A spot welding nugget abnormality diagnosis method and system based on dynamic resistance

Abstract

The application discloses a spot welding nugget abnormality diagnosis method and system based on dynamic resistance, comprising the following steps: acquiring dynamic resistance data collected in real time during spot welding; performing normalization processing on the dynamic resistance data to obtain normalized dynamic resistance data; inputting the normalized dynamic resistance data into a pre-trained spot welding nugget diagnosis model to output a nugget diameter prediction value; and determining whether the current welding spot is abnormal according to the comparison result of the nugget diameter prediction value and a preset qualified threshold. The method and system combine the dynamic resistance curve with the deep learning model, realize online real-time prediction and abnormality diagnosis of the nugget diameter, overcome the problems of the traditional detection methods, such as incomplete detection, detection lag and low precision, can automatically trigger an alarm and parameter compensation, and form quality closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of resistance spot welding quality inspection technology, and in particular to a method and system for diagnosing abnormal spot weld nuggets based on dynamic resistance. Background Technology

[0002] Resistance spot welding is a primary process for joining thin metal sheets and is widely used in automotive body welding. The weld nugget diameter and penetration depth are core indicators for evaluating spot weld quality, directly determining the mechanical properties and connection strength of the weld. Currently, methods for inspecting resistance spot weld quality are mainly divided into two categories: destructive testing and non-destructive testing.

[0003] Destructive testing is the core basis for traditional spot weld quality assessment, mainly including metallographic sectioning and peel testing. Metallographic sectioning involves cutting the weldment containing the weld point into metallographic specimens, which are then ground, polished, and etched before being directly observed and measured under a metallographic microscope to determine the effective diameter and penetration depth of the weld nugget. Peel testing involves mechanically tearing apart the two overlapping metal layers to observe the actual shape of the weld nugget and measure its diameter. While these two methods can accurately obtain weld nugget dimensions, they require damaging the weldment and cannot achieve full inspection; they are only suitable for process verification and batch sampling.

[0004] Non-destructive testing is mainly used for quickly screening obviously defective welds, including percussion testing and ultrasonic testing. Percussion testing uses a small hammer to tap the weld area, indirectly judging the weld nugget formation by the crispness or dullness of the sound. This method relies on operator experience, is highly subjective, and cannot provide quantitative measurement. Ultrasonic testing utilizes the reflection characteristics of ultrasound at the interface between the weld nugget and the base material, roughly determining the presence of a weld nugget and estimating its diameter through waveform and image analysis. However, its accuracy is lower due to factors such as weld structure and plate thickness combination; qualified parts still need to be verified in conjunction with destructive testing.

[0005] The aforementioned traditional testing methods have the following shortcomings: First, destructive testing cannot achieve full inspection due to sample scrapping, while non-destructive testing has low accuracy and strong subjectivity. The risk of missed detections and misjudgments is high under full inspection mode. Currently, the commonly used model is "first-piece destructive testing plus batch non-destructive sampling inspection," which creates blind spots in quality control. Second, most testing is conducted offline. Metallographic sections need to be completed in the laboratory, and peel tests are performed on the workshop sampling station. Test results cannot be fed back to the welding production line in real time. When welding parameters drift or electrode wear causes weld nugget dimensions to be unqualified, adjustments can only be made after sampling inspection. During this period, a large number of defective products may have already been generated, resulting in a serious lag in quality control.

[0006] Therefore, how to achieve real-time, online, non-destructive testing of the diameter of resistance spot weld nuggets has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] To address the technical problems existing in the background art, this invention proposes a method and system for diagnosing abnormal spot weld nuggets based on dynamic resistance.

[0008] This invention proposes a method for diagnosing abnormalities in spot weld nuggets based on dynamic resistance, comprising the following steps: S1. Acquire dynamic resistance data collected in real time during the spot welding process; S2. Normalize the dynamic resistance data to obtain normalized dynamic resistance data. S3. Input the normalized dynamic resistance data into the pre-trained spot weld nugget diagnostic model and output the predicted value of the weld nugget diameter. S4. Based on the comparison between the predicted weld nugget diameter and the preset qualified threshold, determine whether the current weld point is abnormal.

[0009] Preferably, the spot weld nugget diagnostic model includes: The input layer is used to receive normalized dynamic resistance data. A data augmentation layer, connected between the input layer and the first convolutional layer, is used to enhance the features of the normalized dynamic resistance data. The data augmentation layer employs a structure combining sliding window differencing and temporal attention. It calculates the first and second differences of the normalized dynamic resistance data through a sliding window, representing the resistance change rate and acceleration, respectively. A temporal attention mechanism is introduced, assigning different attention weights to the normalized dynamic resistance data, the first-order difference data, and the second-order difference data to highlight the features of resistance abrupt change points. The output dimension of the data augmentation layer is consistent with the input dimension of the input layer, and the enhanced feature data is passed to subsequent convolutional layers. Multiple sequentially connected convolutional and pooling layers are used to extract local features from the enhanced feature data; The fully connected layer, connected to the last pooling layer, is used to integrate the extracted local features; The output layer, connected to the fully connected layer, is used to output the predicted value of the weld nugget diameter.

[0010] Preferably, the training process of the spot weld nugget diagnostic model specifically includes: Construct a sample set, which includes multiple historical dynamic resistance data and the actual weld nugget diameter corresponding to each historical dynamic resistance data; The historical dynamic resistance data in the sample set is preprocessed to remove outlier and duplicate data, and the retained historical dynamic resistance data is normalized. An initial spot weld nugget diagnostic model was built, using normalized historical dynamic resistance data as input and the corresponding actual weld nugget diameter as the output target. The initial spot weld nugget diagnostic model was then trained to obtain a trained spot weld nugget diagnostic model.

[0011] Preferably, the preprocessing of historical dynamic resistance data in the sample set specifically includes: The first set of screening data is obtained by removing dynamic resistance curves without peak values ​​and dynamic resistance data with abrupt changes in resistance. The second screening data is obtained by removing duplicate samples from the first screening data whose process parameters are exactly the same as the dynamic resistance curve; The dynamic resistance data in the second screening data is normalized to obtain the preprocessed historical dynamic resistance data.

[0012] Preferably, when constructing the sample set, each historical dynamic resistance data is bound to the corresponding solder joint unique identifier, and the actual solder joint diameter is obtained by metallographic section measurement of the solder joint. During the measurement, the average value of three measurements of the maximum inscribed circle diameter of the weld joint is taken.

[0013] Preferably, the sample set is divided into training set, validation set and test set according to the ratio of qualified solder joints to unqualified solder joints and the batch of base material, for training, validation and testing of the model.

[0014] Preferably, the initial spot weld nugget diagnostic model extracts feature parameters of the dynamic resistance curve during training. These feature parameters include peak resistance, peak occurrence time, slope of the falling phase, initial resistance, final resistance, total resistance change, thermal accumulation resistance integral, and duration of the resistance plateau. The falling phase slope is the ratio of the resistance change from the peak resistance point to the welding end point to the time change. The total resistance change is the difference between the peak resistance and the final resistance. The thermal accumulation resistance integral is the area enclosed by the dynamic resistance curve and the time axis.

[0015] Preferably, step S4 specifically includes: If the predicted weld nugget diameter is within the preset qualified threshold range, it is determined to be a qualified weld point, and the weld point will proceed normally. If the predicted weld nugget diameter exceeds the preset qualified threshold but does not exceed the preset serious unqualified threshold, it is judged as a slightly unqualified weld nugget, an audible and visual alarm is issued, and the predicted weld nugget diameter is recorded. If the predicted weld nugget diameter exceeds the preset critical non-compliance threshold, it is determined to be a critical non-compliance weld, a shutdown alarm is issued, and the parameter compensation operation of the spot welding controller is triggered. The parameter compensation operation includes automatically increasing the welding current or extending the welding time until the predicted weld nugget diameter of subsequent welds returns to the preset qualified threshold range.

[0016] Preferably, step S2 specifically includes: Calculate the mean and standard deviation of the dynamic resistance data; The normalized dynamic resistance data is obtained by subtracting the mean from each value in the dynamic resistance data and then dividing by the standard deviation.

[0017] This invention proposes a spot weld nugget anomaly diagnosis system based on dynamic resistance, comprising: The data acquisition module is used to collect dynamic resistance data in real time during the spot welding process; The data preprocessing module is used to normalize the dynamic resistance data and output normalized dynamic resistance data. The model inference module is internally deployed with a pre-trained spot weld core diagnostic model, which is used to input the normalized dynamic resistance data into the spot weld core diagnostic model and output the predicted value of the weld core diameter. The judgment and feedback module is used to generate a judgment signal based on the comparison result between the predicted weld diameter and the preset qualified threshold, and to control the alarm device to perform an alarm operation and / or control the spot welding controller to perform a parameter compensation operation based on the judgment signal.

[0018] The proposed method and system for abnormal spot welding nugget diagnosis based on dynamic resistance in this invention achieves online prediction and anomaly determination of the nugget diameter by real-time acquisition of dynamic resistance data during the spot welding process and combining it with a pre-trained spot welding nugget diagnosis model. This overcomes the shortcomings of traditional destructive testing, which cannot perform full inspection, and non-destructive testing, which has low accuracy. It can detect abnormal nugget size in real time during the welding process, avoiding the generation of batches of defective products. At the same time, this invention does not require the addition of complex hardware equipment; it can complete quality diagnosis using only the electrical signals of the welding process itself, which is low-cost and easy to deploy on existing production lines. In addition, it can automatically trigger audible and visual alarms or parameter compensation operations based on the prediction results, realizing closed-loop quality control and significantly improving the intelligence level and quality control capabilities of the spot welding production line. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the workflow of a spot weld nugget anomaly diagnosis method based on dynamic resistance proposed in this invention. Figure 2 This is a schematic diagram of the architecture of a spot weld nugget diagnosis model for a spot weld nugget anomaly diagnosis method based on dynamic resistance proposed in this invention. Figure 3 This is a schematic diagram of the system architecture of a spot weld nugget anomaly diagnosis system based on dynamic resistance proposed in this invention. Detailed Implementation

[0020] Reference Figure 1 and Figure 2The present invention proposes a method for diagnosing abnormalities in spot weld nuggets based on dynamic resistance, comprising the following steps: S1. Acquire dynamic resistance data collected in real time during the spot welding process.

[0021] It should be noted that on the resistance spot welding production line, the welding current and inter-electrode voltage of each weld point are collected in real time using current and voltage sensors built into the welding controller. Based on Ohm's law, the resistance value R=U / I at each sampling moment is calculated, generating a dynamic resistance curve with time on the x-axis and resistance value on the y-axis—this is the dynamic resistance data. Each dynamic resistance data point is uniquely linked to its corresponding weld point for subsequent quality traceability.

[0022] S2. Normalize the dynamic resistance data to obtain normalized dynamic resistance data.

[0023] In this embodiment, step S2 specifically includes: Calculate the mean and standard deviation of the dynamic resistance data; The normalized dynamic resistance data is obtained by subtracting the mean from each value in the dynamic resistance data and then dividing by the standard deviation.

[0024] It should be noted that due to the different combinations of materials and plate thicknesses in the spot welding process, the numerical range of the dynamic resistance varies. In order to eliminate the influence of dimensions and improve the stability of model inference, the original dynamic resistance data needs to be normalized.

[0025] S3. Input the normalized dynamic resistance data into the pre-trained spot weld nugget diagnostic model and output the predicted value of the weld nugget diameter.

[0026] In this embodiment, as Figure 2 As shown, the spot weld nugget diagnostic model includes: The input layer is used to receive normalized dynamic resistance data. A data augmentation layer, connected between the input layer and the first convolutional layer, is used to enhance the features of the normalized dynamic resistance data. The data augmentation layer employs a structure combining sliding window differencing and temporal attention. It calculates the first and second differences of the normalized dynamic resistance data through a sliding window, representing the resistance change rate and acceleration, respectively. A temporal attention mechanism is introduced, assigning different attention weights to the normalized dynamic resistance data, the first-order difference data, and the second-order difference data to highlight the features of resistance abrupt change points. The output dimension of the data augmentation layer is consistent with the input dimension of the input layer, and the enhanced feature data is passed to subsequent convolutional layers. Multiple sequentially connected convolutional and pooling layers are used to extract local features from the enhanced feature data; The fully connected layer, connected to the last pooling layer, is used to integrate the extracted local features; The output layer, connected to the fully connected layer, is used to output the predicted value of the weld nugget diameter.

[0027] Specifically, such as Figure 2 As shown, the first and second differences of the dynamic resistance data are calculated using a sliding window with a window size of 50. The first difference represents the resistance change rate, corresponding to the temperature change rate during weld nugget formation; the second difference represents the resistance change acceleration, corresponding to the trend of weld nugget growth rate. A temporal attention mechanism is then introduced. This mechanism uses a simplified structure of a compressed excitation network, generating channel statistics through global average pooling. Attention weights are learned through two fully connected layers, assigning different attention weights to the original dynamic resistance data, first-order difference data, and second-order difference data. This highlights the characteristics of resistance abrupt change points, which correspond to the critical stages of weld nugget initiation and rapid growth, while suppressing the influence of irrelevant noise data. The output dimension of the data augmentation layer remains consistent with the input dimension to ensure compatibility with subsequent convolutional layers. It only enhances the feature expression without changing the data dimension, allowing the augmented feature data to be seamlessly transferred to subsequent convolutional layers for deep feature extraction.

[0028] In this embodiment, the training process of the spot weld nugget diagnostic model specifically includes: Construct a sample set, which includes multiple historical dynamic resistance data and the actual weld nugget diameter corresponding to each historical dynamic resistance data; The historical dynamic resistance data in the sample set is preprocessed to remove outlier and duplicate data, and the retained historical dynamic resistance data is normalized. An initial spot weld nugget diagnostic model was built, using normalized historical dynamic resistance data as input and the corresponding actual weld nugget diameter as the output target. The initial spot weld nugget diagnostic model was then trained to obtain a trained spot weld nugget diagnostic model.

[0029] Specifically, when constructing the sample set, each historical dynamic resistance data is bound to the corresponding unique identifier of the solder joint. The actual solder core diameter is obtained by metallographic section measurement of the solder joint, and the average value of three measurements of the maximum inscribed circle diameter of the weld core area is taken during the measurement.

[0030] Specifically, multiple complete dynamic resistance curves are collected from historical production data, with each curve corresponding to a solder joint. Simultaneously, the actual weld nugget diameter for each solder joint is obtained using metallographic sectioning, serving as the sample's label value. The specific procedure involves cutting the weldment containing the solder joint into metallographic samples, grinding and polishing them, etching them with a nitric acid-alcohol solution, observing the weld nugget area under a metallographic microscope, measuring the maximum inscribed circle diameter of the weld nugget, repeating the measurement three times, and taking the average of the three measurements as the actual weld nugget diameter for that solder joint. Each dynamic resistance curve is uniquely linked to its corresponding solder joint identifier, ensuring a one-to-one correspondence between the resistance curve and the weld nugget size.

[0031] When constructing the sample set, it is necessary to cover solder joint data with different process parameters, different batches of base materials, and different electrode wear stages, and ensure that at least 10% of the samples are unqualified solder joints in order to improve the model's ability to identify abnormal states.

[0032] In this embodiment, the historical dynamic resistance data in the sample set is preprocessed, specifically including: The first set of screening data is obtained by removing dynamic resistance curves without peak values ​​and dynamic resistance data with abrupt changes in resistance. The second screening data is obtained by removing duplicate samples from the first screening data whose process parameters are exactly the same as the dynamic resistance curve; The dynamic resistance data in the second screening data is normalized to obtain the preprocessed historical dynamic resistance data.

[0033] Specifically, preprocessing includes three sub-steps: anomaly removal, deduplication, and normalization. First, abnormal data are eliminated. Dynamic resistance curves without significant peak values ​​(indicating that the melt nucleus has not formed normally), curves with sudden changes in resistance (indicating poor contact or abnormal data acquisition), and samples with large corresponding metallographic measurement errors are eliminated to obtain the first screening data.

[0034] Secondly, duplicate data is eliminated. From the first set of data, samples with completely identical process parameters and dynamic resistance curve shapes are identified as duplicate samples. Only one of these samples is retained to avoid data redundancy that could lead to model overfitting, thus obtaining the second set of data.

[0035] Finally, normalization is performed; for each dynamic resistance curve in the second screening data, the Z-score normalization method is used to normalize it, and the preprocessed historical dynamic resistance data is obtained.

[0036] In this embodiment, the sample set is divided into training set, validation set and test set according to the ratio of qualified and unqualified solder joints and the batch of base material, for the training, validation and testing of the model.

[0037] Specifically, the preprocessed sample set is stratified into training, validation, and test sets based on the ratio of qualified to unqualified solder joints and the batch of base material. For example, the sets are randomly divided at a ratio of 70%, 15%, and 15%, while ensuring that the distribution ratio of qualified to unqualified samples in each set is basically consistent with that of the original sample set to avoid sample bias.

[0038] In this embodiment, the initial spot weld nugget diagnostic model extracts feature parameters of the dynamic resistance curve during the training process. The feature parameters include the resistance peak, peak occurrence time, slope of the falling phase, initial resistance, final resistance, total resistance change, thermal accumulation resistance integral, and duration of the resistance plateau. Among them, the slope of the falling phase is the ratio of the resistance change from the resistance peak to the welding end point to the time change, the total resistance change is the difference between the resistance peak and the final resistance, and the thermal accumulation resistance integral is the area enclosed by the dynamic resistance curve and the time axis.

[0039] S4. Based on the comparison between the predicted weld nugget diameter and the preset qualified threshold, determine whether the current weld point is abnormal.

[0040] In this embodiment, step S4 specifically includes: If the predicted weld nugget diameter is within the preset qualified threshold range, it is determined to be a qualified weld point, and the weld point will proceed normally. If the predicted weld nugget diameter exceeds the preset qualified threshold but does not exceed the preset serious unqualified threshold, it is judged as a slightly unqualified weld nugget, an audible and visual alarm is issued, and the predicted weld nugget diameter is recorded. If the predicted weld nugget diameter exceeds the preset critical non-conformity threshold, it is determined to be a critical non-conformity weld, a shutdown alarm is issued, and the parameter compensation operation of the spot welding controller is triggered. The parameter compensation operation includes automatically increasing the welding current or extending the welding time until the predicted weld nugget diameter of subsequent welds returns to the preset qualified threshold range.

[0041] In this embodiment, an online model update step is also included: We regularly sample weld points from the production line for metallographic testing to obtain new actual weld nugget diameters, and supplement the sample set with the corresponding dynamic resistance data as new samples. The spot weld nugget diagnostic model was incrementally trained using the updated sample set to update the model parameters.

[0042] Reference Figures 1-3 The present invention proposes a spot weld nugget anomaly diagnosis system based on dynamic resistance, comprising: The data acquisition module is used to collect dynamic resistance data in real time during the spot welding process; The data preprocessing module is used to normalize the dynamic resistance data and output normalized dynamic resistance data. The model inference module contains a pre-trained spot weld core diagnostic model, which is used to input normalized dynamic resistance data into the spot weld core diagnostic model and output the predicted value of the weld core diameter. The judgment and feedback module is used to generate a judgment signal based on the comparison result between the predicted value of the weld nugget diameter and the preset qualified threshold, and to control the alarm device to perform alarm operation and / or control the spot welding controller to perform parameter compensation operation based on the judgment signal.

[0043] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing abnormalities in spot weld nuggets based on dynamic resistance, characterized in that, Includes the following steps: S1. Acquire dynamic resistance data collected in real time during the spot welding process; S2. Normalize the dynamic resistance data to obtain normalized dynamic resistance data. S3. Input the normalized dynamic resistance data into the pre-trained spot weld nugget diagnostic model and output the predicted value of the weld nugget diameter. S4. Based on the comparison between the predicted weld nugget diameter and the preset qualified threshold, determine whether the current weld point is abnormal.

2. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 1, characterized in that, The spot weld nugget diagnostic model includes: The input layer is used to receive normalized dynamic resistance data. A data augmentation layer, connected between the input layer and the first convolutional layer, is used to enhance the features of the normalized dynamic resistance data. The data augmentation layer employs a structure combining sliding window differencing and temporal attention. It calculates the first and second differences of the normalized dynamic resistance data through a sliding window, representing the resistance change rate and acceleration, respectively. A temporal attention mechanism is introduced, assigning different attention weights to the normalized dynamic resistance data, the first-order difference data, and the second-order difference data to highlight the features of resistance abrupt change points. The output dimension of the data augmentation layer is consistent with the input dimension of the input layer, and the enhanced feature data is passed to subsequent convolutional layers. Multiple sequentially connected convolutional and pooling layers are used to extract local features from the enhanced feature data; The fully connected layer, connected to the last pooling layer, is used to integrate the extracted local features; The output layer, connected to the fully connected layer, is used to output the predicted value of the weld nugget diameter.

3. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 1, characterized in that, The training process of the spot weld nugget diagnostic model specifically includes: Construct a sample set, which includes multiple historical dynamic resistance data and the actual weld nugget diameter corresponding to each historical dynamic resistance data; The historical dynamic resistance data in the sample set is preprocessed to remove outlier and duplicate data, and the retained historical dynamic resistance data is normalized. An initial spot weld nugget diagnostic model was built, using normalized historical dynamic resistance data as input and the corresponding actual weld nugget diameter as output target. The initial spot weld nugget diagnostic model was then trained to obtain a trained spot weld nugget diagnostic model.

4. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 3, characterized in that, The preprocessing of historical dynamic resistance data in the sample set specifically includes: The first set of screening data is obtained by removing dynamic resistance curves without peak values ​​and dynamic resistance data with abrupt changes in resistance. The second screening data is obtained by removing duplicate samples from the first screening data whose process parameters are exactly the same as the dynamic resistance curve; The dynamic resistance data in the second screening data is normalized to obtain the preprocessed historical dynamic resistance data.

5. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 3, characterized in that, When constructing the sample set, each historical dynamic resistance data is bound to the corresponding solder joint unique identifier. The actual solder joint diameter is obtained by metallographic section measurement of the solder joint. During the measurement, the average value of three measurements of the maximum inscribed circle diameter of the weld nugget area is taken.

6. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 3, characterized in that, The sample set is divided into training set, validation set and test set according to the ratio of qualified and unqualified solder joints and the batch of base material, for the training, validation and testing of the model.

7. The method for abnormal spot weld nugget diagnosis based on dynamic resistance according to claim 3, characterized in that, The initial spot weld nugget diagnostic model extracts feature parameters of the dynamic resistance curve during training. These feature parameters include peak resistance, peak occurrence time, slope of the falling phase, initial resistance, final resistance, total resistance change, thermal accumulation resistance integral, and duration of the resistance plateau. The falling phase slope is the ratio of the resistance change from the peak resistance point to the welding end point to the time change. The total resistance change is the difference between the peak resistance and the final resistance. The thermal accumulation resistance integral is the area enclosed by the dynamic resistance curve and the time axis.

8. The method for diagnosing spot weld nugget anomalies based on dynamic resistance according to claim 1, characterized in that, Step S4 specifically includes: If the predicted weld nugget diameter is within the preset qualified threshold range, it is determined to be a qualified weld point, and the weld point will proceed normally. If the predicted weld nugget diameter exceeds the preset qualified threshold but does not exceed the preset serious unqualified threshold, it is judged as a slightly unqualified weld nugget, an audible and visual alarm is issued, and the predicted weld nugget diameter is recorded. If the predicted weld nugget diameter exceeds the preset critical non-compliance threshold, it is determined to be a critical non-compliance weld, a shutdown alarm is issued, and the parameter compensation operation of the spot welding controller is triggered. The parameter compensation operation includes automatically increasing the welding current or extending the welding time until the predicted weld nugget diameter of subsequent welds returns to the preset qualified threshold range.

9. The method for diagnosing spot weld nugget anomalies based on dynamic resistance according to claim 1, characterized in that, Step S2 specifically includes: Calculate the mean and standard deviation of the dynamic resistance data; The normalized dynamic resistance data is obtained by subtracting the mean from each value in the dynamic resistance data and then dividing by the standard deviation.

10. A spot weld nugget anomaly diagnosis system based on dynamic resistance, characterized in that, include: The data acquisition module is used to collect dynamic resistance data in real time during the spot welding process; The data preprocessing module is used to normalize the dynamic resistance data and output normalized dynamic resistance data. The model inference module is internally deployed with a pre-trained spot weld core diagnostic model, which is used to input the normalized dynamic resistance data into the spot weld core diagnostic model and output the predicted value of the weld core diameter. The judgment and feedback module is used to generate a judgment signal based on the comparison result between the predicted weld diameter and the preset qualified threshold, and to control the alarm device to perform an alarm operation and / or control the spot welding controller to perform a parameter compensation operation based on the judgment signal.

Images