamp of the current welding defect identified by the deep convolutional neural network can be extracted. This timestamp shares the same clock source as the timestamp of the first welding current waveform data. Using the defect identification timestamp as the endpoint, a preset time backtracking length is used to obtain the starting timestamp of the defect occurrence. The time interval between the starting timestamp and the defect identification timestamp is the time range in which the current welding defect type occurs. The preset time backtracking length can be set according to the solidification time of the molten pool for different materials, as the solidification time varies. For example, the solidification time of a low-carbon steel molten pool is approximately 1-3 seconds, and for stainless steel, it is approximately 2-5 seconds. The time backtracking length needs to cover the entire process from the formation of the molten pool to complete solidification. Next, using the starting and ending timestamps of the defect occurrence time range as indexes, precise positioning is performed in the circular buffer of the first welding current waveform data to determine the starting and ending storage addresses of the corresponding data segments. Based on the located storage addresses, all current sampling point data within the corresponding interval are extracted to form the second welding current waveform data.

[0041] For example, the timestamp for identifying the current welding defect as porosity is extracted as 2023-10-15 14:30:25.123. Based on a preset time backtracking length of 10 seconds, the starting timestamp for the occurrence of porosity is calculated to be 2023-10-15 14:30:15.123. Therefore, the time interval for the occurrence of the current porosity defect is from 2023-10-15 14:30:15.123 to 2023-10-15 14:30:25.123. Subsequently, using the two timestamps within this time interval as indices, the corresponding data segment is located in the annular buffer of the first welding current waveform data. All current sampling points within this interval are extracted, resulting in a total of 100,000 sampling points. After smoothing through a sliding window, the second welding current waveform data is obtained. This data completely records the real-time changes in welding current during the formation of porosity defects.

[0042] In one embodiment of this application, in step S150, historical defect feature vectors and historical welding current waveform data are obtained, and a welding current deviation prediction model is established based on the historical defect feature vectors and historical welding current waveform data.

[0043] Specifically, historical defect feature vectors are standardized high-dimensional feature vectors stored in the industrial control system, corresponding to two typical defects in historical welding processes: incomplete penetration and porosity. These historical defect feature vectors can include normalized features such as the energy entropy values ​​of sub-signals in each frequency band, the dominant frequency and amplitude of the intrinsic mode function, and can comprehensively characterize the signal characteristics of the defects. Historical welding current waveform data refers to the welding current time-series data within the defect occurrence time interval, corresponding one-to-one with the historical defect feature vectors. This data is synchronously acquired by Hall current sensors, with sampling rate and accuracy completely consistent with the first welding current waveform data of the current welding process. The historical defect feature vectors can be used as input to a deep neural network model, and the corresponding historical current deviation values ​​can be used as labels for the deep neural network model. Through iterative training via forward and backward propagation, the weights and biases of the deep neural network are continuously optimized until the loss function converges, ultimately generating a welding current deviation prediction model that can be deployed in the industrial control system for subsequent prediction of current deviations corresponding to the current welding defect type.

[0044] In one embodiment of this application, step S160 involves obtaining the defect feature vector corresponding to the current welding defect type, and inputting the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value. The method also includes the following steps: Figure 5 As shown, the specific content is as follows: Step S510: If the current welding defect type is incomplete penetration, the defect feature vector corresponding to incomplete penetration includes the length and depth of incomplete penetration; Step S520: If the current welding defect type is porosity, then the defect feature vector corresponding to porosity is the porosity diameter and porosity location.

[0045] Specifically, incomplete penetration refers to the defect where the root of the weld is not fully fused and connected during welding, caused by insufficient heat input. If the current welding defect type is incomplete penetration, the corresponding defect feature vector will always include incomplete penetration length and incomplete penetration depth. Incomplete penetration length refers to the extension dimension of the unfused area at the weld root along the weld length direction, representing the coverage area of ​​the incomplete penetration defect. Incomplete penetration depth refers to the vertical depth of the unfused area along the weld cross-section, a parameter characterizing the severity of the incomplete penetration defect, obtained through inversion of the dominant frequency characteristics of the acoustic emission intrinsic mode functions.

[0046] Furthermore, porosity refers to voids formed by air bubbles failing to escape from the molten pool during welding. It is caused by current fluctuations and abnormal shielding gas during the welding process. If the current welding defect type is porosity, the corresponding defect feature vector includes the porosity diameter and porosity location. The porosity diameter refers to the equivalent circular diameter of a single porosity, a parameter characterizing the size of the porosity defect. The porosity location refers to the lateral offset distance of the porosity center relative to the weld centerline, a parameter characterizing the distribution location of the porosity defect.

[0047] Furthermore, the defect feature vector and the second welding current waveform data are simultaneously input into the welding current deviation prediction model. Through the forward inference operation of the welding current deviation prediction model, the current welding current deviation prediction value is output. The current welding current deviation prediction value refers to the difference between the actual welding current that causes the current welding defect and the standard process welding current.

[0048] For example, if the current porosity defect has a pore diameter of 0.8 mm and a pore location 1.2 mm to the right of the weld center, the pore diameter of 0.8 mm and the pore location of 1.2 mm are combined to form a two-dimensional defect feature vector corresponding to the pore. This defect feature vector, along with the second welding current waveform data, is then synchronously input into the welding current deviation prediction model. Through forward inference calculation using an LSTM (Long Short-Term Memory) network, the final output welding current deviation prediction value is -18A, indicating that the actual welding current is 18A lower than the standard process current when the current porosity defect occurs. If the current welding defect type is incomplete penetration, with an incomplete penetration length of 3.2 mm and an incomplete penetration depth of 1.1 mm, this is combined to form the corresponding defect feature vector. After inputting this vector into the welding current deviation prediction model, the output welding current deviation prediction value is +22A, indicating that the actual welding current is 22A higher than the standard process current.

[0049] In one embodiment of this application, step S170, based on the current predicted welding current deviation value, uses a gradient descent optimization algorithm to determine the welding current compensation amount and the welding current waveform correction function, and further includes the following steps: Figure 6 As shown, the specific content is as follows: Step S610: Obtain the welding current waveform correction function corresponding to the current welding defect type; Step S620: With the goal of eliminating incomplete penetration or porosity, the relationship between the defect incidence rate and the welding current compensation amount is used as the loss function. The gradient descent optimization algorithm is used to continuously update the predicted value of welding current deviation and the welding current waveform correction function until the loss function converges. Step S630: Use the predicted value of welding current deviation corresponding to the converged loss function as the welding current compensation amount, and use the welding current waveform correction function corresponding to the converged loss function as the final welding current waveform correction function.

[0050] Specifically, the welding current waveform correction function is used to dynamically correct the original welding current waveform data and match the timing of defect formation. Different welding defects have different causes, and therefore different initial welding current waveform correction functions. For example, incomplete penetration is caused by insufficient welding heat input and insufficient penetration depth; the initial waveform correction function uses low-frequency DC compensation as its core to smoothly increase heat input. Porosity is caused by drastic current fluctuations and the inability of bubbles in the molten pool to escape in time; the initial waveform correction function uses high-frequency damping current stabilization as its core to suppress current oscillations and stabilize the molten pool state. The initial welding current waveform correction function corresponding to incomplete penetration is as follows: ,in It is a low-frequency frequency, matching the solidification cycle of the molten pool; This is a constant, preset to 0.1 to avoid overcompensation. The correction function for the initial welding current waveform corresponding to porosity is: ,in The attenuation coefficient is... For high frequency, It is a constant, preset to 0.05, to adapt to the dynamic characteristics of bubble escape.

[0051] Furthermore, with the optimization objective of completely eliminating incomplete penetration or porosity defects, a standardized loss function is constructed. ,in The historical incidence rate of similar defects, This is the predicted value for the current welding current deviation. This is the theoretically optimal welding current compensation amount. where is the weighting coefficient. Following the backpropagation rule of the gradient descent algorithm, all parameters of the current predicted welding current deviation and the welding current waveform correction function are updated synchronously, continuously reducing the value of the loss function until the loss function meets the convergence condition. The convergence threshold of the loss function is set to . Convergence is defined as the difference between the loss functions of two consecutive iterations being less than a certain threshold. When the loss function reaches convergence, the corresponding parameters have reached the optimal state for eliminating defects. At this point, the predicted value of the current welding current deviation after convergence is directly determined as the welding current compensation amount; at the same time, the welding current waveform correction function after convergence is directly determined as the final welding current waveform correction function.

[0052] For example, the initial welding current waveform correction function corresponding to porosity defects can be invoked through the industrial control system. Subsequently, with the elimination of porosity defects as the optimization objective, a loss function was adopted. Set the learning rate to 0.01, the maximum number of iterations to 100, and the convergence threshold to... The gradient descent optimization algorithm is used to iteratively update the parameters of the current predicted welding current deviation and the initial welding current waveform correction function. At the 42nd iteration, the difference in the loss function is... If the value is less than the convergence threshold, the loss function is considered convergent. Finally, the predicted current welding current deviation of -20A is used as the welding current compensation amount, and the converged welding current waveform correction function is applied. This serves as the final correction function for the welding current waveform.

[0053] In the above method, by matching the initial welding current waveform correction function to the formation mechanism of two types of defects, namely incomplete penetration and porosity, the optimization starting point can be made to fit the actual cause of the defect, avoiding ineffective iteration. The loss function constructed with the goal of eliminating defects and combining the relationship between the defect occurrence rate and the compensation amount can balance the rationality of compensation while prioritizing the defect elimination effect. The learning rate, number of iterations and convergence threshold of the gradient descent algorithm are preset according to the requirements of industrial real-time control and accuracy, which can complete iterative optimization and achieve stable convergence in milliseconds. The final welding current compensation amount and the final welding current waveform correction function can accurately match the current anomaly characteristics corresponding to the defect.

[0054] In one embodiment of this application, step S180, which determines the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function, further includes the following steps: Figure 7 As shown, the specific content is as follows: Step S710: Determine the welding current correction amount based on the welding current waveform correction function; Step S720: Obtain the welding current value of the currently unfinished welding area, and sum the current welding current value, welding current correction amount, and welding current compensation amount respectively; Step S730: Use the result of the summation operation as the adjusted welding current value, and generate a welding current adjustment command for the unfinished welding area based on the adjusted welding current value.

[0055] Specifically, the time value of the current welding moment in the unfinished welding area can be substituted into the final welding current waveform correction function. According to the function's operation rules, the instantaneous welding current correction amount at the corresponding moment is calculated. This correction amount is a numerical value with a positive and negative sign; a positive value indicates that the welding current needs to be increased, and a negative value indicates that the welding current needs to be decreased. This allows for bidirectional dynamic adjustment of the welding current. Therefore, the adjusted welding current value = welding current value of the current unfinished welding area + welding current compensation amount + welding current correction amount. The welding current compensation amount is a static deviation correction value used to eliminate welding defects and correct overall current deviation; the welding current correction amount is a time-series dynamic fine-tuning value used to correct instantaneous current fluctuations. The summation of these two values ​​achieves a dual adjustment effect of static correction and dynamic current stabilization. Finally, the adjusted welding current value obtained from the summation operation is encapsulated according to a preset industrial communication protocol format to generate a complete welding current adjustment command containing the target current, verification information, and execution time. This command can be directly parsed and executed by the welding power controller for real-time control of the welding current in the unfinished welding area.

[0056] For example, in an industrial application scenario of butt welding of low-carbon steel plates, the thickness of the plates to be welded is 8mm, the welding speed is 5mm / s, and the reference welding current value for the currently incomplete welding area stored in the industrial control system is 200A. The converged welding current compensation is -20A, and the final welding current waveform correction function is... The current welding time of the incomplete welding area is recorded according to a preset time sampling step of 1 millisecond. Substituting into the welding current waveform correction function, the welding current correction amount is calculated as follows: Then, the baseline welding current value of 200A for the currently incomplete welding area is obtained and summed as follows: Using 160.07A as the adjusted welding current value, and following industrial communication protocol specifications, a welding current adjustment command for the incomplete welding area is generated, including the target welding current value of 160.07A, a CRC16 checksum, and an execution timestamp of 2023-10-15 14:30:25.124. This command can be transmitted to the welding power controller in real time via industrial Ethernet, and the controller will immediately execute the current adjustment operation after parsing it.

[0057] In one embodiment of this application, after determining the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function in step S180, the following steps are further included: Figure 8 As shown, the specific content is as follows: Step S810: Obtain the arc length, molten pool center temperature, and molten pool edge temperature during the welding process; Step S820: If the arc length does not exceed the preset arc length threshold, increase the welding voltage during the welding process; if the arc length exceeds the preset arc length threshold, decrease the welding voltage during the welding process. Step S830: If the temperature deviation between the center temperature and the edge temperature of the molten pool exceeds the preset temperature deviation threshold, the welding current during the welding process is reduced; if the temperature deviation between the center temperature and the edge temperature of the molten pool does not exceed the preset temperature deviation threshold, the welding current during the welding process is increased.

[0058] Specifically, the arc length can be indirectly acquired through the arc voltage feedback module built into the welding power source, and the arc length and arc voltage are linearly positively correlated. The temperature at the center of the molten pool can be acquired by a first infrared temperature sensor targeting the highest temperature point at the center of the molten pool, reflecting the overall melting degree and heat input level of the molten pool. The temperature at the edge of the molten pool can be acquired by a second infrared temperature sensor targeting the interface between the molten pool and the solid base material, reflecting the solidification boundary and thermal diffusion state of the molten pool.

[0059] Furthermore, the real-time arc length is compared with a preset arc length threshold. If the arc length does not exceed the preset threshold, the arc is considered too short. An excessively short arc can easily cause short circuits, sticking, and uneven penetration. Therefore, the industrial control system outputs a voltage adjustment command, increasing the welding voltage in 0.5V increments until the arc length returns to within the preset threshold. If the arc length exceeds the preset threshold, the arc is considered too long. An excessively long arc can easily cause arc breakage, porosity, and increased spatter. Therefore, the industrial control system outputs a voltage adjustment command, decreasing the welding voltage in 0.5V increments until the arc length returns to within the preset threshold. The preset arc length threshold needs to be set according to the commonly used gas metal arc welding (GMAW) process standards for different welding materials, and also needs to match the voltage adjustment accuracy of the industrial welding power supply to ensure that the adjustment step can accurately control the arc length change.

[0060] Furthermore, the temperature deviation value is equal to the absolute value of the difference between the temperature at the center of the molten pool and the temperature at the edge of the molten pool. The calculated temperature deviation value is compared with a preset temperature deviation threshold. If the temperature deviation value exceeds the preset threshold, it indicates that the thermal field of the molten pool is uneven. Excessive temperature difference can lead to overheating at the center of the molten pool and incomplete fusion at the edges, easily resulting in defects such as incomplete penetration, porosity, and cracks. Therefore, the industrial control system outputs a current adjustment command, reducing the welding current during the welding process in steps of 2A / time to reduce the overall heat input and narrow the temperature difference between the center and edge of the molten pool. If the temperature deviation value does not exceed the preset temperature deviation threshold, it indicates that a small temperature difference will lead to poor molten pool fluidity, rough weld formation, and insufficient penetration. Therefore, the industrial control system outputs a current adjustment command, appropriately increasing the welding current during the welding process in steps of 1A / time to enhance molten pool fluidity and penetration consistency, and improve weld density. The setting of the preset temperature deviation threshold needs to match the measurement accuracy of the infrared temperature sensor to avoid incorrect adjustments due to measurement errors.

[0061] For example, by synchronously acquiring data through the welding power supply voltage feedback module and dual-channel infrared temperature sensors, the real-time arc length was obtained as 2.4 mm, the center temperature of the molten pool was 1680℃, and the edge temperature of the molten pool was 1430℃. Subsequently, the arc length of 2.4 mm was compared with the preset arc length threshold of 3 mm. Since 2.4 mm did not exceed 3 mm, the arc was determined to be too short. The industrial control system increased the welding voltage in steps of 0.5 V / cycle, raising the welding voltage from 25 V to 26.5 V. After adjustment, the arc length stabilized at 3.1 mm, which is within the optimal working range. Next, the temperature deviation value is calculated as 1680℃ minus 1430℃, which equals 250℃. 250℃ is compared with the preset temperature deviation threshold of 200℃. Since 250℃ exceeds 200℃, it is determined that the thermal field of the molten pool is not uniform. The industrial control system reduces the welding current in steps of 2A / cycle, reducing the welding current by 12A from 179.68A to 167.68A. After adjustment, the temperature at the center of the molten pool drops to 1620℃, and the temperature at the edge of the molten pool rises to 1435℃. The temperature deviation is reduced to 185℃, returning to the preset temperature deviation threshold range, thus achieving stable arc combustion and uniform control of the thermal field of the molten pool.

[0062] In the above method, by collecting the arc length, molten pool center temperature, and molten pool edge temperature during the welding process, the stability of the welding arc and the distribution characteristics of the molten pool thermal field can be reflected in real time and accurately. The welding voltage can be automatically adjusted according to the arc length, which can continuously maintain the arc length within the optimal working range, effectively avoiding defects such as short circuits, arc breaks, excessive spatter, and porosity. The welding current can be automatically adjusted according to the temperature deviation between the molten pool center and edge, which can keep the molten pool thermal field in a uniform and reasonable state, improve the fluidity of the molten pool and the consistency of the weld depth, effectively suppress problems such as incomplete penetration, lack of fusion, overheating, and cracks, further improving the stability of the welding process, the weld formation quality, and the defect suppression capability, and significantly improving the quality consistency and reliability of industrial welding production.

[0063] This application also provides an intelligent welding current adaptive adjustment system, which may include a first acquisition module, a first determination module, a second determination module, a second acquisition module, a construction module, a third determination module, a fourth determination module, and a fifth determination module. The first acquisition module is used to acquire acoustic emission signals and first welding current waveform data during the welding process; the first determination module is used to determine the energy entropy values ​​of each frequency band sub-signal based on the acoustic emission signals using a multi-scale decomposition method, and to determine a high-dimensional feature set based on the energy entropy values; the second determination module is used to determine one of the welding defect types in the currently completed welding area based on the high-dimensional feature set using a deep convolutional neural network, where the welding defect types include incomplete penetration and porosity; the second acquisition module is used to acquire the time interval of the current welding defect type and extract the corresponding time interval of second welding current waveform data from the first welding current waveform data; the construction module is used to acquire historical defect characteristics... The system employs five modules: a first module to obtain the defect feature vector and historical welding current waveform data, and a second module to establish a welding current deviation prediction model based on these data; a third module to obtain the defect feature vector corresponding to the current welding defect type, and inputs the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value; a fourth module to determine the welding current compensation amount and welding current waveform correction function based on the current welding current deviation prediction value using a gradient descent optimization algorithm; and a fifth module to determine the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function.

[0064] It should be noted that the embodiments of the intelligent welding current adaptive adjustment system provided in this application can be used to execute the processing flow of the embodiments of the intelligent welding current adaptive adjustment method in the above embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above method embodiments.

[0065] This application also provides an electronic device including one or more processors and memory resources represented by memory for storing instructions executable by the processor, such as application programs. The application programs stored in the memory may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor is configured to execute instructions to perform the aforementioned intelligent welding current adaptive adjustment method. The electronic device may also include a power supply component configured to perform power management of the electronic device, a wired or wireless network interface configured to connect the electronic device to a network, and an input / output (I / O) interface. The electronic device can be operated based on operating devices stored in the memory, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0066] In one embodiment, a computer device, which may be a server, is also provided. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device stores data. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent welding current adaptive adjustment method.

[0067] In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an intelligent welding current adaptive adjustment method. The display unit of the computer device is used to form a visually visible image and may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0068] This application also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of the electronic device, enables the electronic device to perform an intelligent welding current adaptive adjustment method.

[0069] This application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0070] It should be noted that although the steps of the intelligent welding current adaptive adjustment method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps, such as omitting certain steps, combining multiple steps into one step, and / or breaking down one step into multiple steps, should all be considered part of this application.

[0071] It should be understood that this application is not limited to the detailed structure and arrangement of the modules of the intelligent welding current adaptive adjustment system proposed in this specification. This application can have other embodiments and can be implemented and executed in various ways. The foregoing variations and modifications fall within the scope of this application. It should be understood that the invention and definition of this application extend to all alternative combinations of two or more individual features mentioned or apparent in the text and / or drawings. All these different combinations constitute multiple alternative aspects of this application. The embodiments described in this specification illustrate the best known mode for implementing this application and will enable those skilled in the art to utilize this application.

[0072] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for intelligent adaptive adjustment of welding current, characterized in that, include: Acquire acoustic emission signals and first welding current waveform data during the welding process; Based on the acoustic emission signal, the energy entropy value of each frequency band sub-signal is determined by the multi-scale decomposition method, and a high-dimensional feature set is determined according to each energy entropy value; Based on the high-dimensional feature set, a deep convolutional neural network is used to determine one of the welding defect types in the currently completed welding area, the welding defect types including incomplete penetration and porosity; Obtain the time interval in which the current welding defect type occurs, and extract the second welding current waveform data corresponding to the time interval from the first welding current waveform data; Obtain historical defect feature vectors and historical welding current waveform data, and establish a welding current deviation prediction model based on the historical defect feature vectors and historical welding current waveform data; Obtain the defect feature vector corresponding to the current welding defect type, and input the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value; Based on the current predicted value of welding current deviation, the gradient descent optimization algorithm is used to determine the welding current compensation amount and the welding current waveform correction function. The welding current adjustment command for the unfinished welding area is determined based on the welding current compensation amount and the welding current waveform correction function.

2. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The determination of the energy entropy value of each frequency band sub-signal using a multi-scale decomposition method based on the acoustic emission signal includes: Based on the acoustic emission signal, a multi-scale decomposition method is used to determine multiple frequency band sub-signals with the same bandwidth; Calculate the frequency band energy ratio of each frequency band sub-signal, and based on the frequency band energy ratio, use the Shannon energy entropy algorithm to determine the energy entropy value of each frequency band sub-signal.

3. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The determination of the high-dimensional feature set based on each of the energy entropy values ​​includes: Based on the acoustic emission signal, multiple intrinsic mode functions are determined using the empirical mode decomposition method; Extract the dominant frequency and amplitude of each intrinsic mode function, and determine a multidimensional feature vector based on the dominant frequency, the amplitude, and each energy entropy value; The multidimensional feature vectors are normalized to generate the high-dimensional feature set.

4. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, After determining the energy entropy values ​​of each frequency band sub-signal using a multi-scale decomposition method based on the acoustic emission signal, the method further includes: If the energy entropy value of the current frequency band sub-signal exceeds the preset energy entropy value threshold, the frequency band sub-signal is decomposed again using the multi-scale decomposition method to generate multiple frequency band signal units.

5. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The step of determining one type of welding defect in the currently completed welding area using a deep convolutional neural network based on the high-dimensional feature set includes: Based on the high-dimensional feature set, a deep convolutional neural network is used to determine the confidence level corresponding to each welding defect type in the completed welding area. The confidence scores are sorted sequentially from the maximum to the minimum, and the welding defect type corresponding to the maximum confidence score is taken as the welding defect type of the currently completed welding area.

6. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The step of obtaining the defect feature vector corresponding to the current welding defect type includes: If the current welding defect type is incomplete penetration, then the defect feature vector corresponding to the incomplete penetration includes the incomplete penetration length and the incomplete penetration depth; If the current welding defect type is porosity, then the defect feature vector corresponding to the porosity is the porosity diameter and porosity location.

7. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The step of determining the welding current compensation amount and the welding current waveform correction function based on the current predicted welding current deviation value using a gradient descent optimization algorithm includes: Obtain the welding current waveform correction function corresponding to the current welding defect type; With the goal of eliminating incomplete penetration or porosity, the relationship between the defect incidence rate and the welding current compensation amount is used as the loss function. The gradient descent optimization algorithm is used to continuously update the predicted value of the welding current deviation and the welding current waveform correction function until the loss function converges. The predicted value of welding current deviation corresponding to the converged loss function is used as the welding current compensation amount, and the welding current waveform correction function corresponding to the converged loss function is used as the final welding current waveform correction function.

8. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, The step of determining the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function includes: The welding current correction amount is determined based on the welding current waveform correction function. Obtain the welding current value of the currently incomplete welding area, and sum the current welding current value, the welding current correction amount, and the welding current compensation amount respectively; The result of the summation operation is used as the adjusted welding current value, and a welding current adjustment command for the unfinished welding area is generated based on the adjusted welding current value.

9. The intelligent welding current adaptive adjustment method according to claim 1, characterized in that, Following the instruction to adjust the welding current in the unfinished welding area, the method further includes: The arc length, molten pool center temperature, and molten pool edge temperature were obtained during the welding process. If the arc length does not exceed the preset arc length threshold, the welding voltage during the welding process is increased; if the arc length exceeds the preset arc length threshold, the welding voltage during the welding process is decreased. If the temperature deviation between the center temperature and the edge temperature of the molten pool exceeds a preset temperature deviation threshold, the welding current during the welding process is reduced; if the temperature deviation between the center temperature and the edge temperature of the molten pool does not exceed the preset temperature deviation threshold, the welding current during the welding process is increased.

10. An intelligent welding current adaptive adjustment system, characterized in that, include: The first acquisition module is used to acquire acoustic emission signals and first welding current waveform data during the welding process; The first determining module is used to determine the energy entropy value of each frequency band sub-signal based on the acoustic emission signal using a multi-scale decomposition method, and to determine a high-dimensional feature set based on the energy entropy value; The second determining module is used to determine one of the welding defect types in the currently completed welding area based on the high-dimensional feature set and using a deep convolutional neural network. The welding defect types include incomplete penetration and porosity. The second acquisition module is used to acquire the time interval in which the current welding defect type occurs, and to extract the second welding current waveform data corresponding to the time interval from the first welding current waveform data. A construction module is used to acquire historical defect feature vectors and historical welding current waveform data, and to establish a welding current deviation prediction model based on the historical defect feature vectors and historical welding current waveform data. The third determining module is used to obtain the defect feature vector corresponding to the current welding defect type, and input the defect feature vector and the second welding current waveform data into the welding current deviation prediction model to output the current welding current deviation prediction value. The fourth determining module is used to determine the welding current compensation amount and the welding current waveform correction function based on the current predicted value of welding current deviation using a gradient descent optimization algorithm. The fifth determining module is used to determine the welding current adjustment command for the unfinished welding area based on the welding current compensation amount and the welding current waveform correction function.

Images