Riveting strength intelligent recommendation method and system based on machine learning

Abstract

The invention provides a riveting strength intelligent recommendation method and system based on machine learning. The riveting strength intelligent recommendation method comprises the steps that the number of overlapping layers of riveted plates and plate material parameters are input; inputting data including the number of plate lap joint layers and plate material parameters into the rivet model recommendation model, and outputting a rivet model and a bottom model number which meet a qualified metallographic phase; the nail model recommendation model is a multi-output machine learning model; the rivet model, the bottom model number and the mechanical property type output by the rivet model recommendation model are jointly input into a connection strength prediction model, and a load peak value is output; the connection strength prediction model adopts a regression model or a deep learning model; and if multiple groups of data are obtained, result comparison is carried out through imaging. According to the method, the specifications of the rivet and the bottom die for self-piercing riveting can be quickly recommended, the connection strength value is predicted, the operation is simple and convenient, and the prediction accuracy and reliability are improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent automobile design and manufacturing, and in particular relates to a riveting strength intelligent recommendation method and system based on machine learning. Background Art

[0002] With the widespread application of lightweight materials in the field of automobile manufacturing, riveting (including self-piercing riveting SPR, rotary tapping riveting FDS, etc.) has developed rapidly as an efficient connection technology. Riveting refers to the use of rivets in combination with bottom molds to achieve effective connection of multiple layers of dissimilar materials without penetrating the sheet. Selecting the right rivets and bottom molds is crucial to ensure the quality of riveting. The selection of rivets is usually based on the thickness, hardness and required connection strength of the materials to be connected; the design of the bottom mold needs to consider its shape, size and material properties to ensure that a good rivet head shape and sufficient clamping force can be formed during the riveting process.

[0003] In terms of evaluating the mechanical properties of riveting, the main methods currently used can be roughly divided into two categories: (1) Empirical model method The empirical model method is based on a large amount of experimental data and engineers' accumulated experience. It uses data regression or classification analysis to establish a simple formula or rule system to quickly recommend rivet and base die specifications and estimate the mechanical properties of riveting.

[0004] (2) Simulation method The simulation method uses computer-aided engineering (CAE) technology and finite element analysis (FEA) to simulate the riveting process and predict the connection performance and mechanical response of the rivet and bottom die combination. This method is based on the principles of physics and mechanics, combines material properties and geometric information, and replaces actual tests with virtual experiments to achieve efficient and low-cost performance evaluation and optimization.

[0005] Although the empirical model method and simulation method have their own advantages in evaluating the mechanical properties of riveting, they also have some obvious limitations, which limit their extensiveness and effectiveness in practical applications.

[0006] (1) Disadvantages of the empirical model method Limited scope of application: Empirical models are highly dependent on experimental data, which limits their scope of application. Especially when it comes to new materials (such as high-strength steel, composite materials) or complex multilayer plate structures, existing empirical models often fail to provide reliable guidance.

[0007] Lack of nonlinear considerations: Phenomena such as material yield and contact separation involved in the riveting process have significant nonlinear characteristics, and empirical formulas are difficult to fully and accurately capture these complex nonlinear behaviors.

[0008] Poor generalizability: Once the empirical range of a specific material combination or rivet parameters is exceeded, the effectiveness of the model will be greatly reduced, and a large number of experiments need to be carried out again to build a new model, which is not only time-consuming and labor-intensive, but also costly.

[0009] (2) Disadvantages of simulation method High computational cost: In order to obtain high-precision simulation results, a large amount of computing resources and time are often required. This high cost becomes a major challenge, especially in projects with frequent design iterations.

[0010] High professional requirements: The use of simulation tools requires not only deep professional knowledge, but also requires the operator to have a high level of skills, which is a technical threshold that is difficult to cross for many small and medium-sized enterprises.

[0011] Insufficient data coverage: Although simulation can provide detailed predictions for specific scenarios, it is still difficult to quickly adapt to changes in different material combinations and joining processes. The breadth and depth of the simulation database are not enough to meet diverse needs.

[0012] Actual verification requirements: Even under the most ideal conditions, simulation results must be verified by physical tests to ensure their accuracy. This means that simulation cannot completely replace traditional test methods, and the two need to be used in combination to ensure the reliability of connection performance.

[0013] In summary, the existing riveting evaluation methods still have much room for improvement in terms of flexibility, accuracy and economy, and there is an urgent need to develop more intelligent and efficient solutions to overcome these challenges. Summary of the invention

[0014] The present invention proposes a riveting strength intelligent recommendation method and system based on machine learning, which can quickly recommend rivet and bottom mold specifications for riveting and predict the connection strength value. The operation is simple and the accuracy and reliability of the prediction are improved.

[0015] To achieve the above object, the technical solution of the present invention is achieved as follows: A riveting strength intelligent recommendation method based on machine learning, comprising: S1. Input the number of overlapping layers of riveted plates and plate material parameters; S2. Inputting data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and outputting the rivet model and bottom model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model; S3, inputting the rivet model and base model number and mechanical property type output by the nail mold recommendation model into a connection strength prediction model, and outputting the load peak value; the connection strength prediction model adopts a regression model or a deep learning model; S4. If multiple sets of data are obtained, the results can be compared graphically.

[0016] Furthermore, the nail model recommendation model of step S2 is obtained through training, and the training process includes: S201, obtaining historical data of riveting, wherein the historical data includes riveted plate overlap combination parameters, plate material parameters, rivet model and base mold number used for riveting, metallographic detection data of rivets and base molds, mechanical strength data, and process parameters; S202, dividing the acquired historical data into data sets, where the data sets include a training set, a validation set, and a test set; S203, selecting a multi-output machine learning model, using a training set for training, taking the rivet model and the base model number as targets, and taking other data in the training set as input; the multi-output machine learning model is a traditional machine learning model supporting multi-output, or a multi-task learning model; S204. During the model training process, use the validation set to evaluate the model performance under different hyperparameter settings and select the best hyperparameter combination. After completing the model training and hyperparameter tuning, use the test set to conduct a comprehensive performance test on the model to evaluate the model's generalization ability on new data.

[0017] Furthermore, in the training of step S203, a cross entropy loss function is used to process multi-label classification. If the rivet and the bottom mold have multiple categories respectively, a cross entropy loss function is defined for the rivet and the bottom mold respectively, and then the weighted sum is used as the final loss function.

[0018] Furthermore, the connection strength prediction model in step S3 is obtained through training, and the training process includes: S301, acquiring data from a process database and a load performance experiment, including connection strength of a combination of a rivet model and a base model number under different mechanical performance types; the rivet model and the base model number are mapped into continuous features through an Embedding technology; the mechanical performance type is processed through a category code; S302, dividing the acquired data into data sets, where the data sets include a training set, a validation set, and a test set; S303, selecting a regression model or a deep learning model, using a training set for training, taking the rivet model and base model number, and the mechanical property type in the training set as input, and taking the connection strength as the target; S304: Adjust model parameters through the validation set, and test model performance through the test set.

[0019] Furthermore, in the training of step S303, the mean square error is used as the loss function.

[0020] On the other hand, the present invention also proposes a riveting strength intelligent recommendation system based on machine learning, comprising: Input module: input the number of overlapping layers of riveted plates and plate material parameters; Nail mold recommendation module: inputs data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and outputs the rivet model and bottom model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model; Strength prediction module: input the rivet model and base model number and mechanical property type output by the nail mold recommendation model into the connection strength prediction model, and output the load peak value; the connection strength prediction model adopts a regression model or a deep learning model; Comparison module: If multiple sets of data are obtained, the results can be compared graphically.

[0021] Furthermore, the nail mold recommendation module includes: Historical data acquisition unit: acquires historical data of riveting, wherein the historical data includes riveted plate overlap combination parameters, plate material parameters, rivet model and base mold number used for riveting, metallographic detection data of rivets and base molds, mechanical strength data, and process parameters; Division unit: divide the acquired historical data into data sets, which include training set, validation set and test set; Recommended training unit: select a multi-output machine learning model, use the training set for training, use the rivet model and the base model number as targets, and use other data in the training set as input; the multi-output machine learning model is a traditional machine learning model that supports multi-output, or a multi-task learning model; Validation test unit: During the model training process, the validation set is used to evaluate the model performance under different hyperparameter settings and select the best hyperparameter combination. After completing the model training and hyperparameter tuning, the test set is used to conduct a comprehensive performance test on the model to evaluate the model's generalization ability on new data.

[0022] Furthermore, in the nail mold recommendation model training unit, the cross entropy loss function is used to process multi-label classification. If rivets and base molds have multiple categories respectively, a cross entropy loss function is defined for the rivets and the base molds respectively, and then the weighted sum is used as the final loss function.

[0023] Furthermore, the strength prediction module includes: Data acquisition unit: acquires data from the process database and load performance experiments, including the connection strength of the combination of rivet model and base model number under different mechanical performance types; the rivet model and base model number are mapped into continuous features through Embedding technology; the mechanical performance type is processed through category coding; Division unit: divide the acquired data into data sets, which include training set, validation set and test set; Prediction training unit: select regression model or deep learning model, use training set for training, take rivet model and base model number, mechanical property type in the training set as input, and take connection strength as target; Verification and testing unit: adjust model parameters through the verification set and test model performance through the test set.

[0024] Furthermore, in the prediction training unit, the mean square error is used as the loss function.

[0025] Compared with the prior art, the present invention has the following beneficial effects: 1. Improve accuracy: The present invention can more accurately predict the connection strength and reduce human errors through the nail mold recommendation model and the connection strength prediction model.

[0026] 2. Improve efficiency: The present invention realizes automated data processing and model training, greatly reducing the time and resource consumption of small batch experiments.

[0027] 3. Enhanced adaptability: The present invention uses machine learning models to make recommendations and predictions, and can quickly adapt to new materials and new processes without having to re-run a large number of experiments and data analyses.

[0028] 4. Improve consistency: The present invention ensures the consistency and repeatability of recommendation results through standardized data processing and model training.

[0029] 5. Improve the degree of intelligence: The present invention uses intelligent technology in the automobile design process, which can promote technological integration and further promote the deep integration of intelligent technology and the automobile industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the SPR connection strength intelligent recommendation method according to Example 1 of the present invention; Figure 2 This is a diagram of the plate material and parameter information input interface of Example 1 of the present invention; Figure 3 This is a result interface diagram of the recommended rivet model and base model number of Example 1 of the present invention; Figure 4 This is a diagram of the connection strength prediction result interface of Example 1 of the present invention; Figure 5 is a flowchart of comparing graphical results of multiple sets of data in Example 1 of the present invention; Figure 6 This is a diagram of a chart creation interface according to Embodiment 1 of the present invention; Figure 7 It is a schematic diagram for comparing the graphical results of Example 1 of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0032] The design idea of ​​the present invention is to use an artificial intelligence method. After inputting information such as the plate material, it can quickly recommend suitable rivets and base model numbers, and further predict the connection strength values ​​under different mechanical performance types. The different mechanical performance types include shear strength test (cross stretching 0°), peel strength test (cross stretching 90°), oblique shear strength test (cross stretching 45°), peel strength test and tensile shear strength test.

[0033] Based on the design concept, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] Embodiment 1: In Example 1, the connection of self-piercing riveting SPR is taken as an example, and the method proposed by the present invention is applied. The specific steps are as follows: Figure 1 As shown, including: S1. Input the number of overlapped plates and plate material parameters for SPR connection.

[0035] like Figure 2 The following is the interface for inputting parameters, including: Number of overlapping layers of boards: Open the software and select the specific number of overlapping layers in the drop-down option of the number of overlapping layers. The number of overlapping layers can be divided into 2, 3, and 4, representing two-layer boards, three-layer boards, and four-layer boards respectively.

[0036] Sheet material parameters: After selecting the number of layers, select the sheet position, sheet material, enter the grade, thickness, tensile strength, elongation after fracture, total elongation at fracture, maximum plasticity, yield strength, Young's modulus, elastic modulus, Poisson's ratio, strain hardening exponent, plastic strain ratio, etc.

[0037] The basic parameters of the above materials are as follows: Figure 2 After the input is completed, if there are any parameters that do not meet the requirements, click the [Confirm] button, and the software will automatically pop up a window to prompt that the parameters do not meet the process requirements of the SPR connection, and mark the wrong parameter information in red. Re-enter the correct parameters and click [Confirm].

[0038] S2. Input data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and output the rivet model and base model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model.

[0039] 1. Nail model recommendation model architecture and training: (1) A multi-output machine learning model is used to simultaneously predict the rivet model and base model number to capture the relationship between the two.

[0040] Traditional machine learning models: You can use random forests or gradient boosted trees (XGBoost) that support multiple outputs.

[0041] Deep learning model: Transformer that can use multi-task learning.

[0042] (2) Input: number of overlapped layers of plates, plate material parameters, metallographic inspection data of rivets and base molds, mechanical strength data, and process parameters.

[0043] (3) Output: Rivet model and base model number that meet qualified metallographic requirements.

[0044] (4) Source of training data: large-scale historical experimental data of SPR connection, including plate lap combination parameters of SPR connection, plate material parameters, rivet model and base model number used in SPR connection, metallographic inspection data of rivets and base molds, mechanical strength data, and process parameters; after obtaining the data, the data set can be divided into training set, validation set, and test set in a ratio of 6:2:2.

[0045] (5) Principle and derivation: Loss function. For classification problems, the cross entropy loss function can be used to handle multi-label classification. If rivets and base molds have multiple categories, a cross entropy loss function can be defined for each task, and then the weighted sum of these losses is used as the final loss function L.

[0046] ; L: is the final total loss value, which is used to measure the difference between the model prediction and the true label.

[0047] i: The index of the sample, from 1 to N, representing each specific sample.

[0048] N: represents the number of samples, that is, the total number of training data points.

[0049] j: represents the index of the rivet category.

[0050] C1: The number of rivet categories, indicating how many different types of rivets there are.

[0051] k: represents the index of the die category.

[0052] C2: The number of die categories, indicating how many different die categories there are.

[0053] : is an indicator variable. For the i-th sample, if its true label belongs to the j-th category of rivets, then =1 if the value is set to 0, otherwise it is 0.

[0054] : The probability that the model predicts that the i-th sample belongs to the j-th class of rivets.

[0055] : It is also an indicator variable. For the i-th sample, if its true label belongs to the k-th category of the bottom model, then =1 if the value is set to 0, otherwise it is 0.

[0056] : The probability that the i-th sample predicted by the model belongs to the k-th class of the underlying model.

[0057] (6) Data preprocessing: Standardize or normalize the continuous features of plate material parameters (such as thickness and tensile strength). Use One-Hot Encoding or Embedding to represent the categorical features of plate material parameters (such as brand).

[0058] (7) Training process: Use the training set input, the number of plate overlaps and plate material parameters in the training set as input, and the rivet model and base model number as targets; after the model's fully connected layer mapping, generate the predicted values ​​of the rivet model and base model number. For the multi-task model, the two output prediction values ​​are calculated separately and then weighted summed to optimize the overall performance.

[0059] 2. Output results: Figure 3 As shown, using the nail model recommendation model obtained after training and inputting the parameters of step S1, we can get: Rivet model: recommended rivet model Bottom model number: The bottom model number corresponding to each rivet model can be selected in the drop-down box.

[0060] S3. The rivet model number and base model number, as well as the mechanical property type output by the nail mold recommendation model are input into a connection strength prediction model, and the load peak value is output; the connection strength prediction model adopts a regression model or a deep learning model.

[0061] 1. Input parameters: Rivet model: output from the nail mold recommendation model; Bottom model number: It also comes from the output result of the nail mold recommendation model. After selecting a rivet model, select the corresponding bottom model number; Mechanical properties: including shear strength test (cross stretching 0°), peel strength test (cross stretching 90°), oblique shear strength test (cross stretching 45°), peel strength test and tensile shear strength test, etc.

[0062] 2. Connection strength prediction model architecture and training: (1) Model architecture: You can use: Regression models: traditional machine learning models such as random forest, SGBoost, and support vector machine regression (SVR); Deep learning models: Multi-layer perceptron (MLP), convolutional neural network (CNN), Transformer and other models.

[0063] (2) Input: rivet model, base model number, mechanical property type; (3) Output: load peak value (unit: N); (4) Source of training data: from the connection process database and load performance experimental data, including the strength values ​​of the rivet and bottom die combination under different mechanical performance types. After obtaining the data, the data set can be divided into training set, verification set, and test set in a ratio of 6:2:2.

[0064] (5) Principle and derivation: a. Loss function: Use Mean Squared Error (MSE) as the loss function.

[0065] b. Feature Engineering: The rivet model and base model number are mapped as continuous features through Embedding; Mechanical properties test types are handled by category coding (such as One-Hot or Embeddig); If there are more experimental parameters (such as temperature, speed, etc.), they can be used as additional features.

[0066] c. Data Modeling: The rivet model, base model number, and mechanical property type in the training set are used as input, and the connection strength is used as the target; Regression models: such as Random Forest and SGBoost can capture nonlinear relationships between features.

[0067] Deep learning model: Input features learn nonlinear mapping relationships through multi-layer neural networks. If the data has complex contextual relationships (such as time series test data), LSTM or Transformer can be introduced.

[0068] d. Model evaluation: The mean square error (MSE) or root mean square error (RMSE) was used to evaluate the model performance.

[0069] (3) Output results: like Figure 4 Shown are the prediction results output by the connection strength prediction model, including the peak load; Peak load: predict the strength value of the current nail mold combination (rivet model and base model number combination) under the specified mechanical property type.

[0070] S4. If multiple sets of data are obtained, the results can be compared graphically.

[0071] The flowchart of graphical result comparison is as follows Figure 5 As shown, first create a chart. When creating a chart, select the comparison group, mechanical type, and chart type, and then generate the corresponding chart. The specific operations are: In such Figure 6 Click [Create Chart] on the interface shown, select the comparison group, mechanical property type, and comparison chart type, and click [Generate] to generate the corresponding graph. You can add multiple mechanical property test types and comparison chart types to perform multi-dimensional comparison. The generated comparison chart is as follows: Figure 7 shown.

[0072] The method described in this embodiment automatically recommends rivet and bottom mold specifications through a nail mold recommendation model, reduces manual trial and error links, greatly shortens the design cycle, reduces computing costs, and is particularly suitable for frequent design iterations; using a connection strength prediction model based on deep learning, it can comprehensively consider multiple parameters of materials, rivets, and bottom molds, and perform high-precision strength predictions. It can handle various materials (such as high-strength steel, composite materials) and complex multi-layer plate structures, is not limited to specific test data, and has a wider range of applicability. This method is easy to operate, does not require professional knowledge, lowers the threshold for use, and is suitable for all types of companies and engineers. By continuously accumulating and learning new test data, the model can be continuously optimized to improve the accuracy and reliability of the prediction, thereby providing a more reliable connection performance evaluation in practical applications. This embodiment uses intelligent technology in the automotive design process, which can promote technological integration and further promote the deep integration of intelligent technology and the automotive industry.

[0073] Embodiment 2: This embodiment proposes an SPR connection strength intelligent recommendation system, including: Input module: input the number of overlapping layers of the plates connected by SPR and the material parameters of the plates; Nail mold recommendation module: inputs data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and outputs the rivet model and bottom model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model; Strength prediction module: input the rivet model and base model number and mechanical property type output by the nail mold recommendation model into the connection strength prediction model, and output the load peak value; the connection strength prediction model adopts a regression model or a deep learning model; Comparison module: If multiple sets of data are obtained, the results can be compared graphically.

[0074] Among them, the nail mold recommendation module includes: Historical data acquisition unit: acquires historical data of SPR connection, wherein the historical data includes plate overlap combination parameters of SPR connection, plate material parameters, rivet model and bottom mold number used in SPR connection, metallographic detection data of rivets and bottom molds, mechanical strength data, and process parameters; Division unit: divide the acquired historical data into data sets, which include training set, validation set and test set; Recommended training unit: select a multi-output machine learning model, use the training set for training, use the rivet model and the base model number as targets, and use other data in the training set as input; the multi-output machine learning model is a traditional machine learning model that supports multi-output, or a multi-task learning model; Validation test unit: During the model training process, the validation set is used to evaluate the model performance under different hyperparameter settings and select the best hyperparameter combination. After completing the model training and hyperparameter tuning, the test set is used to conduct a comprehensive performance test on the model to evaluate the model's generalization ability on new data.

[0075] In the nail mold recommendation model training unit, the cross entropy loss function is used to process multi-label classification. If rivets and base molds have multiple categories respectively, a cross entropy loss function is defined for the rivets and base molds respectively, and then the weighted sum is used as the final loss function.

[0076] The intensity prediction module includes: Data acquisition unit: acquires data from the process database and load performance experiments, including the connection strength of the combination of rivet model and base model number under different mechanical performance types; the rivet model and base model number are mapped into continuous features through Embedding technology; the mechanical performance type is processed through category coding; Division unit: divide the acquired data into data sets, which include training set, validation set and test set; Prediction training unit: select regression model or deep learning model, use training set for training, take rivet model and base model number, mechanical property type in the training set as input, and take connection strength as target; Verification and testing unit: adjust model parameters through the verification set and test model performance through the test set.

[0077] In the prediction training unit, the mean square error is used as the loss function.

[0078] The SPR connection strength intelligent recommendation system proposed in this embodiment can implement the SPR connection strength intelligent recommendation method described in Example 1, and has the same technical effect as Example 1.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A riveting strength intelligent recommendation method based on machine learning, characterized in that: include: S1. Input the number of overlapping layers of riveted plates and plate material parameters; S2. Inputting data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and outputting the rivet model and bottom model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model; S3, inputting the rivet model and base model number and mechanical property type output by the nail mold recommendation model into a connection strength prediction model, and outputting the load peak value; the connection strength prediction model adopts a regression model or a deep learning model; S4. If multiple sets of data are obtained, the results can be compared graphically.

2. The method for intelligently recommending riveting strength based on machine learning according to claim 1 is characterized in that: The nail model recommendation model of step S2 is obtained through training, and the training process includes: S201, obtaining historical data of riveting, wherein the historical data includes riveted plate overlap combination parameters, plate material parameters, rivet model and base mold number used for riveting, metallographic detection data of rivets and base molds, mechanical strength data, and process parameters; S202, dividing the acquired historical data into data sets, where the data sets include a training set, a validation set, and a test set; S203, selecting a multi-output machine learning model, using a training set for training, taking the rivet model and the base model number as targets, and taking other data in the training set as input; the multi-output machine learning model is a traditional machine learning model supporting multi-output, or a multi-task learning model; S204. During the model training process, use the validation set to evaluate the model performance under different hyperparameter settings and select the best hyperparameter combination. After completing the model training and hyperparameter tuning, use the test set to conduct a comprehensive performance test on the model to evaluate the model's generalization ability on new data.

3. The method for intelligently recommending riveting strength based on machine learning according to claim 2 is characterized in that: In the training of step S203, a cross entropy loss function is used to process multi-label classification. If the rivet and the bottom mold have multiple categories respectively, a cross entropy loss function is defined for the rivet and the bottom mold respectively, and then the weighted sum is used as the final loss function.

4. The method for intelligently recommending riveting strength based on machine learning according to claim 1, characterized in that: The connection strength prediction model in step S3 is obtained through training, and the training process includes: S301, acquiring data from a process database and a load performance experiment, including connection strength of a combination of a rivet model and a base model number under different mechanical performance types; the rivet model and the base model number are mapped into continuous features through an Embedding technology; the mechanical performance type is processed through a category code; S302, dividing the acquired data into data sets, where the data sets include a training set, a validation set, and a test set; S303, selecting a regression model or a deep learning model, using a training set for training, taking the rivet model and base model number, and the mechanical property type in the training set as input, and taking the connection strength as the target; S304: Adjust model parameters through the validation set, and test model performance through the test set.

5. The method for intelligently recommending riveting strength based on machine learning according to claim 2 is characterized in that: In the training of step S303, the mean square error is used as the loss function.

6. A riveting strength intelligent recommendation system based on machine learning, characterized in that: include: Input module: input the number of overlapping layers of riveted plates and plate material parameters; Nail mold recommendation module: inputs data including the number of overlapping layers of the plate and the material parameters of the plate into the nail mold recommendation model, and outputs the rivet model and bottom model number that meet the qualified metallographic structure; the nail mold recommendation model is a multi-output machine learning model; Strength prediction module: input the rivet model and base model number and mechanical property type output by the nail mold recommendation model into the connection strength prediction model, and output the load peak value; the connection strength prediction model adopts a regression model or a deep learning model; Comparison module: If multiple sets of data are obtained, the results can be compared graphically.

7. The riveting strength intelligent recommendation system based on machine learning according to claim 6 is characterized in that: Recommended modules for nail molds include: Historical data acquisition unit: acquires historical data of riveting, wherein the historical data includes riveted plate overlap combination parameters, plate material parameters, rivet model and base mold number used for riveting, metallographic detection data of rivets and base molds, mechanical strength data, and process parameters; Division unit: divide the acquired historical data into data sets, which include training set, validation set and test set; Recommended training unit: select a multi-output machine learning model, use the training set for training, use the rivet model and the base model number as targets, and use other data in the training set as input; the multi-output machine learning model is a traditional machine learning model that supports multi-output, or a multi-task learning model; Validation test unit: During the model training process, the validation set is used to evaluate the model performance under different hyperparameter settings and select the best hyperparameter combination. After completing the model training and hyperparameter tuning, the test set is used to conduct a comprehensive performance test on the model to evaluate the model's generalization ability on new data.

8. The riveting strength intelligent recommendation system based on machine learning according to claim 7 is characterized in that: In the nail mold recommendation model training unit, the cross entropy loss function is used to process multi-label classification. If rivets and base molds have multiple categories respectively, a cross entropy loss function is defined for the rivets and base molds respectively, and then the weighted sum is used as the final loss function.

9. The riveting strength intelligent recommendation system based on machine learning according to claim 6 is characterized in that: The intensity prediction module includes: Data acquisition unit: acquires data from the process database and load performance experiments, including the connection strength of the combination of rivet model and base model number under different mechanical performance types; the rivet model and base model number are mapped into continuous features through Embedding technology; the mechanical performance type is processed through category coding; Division unit: divide the acquired data into data sets, which include training set, validation set and test set; Prediction training unit: select regression model or deep learning model, use training set for training, take rivet model and base model number, mechanical property type in the training set as input, and take connection strength as target; Verification and testing unit: adjust model parameters through the verification set and test model performance through the test set.

10. The riveting strength intelligent recommendation system based on machine learning according to claim 9 is characterized in that: In the prediction training unit, the mean square error is used as the loss function.

Images