Multi-factor coupling welding electrode wear degree online evaluation method

Abstract

The invention relates to the technical field of resistance spot welding monitoring, in particular to a multi-factor coupling welding electrode abrasion degree online evaluation method which comprises the steps that on the basis of current and voltage data in the existing welding process, according to real-time current data and voltage data, dynamic resistance is calculated, and a discrete data set with the resistance changing along with time is obtained; performing dynamic data processing according to the obtained discrete data set, performing static data processing based on the obtained resistance data, calculating peak time delay by using the discrete data set, calculating a dynamic resistance drop rate according to a dynamic resistance curve, and performing normalization processing on the dynamic resistance drop rate; according to the method, the multi-dimensional characteristic factors are fused, and the electrode abrasion state is comprehensively reflected from multiple angles such as the resistance change speed, the curve form, the resistance peak value time delay and the resistance drop rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resistance spot welding monitoring, in particular to a multi-factor coupled welding electrode wear degree online evaluation method. BACKGROUND

[0002] As an important material processing method for automobile body-in-white welding, resistance spot welding is widely used in actual production. In the resistance spot welding process, the electrode material will react with the automobile body-in-white coating. With the increase of the number of welding points, the compounds attached to the electrode surface gradually increase, causing electrode wear. However, the electrode wear process is affected by various factors such as welding materials and welding parameters, and is random. If the electrode is ground too early, it will cause waste of electrode material. If the electrode is not ground in time, it will lead to unqualified joints. Therefore, it is of great significance to realize online evaluation of the wear degree of resistance spot welding.

[0003] At present, the monitoring methods for electrode wear in the industry mainly include single-parameter threshold method, resistance average comparison method and mechanical parameter monitoring method. The single-parameter threshold method, such as judging the wear state by whether the duration of the rising stage of the dynamic resistance curve exceeds the threshold value, or determining according to the electrode pressure amount of the welding clamp and the threshold value of the spatter frequency, only relies on a single process parameter, is easily disturbed by welding material fluctuations or process parameter adjustments, and leads to misjudgment. Secondly, the resistance average comparison method compares the average value of the welding resistance in the recent period with the reference value after grinding, as shown in patent CN111673256B. This method ignores the time sequence characteristics of the dynamic resistance curve, can only reflect the change of the average value of the resistance, cannot capture key wear characteristics such as curve shape and trend, and is difficult to effectively identify when the signal is weak in the early stage of wear. Finally, the mechanical parameter monitoring method evaluates wear through changes in electrode movement stroke, as shown in patent CN1251548A, or a wear amount prediction model. This kind of method relies on the accuracy of mechanical displacement sensors and is not directly related to the physical mechanism of welding electrical parameters and wear. In the multi-variety mixed production scene, the adaptability is poor. At present, a multi-factor coupled welding electrode wear degree online evaluation method is needed. SUMMARY

[0004] In order to solve the problem of high misjudgment rate and weak anti-interference ability due to insufficient parameter correlation mentioned in the background art, the present application provides a multi-factor coupled welding electrode wear degree online evaluation method.

[0005] The multi-factor coupled welding electrode wear degree online evaluation method provided by the present application adopts the following technical scheme: A multi-factor coupled welding electrode wear degree online evaluation method, comprising: On the basis of the current and voltage data in the existing welding process, dynamic resistance is calculated according to real-time current data and voltage data, and discrete data groups of resistance changing with time are obtained; Dynamic data processing is performed according to the obtained discrete data groups, including obtaining dynamic resistance change speed according to resistance change amount and normalizing, and calculating trend change factor based on the normalization result; Static data processing is performed based on the obtained resistance data, including normalizing dynamic resistance data, and introducing weight function to calculate shape change factor; The peak time delay amount is calculated using the discrete data groups, including determining the peak arrival time according to the dynamic resistance curve, and calculating the delay amount according to the peak arrival time; The dynamic resistance drop rate is calculated according to the dynamic resistance curve, and the dynamic resistance drop rate is normalized; The comprehensive change factor is calculated by weighted average, and the electrode wear degree is judged according to the value of the comprehensive change factor.

[0006] Further, the dynamic resistance is calculated according to real-time current data and voltage data, including obtaining current and voltage signals at different time periods in the resistance spot welding process at fixed time intervals, generating an original data group containing n data points, calculating the resistance value of each sampling point based on the original data group using Ohm's law, generating a discrete dynamic resistance data group, and denoising the dynamic resistance data by Kalman filtering algorithm.

[0007] Further, the dynamic data processing according to the obtained discrete data groups includes calculating the difference between the resistances of two adjacent data points in the discrete data group to obtain the dynamic resistance change speed of the entire welding process, normalizing the dynamic resistance change speed using the modified sigmoid function, and calculating the Euclidean distance between the current data group and the standard data group of the un-worn electrode according to the normalization result, determining the trend change factor according to the Euclidean distance, and the expression of the trend change factor is: , Wherein, n is the total number of measured data points, represents the resistance value obtained after normalization of the un-worn electrode, represents the difference between the resistances of two adjacent data points after normalization.

[0008] Further, the dynamic resistance change speed is normalized using the modified sigmoid function, including calculating the variance of the resistance change speed of the first m sampling points in the initial stage of welding, determining the sensitivity coefficient according to the variance of the resistance change speed, and normalizing the dynamic resistance change speed based on the sensitivity coefficient using the modified sigmoid function, and the normalization formula is: , Wherein, sensitivity coefficient, is the difference between the resistance of two adjacent data points.

[0009] Further, the static data processing based on the obtained resistance data includes Z-score normalization processing on the discrete data group of the obtained resistance, then introducing a weight function to reduce the weight of the initial welding data, calculating the Euclidean distance between the current normalized dynamic resistance data group and the standard data group of the unworn electrode, and obtaining a shape change factor, the expression of the shape change factor is: , wherein, is the resistance value of the i-th sampling point after Z-score normalization processing on the dynamic resistance data group of the unworn electrode, and n is the total number of data points, is the resistance value of the i-th sampling point after Z-score normalization processing on the dynamic resistance data group of the k-th welding, is the weight function.

[0010] Further, the introduction of the weight function to reduce the weight of the initial welding data includes determining the exponential term of the weight function according to the offset and the exponential coefficient, calculating the weight function value through the exponential term, using a transfer learning model, and dynamically adjusting the weight function according to the thickness of the welding material oxide film, the expression of the weight function is: , wherein, i represents the sampling point index, is the offset, is the exponential coefficient.

[0011] Further, the calculation of the peak time delay amount using the discrete data group includes determining a dynamic resistance curve according to the discrete data group, calculating a resistance curve gradient sequence according to the dynamic resistance curve, setting a primary threshold and determining candidate points in combination with the resistance curve gradient sequence, performing gradient sign change detection on the candidate points to obtain real peak points, finding the time when the resistance in the middle of the welding reaches the real peak point, and subtracting the peak time of the standard dynamic resistance curve from the time when the resistance in the middle of the welding reaches the real peak point to obtain the peak time delay amount.

[0012] Further, the normalization processing on the dynamic resistance drop rate includes positioning the maximum peak resistance and the resistance at the end of the welding through the gradient detection method, introducing a dynamic benchmark model based on LSTM to update the maximum peak resistance and the resistance at the end of the welding, calculating the dynamic resistance drop rate according to the updated peak resistance and the resistance at the end of the welding, applying weight to the dynamic resistance drop rate according to the time position of the dynamic resistance curve, and performing normalization processing on the dynamic resistance drop rate, the normalization expression is: , wherein, represents the original dynamic resistance drop rate, represents the historical maximum drop rate updated based on the LSTM dynamic reference model, represents the historical minimum drop rate updated based on the LSTM dynamic reference model, represents the time weight function.

[0013] Further, the comprehensive change factor is calculated by weighted average, including weighted average of the trend change factor, the shape change factor, the normalized peak time delay amount and the dynamic resistance drop rate normalized value, to obtain the comprehensive change factor, and the expression of the comprehensive change factor is: , wherein, represents the trend change factor, represents the weight of the trend change factor, represents the shape change factor, represents the dynamic resistance drop rate normalized value, represents the peak time delay amount, is the weight of the shape change factor, is the weight of the dynamic resistance drop rate normalized value, is the weight of the peak time delay amount.

[0014] Further, the weighted average adopts the inverse variance weighting method, the weight of each factor is calculated according to the historical variance of each factor, the coupling of the factors is constrained through the covariance matrix, the objective function is constructed and the weight vector is optimized combined with the constraint condition, the Lagrange multiplier is introduced, the constrained optimization problem is converted into an unconstrained problem, the optimal weight vector is obtained, and the expression of the covariance matrix is: , wherein, n represents the number of data samples, represents the feature factor vector of the kth welding, represents the sample mean vector.

[0015] In summary, the present application has the following beneficial technical effects: 1、The present application fuses the trend change factor, the shape change factor, the peak time delay amount and the dynamic resistance drop rate and other multi-dimensional feature factors, comprehensively reflects the electrode wear state from the resistance change speed, the curve shape, the nugget growth time and the nugget size change and other multiple angles, at the same time, the weighted average is adopted and the inverse variance weighting method and the covariance matrix constraint optimization weight are combined, the limitations of single factors and the information overlap between factors are avoided, the correlation between the comprehensive change factor and the actual wear amount is greatly improved.

[0016] 2、The present application adopts a modified sigmoid function to adaptively determine the sensitivity coefficient according to the variance of the resistance change rate in the initial welding stage, normalizes the dynamic resistance change rate, effectively suppresses the interference in the initial welding stage, introduces a weight function and dynamically adjusts according to the offset, the exponential coefficient and the thickness of the oxide film of the welding material, reduces the data weight in the initial welding stage, enhances the resistance to interference of non-wear factors such as oxide film breakdown, and ensures the reliability and effectiveness of the data.

[0017] 3、The present application updates the normalized reference of the dynamic resistance drop rate in real time based on the dynamic reference model of LSTM, automatically adjusts according to the changes of different welding materials and process parameters, so that the normalization result is not affected by factors such as plate material, thickness and welding current, pressure and the like, and the anti-interference ability of the detection is improved.

[0018] 4、The weight function of the present application dynamically adjusts according to the thickness of the oxide film of the welding material through the transfer learning model, and when the material is switched or the process is adjusted, the algorithm can automatically optimize the weight of each factor and the calculation logic, ensure the stability and accuracy of the evaluation method under different working conditions, and realize adaptive wear evaluation across working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a whole process schematic diagram of a multi-factor coupled welding electrode wear degree online evaluation method of an embodiment of the present application.

[0020] Figure 2 is a schematic diagram of a typical dynamic resistance curve of resistance spot welding of an embodiment of the present application.

[0021] Figure 3 is a calculation step flowchart of a trend change factor of an embodiment of the present application.

[0022] Figure 4 is a calculation step flowchart of a shape change factor of an embodiment of the present application.

[0023] Figure 5 is a calculation step flowchart of a peak time of an embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application will be further described in detail below with reference to the accompanying drawings.

[0025] Embodiment 1 Referring to Figure 1 , a multi-factor coupled welding electrode wear degree online evaluation method of the present embodiment comprises: On the basis of the existing current and voltage data in the welding process, the dynamic resistance is calculated according to the real-time current and voltage data, and the discrete data group of the resistance change with time is obtained; Dynamic data processing is performed on the acquired discrete data set, including obtaining the dynamic resistance change rate based on the resistance change and normalizing it, and calculating the trend change factor based on the normalization result. Static data processing is performed based on the acquired resistance data, including normalizing the dynamic resistance data and introducing a weighting function to calculate the shape change factor. Calculating peak time delay using discrete data sets includes determining peak arrival time based on dynamic resistance curves and calculating delay based on peak arrival time. The dynamic resistance decrease rate is calculated based on the dynamic resistance curve, and then normalized. The degree of electrode wear is determined by calculating the comprehensive change factor through weighted averaging and judging the value of the comprehensive change factor.

[0026] Specifically, a multi-factor coupled online evaluation method for the wear degree of welding electrodes includes the following steps: S1. Obtain current data during the welding process, and calculate the dynamic resistance based on the current data to obtain a discrete data set of resistance changes over time. like Figure 1 As shown, based on the acquired welding process current data set and voltage data group For each sampling point Calculate the dynamic resistance value, where k is the cumulative number of welding points (k=0 represents a brand new electrode), and i is the sampling point index. For each sampling point (i,k), calculate the resistance value according to Ohm's law, using the following formula: , in, Let represent the voltage at the i-th welding point during the k-th welding operation. Let the current at the i-th welding point during the k-th welding be represented. A discrete dynamic resistance data set is generated. The number of sampling points in one welding process is set to n, and the sampling interval is Δt (satisfying Δt ≤ 1 / 10kHz). Through n iterations, a discrete data set of dynamic resistance changing with time is generated. This data set serves as the foundational input for all subsequent data processing; it represents the dynamic resistance data collected during the resistance spot welding process. Inevitably affected by electromagnetic interference and contact noise, these noises are time-varying. Traditional filtering methods (such as mean filtering and Gaussian filtering) are difficult to retain the dynamic characteristics of the resistance curve while suppressing noise. Kalman filtering, as a recursive optimal estimation algorithm, constructs a state-space model and uses a weighted combination of the state estimate from the previous moment and the current observation to achieve real-time tracking of the true resistance state. It is particularly suitable for processing the mixed signal of slow resistance change characteristics and sudden noise during welding.

[0027] The Kalman filter algorithm is used to denoise the original dynamic resistance data, and a state space model is established: , wherein, is a state transition matrix, and is usually set to a unit matrix A=1 in a slow resistance change scenario, and represents the current time resistance state equal to the previous time state, is an observation matrix, which represents a linear relationship between the observation value and the true state, Q is a process noise covariance matrix, and the natural fluctuation of the modeled resistance true value is, and R is an observation noise covariance matrix, representing sensor errors and environmental disturbances, is the observed resistance at the i-th time, and the Kalman filter improves the signal-to-noise ratio of the data through the recursive mechanism of "prediction-update".

[0028] S2, according to the obtained discrete data group, dynamic data processing is carried out, including obtaining dynamic resistance change speed according to resistance change and normalizing, and calculating trend change factor based on the normalized result; As Figure 3 shown, the resistance difference between adjacent data points is calculated for the denoised dynamic resistance data group , , and the difference sequence is the dynamic resistance change speed sequence, which reflects the growth rate of the nugget, and a positive value indicates an increase in resistance, and a negative value indicates a decrease in resistance. Electrode wear will change the shape of the change speed curve, so it can be used as a wear characteristic index.

[0029] The value range of the original change speed sequence is affected by the welding parameters, and the fluctuation range is different under different working conditions, so it needs to be normalized to the [0, 1] interval to eliminate the dimension effect. Select the first m sampling points in the initial welding stage, and calculate the variance of the change speed: , wherein, is the mean of the first m points, and m represents the number of sampling points. The sensitivity coefficient is dynamically adjusted according to the variance value, and the formula is: , wherein, is the reference value, is the adjustment coefficient, when the contact fluctuation is large , , the Sigmoid function is reduced, which enhances the robustness to noise, and vice versa, , the sensitivity to small changes is improved, and for each change speed value , the modified Sigmoid function is applied: , wherein, represents the sensitivity coefficient, represents the difference between the resistances of adjacent data points, and the function is adjusted adaptively to avoid the information compression problem caused by outliers in traditional linear normalization.

[0030] At the full new state of the electrode (k = 0), a plurality of (such as 10 times) welding data are collected, and the average normalized change speed sequence is calculated , which is used as a standard reference, and the Euclidean distance between the normalized sequence of the kth welding and the standard sequence is calculated, and the trend change factor is determined according to the Euclidean distance, and the expression of the trend change factor is: , wherein, n is the total number of measured data points, represents the resistance value obtained after the un-worn electrode is normalized, represents the difference between the resistances of adjacent data points after normalization, quantifies the deviation of the resistance change trend of the current welding process from the standard state, and electrode wear will cause abnormal growth of the nugget (such as slowing down), which will change the shape of the change speed curve, thereby increasing the value of .

[0031] S3, based on the obtained resistance data, static data processing is performed, including normalization of dynamic resistance data, and introduction of a weight function to calculate a shape change factor; As shown in Figure 4 , first, the dynamic resistance data set after denoising is subjected to Z-score normalization, and the formula is: , wherein, is the average resistance of the kth welding, is the standard deviation of the resistance in the data set, and the resistance data under different working conditions are mapped to a standard normal distribution with a mean of 0 and a standard deviation of 1, so as to eliminate the influence of the absolute value difference of the resistance (such as the order of magnitude difference of the resistance of low carbon steel and aluminum alloy) on feature extraction, and and are calculated immediately after each data set is collected, so as to avoid calculation delay caused by accumulation of historical data, Welford algorithm is used to recursively calculate the mean and variance, so as to avoid precision loss under large data volume, and the formula is: , wherein, ​denotes the cumulative mean of dynamic resistance data after the (k-1)th welding, denotes the resistance value at the i th sampling point in the k th welding, k denotes the current cumulative welding times, denotes the cumulative variance of dynamic resistance data after the (k-1)th welding.

[0032] Then the logistic function is used as the weight function basic form: , wherein i denotes the sampling point index, denotes the offset, denotes the exponential coefficient, the electrode and the plate are unstable in contact (such as oxide film breakdown, micro-discharge) within 0-5 ms, the resistance data contains a large number of non-wear interference, and the weight of this stage is set to 0.2-0.5 through the weight function, the resistance change in the nugget formation stage (such as 5-80 ms) is strongly related to the electrode wear, and the weight of this stage is set to 0.8-1.0 through the weight function, the mapping relationship between the welding current-voltage waveform and the oxide film thickness is established through the previous experiment, the pre-trained CNN is used by using the transfer learning model, the features are extracted from the welding current-voltage waveform through the pre-trained CNN model, and the oxide film thickness estimation value d is mapped, and then the weight function parameters are adjusted: , 10 groups of welding data of a brand new electrode (k=0) are collected, normalized by Z-score and averaged to obtain a standard resistance sequence , as a reference benchmark, the normalized sequence of the k th welding is calculated, and the weighted Euclidean distance with the standard sequence is calculated, and the shape change factor is determined according to the Euclidean distance, and the shape change factor expression is: , wherein, denotes the resistance value of the i th sampling point of the dynamic resistance data group of the un-worn electrode after Z-score normalization processing, and n is the total number of data points, denotes the resistance value of the i th sampling point of the dynamic resistance data group of the k th welding after Z-score normalization processing, denotes the weight function.

[0033] S4, calculating the peak time delay amount by using discrete data groups, including determining the peak arrival time according to the dynamic resistance curve, and calculating the delay amount according to the peak arrival time; As shown in Figure 5 , the curve smoothing processing is performed on the denoised dynamic resistance data group, specifically, the Savitzky-Golay filtering is performed, a 5-point sliding window is used to fit a quadratic polynomial, and the formula is: , where, is the filter coefficient, which suppresses high-frequency noise while preserving the curve shape feature, for the smoothed resistance curve, the gradient sequence is calculated using the central difference method , where, is the sampling interval, the positive value of the gradient indicates the resistance rise (nugget growth stage), the negative value indicates the resistance drop (nugget stable or cooling stage), the peak value of the absolute value of the gradient corresponds to the inflection point of the resistance curve, which is related to the maximum value of the nugget growth rate, then, the mean value of the gradient sequence is calculated and the standard deviation , the primary threshold is set to , traverse the gradient sequence, extract the points that satisfy as candidate points, that is, the points where the gradient jumps from below the threshold to above the threshold, ensuring that the candidate points are located on the rising edge of the resistance.

[0034] For each candidate point i, search backward to the first point j where the gradient sign changes from positive to negative, that is , the interval [i,j] corresponds to the resistance rise segment, and the real peak point is the maximum resistance point in the interval, 20 groups of welding data of the fresh electrode (k=0) are collected, and the average peak time is calculated as the reference: , where, represents the peak time of the mth fresh electrode welding, for the kth welding, the real peak time is: , where, represents the time when the dynamic resistance curve reaches the real peak point during the kth welding, represents the index of the real peak point in the discrete data group, is the sampling interval, and the peak time delay is: , to eliminate the reference difference of welding of different materials, the delay is normalized: , where, and are the historical mean and standard deviation of the delay under the current material.

[0035] S5, calculate the dynamic resistance drop rate according to the dynamic resistance curve, and normalize the dynamic resistance drop rate; as shown in Figure 2 , the resistance value corresponding to the real peak point determined by step S4, that is , the dynamic resistance curve of resistance spot welding changes as shown in Figure 1 , the resistance value gradually decreases from the peak value reaching the intermediate C point until the end of welding. After the C point, the material strength decreases, and under the action of electrode pressure, the thickness of the joint area material decreases, resulting in a decrease in the total resistance. The dynamic resistance drop rate in the CD segment is positively correlated with the nugget size. Therefore, the dynamic resistance drop rate can be calculated, and the nugget size can be indirectly obtained through its value, and then the electrode wear condition can be judged. The dynamic resistance drop rate is defined as: , which represents the degree of resistance attenuation of the nugget from the maximum size to the cooling stage, and is related to the nugget solidification shrinkage and the contact state between the electrode and the plate. Electrode wear can cause the contact area to abnormally increase, resulting in increasing, and thus decreasing.

[0036] A single-layer LSTM (128 neurons) + fully connected layer is used, with the input being the drop rate sequence of the last 50 welds , and the output being the reference range of the current welding . The model is pre-trained using 200 weld data with a new electrode, and after each welding, the current drop rate is added to the training set. Gradient descent method is used to fine-tune the model parameters.

[0037] According to the LSTM, the extreme value update calculation is as follows: , where is the standard deviation of the historical drop rate, which is calculated recursively using the Welford algorithm, and a triangular window function is used in combination with the weight function of the welding stage division: , where represents the end of the initial interference stage, represents the beginning of the stable nugget growth period, represents the end of the nugget growth period, represents the beginning of the cooling stage, and the normalized expression of the dynamic resistance drop rate is: , where is the original dynamic resistance drop rate, is the historical maximum drop rate updated based on the LSTM dynamic reference model, is the historical minimum drop rate updated based on the LSTM dynamic reference model, is the time weight function, which ensures that the normalized result reflects the drop rate characteristics of the effective stage.

[0038] S6, calculating the comprehensive change factor by weighted average, judging the electrode wear degree according to the value of the comprehensive change factor.

[0039] The comprehensive change factor is calculated by weighted average based on each factor, and the linear weighted summation is adopted to integrate the trend change factor , the shape change factor , the normalized peak time delay amount and the normalized value of dynamic resistance drop rate Four-dimensional characteristics, the formula is: , Among them, indicates the trend change factor, indicates the weight of the trend change factor, indicates the shape change factor, indicates the normalized value of dynamic resistance drop rate, indicates the peak time delay amount, is the weight of the shape change factor, is the weight of the normalized value of dynamic resistance drop rate, is the weight of the peak time delay amount, by weighted fusion of multi-dimensional characteristics, the complex electrode wear state is converted into a single quantitative index.

[0040] In order to ensure that the factors with high reliability occupy a larger weight in the comprehensive evaluation, the inverse variance weighting strategy is adopted, first calculating the historical variance of each characteristic factor: for the last m welding data, the variance of each factor is calculated, and the initial weight of each factor is calculated according to the variance of each factor, taking the trend change factor as an example, the initial weight is Among them, indicates the historical variance of the trend change factor, indicates the sum of the inverse of the variance of the four characteristic factors (trend change factor, shape change factor, peak time delay amount, drop rate normalization value), this method makes the factor with small variance (i.e. small fluctuation, high reliability) obtain higher weight, in order to avoid the information overlap between factors leading to evaluation deviation, the covariance matrix of is constructed, and the expression of the covariance matrix is: , Among them, n indicates the number of data samples, indicates the characteristic factor vector of the kth welding, indicates the sample mean vector.

[0041] Finally, the objective function is constructed, wherein is a weight vector, the objective function reduces the coupling effect between factors by minimizing the weighted covariance, and meanwhile, constraint conditions are introduced to ensure the weight normalization and avoid the single factor dominating the evaluation result, the constraint optimization problem is converted into an unconstrained problem by using the Lagrange multiplier method: wherein, is a Lagrange multiplier, that is, an auxiliary variable introduced for converting the constrained optimization problem into an unconstrained problem, is a full one vector, is a covariance matrix, and partial derivatives of and are calculated and set to 0, the equation groups are solved to obtain the optimal weight vector, and then the comprehensive change factor .

[0042] The mapping relationship between the comprehensive change factor and the electrode wear amount is established, the wear grade threshold is divided, and then the electrode wear degree is judged.

[0043] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, so: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.​​

Claims

1. A method for online evaluation of the wear degree of welding electrodes using multi-factor coupling, characterized in that, include: Based on the existing current and voltage data in the welding process, the dynamic resistance is calculated according to the real-time current and voltage data to obtain a discrete data set of resistance changing with time. Dynamic data processing is performed on the acquired discrete data set, including obtaining the dynamic resistance change rate based on the resistance change and normalizing it, and calculating the trend change factor based on the normalization result. Static data processing is performed based on the acquired resistance data, including normalizing the dynamic resistance data and introducing a weighting function to calculate the shape change factor. Calculating peak time delay using discrete data sets includes determining peak arrival time based on dynamic resistance curves and calculating delay based on peak arrival time. The dynamic resistance decrease rate is calculated based on the dynamic resistance curve, and then normalized. The degree of electrode wear is determined by calculating the comprehensive change factor through weighted averaging and judging the value of the comprehensive change factor.

2. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling as described in claim 1, characterized in that, The process of calculating dynamic resistance based on real-time current and voltage data includes acquiring current and voltage signals at fixed time intervals during resistance spot welding, generating an original data set containing n data points, calculating the resistance value for each sampling point based on Ohm's law using the original data set, generating a discrete dynamic resistance data set, and denoising the dynamic resistance data using a Kalman filter algorithm.

3. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 1, characterized in that, The dynamic data processing based on the acquired discrete data set includes calculating the resistance difference between two adjacent data points in the discrete data set to obtain the dynamic resistance change rate throughout the welding process, normalizing the dynamic resistance change rate using a modified sigmoid function, calculating the Euclidean distance between the current data set and the standard data set of the unworn electrode based on the normalization result, and determining the trend change factor based on the Euclidean distance. The expression for the trend change factor is as follows: ， Where n is the total number of data points measured. This is expressed as the resistance value obtained after normalization of the unworn electrode. It is expressed as the difference in resistance between two adjacent data points after normalization.

4. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling as described in claim 3, characterized in that, The normalization of the dynamic resistance change rate using the modified sigmoid function includes calculating the variance of the resistance change rate for the first m sampling points in the initial welding stage, determining a sensitivity coefficient based on the variance of the resistance change rate, and normalizing the dynamic resistance change rate using the modified sigmoid function based on the sensitivity coefficient. The normalization formula is as follows: ， in, Represents the sensitivity coefficient. It is expressed as the difference in resistance between two adjacent data points.

5. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 1, characterized in that, The static data processing based on the acquired resistance data includes Z-score normalization of the discrete resistance data set, followed by the introduction of a weighting function to reduce the weight of the data in the initial welding stage. The Euclidean distance between the current normalized dynamic resistance data set and the standard data set of the unworn electrode is calculated to obtain the shape change factor. The expression for the shape change factor is: ， in, This represents the resistance value of the i-th sampling point after Z-score normalization of the dynamic resistance data set of the unworn electrode, where n is the total number of data points. This represents the resistance value of the i-th sampling point after Z-score normalization of the dynamic resistance data set from the k-th welding operation. It is represented as a weight function.

6. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 5, characterized in that, The introduction of a weighting function to reduce the weight of initial welding data includes determining the exponential term of the weighting function based on the offset and exponential coefficient, calculating the weighting function value through the exponential term, and using a transfer learning model to dynamically adjust the weighting function based on the oxide film thickness of the welding material. The expression for the weighting function is as follows: ， Where i represents the sampling point index, Represented as offset, It is represented as an exponential coefficient.

7. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 1, characterized in that, The method of calculating the peak time delay using discrete data sets includes determining a dynamic resistance curve based on the discrete data sets, calculating a resistance curve gradient sequence based on the dynamic resistance curve, setting a primary threshold and determining candidate points by combining the resistance curve gradient sequence, detecting gradient sign changes on the candidate points to obtain the true peak point, finding the time when the resistance reaches the true peak point during the welding process, and subtracting the time of the true peak of the resistance during the welding process from the peak time of the standard dynamic resistance curve to obtain the peak time delay.

8. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 1, characterized in that, The normalization process for the dynamic resistance decrease rate includes locating the maximum peak resistance and the resistance at the end of welding using a gradient detection method, updating the maximum peak resistance and the resistance at the end of welding using an LSTM-based dynamic baseline model, calculating the dynamic resistance decrease rate based on the updated peak resistance and the resistance at the end of welding, applying a weight to the dynamic resistance decrease rate according to the time position of the dynamic resistance curve, and normalizing the dynamic resistance decrease rate. The normalization expression is as follows: ， in, This is expressed as the rate of decrease in the original dynamic resistance. Represented as the historical maximum rate of decline updated based on the LSTM dynamic baseline model. Represented as the historical minimum rate of decline updated based on the LSTM dynamic benchmark model. This is represented as a time weighting function.

9. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 1, characterized in that, The calculation of the comprehensive change factor by weighted average includes weighting the trend change factor, shape change factor, normalized peak time delay, and normalized value of dynamic resistance decrease rate to obtain the comprehensive change factor. The expression for the comprehensive change factor is as follows: ， in, Represented as trend change factor, This is represented by the weight of the trend change factor. Represented as shape change factor, Expressed as the normalized value of the rate of decrease in dynamic resistance. This is expressed as the peak time delay. The weights of the shape change factor, The weights for the normalized values ​​of the dynamic resistance decrease rate. The weight of the peak time delay.

10. The online evaluation method for the wear degree of welding electrodes using multi-factor coupling according to claim 9, characterized in that, The weighted average employs an inverse variance weighting method, calculating the weights of each factor based on its historical variance. The factor coupling is constrained by the covariance matrix, and an objective function is constructed. The weight vector is then optimized using these constraints. By introducing Lagrange multipliers, the constrained optimization problem is transformed into an unconstrained problem, yielding the optimal weight vector. The covariance matrix expression is as follows: ， Where n represents the number of data samples, Let it be represented as the feature factor vector of the k-th welding. It is represented as a sample mean vector.

Images