An intelligent monitoring method and system for full-process automatic welding

Abstract

The application discloses an intelligent monitoring method and system for full-process automatic welding, and relates to the technical field of welding monitoring. The method comprises the following steps: S1, feature extraction: through a set five-channel synchronous acquisition system, multi-modal data are synchronously acquired, and dynamic features are determined; S2, defect determination: according to the multi-modal data and the dynamic features, a collaborative matrix is constructed, and a double attention network is set to determine defect positions and probabilities; and S3, parameter adjustment: according to the defect probabilities, parameters in a welding process are dynamically adjusted. The application adjusts welding current slopes, protective gas flow and cooling gas flow in real time through the defect probabilities, and controls a molten pool cooling rate through a PID, thereby forming a closed loop system of "monitoring-decision-adjustment", so that the generation of defects is effectively inhibited, and the stability of welding quality is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of welding monitoring, in particular to an intelligent monitoring method and system for full-process automatic welding. BACKGROUND

[0002] As a core manufacturing process in modern manufacturing industry, especially in the fields of automobile, aerospace, shipbuilding, energy equipment, heavy machinery, etc., the quality of welding technology directly determines the structural strength, safety and service life of the product. With the rapid development of industrial automation and intelligent manufacturing, full-process automatic welding (i.e. automation of the whole process from workpiece clamping, positioning, welding to post-welding treatment) has become a key means to improve production efficiency, ensure product quality stability and reduce labor costs.

[0003] In the full-process automatic welding process, welding quality is influenced by a variety of complex factors, including but not limited to: welding process parameters (current, voltage, speed, wire feed speed, protective gas flow, etc.), welding gun posture and position, matching of base material and welding material, environmental conditions (temperature, humidity, airflow), equipment status (welding machine, robot, fixture), and pre-welding preparation (groove cleaning, assembly accuracy), etc. Any slight deviation in any link may lead to the generation of welding defects, such as porosity, slag inclusion, incomplete fusion, incomplete penetration, cracks, undercut, and welding bumps, etc. These defects not only weaken the mechanical properties of the weld, but also may become the starting point of structural failure, causing serious safety hazards and huge economic losses.

[0004] A Chinese invention patent with publication number CN113984837A discloses a resistance spot welding spot quality detection method based on welding spot feature information fusion, which includes the following steps: S1, setting the secondary current pulse to lag behind the main current pulse of resistance spot welding, and the secondary current pulse period is located in the electrode pressure maintaining stage; S2, clamping and fixing the workpiece to be welded, and starting welding by electrode pressing; S3, during the welding process, collecting the electrode voltage signal and current pulse signal generated by the secondary current pulse, and performing division operation on the collected electrode voltage signal and current pulse signal generated by the secondary current pulse to obtain the welding spot resistance characteristic value; S4, after the welding is completed, collecting the image information of the welding spot surface indentation, and obtaining the welding spot indentation image characteristic value based on machine vision technology; S5, taking the welding spot resistance characteristic value and the welding spot indentation image characteristic value as input layer neuron input to an artificial neural network model to calculate the welding spot tensile strength detection value. The welding spot tensile strength detection value can be obtained without damaging the welding spot, and the accuracy is high.

[0005] In the process of automatic welding of aluminum alloy sheet, the above detection scheme can obtain the resistance characteristic value through the secondary current pulse, but due to the dependence on the arithmetic mean or peak value of low-frequency signal, the high-frequency arc instability phenomenon in the welding process cannot be captured. At the same time, the alarm mechanism based on the static threshold setting is difficult to capture the transient abnormality in the welding process, and cannot quantify the synergistic effect among current, voltage and resistance, resulting in high defect detection rate of porosity and process parameter adjustment lag. SUMMARY

[0006] The purpose of the present application is to provide an intelligent monitoring method and system for full-process automatic welding to solve the problems raised in the background art.

[0007] To achieve the above purpose, the present application provides the following technical solution: an intelligent monitoring method for full-process automatic welding, comprising:

[0008] S1: feature extraction: through the set five-channel synchronous acquisition system, the multi-modal data are synchronously acquired, and the dynamic features are determined;

[0009] S2: defect determination: according to the multi-modal data and dynamic features, a synergistic matrix is constructed, and a defect position and probability are determined through the set double-attention network;

[0010] S3: parameter adjustment: according to the defect probability, the parameters in the welding process are dynamically adjusted, comprising:

[0011] S3.1: current adjustment: according to the defect probability, the correlation between the current rise slope and the defect probability is determined;

[0012] S3.2: gas flow control: according to the welding material, a reference flow of protective gas and a probability weight are set, and according to the reference flow of protective gas and the probability weight, a correction model between the protective gas flow and the defect probability is determined;

[0013] S3.3: heat management optimization: the center temperature of the molten pool is monitored by an infrared thermometer, and according to the center temperature of the molten pool and the monitoring time, the cooling rate is determined, and according to the cooling rate error, the PID controller is adjusted to determine the flow of cooling gas.

[0014] Further, the dynamic features are determined, comprising:

[0015] S1.1: Multimodal signal acquisition: through a differential current sensor, a high-voltage isolation voltage probe, a contact resistance measurement module and an acoustic emission sensor array, a five-channel synchronous acquisition system is set up, and current, voltage, contact resistance and acoustic emission data signals are obtained, and at the same time, through a hyperspectral camera, a hyperspectral imaging data signal is obtained;

[0016] S1.2: Feature extraction: according to the processing of the current signal through the complex Morlet wavelet transform algorithm, the instantaneous amplitude entropy and phase coherence are extracted, and at the same time, through the microwave reflection method, the change rate of the molten pool dielectric constant is obtained.

[0017] Further, obtaining the change rate of the molten pool dielectric constant comprises:

[0018] S1.2.1: Current signal processing: through the set MAC unit, the normalized energy and phase angle size of each scale are determined, and through the normalized energy and phase angle size, the instantaneous amplitude entropy and phase coherence are obtained;

[0019] S1.2.2: Molten pool state monitoring: through the continuous wave emitted by the microwave transmitter, the reflected signal and the reference signal are mixed to divide into in-phase component signal and quadrature component signal, and the dielectric constant size is determined, and at the same time, the dielectric constant is subjected to 5-point sliding average filtering to determine the center difference size, and the dielectric constant change rate is obtained.

[0020] Further, obtaining the instantaneous amplitude entropy and phase coherence comprises:

[0021] S1.2.1.1: Wavelet transform: according to the set center frequency and target frequency band, the scale boundary of the MAC unit is determined, and at the same time, according to the scale boundary of the MAC unit, the wavelet coefficient is set;

[0022] S1.2.1.2: Determine the instantaneous amplitude entropy: according to the wavelet coefficient, the normalized energy of each scale is obtained, and the instantaneous amplitude entropy is determined;

[0023] S1.2.1.3: Determine the phase coherence: according to the wavelet coefficient, the phase angle size of the wavelet coefficient of each scale is determined, and according to the sampling point data in the sliding window, the phase coherence is determined.

[0024] Further, determining the defect position and probability comprises:

[0025] S2.1: Multiscale field coupling modeling: according to the multimodal data and dynamic characteristics, a coordination matrix is constructed, and the condition number of the coordination matrix is determined, and at the same time, the condition number is compared with the warning threshold, and according to the comparison result, the triggering state of the warning signal is determined, specifically:

[0026] When the condition number of the coordination matrix is greater than the early warning threshold, an early warning signal is triggered, and the next step S2.2 is executed; otherwise, no early warning signal is triggered.

[0027] S2.2: Defect decision: the normalized current signal is taken as the input of the time domain branch in the double attention network, and the modulus value of the wavelet coefficient is taken as the input of the frequency domain branch in the double attention network, the five types of defect probabilities are output, the five types of defect probabilities are compared, the maximum defect probability is determined, the maximum defect probability is the final defect probability, and the defect type corresponding to the maximum defect probability is the final defect type, and the acoustic emission signal is received according to the set sensor array, and the defect position is determined according to the receiving time difference of the acoustic emission signal.

[0028] Further, according to the difference of the welding material, relevant tests are carried out to obtain the early warning threshold, which is specifically:

[0029] The aluminum alloy material is subjected to high-frequency fatigue test, and the condition number at the time of crack initiation is set as the early warning threshold;

[0030] The high-strength steel material is subjected to phase transformation induced plasticity test, and the matrix condition number at the martensite phase transformation critical point is set as the early warning threshold.

[0031] Further, when the defect type is a pore defect, the pore defect is verified by hyperspectral imaging data, including:

[0032] S2.2.1: Hyperspectral verification: according to the original spectral data in the reference waveband of 500-550nm in the hyperspectral imaging data, a quadratic polynomial fitting is carried out to obtain a baseline absorbance curve, and the absorbance in the reference waveband of 500-550nm is corrected according to the baseline absorbance curve, and the corrected absorbance is compared with the absorbance early warning threshold, and the pore defect is determined according to the comparison result; specifically:

[0033] When the corrected absorbance is greater than the absorbance early warning threshold, the current defect is a pore defect, and step S2.2.2 is executed to determine the adjusted welding pressure; otherwise, return to step S1 to determine the defect again;

[0034] S2.2.2: Dynamic pressure adjustment: the relationship between the pore size and the failure probability is determined through material fatigue test data, and the probability-pressure mapping relationship is determined, which is specifically:

[0035]

[0036] Wherein: is the adjusted welding pressure, a standard pressure in a defect-free state, a porosity defect probability;

[0037] S2.2.3: ultrasonic secondary verification: according to the adjusted welding pressure, the welding pressure is adjusted, and the welding point after the welding pressure adjustment is detected through the ultrasonic probe, the second harmonic amplitude and the fundamental wave amplitude are obtained, and the amplitude ratio is determined, and the amplitude ratio is compared with the preset ratio threshold range, and according to the comparison result, the adjusted defect type is determined, specifically:

[0038] When the amplitude ratio is less than the lower limit threshold of the preset ratio threshold range, the corresponding defect type is incomplete fusion; when the amplitude ratio is within the preset ratio threshold range, it is normal penetration; when the amplitude ratio is greater than the upper limit threshold of the preset ratio threshold range, the corresponding defect type is porosity.

[0039] Further, according to the critical value of the minimum porosity defect absorbance that the reference thickness material can detect, the absorbance reference threshold is set, and according to the absorbance reference threshold and the material thickness, the absorbance early warning threshold is set, specifically:

[0040]

[0041] wherein: is the absorbance early warning threshold, is the absorbance reference threshold, is the material thickness, is the reference thickness.

[0042] An intelligent monitoring system for full-process automatic welding uses any one of the intelligent monitoring methods for full-process automatic welding described above, comprising:

[0043] Data acquisition module: through a five-channel synchronous acquisition system, multi-source data in the welding process are collected and acquired, including welding current signal, welding voltage signal, welding point contact resistance data, acoustic emission signal and spectral data of the molten pool area;

[0044] Feature extraction module: according to the multi-source data, the dynamic characteristics of the welding process are determined, including instantaneous amplitude entropy, phase coherence and molten pool dielectric constant change rate;

[0045] Collaborative decision module: through the multi-source data and dynamic characteristics, a collaborative matrix is constructed, and through a double attention network, a defect position and probability are determined;

[0046] The parameter adjustment module adjusts the current rising slope according to the defect position and probability, sets the reference flow of the protective gas and the probability weight according to the welding material type, adjusts the gas flow through the correction model, and adjusts the cooling gas flow according to the cooling rate of the molten pool center temperature and the PID controller.

[0047] Compared with the prior art, the present application has the following advantages:

[0048] Firstly, the present application adjusts the welding current slope, the protective gas flow and the cooling gas flow in real time through the defect probability, and forms a closed-loop system of "monitoring-decision-adjustment" by PID control of the molten pool cooling rate, thereby effectively suppressing the generation of defects and improving the stability of the welding quality. At the same time, the present application corrects the absorbance by hyperspectral imaging to perform secondary verification on the porosity defects, and combines ultrasonic second harmonic analysis to further confirm the defect type, thereby improving the robustness of the detection result.

[0049] Secondly, the present application can synchronously acquire the current, voltage, contact resistance, acoustic emission and hyperspectral imaging data through the five-channel synchronous acquisition system, and extracts the instantaneous amplitude entropy, phase coherence and dielectric constant change rate dynamic characteristics by combining the complex Morlet wavelet transform and the microwave reflection method, thereby realizing multi-dimensional monitoring of the welding process and improving the accuracy and reliability of the defect detection.

[0050] Thirdly, the present application can not only accurately identify five types of different defects such as porosity, crack and slag inclusion, but also determine the defect position by performing collaborative analysis on the multi-modal data through the synergy matrix and the time / frequency domain branch in the dual attention network, thereby reducing the missed detection rate and the misjudgment rate. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a flowchart of the intelligent monitoring method in the present application.

[0052] Figure 2 It is a time domain input signal diagram in the present application.

[0053] Figure 3 It is a frequency domain input signal diagram in the present application.

[0054] Figure 4 It is a defect probability output diagram in the present application. DETAILED DESCRIPTION

[0055] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0056] In the process of automatic welding of aluminum alloy sheet, although the existing monitoring scheme can obtain the resistance characteristic value through a secondary current pulse, it cannot capture the high-frequency arc instability phenomenon in the welding process because it only relies on the calculation of the arithmetic mean or peak value of low-frequency signals. Meanwhile, the alarm mechanism based on static threshold setting cannot capture transient abnormalities in the welding process, nor can it quantify the synergistic effect among current, voltage and resistance, resulting in high defect detection rate of porosity and process parameter adjustment lag. The technical scheme of the present application obtains multi-modal data of current, voltage, contact resistance, acoustic emission and hyperspectral imaging through a five-channel synchronous acquisition system, and extracts dynamic characteristics of instantaneous amplitude entropy and phase coherence using complex Morlet wavelet transform. The change rate of the molten pool dielectric constant is obtained by microwave reflection method. At the same time, a synergy matrix is constructed based on multi-modal data, and the defect position and probability are determined by a double attention network. The welding current slope, protective gas flow and cooling gas flow parameters are dynamically adjusted according to the determined defect probability, thereby realizing closed-loop control of the welding process and improving the accuracy of welding defect detection and the real-time performance of parameter adjustment.

[0057] Embodiment 1

[0058] Reference Figures 1-4 The present embodiment provides an intelligent monitoring method for full-process automatic welding, which comprises the following steps:

[0059] Step S1: feature extraction. That is, through the set five-channel synchronous acquisition system, the current, voltage, contact resistance, acoustic emission and hyperspectral imaging data are synchronously acquired, and the dynamic characteristics are determined from the obtained multi-modal data. Specifically as follows:

[0060] Step S1.1: multi-modal signal acquisition. That is, through the differential current sensor, the acquisition channels 1 and 2 are set to acquire the current signal. Through the high-voltage isolation voltage probe, the acquisition channel 3 is set to acquire the voltage signal. Through the contact resistance measurement module, the acquisition channel 4 is set to acquire the contact resistance signal. Through the acoustic emission sensor array, the acquisition channel 5 is set to acquire the acoustic emission signal. At the same time, the hyperspectral camera is synchronously triggered with the five-channel synchronous acquisition system to acquire the hyperspectral imaging data signal.

[0061] In the embodiment, according to the differential Hall current sensor with a bandwidth not less than 2 MHz and a common-mode rejection ratio not less than 120 dB, channel 1 and channel 2 in the five-channel synchronous acquisition system are set. According to the high-voltage isolation voltage probe with an input impedance of 1 MΩ‖2 pF and a rise time greater than 5 ns, channel 3 in the five-channel synchronous acquisition system is set. According to the four-wire contact resistance measurement module with an excitation current of 10 mA@1 kHz and a resolution of 0.1 mΩ, channel 4 in the five-channel synchronous acquisition system is set. According to the acoustic emission sensor array with a center frequency of 150 kHz and a sensitivity not less than 80 dB (V / m·s -1 ), channel 5 in the five-channel synchronous acquisition system is set. According to the snapshot hyperspectral camera with a spectral resolution of 5 nm and a spatial resolution of 0.1 mm / pixel, the hyperspectral channel is set. That is, through the five channels and the hyperspectral channel in the set five-channel synchronous acquisition system, multi-modal data in the welding process, i.e., including current, voltage, contact resistance, acoustic emission and hyperspectral imaging data, are synchronously acquired.

[0062] Step S1.2: feature extraction. That is, according to the multi-modal data acquired in step S1.1, the current signal is processed by the complex Morlet wavelet transform algorithm, the instantaneous amplitude entropy and the phase coherence are extracted, and the change rate of the molten pool dielectric constant is calculated by combining with the microwave reflection method. Specifically as follows:

[0063] Step S1.2.1: current signal processing. That is, through the complex Morlet wavelet transform algorithm, the parallel MAC unit is set, and according to the set parallel MAC unit, the normalized energy and the phase angle size of each scale are determined, and according to the normalized energy of each scale, the instantaneous amplitude entropy is acquired. Specifically:

[0064]

[0065] Wherein: is the instantaneous amplitude entropy, is the normalized probability of the energy of the i th scale, is the total grade of the scale, is the index of the scale energy.

[0066] Further, according to the determined phase angle size, the phase coherence is acquired. Specifically:

[0067]

[0068] Wherein: is the phase coherence coefficient, is the number of sampling points in the sliding window, is the phase angle of the k th sampling point, The average phase within the window, This is the index of the sampling point.

[0069] Step S1.2.2: Molten Pool Status Monitoring. This involves transmitting a 24GHz continuous wave using a microwave transmitter with an output power of 10mW and an antenna gain of 15dBi. The corresponding rate of change of dielectric constant is obtained through the reflected phase difference. Furthermore, the reflected signal and the reference signal are mixed to generate in-phase and quadrature component signals. In other words, at a sampling rate of 1MHz, each molten pool status data packet contains 1000 in-phase and quadrature component signals.

[0070] In this embodiment, the corresponding dielectric constant is determined based on the in-phase component signal and the quadrature component signal, specifically as follows:

[0071]

[0072] in: Where is the dielectric constant. It is an in-phase component signal. These are orthogonal component signals.

[0073] Furthermore, based on the determined dielectric constant, a 5-point moving average filter is applied to determine the corresponding center difference, specifically:

[0074]

[0075] in: The rate of change of dielectric constant, For a moment The filtered dielectric constant, For a moment The filtered dielectric constant, The time difference step size.

[0076] Step S2: Defect Determination. Based on the multimodal data and dynamic features obtained in Step S1, a collaboration matrix is ​​constructed. Then, using the configured dual-attention network and collaboration matrix, the defect location and probability are determined. Details are as follows:

[0077] Step S2.1: Multiphysics coupling modeling. Based on the multimodal data obtained in Step S1.1 (current, voltage, contact resistance, and acoustic emission) and the dynamic characteristics obtained in Step S1.2 (instantaneous amplitude entropy, phase coherence, and rate of change of dielectric constant), a cooperative matrix is ​​constructed, specifically as follows:

[0078]

[0079] in: For the collaborative matrix, the current-voltage transient slope, the contact resistance change rate, the integral of the acoustic emission and current product, the dielectric constant change rate, the transient amplitude entropy, the phase coherence coefficient.

[0080] In the process of specific implementation, the corresponding cooperation matrix at 10 ms is specifically , after normalizing each variable to the interval [0, 1], the corresponding normalized cooperation matrix is specifically .

[0081] Further, according to the set cooperation matrix, the corresponding condition number size is determined, which is specifically:

[0082]

[0083] Among them: the condition number of the cooperation matrix, the 2-norm of the cooperation matrix, the 2-norm of the inverse matrix of the cooperation matrix.

[0084] In this embodiment, according to the difference of welding materials, relevant tests are carried out to obtain the corresponding early warning threshold. Specifically, the aluminum alloy material 6061 is subjected to high-frequency fatigue test to monitor and obtain the condition number when the crack is generated. The high-strength steel 22MnB5 material is subjected to phase transformation induced plasticity test to record the matrix condition number of the martensite phase transformation critical point. Thus, the corresponding early warning threshold of different welding materials is determined.

[0085] Further, the condition number of the obtained cooperation matrix is compared with the early warning threshold, and the triggering state of the early warning signal is determined according to the comparison result. Specifically:

[0086] When the condition number of the obtained cooperation matrix is greater than the early warning threshold, the early warning signal is triggered, and the next step S2.2 is executed. On the contrary, when the condition number of the obtained cooperation matrix is not greater than the early warning threshold, the early warning signal is not triggered.

[0087] Step S2.2: defect decision. That is, the current signal of 1 ms time window obtained in step S1.1 is normalized to 1000-dimensional vector as the input of time domain branch in double attention network, and the modulus value of set wavelet coefficient is input as the input of frequency domain branch in double attention network. Referring to Figure 2 and Figure 3 , the time domain signal can successfully capture the characteristics of unfused defects, and the frequency domain wavelet coefficient can accurately identify the characteristics of pores.

[0088] Further, according to the calculation of the time domain attention in the time domain branch and the calculation of the frequency domain attention in the frequency domain branch, the corresponding five types of defect probabilities are output, wherein the five types of defects include the porosity defect, the crack defect, the slag inclusion defect, the incomplete fusion defect and the oxidation defect. That is, according to the comparison of the defect probability corresponding to each type of defect, the maximum defect probability is determined, and the defect type corresponding to the maximum defect probability is the finally determined defect type.

[0089] In the process of specific implementation, reference is made to Figure 4 , the five types of defect probabilities output by the dual attention network are respectively: porosity defect: 0.82, crack defect: 0.05, slag inclusion defect: 0.03, incomplete fusion defect: 0.08, and oxidation defect: 0.02, and the corresponding final defect type is the porosity defect.

[0090] Further, according to the set 4*4 sensor array, the acoustic emission signal is received, so that the corresponding defect position can be determined according to the time difference size, and specifically:

[0091]

[0092] Among them: is the time difference of the arrival of the acoustic wave between the sensor m and the sensor n, is the propagation speed of the acoustic wave in the material, is the spatial coordinates of the defect source, is the coordinates of the mth acoustic emission sensor, is the coordinates of the nth acoustic emission sensor.

[0093] Step S3: parameter adjustment. That is, according to the defect probability determined in step S2.2, the parameters in the welding process are dynamically adjusted. Specifically as follows:

[0094] Step S3.1: current adjustment. That is, according to the defect probability determined in step S2.2, the correlation between the current rise slope and the defect probability is determined, and specifically:

[0095]

[0096] Among them: is the current rise slope, is the defect probability.

[0097] That is, in the embodiment, the basic slope is set to 2A / μs, and the adjustment step is set to 0.1 A / μs. At the same time, the maximum slope in the embodiment can be set to 2 A / μs, and the minimum slope is set to 1 A / μs, so as to not only prevent voltage breakdown, but also maintain the stability of the arc.

[0098] Step S3.2: gas flow control. That is, according to the different welding materials, the corresponding probability weight size is set. Specifically, when the welding material is aluminum alloy 6061, the corresponding reference flow is 15 L / min, and the corresponding probability weight size is 0.3. When the welding material is high-strength steel, the corresponding reference flow is 20 L / min, and the corresponding probability weight size is 0.2. When the welding material is magnesium alloy, the corresponding reference flow is 12 L / min, and the corresponding probability weight size is 0.5.

[0099] In this embodiment, according to the reference flow and the probability weight size corresponding to different welding materials, the correction model between the protective gas flow and the defect probability corresponding to different materials is determined, which is specifically:

[0100]

[0101] Wherein: is the adjusted protective gas flow, is the reference flow of the protective gas, is the probability weight, is the defect probability.

[0102] Step S3.3: heat management optimization. That is, the infrared thermometer of 5-10 μm wave band is used to monitor the center temperature of the molten pool to obtain the center temperature of the molten pool. At the same time, according to the monitored center temperature of the molten pool and the corresponding monitoring time, the corresponding cooling rate is determined, which is specifically:

[0103]

[0104] Wherein: is the cooling rate of the molten pool, is the center temperature of the molten pool at time t, is the center temperature of the molten pool at time t-△t, is the time difference step.

[0105] Further, according to the determined cooling rate of the molten pool, the corresponding cooling rate error is obtained, which is specifically:

[0106]

[0107] Wherein: is the cooling rate error, is the standard cooling rate, is the actual cooling rate.

[0108] Further, according to the determined cooling rate error, the PID controller is adjusted to determine the flow size of the cooling gas, which is specifically:

[0109]

[0110] wherein: is the flow rate of the cooling gas, is the proportional gain coefficient, is the integral gain coefficient, is the derivative gain coefficient, is the cooling rate error.

[0111] The embodiment also provides an intelligent monitoring system for full-process automatic welding, which uses the intelligent monitoring method for full-process automatic welding. The intelligent monitoring system specifically comprises a data acquisition module, a feature extraction module, a collaborative decision module, and a parameter adjustment module. The data acquisition module is configured to acquire multi-source data in a welding process, i.e., a welding current signal, a welding voltage signal, a contact resistance data of a welding point, an acoustic emission signal, and spectral data of a molten pool region, through a five-channel synchronous acquisition system. The feature extraction module is configured to determine an instantaneous amplitude entropy and a phase coherence coefficient by processing the current signal in the multi-source data acquired by the data acquisition module through a complex Morlet wavelet transform algorithm, so as to represent arc stability and molten pool oscillation characteristics. Meanwhile, the molten pool dielectric constant change rate is acquired through a microwave reflection method (for example, a 24 GHz continuous wave), so that the dynamic change of the molten pool state can be determined.

[0112] Further, the collaborative decision module is configured to construct a corresponding collaborative matrix according to the multi-source data acquired by the data acquisition module and the dynamic features acquired by the feature extraction module, determine a corresponding condition number through the collaborative matrix, and compare the condition number with a material-related early warning threshold, so as to determine a trigger signal of an early warning signal. Meanwhile, the normalized current signal is taken as an input of a time domain branch in a double attention network, and the wavelet coefficient modulus is taken as an input of a frequency domain branch in the double attention network, so as to capture defect features such as incomplete fusion and identify frequency domain features such as pores, thereby outputting corresponding defect types and probabilities. Meanwhile, the time difference positioning algorithm of the acoustic emission sensor array is used to determine corresponding defect spatial coordinates, i.e., to determine the defect position.

[0113] Further, the parameter adjustment module dynamically adjusts the current rise slope according to the defect probability determined by the collaborative decision module, so as to maintain the arc stability. Meanwhile, the reference flow rate of the protective gas and the probability weight are set according to the welding material type, and the gas flow is adjusted through a correction model. Meanwhile, the molten pool center temperature is monitored through an infrared thermometer to determine the corresponding cooling rate, and the cooling gas flow is adjusted through a PID controller according to the cooling rate error, so as to control the molten pool cooling process.

[0114] Embodiment 2

[0115] This embodiment provides an intelligent monitoring method for fully automated welding, which is implemented in the same way as in Embodiment 1. The difference is that the MAC unit in the FPGA parallel architecture is determined by the complex Morlet wavelet transform algorithm, and the instantaneous amplitude entropy and phase coherence are obtained based on the MAC unit. The invention will be illustrated below with specific examples of the implementation of this embodiment.

[0116] In this embodiment, the instantaneous amplitude entropy and phase coherence are obtained based on the MAC unit, including the following steps:

[0117] Step S1.2.1.1: Wavelet Transform. Since the measured natural oscillation frequency of the aluminum alloy molten pool under pulsed arc is 48-52kHz, the center frequency is set to 50kHz in this embodiment. Simultaneously, through calibration experiments, a 1μs pulse interference is added to the 50kHz standard sinusoidal signal, and the detection results are shown in Table 1 below.

[0118] Table 1: Test Results Table

[0119] Bandwidth parameter Pulse detection rate Frequency offset error 1 92% ±0.8 kHz 1.5 98% ±0.5 kHz 2 95% ±1.2 kHz

[0120] In other words, according to the pulse detection rate and frequency offset error in Table 1, the optimal bandwidth parameter is 1.5. Therefore, in this embodiment, the bandwidth parameter is set to 1.5. This corresponds to a time resolution of 0.87 μs and a frequency resolution of 1.15 kHz.

[0121] Furthermore, based on the set center frequency of 50kHz and the target frequency band of 10-100kHz, the scale boundary corresponding to the MAC unit is determined, specifically as follows:

[0122]

[0123] in: For the smallest scale, For the largest scale.

[0124] Furthermore, based on the determined scale boundary [0.5, 5], it is divided into 20 scales according to a logarithmic even distribution to avoid oversampling in the high-frequency band. It is worth noting that each MAC unit corresponds to one scale; that is, this embodiment uses 20 MAC units. Therefore, in this embodiment, the wavelet coefficients are specifically set as follows:

[0125]

[0126] in: For scale Time offset wavelet coefficients at the location, is a discrete time index, is the jth sample value of the welding current, is the jth sample value of the welding current, is a conjugate function of the complex Morlet wavelet, is a time shift parameter, is the jth scale parameter, is an analysis window length.

[0127] Step S1.2.1.2: Determine the instantaneous amplitude entropy. That is, according to the wavelet coefficients set in step S1.2.1.1, the normalized energy of each scale is obtained, specifically:

[0128]

[0129] wherein: is the normalized probability of the ith scale energy, is the wavelet coefficient at scale , time shift , is the wavelet coefficient at scale , time shift , is the index of the scale. Further, according to the normalized energy of each scale, the instantaneous amplitude entropy is determined, specifically:

[0130]

[0131]

[0132] wherein: is the instantaneous amplitude entropy, is the normalized probability of the ith scale energy, is the index of the scale energy.

[0133] Step S1.2.1.3: Determine the phase coherence. That is, according to the wavelet coefficients set in step S1.2.1.1, the phase angle size corresponding to the wavelet coefficient of each scale is determined, specifically:

[0134]

[0135] wherein: is the phase angle of the kth sample point, is the wavelet coefficient at scale , time shift .

[0136] In the sliding window in this embodiment, 100 sample points are set, and the corresponding phase coherence is specifically:

[0137] ​​

[0138] wherein: is a phase coherence coefficient, is a phase angle of the kth sampling point, is an average phase within a window, is an index of the sampling point.

[0139] Embodiment 3

[0140] The embodiment provides an intelligent monitoring method for full-process automatic welding, and the specific implementation method is the same as that of Embodiment 1, and the difference is that when the defect type is a gas hole defect, the gas hole defect is verified according to the hyperspectral imaging data obtained in step S1.1. The application will be illustrated below in combination with the specific implementation of the embodiment.

[0141] In the embodiment, the gas hole defect is verified by hyperspectral imaging data, and the specific steps include the following steps:

[0142] Step S2.2.1: Hyperspectral verification. That is, an imaging spectrometer is installed at a position 30 cm behind the welding torch and at an angle of 45° between the optical axis and the welding direction to obtain corresponding hyperspectral imaging data. Further, in a circular region with the center of the molten pool as the origin and a radius of 3 mm, a reference wavelength band of 500-550 nm is selected, and the original spectral data in the 500-550 nm interval is fitted by a second-order polynomial, specifically:

[0143]

[0144] wherein: is a baseline absorbance curve, is a quadratic coefficient, is a linear coefficient, is a constant term, is a wavelength.

[0145] In the process of specific implementation, the data table shown in Table Two is set, specifically:

[0146] Table Two: Data Table

[0147] Wavelength / nm Raw absorbance Fitted value Residuals 500 0.08 0.079 0.001 520 0.07 0.071 -0.001 550 0.06 0.062 -0.002

[0148] According to the data in Table Two, the quadratic coefficient is 2.1*10 -6 , the linear coefficient is -0.0023, and the constant term is 1.15.

[0149] Further, according to the constructed baseline absorbance curve, the selected 500-550 nm wavelength band interval is corrected for absorbance, specifically:

[0150]

[0151] wherein: is the corrected absorbance, is the original absorbance, is the baseline absorbance curve.

[0152] In this embodiment, the absorbance reference threshold is set according to the critical value of the absorbance of the minimum pore defect (e.g. 0.1 mm) that the material of the reference thickness (e.g. 1 mm) can detect. At the same time, the corresponding absorbance warning threshold is determined according to the absorbance reference threshold and the thickness of the material, which is specifically:

[0153]

[0154] wherein: is the absorbance warning threshold, is the absorbance reference threshold, is the thickness of the material, is the reference thickness.

[0155] Further, the corrected absorbance obtained is compared with the set absorbance warning threshold, and the pore defect is determined according to the comparison result. Specifically:

[0156] When the corrected absorbance obtained is greater than the absorbance warning threshold, the current defect is a pore defect, and step S2.2.2 is performed to determine the adjusted welding pressure. Otherwise, when the corrected absorbance obtained is not greater than the absorbance warning threshold, return to step S1 to re-determine the defect.

[0157] Step S2.2.2: Dynamic pressure adjustment. That is, according to the relationship between the pore size and the failure probability, the relationship between the pore size and the failure probability is determined according to the material fatigue test data, which is specifically:

[0158]

[0159] wherein: is the pore defect probability, is the base of the exponential function, is the equivalent diameter of the pore, is the characteristic scale parameter, is the shape parameter.

[0160] Further, according to the relationship between the pore size and the failure probability, the probability-pressure mapping relationship is determined, which is specifically:

[0161]

[0162] wherein: is the adjusted welding pressure, is a standard pressure in a defect-free state, is a porosity defect probability.

[0163] Step S2.2.3: ultrasonic secondary verification. That is, according to the adjusted welding pressure determined in step S2.2.2, the welding pressure in the automatic welding process is adjusted, and at the same time the welding point after adjusting the welding pressure is detected through the ultrasonic probe to obtain the corresponding second harmonic amplitude and fundamental wave amplitude, specifically:

[0164]

[0165] Wherein: is a fundamental wave signal amplitude, is a second harmonic signal amplitude, is the number of sampling points in the analysis window, is a fundamental wave time domain signal, is a second harmonic time domain signal, is the index of the sampling point in the analysis window.

[0166] Further, according to the obtained fundamental wave signal amplitude and second harmonic signal amplitude, the corresponding amplitude ratio is determined, specifically:

[0167]

[0168] Wherein: is an amplitude ratio, is a fundamental wave signal amplitude, is a second harmonic signal amplitude.

[0169] In this embodiment, the obtained amplitude ratio is compared with the preset ratio threshold range (for example, 0.1-0.3), and according to the comparison result, the adjusted defect type is determined, specifically:

[0170] When the obtained amplitude ratio is less than the lower threshold value of the preset ratio threshold range, that is, 0.1, the corresponding defect type is incomplete fusion. When the obtained amplitude ratio is within the preset ratio threshold range, that is, 0.1-0.3, it is normal penetration. When the obtained amplitude ratio is greater than the upper threshold value of the preset ratio threshold range, that is, 0.3, the corresponding defect type is porosity.

[0171] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for fully automated welding, characterized in that, Including: S1: Feature Extraction: Through a five-channel synchronous acquisition system, multimodal data is acquired synchronously, and dynamic features are determined, including: S1.1: Multimodal signal acquisition: A five-channel synchronous acquisition system is set up using a differential current sensor, a high-voltage isolation voltage probe, a contact resistance measurement module, and an acoustic emission sensor array to acquire current, voltage, contact resistance, and acoustic emission data signals. At the same time, a hyperspectral imaging data signal is acquired using a hyperspectral camera. S1.2: Feature Extraction: Based on the processing of the current signal using the complex Morlet wavelet transform algorithm, the instantaneous amplitude entropy and phase coherence are extracted. Simultaneously, the rate of change of the dielectric constant of the molten pool is obtained using the microwave reflection method, including: S1.2.1: Current signal processing: Through the set MAC unit, the normalized energy and phase angle of each scale are determined, and the instantaneous amplitude entropy and phase coherence are obtained through the normalized energy and phase angle. S1.2.2: Molten pool condition monitoring: The continuous wave emitted by the microwave transmitter mixes the reflected signal and the reference signal, divides them into in-phase component signals and quadrature component signals, determines the magnitude of the dielectric constant, and simultaneously performs a 5-point moving average filter on the dielectric constant to determine the magnitude of the center difference and obtain the rate of change of the dielectric constant. S2: Defect determination: Based on the multimodal data and dynamic features, a collaborative matrix is ​​constructed, and the defect location and probability are determined through a set dual attention network; S3: Parameter Adjustment: Based on the defect probability, the parameters during the welding process are dynamically adjusted, including: S3.1: Current adjustment: Based on the defect probability, determine the correlation between the current rise slope and the defect probability; S3.2: Gas flow control: Based on the welding material, set the reference flow rate and probability weight of the protective gas, and determine the correction model between the protective gas flow rate and the defect probability based on the reference flow rate and probability weight of the protective gas. S3.3: Thermal Management Optimization: The temperature of the center of the molten pool is monitored by an infrared thermometer, and the cooling rate is determined based on the temperature of the center of the molten pool and the monitoring time. The cooling rate error is obtained based on the cooling rate, and the PID controller is adjusted based on the cooling rate error to determine the flow rate of the cooling gas.

2. The intelligent monitoring method for fully automated welding according to claim 1, characterized in that, The instantaneous amplitude entropy and phase coherence are obtained, including: S1.2.1.1: Wavelet transform: Determine the scale boundary of the MAC unit based on the set center frequency and target frequency band, and set the wavelet coefficients based on the scale boundary of the MAC unit; S1.2.1.2: Determine the instantaneous amplitude entropy: Based on the wavelet coefficients, obtain the normalized energy of each scale and determine the instantaneous amplitude entropy; S1.2.1.3: Determine phase coherence: Based on the wavelet coefficients, determine the phase angle of the wavelet coefficients at each scale, and determine the phase coherence based on the sampling point data within the sliding window.

3. The intelligent monitoring method for fully automated welding according to claim 1, characterized in that, Determining the location and probability of defects includes: S2.1: Multiphysics Coupling Modeling: Based on the multimodal data and dynamic characteristics, a coordination matrix is ​​constructed, and the condition number of the coordination matrix is ​​determined. Simultaneously, the condition number is compared with a warning threshold, and based on the comparison result, the triggering state of the warning signal is determined. Specifically: When the condition number of the cooperative matrix is ​​greater than the warning threshold, a warning signal is triggered and the next step S2.2 is executed; otherwise, a warning signal is not triggered. S2.2: Defect Decision: The normalized current signal is used as the input to the time-domain branch of the dual-attention network, and the magnitude of the wavelet coefficients is used as the input to the frequency-domain branch of the dual-attention network. The output obtains five types of defect probabilities. At the same time, the five types of defect probabilities are compared to determine the maximum defect probability. The maximum defect probability is the final defect probability, and the defect type corresponding to the maximum defect probability is the final defect type. Meanwhile, the acoustic emission signal is received according to the set sensor array, and the defect location is determined by the reception time difference of the acoustic emission signal.

4. The intelligent monitoring method for fully automated welding according to claim 3, characterized in that, Depending on the welding material, relevant tests are conducted to obtain the warning threshold, specifically: High-frequency fatigue tests were conducted on aluminum alloy materials, and the condition number at which crack initiation occurred was set as the early warning threshold. Phase transformation induced plasticity tests were conducted on high-strength steel materials, and the matrix condition number of the martensitic phase transformation critical point was set as the warning threshold.

5. The intelligent monitoring method for fully automated welding according to claim 3, characterized in that, When the defect type is porosity, the porosity defect is verified using hyperspectral imaging data, including: S2.2.1: Hyperspectral Verification: Based on the original spectral data within the 500-550nm reference band of the hyperspectral imaging data, a quadratic polynomial fitting is performed to obtain the baseline absorbance curve. Simultaneously, based on the baseline absorbance curve, the absorbance within the 500-550nm reference band is corrected, and the corrected absorbance is compared with the absorbance warning threshold. Based on the comparison result, porosity defects are determined; specifically: When the corrected absorbance is greater than the absorbance warning threshold, the current defect is a porosity defect, and step S2.2.2 is executed to determine the adjusted welding pressure; otherwise, return to step S1 and re-determine the defect. S2.2.2: Dynamic Pressure Adjustment: Using material fatigue test data, the relationship between pore size and failure probability is determined, and the probability-pressure mapping relationship is established, specifically: ； in: For the adjusted welding pressure, The standard pressure under defect-free conditions. This represents the probability of porosity defects. S2.2.3: Ultrasonic Secondary Verification: Based on the adjusted welding pressure, the welding pressure is adjusted, and the weld point after the adjustment is detected by an ultrasonic probe to obtain the second harmonic amplitude and the fundamental amplitude, and the amplitude ratio is determined. Simultaneously, the amplitude ratio is compared with a preset ratio threshold range, and based on the comparison result, the adjusted defect type is determined, specifically: When the amplitude ratio is less than the lower limit of the preset ratio threshold range, the corresponding defect type is incomplete fusion; when the amplitude ratio is within the preset ratio threshold range, normal fusion is achieved; when the amplitude ratio is greater than the upper limit of the preset ratio threshold range, the corresponding defect type is porosity.

6. The intelligent monitoring method for fully automated welding according to claim 5, characterized in that, An absorbance baseline threshold is set based on the absorbance critical value at which the smallest detectable porosity defect occurs in the material with a reference thickness. An absorbance warning threshold is then set based on the absorbance baseline threshold and the material thickness, specifically as follows: ； in: The absorbance warning threshold is... The absorbance reference threshold, For material thickness, The reference thickness.

7. An intelligent monitoring system for fully automated welding, characterized in that, The intelligent monitoring method for fully automated welding as described in any one of claims 1-6 includes: Data acquisition module: Through a five-channel synchronous acquisition system, it acquires multi-source data during the welding process, including welding current signal, welding voltage signal, weld contact resistance data, acoustic emission signal and spectral data of the molten pool area; Feature extraction module: Based on the multi-source data, determine the dynamic features of the welding process, including instantaneous amplitude entropy, phase coherence, and rate of change of the dielectric constant of the molten pool; Collaborative decision-making module: Constructs a collaborative matrix using the multi-source data and dynamic features, and determines the defect location and probability using a dual attention network; Parameter adjustment module: Adjusts the current rise slope according to the defect location and probability, sets the shielding gas reference flow rate and probability weight according to the welding material type, and adjusts the gas flow rate through a correction model. At the same time, it adjusts the cooling gas flow rate according to the cooling rate of the molten pool center temperature and the PID controller.

Images