A method and system for predicting tensile shear force in resistance spot welding

Abstract

The application discloses a resistance spot welding tensile shear force prediction method and system based on physical feature stability screening and Monte Carlo cross-validation. Welding current, welding time, electrode force and plate thickness are obtained, and physical intermediate features are generated through transient heat conduction simulation; under multiple rounds of same random division, the prediction performance difference distribution of the complete feature set and the feature set removed item by item is compared, and a simplified feature set is constructed in combination with correlation; a physical anchor model is established with the simplified features, a tensile shear force prediction distribution is formed by summarizing the output of multiple rounds of models, and a compliance probability and uncertainty are calculated; and according to the result, a batch release, expanded sampling, batch isolation, nondestructive detection review or sample tensile shear verification instruction is output. The application reduces the prediction fluctuation caused by single division and noise features through multiple rounds of paired random division, provides an interpretable basis for small sample spot quality control, and converts model fluctuation into compliance probability and dispersion to provide inspection instructions matched with prediction risks for manufacturing sites.

Description

Technical Field

[0001] This invention relates to the field of resistance spot welding quality inspection technology, and specifically discloses a method and system for predicting the tensile shear force of resistance spot welding. Background Technology

[0002] Resistance spot welding is widely used in the manufacture of automotive body-in-white and thin-sheet structural components. During resistance spot welding, the quality of the weld directly affects the strength and service life of the welded structure. One of the key indicators of weld quality is the tensile shear force of the weld, which is the maximum shear force the weld can withstand under shear stress, reflecting its ability to resist deformation and fracture. Accurately predicting the tensile shear force of the weld is particularly important to ensure welding quality. Tensile shear force is a crucial indicator for evaluating the load-bearing capacity of a weld; however, tensile shear tests can damage the specimen, and on the production floor, indirect quality control is typically achieved through witness testing, sampling inspection, or non-destructive testing.

[0003] Existing data-driven prediction methods are susceptible to the impact of training-test splitting when sample size is limited. Physical simulations can generate intermediate quantities such as melt core geometry, heat input, and energy utilization, but directly inputting all intermediate quantities into the prediction model may introduce noisy and highly correlated features. Evaluating feature contributions and model performance using only a single data splitting can also allow accidental data composition to simultaneously influence feature selection and performance conclusions.

[0004] Therefore, a technical solution is needed that connects physical feature screening, repeated random division evaluation, tensile shear force prediction distribution and on-site inspection actions, so that the prediction system can identify stable and effective physical signals and transform prediction uncertainty into executable quality review instructions. Summary of the Invention

[0005] (a) Technical problems to be solved Based on this, the present invention aims to reduce the impact of noise physical characteristics and single random partitioning on the prediction of tensile and shear forces in small-sample resistance spot welding under small sample conditions, and to provide inspection instructions that match the predicted risks to the manufacturing site.

[0006] (II) Technical Solution To address the aforementioned technical problems, this invention proposes a method for predicting the tensile and shear force of resistance spot welding. This method is used to predict the tensile and shear force of resistance spot welding under small sample conditions and output the probability of achieving a preset tensile and shear force standard. The method includes: S1. Data acquisition: Acquire the welding current, welding time, electrode force, plate thickness, and tensile shear force corresponding to each resistance spot welding sample for multiple resistance spot welding samples; S2. Physical simulation: Input the process parameters of each resistance spot welding sample obtained in step S1 into the transient heat conduction physical simulation model to obtain multiple candidate physical intermediate features. S3. Paired random partitioning: A multi-round random partitioning sequence is preset, and the resistance spot welding samples are repeatedly divided into training set and test set. In each round of partitioning, the same training set and test set are used to evaluate the prediction performance of the complete feature set and the prediction performance after removing the candidate physical intermediate features one by one, so as to obtain the pairing performance difference distribution of each candidate physical intermediate feature. S4. Stability screening: Construct a simplified physical feature set based on the correlation between the pairing performance difference distribution and the candidate physical intermediate features; S5. Physical Anchoring Prediction Model: The physical anchoring prediction model is trained based on the process parameters obtained in step S1 and the simplified physical feature set. S6. Uncertainty assessment: Summarize the tensile and shear force prediction results of multiple physical anchoring prediction models for the target weld point to form a tensile and shear force prediction distribution, and calculate the prediction mean, prediction dispersion and the probability of meeting the preset tensile and shear force standard for each resistance spot welding sample. S7. Based on the compliance probability and predicted dispersion of each resistance spot welding sample, output at least one of the following inspection instructions through the inspection instruction output interface: batch release, expanded sampling, batch isolation, non-destructive testing verification, or sample tensile shear verification.

[0007] Preferably, in step S3, the multi-round random partitioning uses the same random seed sequence to evaluate the complete feature set and the feature set removed item by item, so that the performance difference of each candidate physical intermediate feature in each partitioning round is obtained by pairing the same training samples and the same test samples.

[0008] Preferably, in step S3, the inverse value of the coefficient of determination or error index is used as the unified performance score; when the mean of the pairing performance difference after removing the candidate physical intermediate features is greater than the positive threshold, the candidate physical intermediate features are determined to be positive features; when the mean of the pairing performance difference after removing the candidate physical intermediate features is not greater than zero, the candidate physical intermediate features are determined to be noise features, and the stability is determined based on the standard deviation or quantile interval of the pairing performance difference.

[0009] Preferably, when the absolute value of the correlation coefficient between two candidate physical intermediate features is greater than a preset correlation threshold, both candidate physical intermediate features are determined to be redundant features, and at least one feature with high performance contribution, low computational cost, or stable physical meaning is retained.

[0010] Preferably, the simplified physical feature set includes melt core diameter, melt penetration rate, normalized indentation, energy efficiency, total input energy, and effective heat of fusion.

[0011] Preferably, for deployment scenarios with limited computing resources, energy efficiency, melting penetration, normalized indentation, and effective heat of fusion are selected from the simplified physical feature set to form a deployment feature subset.

[0012] Preferably, for a predicted new solder joint, each prediction model obtained through the multi-round random partitioning training outputs a prediction value; when verifying historical samples, the prediction values ​​generated in the rounds in which the historical sample did not participate in the corresponding model training are summarized.

[0013] Preferably, the compliance probability is the proportion of the number of predicted values ​​that are not lower than the preset tensile shear force standard in the tensile shear force prediction distribution to the total number of predicted values; when the compliance probability is lower than the first threshold or the prediction dispersion is higher than the dispersion threshold, the batch isolation, expanded sampling, non-destructive testing verification or sample tensile shear verification command is output.

[0014] This invention also provides a resistance spot welding quality prediction system, which includes a processor, a memory, a data interface, and an inspection command output interface. The memory stores a program, and when the processor executes the program, it implements the above-mentioned resistance spot welding tensile shear force prediction method. The data interface is used to receive process parameters output by the welding controller, and the inspection command output interface is used to send inspection commands to the manufacturing execution system, non-destructive testing equipment, or sampling management terminal.

[0015] Preferably, the data interface is further used to receive at least one of the following operating condition information: material grade, coating state, electrode end face size, and welding equipment identification, and to use the operating condition information to select the corresponding physical simulation parameter group or prediction model.

[0016] (III) Beneficial Effects Compared with the prior art, the resistance spot welding tensile shear force prediction method of the present invention has the following advantages: (1) The complete feature set and the feature set removed item by item use the same training samples and test samples, so that the performance difference can mainly reflect the contribution of the removed physical features.

[0017] (2) By using multi-round performance difference distribution instead of single difference for screening, we can obtain both contribution size and stability information at the same time.

[0018] (3) Physical anchoring models provide a prediction benchmark when training small sample neural networks is unstable. Their role is to reduce the risk of failure, rather than to guarantee that all machine learning models will achieve improved accuracy.

[0019] (4) The sample-level prediction distribution converts model fluctuations into the probability of achieving the target and the dispersion, and further converts them into executable test instructions.

[0020] (5) Simplifying the physical feature set reduces invalid simulation outputs participating in deployment calculations, which helps to reduce the amount of computation and improve interpretability.

[0021] In summary, the resistance spot welding tensile shear force prediction method provided by this invention constructs a simplified physical feature set by filtering intermediate quantities generated by physical simulation through paired repeated random partitioning, and then inputs it into the physical anchoring prediction model. This avoids the introduction of noise features and highly correlated features that would interfere with the prediction caused by directly inputting all intermediate quantities into the prediction model. Furthermore, its multi-round random partitioning can also avoid the impact of a single prediction on small sample resistance spot welding tensile shear force. The sample-level prediction distribution converts model fluctuations into compliance probabilities and dispersion, which further provides inspection instructions that match the predicted risks to the manufacturing site, facilitating automated inspection implementation in the actual inspection process. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the resistance spot welding tensile shear force prediction method of the present invention; Figure 2 This is a data flow diagram of the resistance spot welding tensile shear force prediction system of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] The following is in conjunction with the appendix Figure 1-2 The resistance spot welding tensile shear force prediction method and system of the present invention will be further described.

[0026] Please refer to this carefully. Figure 1-2 This invention discloses a method for predicting the tensile and shear force of resistance spot welding. This method is used to predict the tensile and shear force of resistance spot welding under small sample conditions and output the probability of achieving a preset tensile and shear force standard. It includes: S1. Data Acquisition: Acquire the welding current I, welding time tw, electrode force Fe, plate thickness h, and the tensile and shear force y corresponding to each resistance spot welding sample for multiple resistance spot welding samples; S2. Physical simulation: Based on the I, tw, Fe and h of each resistance spot welding sample obtained in step S1, input them into the transient heat conduction physical simulation model to obtain multiple candidate physical intermediate features. In step S2, the transient heat conduction physical simulation model includes at least two physical processes from Joule heat source, contact resistance, the effect of electrode force on contact state, latent heat of phase change, electrode indentation, and heat dissipation compensation. Its candidate physical intermediate features include melt core diameter, melt penetration, peak temperature, normalized indentation, energy efficiency, effective resistance, total input energy, and effective heat of fusion.

[0027] S3. Paired random partitioning: Predetermine B random seeds. In the b-th round, use seed sb to form training set Db,tr and test set Db,te. The complete feature set and each feature set removed item by item share Db,tr and Db,te, and obtain the pairing performance difference Δj,b. In step S3, the inverse value of the coefficient of determination or error index is used as the unified performance score. When the mean of the pairing performance difference after removing candidate physical intermediate features is greater than the positive threshold, the candidate physical intermediate features are judged as positive features. When the mean of the pairing performance difference after removing candidate physical intermediate features is not greater than zero, the candidate physical intermediate features are judged as noise features. The stability is judged based on the standard deviation or quantile interval of the pairing performance difference.

[0028] The complete feature set and the feature set removed item by item in this resistance spot welding tensile shear force prediction method both use the same training and test samples, so that the performance difference mainly reflects the contribution of the removed physical features and avoids the influence of training and testing division. Furthermore, the use of multi-round performance difference distribution instead of single difference for screening in step S3 allows it to simultaneously obtain information on contribution magnitude and stability.

[0029] S4. Stability Screening: Based on the correlation between the pairing performance difference distribution Δj,b and the candidate physical intermediate features, a simplified physical feature set is constructed. When the absolute value of the correlation coefficient of two candidate physical intermediate features is greater than the preset correlation threshold, both candidate physical intermediate features are determined to be redundant features, and at least one feature with high performance contribution, low computational cost, or stable physical meaning is retained.

[0030] Furthermore, its simplified physical feature set includes melt core diameter, melt penetration rate, normalized indentation, energy efficiency, total input energy, and effective heat of fusion. When considering deployment scenarios with limited computational resources, energy efficiency, melt penetration rate, normalized indentation, and effective heat of fusion are selected from the simplified physical feature set to form a deployment feature subset. In this step, the construction of the simplified physical feature set effectively reduces the participation of invalid simulation outputs in deployment calculations, which helps to reduce computational load and improve interpretability.

[0031] S5. Physical Anchoring Prediction Model: A physical anchoring prediction model is trained based on the process parameters and simplified physical feature set obtained in step S1. This model includes ridge regression, lasso regression, elastic network regression, or other regularized regression models. Furthermore, after training the physical anchoring prediction model, a gradient boosting tree, random forest, support vector machine, or neural network bias correction model is trained using the difference between the actual tensile shear force and the predicted physical anchoring value as the training objective. The bias estimate is then added to the predicted physical anchoring value to obtain the corrected prediction value.

[0032] In this step, the resistance spot welding tensile shear force prediction method utilizes a physical anchoring model to provide a prediction benchmark when the training of a small sample neural network is unstable, thereby effectively reducing the risk of failure.

[0033] S6. Uncertainty Assessment: Summarize the tensile and shear force prediction results of the B physical anchoring prediction models for the target weld point to form a tensile and shear force prediction distribution, and calculate the prediction mean, prediction dispersion and the probability of meeting the preset tensile and shear force standard for each resistance spot welding sample. In this implementation step, for a predicted new weld point, each prediction model obtained through multiple rounds of random partitioning training outputs a predicted value. When verifying historical samples, only the predicted values ​​generated by the historical samples in rounds that did not participate in the corresponding model training are summarized to avoid leakage of training information. Furthermore, the compliance probability is the proportion of the number of predicted values ​​in the tensile shear force prediction distribution that are not lower than the preset tensile shear force standard to the total number of predicted values. When the compliance probability is lower than the first threshold or the prediction dispersion is higher than the dispersion threshold, batch isolation, expanded sampling, non-destructive testing verification, or sample tensile shear verification instructions are output.

[0034] In this step, the resistance spot welding tensile shear force prediction method converts the fluctuations of the physical anchoring model into compliance probabilities and dispersion, thereby transforming its model fluctuations into inspection instructions that can be executed by the prediction system.

[0035] S7. Based on the pass rate and predicted dispersion of each resistance spot welding sample, output at least one of the following inspection instructions through the inspection instruction output interface: batch release, expanded sampling, batch isolation, non-destructive testing verification, or sample tensile shear verification.

[0036] This invention also provides a resistance spot welding quality prediction system, which includes a processor, a memory, a data interface, and an inspection command output interface. The memory stores a program, and when the processor executes the program, it implements the aforementioned resistance spot welding tensile-shear force prediction method. The data interface is used to receive process parameters output by the welding controller, and the inspection command output interface is used to send inspection commands to a manufacturing execution system, non-destructive testing equipment, or a sampling management terminal. The data interface is also used to receive at least one operating condition information from material grade, coating state, electrode end face size, and welding equipment identification, and uses this operating condition information to select the corresponding physical simulation parameter group or prediction model.

[0037] Based on the above-described method and system for predicting resistance spot welding tensile shear force, in specific embodiment one, the data and physical intermediate characteristic steps are as follows: This embodiment uses 493 valid weld point samples from the RSW Insights public dataset. Each sample includes at least the welding current I, welding time tw, electrode force Fe, average plate thickness h, and tensile-shear force y. Here, Fe represents force in kN. The one-dimensional transient heat conduction model is expressed as follows: Where ρ is density, cp is specific heat capacity, T is temperature, k is thermal conductivity, τ is physical time, z is the plate thickness coordinate, and qv is the Joule heat source. The model outputs the candidate physical intermediate features shown in Table 1.

[0038] Table 1. Intermediate physical features of candidates

[0039] In a specific embodiment, the paired-repeating random partitioning and stability screening steps are as follows: Let Sfull,b be the unified performance score of the complete feature set on the b-th round of the test set, and Sj,b be the score after removing the j-th candidate physical feature. Then, the pairwise contribution of the j-th feature is Δj,b = Sfull,bSj,b. For metrics such as R², where a larger value is better, this formula can be directly used; for metrics such as MAE, where a smaller value is better, the opposite of the value can be taken first to form the unified performance score.

[0040] In this embodiment, B=50, and the random seeds are sequentially from 42 to 91. In each round, 80% of the samples are used as the training set and 20% as the test set. The complete feature set and all item-by-item removal configurations share the same partition in the same round, and then the mean and standard deviation of Δj,b are calculated. This protocol is a repeated random partitioning, also known as Monte Carlo cross-validation.

[0041] When the mean of Δj,b is positive, removing the feature degrades performance as it contains positive predictive information; when the mean is not greater than zero, it is marked as a noisy candidate; when the contribution is small but positive, it is marked as a low-contribution candidate. Further, the correlation coefficients between candidate features are calculated, and highly correlated features are selected based on performance contribution, computational cost, and physical stability.

[0042] In this specific embodiment, the physical anchoring prediction and uncertainty assessment steps are as follows: The original process parameters and simplified physical characteristics are input into the ridge regression model to obtain the physical anchoring prediction values. When bias correction is needed, construct a new training set. And train the bias correction function f(x), and the final predicted value is .

[0043] For each new solder joint to be predicted, the model obtained from each training round outputs a predicted value. The predicted mean is The prediction standard deviation is Given the standard tensile-shear force Y0, the probability of meeting the standard is: .

[0044] When P is high and σ is below the dispersion threshold, a routine batch release recommendation can be output; when P is in the intermediate range or σ is above the dispersion threshold, an expanded sampling or non-destructive testing verification instruction can be output; when P is below the risk threshold, a batch isolation and sample tensile-shear verification instruction can be output. The dispersion threshold and risk threshold are determined by the material system, weld joint grade, and the company's quality risk requirements.

[0045] In this specific embodiment, the quantitative verification step is as follows: All tables in this embodiment use the aforementioned 50-round 80% / 20% paired repeated random partitioning protocol. Table 2 shows the feature removal results, where ΔR² is the full configuration R² minus the R² after feature removal, and a positive value indicates that the feature contains positive prediction information.

[0046] Table 2. Experimental results of feature-by-feature removal

[0047] In this embodiment, Table 2 shows that η, Ppen, Inorm, and Qeff make high contributions, Dnugget and Etotal still have positive contributions, and Reff and Tmax do not provide positive gains under this dataset and model configuration. This conclusion does not mean that Reff and Tmax are ineffective under all material and device conditions.

[0048] Table 3 Experimental Results of Group Characteristics

[0049] Raw+Phys 10D improves R² by 0.366 compared to the configuration using only four process parameters, indicating that the selected physical intermediate features provide measurable predictive information. The Pearson correlation coefficient between Dnugget and Qeff in the synthetic parameter space is 0.981, suggesting that either can be retained in resource-constrained deployments.

[0050] Table 4. Stability of repeated random partitioning for different models

[0051] Direct MLP is highly sensitive to data partitioning; Physically anchored MLP is basically the same as Ridge Physical Anchor, indicating that this framework mainly provides a fallback for predictions in this embodiment. The mean values ​​of Physically Anchored XGBoost and Direct XGBoost are close, so it cannot be concluded that Physical Anchor necessarily improves the accuracy of the tree model.

[0052] Table 5 Comparison of Paired Models

[0053] The above results support describing the technical effects of the present invention as feature selection stability, prediction fluctuation characterization, and small sample failure fallback, rather than describing bias correction as a precision amplification mechanism that applies to all learners.

[0054] In this specific embodiment, the system structure and verification execution steps are as follows: The system includes a processor, memory, a data interface, and an inspection instruction output interface. The data interface reads I, tw, Fe, and h from the welding controller or manufacturing execution system; the processor sequentially calls... Figure 2 Modules 101 to 108 are shown; the inspection instruction output interface sends the risk classification results to the manufacturing execution system, non-destructive testing equipment, or sampling management terminal.

[0055] Table 6 System Number and Function

[0056] In this specific embodiment, the applicable boundaries are as follows: the quantitative results of this embodiment are derived from 493 samples of weld points on low-carbon steel thin plates and a given one-dimensional physical model. When there are significant changes in material grade, coating, electrode end face, equipment dynamic characteristics, or plate thickness range, the physical simulation parameters should be recalibrated, and paired repeated random partitioning and model evaluation should be re-executed. The four-feature deployment subset is a preferred solution for computationally limited scenarios, and the six-feature set corresponds to the complete set with positive contributions in this embodiment.

Claims

1. A method for predicting the tensile shear force of resistance spot welding, wherein the method is used to predict the tensile shear force of resistance spot welding under small sample conditions and output the probability of achieving a preset tensile shear force standard, characterized in that, The method for predicting the tensile shear force in resistance spot welding includes: S1. Data acquisition: Acquire the welding current, welding time, electrode force, plate thickness, and tensile shear force corresponding to each resistance spot welding sample for multiple resistance spot welding samples; S2. Physical simulation: Input the process parameters of each resistance spot welding sample obtained in step S1 into the transient heat conduction physical simulation model to obtain multiple candidate physical intermediate features. S3. Paired random partitioning: A multi-round random partitioning sequence is preset, and the resistance spot welding samples are repeatedly divided into training set and test set. In each round of partitioning, the same training set and test set are used to evaluate the prediction performance of the complete feature set and the prediction performance after removing the candidate physical intermediate features one by one, so as to obtain the pairing performance difference distribution of each candidate physical intermediate feature. S4. Stability screening: Construct a simplified physical feature set based on the correlation between the pairing performance difference distribution and the candidate physical intermediate features; S5. Physical Anchoring Prediction Model: The physical anchoring prediction model is trained based on the process parameters obtained in step S1 and the simplified physical feature set. S6. Uncertainty assessment: Summarize the tensile and shear force prediction results of multiple physical anchoring prediction models for the target weld point to form a tensile and shear force prediction distribution, and calculate the prediction mean, prediction dispersion and the probability of meeting the preset tensile and shear force standard for each resistance spot welding sample. S7. Based on the compliance probability and predicted dispersion of each resistance spot welding sample, output at least one of the following inspection instructions through the inspection instruction output interface: batch release, expanded sampling, batch isolation, non-destructive testing verification, or sample tensile shear verification.

2. The method for predicting resistance spot welding tensile shear force according to claim 1, characterized in that, In step S3, the multi-round random partitioning uses the same random seed sequence to evaluate the complete feature set and the feature set removed item by item, so that the performance difference of each candidate physical intermediate feature in each partitioning round is obtained by pairing the same training samples and the same test samples.

3. The method for predicting resistance spot welding tensile shear force according to claim 2, characterized in that, In step S3, the inverse value of the coefficient of determination or error index is used as the unified performance score; when the mean of the pairing performance difference after removing the candidate physical intermediate features is greater than the positive threshold, the candidate physical intermediate features are determined to be positive features; when the mean of the pairing performance difference after removing the candidate physical intermediate features is not greater than zero, the candidate physical intermediate features are determined to be noise features, and the stability is determined based on the standard deviation or quantile interval of the pairing performance difference.

4. The method for predicting resistance spot welding tensile shear force according to claim 3, characterized in that, When the absolute value of the correlation coefficient between two candidate physical intermediate features is greater than a preset correlation threshold, both candidate physical intermediate features are determined to be redundant features, and at least one feature with high performance contribution, low computational cost, or stable physical meaning is retained.

5. The method for predicting resistance spot welding tensile shear force according to claim 1, characterized in that, The simplified physical feature set includes melt core diameter, melt penetration, normalized indentation, energy efficiency, total input energy, and effective heat of fusion.

6. The method for predicting resistance spot welding tensile shear force according to claim 5, characterized in that, For deployment scenarios with limited computing resources, energy efficiency, melting penetration, normalized indentation, and effective heat of fusion are selected from the simplified physical feature set to form a deployment feature subset.

7. The method for predicting resistance spot welding tensile shear force according to claim 1, characterized in that, For a predicted new solder joint, each prediction model obtained through the multi-round random partitioning training outputs a prediction value; when verifying historical samples, the prediction values ​​generated by the rounds in which the historical sample did not participate in the corresponding model training are summarized.

8. The method for predicting resistance spot welding tensile shear force according to claim 1, characterized in that, The compliance probability is the proportion of the number of predicted values ​​that are not lower than the preset tensile shear force standard in the tensile shear force prediction distribution to the total number of predicted values; when the compliance probability is lower than the first threshold or the prediction dispersion is higher than the dispersion threshold, the batch isolation, expanded sampling, non-destructive testing verification or sample tensile shear verification command is output.

9. A resistance spot welding quality prediction system, characterized in that, The resistance spot welding quality prediction system includes a processor, a memory, a data interface, and an inspection command output interface; the memory stores a program, and when the processor executes the program, it implements the resistance spot welding tensile shear force prediction method according to any one of claims 1 to 8; the data interface is used to receive process parameters output by the welding controller, and the inspection command output interface is used to send inspection commands to the manufacturing execution system, non-destructive testing equipment, or sampling management terminal.

10. The resistance spot welding quality prediction system according to claim 9, characterized in that, The data interface is also used to receive at least one of the following operating condition information: material grade, coating state, electrode end face size, and welding equipment identification, and to use the operating condition information to select the corresponding physical simulation parameter group or prediction model.

Images