Metal additive manufacturing forming quality prediction method and device based on ensemble learning

Abstract

The invention discloses a metal additive manufacturing forming quality prediction method based on integrated learning, and the method comprises the steps: collecting key technological parameters which affect the forming quality of a WAAM process, and constructing an original data set; the method comprises the following steps: expanding an original data set based on a multi-modal data enhancement architecture of Gaussian distribution fusion to obtain an expanded training data set; selecting a plurality of basic machine learning models, training each basic machine learning model based on the expanded training data set, and optimizing hyper-parameters of each basic machine learning model; constructing an initial integration model based on the optimized basic machine learning model; optimizing the weight of each basic machine learning model in the integrated model by adopting a samson swarm algorithm to obtain an optimized integrated model; and the metal additive manufacturing forming quality is predicted based on the optimized integrated model, and a prediction result is obtained. Therefore, the AM forming quality and the material utilization efficiency are remarkably improved, and the production cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of metal additive manufacturing, and particularly relates to a method and device for predicting the forming quality of metal additive manufacturing based on ensemble learning. Background Art

[0002] With the continuous growth of the demand for high-performance, lightweight, and complex structural components in modern industry, metal additive manufacturing (AM) technology, especially wire arc additive manufacturing (WAAM) technology, has become a key technology. As a typical metal material, aluminum alloy has been widely used in many fields such as aerospace, automotive manufacturing, and marine engineering due to its excellent specific strength, good thermal conductivity, and corrosion resistance. However, in the AM process, especially in aluminum alloy WAAM, due to the complex non-linear coupling relationship between process parameters and the forming quality of structural components, it poses many severe challenges to achieve precise control during the forming process.

[0003] Most of the existing forming quality prediction methods rely on empirical formulas and physical models. Although empirical formulas have the advantage of being simple and easy to use, they are often limited by specific experimental conditions and can only consider a few process parameters, so they lack sufficient generality and are difficult to comprehensively and accurately describe the complex additive manufacturing process. In contrast, although the physical model-based method can quantitatively analyze the forming manufacturing process from a theoretical level, due to involving a large number of assumptions and simplifications, and having very strict requirements for material parameters and boundary conditions, its practical application is restricted to a certain extent.

[0004] Different from traditional forming quality prediction methods, the prediction method based on machine learning can automatically mine potential laws from massive data and model the complex additive manufacturing process. Nevertheless, machine learning methods also face some challenges. First, machine learning models usually require a large amount of training data. Under the condition of small samples, a single machine learning model often has insufficient learning ability for complex manufacturing processes, is extremely prone to overfitting, resulting in insufficient generalization ability and being unable to cope with changes under different working conditions. Second, the black-box characteristic of machine learning models makes them lack sufficient interpretability when providing process parameter optimization suggestions, thus restricting their application in engineering practice. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting the forming quality of metal additive manufacturing based on ensemble learning to achieve high-precision prediction of the forming quality of metal additive manufacturing.

[0006] To solve the above problems, the technical solution of the present invention is as follows: A method for predicting the forming quality of metal additive manufacturing based on ensemble learning, comprising: Collect key process parameters that affect the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and construct an original data set; Based on a multi-modal data augmentation architecture fused with Gaussian distribution, augment the original data set to obtain an augmented training data set; Select a variety of basic machine learning models, and based on the augmented training data set, train each basic machine learning model and optimize the hyperparameters of each basic machine learning model; Based on the optimized basic machine learning models, construct an initial ensemble model; use the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model to obtain an optimized ensemble model; Based on the optimized ensemble model, predict the forming quality of metal additive manufacturing to obtain a prediction result.

[0007] According to an embodiment of the present invention, augmenting the data set based on the multi-modal data augmentation architecture fused with Gaussian distribution further includes: Use the expectation maximization algorithm to perform a preliminary estimation of the parameters of the multi-modal data augmentation architecture fused with Gaussian distribution; the parameters include the mean, covariance matrix, and weight; In each iteration, dynamically adjust the parameters of the multi-modal data augmentation architecture fused with Gaussian distribution according to the distribution characteristics of the samples in the original data set until convergence to the optimal solution; According to the fitted multi-modal data augmentation architecture fused with Gaussian distribution, generate new data samples and merge them with the original data set to form an augmented training data set.

[0008] According to an embodiment of the present invention, perform data standardization processing on the augmented training data set to improve the stability and accuracy of model training.

[0009] According to an embodiment of the present invention, selecting a variety of basic machine learning models and training each basic machine learning model based on the augmented training data set further includes: Select basic machine learning models including the extreme gradient boosting model, extreme random tree model, and support vector machine, and based on the standardized training data set, perform training in the manner of K-fold cross-validation to optimize the hyperparameters of each basic machine learning model.

[0010] According to an embodiment of the present invention, during the training process, use the Optuna algorithm to automatically search for and optimize the hyperparameters of each basic machine learning model; In each cross-validation, use the optimized hyperparameters to train the basic machine learning model and evaluate the performance of the trained model on the validation set until the basic machine learning model with the minimum RMSE is obtained.

[0011] According to an embodiment of the present invention, further optimizing the weights of each basic machine learning model in the ensemble model by using the sand cat swarm algorithm includes: Taking the weight combination of the basic machine learning models as the optimization object of the sand cat swarm algorithm, taking the RMSE of the ensemble model on the validation set as the fitness function, and obtaining the weight combination when the fitness function is minimized through the sand cat swarm algorithm.

[0012] According to an embodiment of the present invention, a preset number of sand cat individuals are randomly generated, and each sand cat individual represents a group of weight combinations; Calculating the fitness value of each individual by using the fitness function; According to the search characteristics of the sand cat swarm algorithm, update the weight combination through the following formula: Wherein, represents the weight vector of the current individual, represents the weight vector of the individual with the minimum fitness, represents the weight vector of a randomly selected individual, and represent random coefficients with values in [0, 1]; When the sand cat individual reaches the preset condition, further adjust the weight combination through the following formula to quickly converge to the optimal solution: Wherein, represents the weight vector of a randomly selected individual, represents the weight vector of another randomly selected individual, and represent random coefficients with values in [0, 1], represents a random angle.

[0013] According to an embodiment of the present invention, based on the weight combination optimized by the sand cat swarm algorithm, construct an ensemble model and obtain the prediction formula of the ensemble model: Wherein, represents the prediction result of the ensemble model, represents the dynamic weight allocation of different basic machine learning models, represents the prediction result of the basic machine learning model on the validation set.

[0014] According to an embodiment of the present invention, after obtaining the integrated model, a model interpretation strategy based on game feature contribution quantification is used to interpret the integrated model, and the contribution values of different key process parameters affecting the forming quality of the WAAM process to the prediction result of the integrated model are calculated to improve the interpretability of the integrated model.

[0015] A device for predicting the forming quality of metal additive manufacturing based on ensemble learning, comprising: A data acquisition module, configured to acquire key process parameters affecting the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and construct an original data set; A data augmentation module, configured to augment the original data set based on a multi-modal data augmentation architecture fused with Gaussian distribution to obtain an augmented training data set; A model optimization module, configured to select a variety of basic machine learning models, and based on the augmented training data set, train each basic machine learning model and optimize the hyperparameters of each basic machine learning model; An ensemble learning module, configured to construct an initial integrated model based on the optimized basic machine learning models; and optimize the weights of each basic machine learning model in the integrated model by using a sand cat swarm algorithm to obtain an optimized integrated model; A prediction module, configured to predict the forming quality of metal additive manufacturing based on the optimized integrated model to obtain a prediction result.

[0016] Due to the adoption of the above technical solutions, the present invention has the following advantages and positive effects compared with the prior art: In an embodiment of the present invention, a method for predicting the forming quality of metal additive manufacturing based on ensemble learning acquires key process parameters affecting the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and constructs an original data set; based on a multi-modal data augmentation architecture fused with Gaussian distribution, the original data set is augmented to obtain an augmented training data set; Select a variety of basic machine learning models, and based on the augmented training data set, train each basic machine learning model and optimize the hyperparameters of each basic machine learning model; based on the optimized basic machine learning models, construct an initial integrated model; optimize the weights of each basic machine learning model in the integrated model by using a sand cat swarm algorithm to obtain an optimized integrated model; predict the forming quality of metal additive manufacturing based on the optimized integrated model to obtain a prediction result. Thereby effectively overcoming the problems of low prediction accuracy, poor robustness, and large optimization blindness in the prior art, significantly improving the forming quality and material utilization efficiency of AM, and reducing production costs. Description of the Drawings

[0017] Figure 1It is a flowchart of a method for predicting the forming quality of metal additive manufacturing based on ensemble learning in an embodiment of the present invention; Figure 2 It is a flowchart for constructing an ensemble model in an embodiment of the present invention; Figure 3 It is a comparison diagram for verifying the prediction of the forming width in an embodiment of the present invention; Figure 4 It is a comparison diagram for verifying the prediction of the forming layer height in an embodiment of the present invention; Figure 5 It is an explanatory diagram of the model based on the quantification of game feature contributions in an embodiment of the present invention. Specific embodiments

[0018] The following further describes in detail a method and device for predicting the forming quality of metal additive manufacturing based on ensemble learning proposed by the present invention in combination with the accompanying drawings and specific embodiments. The advantages and features of the present invention will be clearer according to the following description and the claims.

[0019] This embodiment provides a method for predicting the forming quality of metal additive manufacturing based on ensemble learning, including the following steps: Collect key process parameters that affect the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and construct an original data set; Based on a multi-modal data augmentation architecture based on Gaussian distribution fusion, augment the original data set to obtain an augmented training data set; Select a variety of basic machine learning models, and based on the augmented training data set, train each basic machine learning model and optimize the hyperparameters of each basic machine learning model; Based on the optimized basic machine learning models, construct an initial ensemble model; use the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model to obtain an optimized ensemble model; Based on the optimized ensemble model, predict the forming quality of metal additive manufacturing to obtain a prediction result.

[0020] This method realizes high-precision prediction of the forming quality through in-depth mining and analysis of the process parameter data in the AM process, and uses advanced machine learning algorithms and model integration strategies. At the same time, a multi-modal data augmentation architecture based on Gaussian distribution fusion is used to solve the problem of limited data of aluminum alloy WAAM, thereby effectively overcoming the problems of low prediction accuracy, poor robustness, and large blindness in optimization in the prior art, significantly improving the forming quality and material utilization efficiency of AM, and reducing production costs.

[0021] Taking the WAAM experiment of aluminum alloy as an example, this paper specifically illustrates how to precisely control various process parameters in the WAAM process of aluminum alloy to achieve high-quality aluminum alloy WAAM forming experiments. In this experiment, the robot system consists of a FANUC M-10iD / 8L model robot, its supporting robot control cabinet, welding torch, and wire feeding mechanism. The robot carries the welding torch and wire feeding mechanism, and through precise motion control, the welding torch performs welding operations along the preset teaching trajectory.

[0022] The welding system consists of a high-performance welding device of the MEGMEET Artsen Plus 500 model, a welding torch, and a pure argon gas protection device. This welding system is used to precisely control the key process parameters during welding to meet the requirements of different experimental conditions.

[0023] The data processing system consists of an industrial control computer equipped with the Windows 11 operating system. In this industrial control computer, the Python programming language is used, combined with relevant machine learning libraries such as Scikit-learn, to realize the development of models such as Extreme Gradient Boosting (XGBoost), Extreme Random Trees (ET), and Support Vector Machines (SVM). Thus, data preprocessing, model training, integrated model development, prediction result interpretation, and evaluation index calculation are carried out on the data in the WAAM process, providing precise prediction and control for the aluminum alloy WAAM forming experiment.

[0024] Among them, the data processing system is the implementation system of the metal additive manufacturing forming quality prediction method based on ensemble learning in this embodiment. Specifically, please refer to Figure 1 This metal additive manufacturing forming quality prediction method based on ensemble learning includes the following steps: Data acquisition and preprocessing: Data acquisition: First, a variety of key process parameters affecting the forming quality of the WAAM process were systematically identified and determined, and a reasonable value range was set for each parameter. Specifically, these process parameters include: welding current (30 - 10 A), welding speed (18 - 35 cm / min), swing amplitude (0 - 6 Hz), swing frequency (0 - 8 mm), and cooling time (0 - 180 s). Secondly, according to the predetermined process parameter combinations, WAAM experiments were carried out, and key forming quality indicators such as the width of the additive weld bead and the height of the weld bead layer were recorded under the corresponding process parameter combinations. By collecting experimental data of multiple different process parameter combinations, a total of 25 groups of original data sets were obtained, which will provide a data basis for subsequent data augmentation and the construction of machine learning models.

[0025] Data augmentation: To expand the training data of the machine learning model, the collected original data set is divided into an original training set and an original test set according to a ratio of 7:3, and the original training set is expanded using a multi-modal data augmentation architecture based on Gaussian distribution fusion. By fusing multiple Gaussian distributions, new data samples are generated to enhance the diversity of the data.

[0026] Suppose the original training data set is , where represents each data sample.

[0027] The multi-modal data augmentation architecture based on Gaussian distribution fusion can be expressed as: Among them, represents the number of Gaussian distributions, represents the weight coefficient of the th Gaussian distribution, and satisfies , represents a Gaussian distribution with a mean of and a covariance matrix of .

[0028] When using the multi-modal data augmentation architecture based on Gaussian distribution fusion, the number of Gaussian distributions is set to K = 5, and the maximum number of iterations is 200.

[0029] When performing probability distribution fitting on the original training data set , first use the Expectation-Maximization (EM) algorithm to preliminarily estimate the parameters (mean , covariance matrix , weight ) of the multi-modal data augmentation architecture based on Gaussian distribution fusion. In each iteration, according to the sample distribution characteristics in the original training data set , dynamically adjust the parameters of the multi-modal data augmentation architecture based on Gaussian distribution fusion until convergence to the optimal solution.

[0030] Among them, the Expectation-Maximization (EM) algorithm is an iterative optimization method for parameter estimation in statistical models. The core idea is to gradually optimize the model parameters by alternately executing two steps: the Expectation step (E-step) and the Maximization step (M-step). These two steps are alternated in each iteration until convergence. In the E-step, the algorithm calculates the conditional expectation of the observed data based on the current parameter estimates. Specifically, it calculates the expected values of the latent variables, which are unobserved but can be deduced from the current parameter estimates. For example, in a Gaussian mixture model, the E-step calculates the probability that each observed data belongs to a certain Gaussian distribution. In the M-step, the algorithm maximizes the log-likelihood function using the expected values obtained in the E-step, thereby updating the model parameters. This step usually involves taking the derivative of the parameters and finding the parameter values that maximize the likelihood function. Through this alternating iterative approach, the EM algorithm can gradually approach the maximum likelihood estimate or the maximum a posteriori estimate of the model.

[0031] According to the multi-modal data augmentation architecture based on Gaussian distribution fusion obtained by fitting, new data samples are generated and combined with the original training data set to form the augmented training data set together. Among them, generating new data samples by the multi-modal data augmentation architecture based on Gaussian distribution fusion includes the following steps: Randomly select a Gaussian distribution according to the weight of each Gaussian distribution; Randomly draw a data sample from the selected Gaussian distribution, and this data sample is the new data sample.

[0032] Repeat the above steps to obtain a new data set, and merge it with the original training data set to obtain the augmented training data set.

[0033] Compared with existing data augmentation methods such as SMOTE and GAN, the multi-modal data augmentation method based on Gaussian distribution fusion generates new data, which can better simulate the real data distribution and maintain the consistency and coherence of the data. During training, the central tendency and dispersion degree of the data can be controlled, effectively enhancing the usability of small sample data sets and improving the prediction accuracy and reliability of the model.

[0034] For the augmented training data set Perform data standardization processing to avoid model training bias, thereby improving the stability and accuracy of model training. The method of data standardization processing is: Among them, represents the data set data sample, and respectively represent the dataset the minimum value and the maximum value, represent the standardized data samples.

[0035] Base model construction and training: Select multiple basic machine learning models with different learning abilities and characteristics, including Extreme Gradient Boosting (XGBoost), Extra Trees (ET), Support Vector Machine (SVM), etc. These models have their own advantages and characteristics when dealing with different types of data and problems, and can learn and model data from different perspectives. The Extreme Gradient Boosting (XGBoost), Extra Trees (ET), and Support Vector Machine (SVM) are all existing and relatively mature models, and their specific structures will not be introduced in detail here.

[0036] Divide the standardized training set according to the five-fold cross-validation method, that is, randomly divide the training set into five subsets, and each subset contains approximately the same number of samples. In each cross-validation, use four of the subsets as the training set, and the remaining one subset as the validation set, and calculate the Root Mean Square Error (RMSE) of the model on the validation set.

[0037] For each basic model, first train it on the training subset. During the training process, use the Optuna optimization algorithm to automatically search for and optimize the hyperparameters of the model. Specifically, for XGBoost, optimize its learning rate, maximum depth and other hyperparameters; for ET, optimize the number of trees, maximum depth and other hyperparameters; for SVM, optimize the kernel function, regularization parameter, etc. Optuna automatically explores the optimal hyperparameter combination by defining the search space and the objective function to minimize the RMSE in cross-validation. By adjusting these hyperparameters, the model can better learn the features and patterns from the training data. Optuna is an open-source Python library for hyperparameter optimization, designed specifically for machine learning and deep learning tasks, and its structure will not be introduced in detail here.

[0038] In each cross-validation, use the hyperparameters optimized by Optuna to train the model, and evaluate the performance of the trained model on the validation set, and find the basic model that minimizes the calculated RMSE. The smaller the RMSE, the better the prediction performance of the model. Its calculation formula is: where, represents the number of samples in the validation set in cross-validation, represents the true value, represents the model prediction value.

[0039] Ensemble learning model construction: For the three optimized basic models, randomly initialize their weights in the ensemble model.

[0040] Use the sand cat swarm algorithm to optimize the dynamic weights of the three basic models. The sand cat swarm algorithm is an optimization algorithm that simulates the predation behavior of sand cats. By simulating the search and predation behavior of sand cats in the desert, it searches for the optimal solution. In this embodiment, the weight combination of the basic models is used as the optimization object of the sand cat swarm algorithm, and the fitness function is defined as the RMSE of the ensemble model on the validation set, with the goal of minimizing the fitness function. The specific steps are as follows, and the algorithm processing flow is as Figure 2 shown.

[0041] Randomly generate a certain number of sand cat individuals, and each individual represents a group of weight combinations of the basic models. At the same time, the population size is set to 50, and the number of iterations is set to 20 - 50.

[0042] Calculate the fitness value of each individual, that is, the RMSE of the ensemble model on the validation set. The smaller the fitness value, the better the weight combination corresponding to the individual.

[0043] Prey search stage This algorithm updates the position (weight combination) of the individual by simulating the search behavior of sand cats and based on the following formula.

[0044] Among them, represents the weight vector of the current individual, represents the weight vector of the individual with the minimum fitness (the best prey), represents the weight vector of a randomly selected individual, and represent random coefficients with values in [0, 1].

[0045] Prey attack stage When a sand cat individual approaches the best prey (i.e., the individual with the minimum fitness value ) to a certain extent (such as reaching the preset range), it enters the prey attack stage. In this stage, the weight combination is further adjusted to converge to the optimal solution faster, and the position update can be expressed as: Among them, represents the weight vector of a randomly selected better individual (prey), represents the weight vector of another randomly selected individual, and represent random coefficients with values in [0, 1], represents a random angle.

[0046] Repeat the above process of searching for and attacking prey until the preset termination condition is met, i.e., minimizing the fitness function.

[0047] After being optimized by the sand cat swarm algorithm, the optimal combination of the weights of the basic models is obtained. According to these weights, an ensemble model is constructed, and the final prediction result of the ensemble model is obtained. The prediction result of the ensemble model in cross-validation can be expressed as: Where, represents the prediction result of the ensemble model, represents the dynamic weight allocation of different basic machine learning models, represents the prediction result of the basic machine learning model on the validation set.

[0048] Perform performance evaluation on the constructed ensemble model on the validation set, and calculate its RMSE index. If the RMSE does not meet the requirements, the parameters of the sand cat swarm algorithm can be adjusted until the RMSE of the ensemble model on the validation set meets the requirements.

[0049] Model Testing and Interpretation Use an independent test set to test the final ensemble model. The test set is a data set that has never participated in model training and validation, and can objectively evaluate the generalization ability and practical application effect of the model. During the test, the minimum and maximum values of the previous training set are used to standardize the test set data. Then, the standardized test set data is input into the ensemble model to obtain the prediction result of the model. At the same time, calculate indexes such as prediction error, such as mean absolute error (MAE), RMSE, and coefficient of determination (R 2 ) etc., to evaluate the prediction accuracy of the model. Among them, MAE and R 2 are expressed using the following formulas. The effects of the present invention on width and floor height prediction are as Figure 3 and Figure 4 shown.

[0050] Where, represents the number of test set samples, represents the true value, represents the model prediction value, represents the mean of the true values.

[0051] After that, an interpretation strategy based on game feature contribution quantification is adopted to interpret the ensemble model. This interpretation strategy is based on game theory principles, which can accurately calculate the contribution values of different process parameters to the model prediction results and clearly reveal the interaction relationships between different process parameters, thus significantly improving the transparency and interpretability of the model. The corresponding interpretation results are as Figure 5 shown. In this interpretation strategy, assume that the ensemble model is , for each process parameter in the feature vector , its contribution value to the model prediction result can be calculated through the following steps: Define a feature subset , representing the combination of other parameters except parameter . Define to represent the predicted value of the model under the feature subset .

[0052] Then the contribution value of parameter can be given by the following formula: where represents the number of elements in the feature subset .

[0053] By calculating the value of each parameter, the importance and contribution degree of each process parameter to the model prediction result can be quantitatively obtained.

[0054] In this embodiment, the multi-modal data augmentation architecture based on Gaussian distribution fusion can effectively enhance the diversity of WAAM data samples and reduce the prediction error of the model. After adopting this data augmentation architecture, compared with the case without enhanced data, the MAE of a single machine learning model is reduced by 23.55% and 29.34% on average, and the MSE is reduced by 36.44% and 40.61% on average.

[0055] When using the ensemble model in the aluminum alloy WAAM experiment, it shows high prediction accuracy and robustness. Compared with a single machine learning prediction model, this method reduces the MAE by 12.51% and 2.36% respectively, and reduces the RMSE by 9.98% and 3.00% respectively, and the 2 R increases by 1.21% and 1.50% in turn. And compared with the mainstream average and stacking ensemble methods, this method reduces the MAE by 7.83% on average and the RMSE by 13.48% on average in the forming size prediction, significantly improving the robustness and generalization ability of the model.

[0056] The model interpretation strategy based on game feature contribution quantification can quantify the decision-making process of the model, clarify the influence degree and interaction relationship of different process parameters on the forming quality, and can carry out process regulation according to the model interpretation results.

[0057] In summary, this method is not only applicable to aluminum alloy WAAM, but also can be extended to the WAAM process of other metal materials after appropriately adjusting the data set, with good adaptability and generality, providing strong support for the wide application and development of metal additive manufacturing technology.

[0058] Based on the same concept, this embodiment also provides a device for predicting the forming quality of metal additive manufacturing based on ensemble learning, including: A data acquisition module, used to collect key process parameters affecting the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, cooling time, etc., and construct an original data set; A data enhancement module, used to expand the original data set based on a multi-modal data enhancement architecture fused with Gaussian distribution to obtain an expanded training data set; A model optimization module, used to select a variety of basic machine learning models, and train each basic machine learning model based on the expanded training data set to optimize the hyperparameters of each basic machine learning model; An ensemble learning module, used to construct an initial ensemble model based on the optimized basic machine learning models; use the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model to obtain an optimized ensemble model; A prediction module, used to predict the forming quality of metal additive manufacturing based on the optimized ensemble model to obtain a prediction result.

[0059] This device is used to implement the above-mentioned method for predicting the forming quality of metal additive manufacturing based on ensemble learning, and its specific implementation is similar and will not be repeated here.

[0060] The above has described the embodiments of the present invention in detail with reference to the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalent technologies, they still fall within the protection scope of the present invention.

Claims

1. A method for predicting the forming quality of metal additive manufacturing based on ensemble learning, characterized in that Including: Collect key process parameters that affect the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and construct an original data set. Based on a multi-modal data augmentation architecture fused with Gaussian distribution, augment the original data set to obtain an augmented training data set. Select multiple basic machine learning models, and based on the augmented training data set, train each basic machine learning model and optimize the hyperparameters of each basic machine learning model. Based on the optimized basic machine learning models, construct an initial ensemble model. Use the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model to obtain an optimized ensemble model. Based on the optimized ensemble model, predict the forming quality of metal additive manufacturing to obtain a prediction result.

2. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 1, characterized in that Based on the multi-modal data augmentation architecture fused with Gaussian distribution, further augmenting the data set includes: Use the expectation maximization algorithm to preliminarily estimate the parameters of the multi-modal data augmentation architecture fused with Gaussian distribution; the parameters include mean, covariance matrix, and weight. In each iteration, dynamically adjust the parameters of the multi-modal data augmentation architecture fused with Gaussian distribution according to the distribution characteristics of the samples in the original data set until convergence to the optimal solution. According to the fitted multi-modal data augmentation architecture fused with Gaussian distribution, generate new data samples and merge them with the original data set to form an augmented training data set.

3. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 2, wherein Perform data standardization processing on the augmented training data set to improve the stability and accuracy of model training.

4. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 3, wherein Select multiple basic machine learning models, and based on the augmented training data set, further training each basic machine learning model includes: Select basic machine learning models including extreme gradient boosting model, extreme random tree model, and support vector machine, and based on the standardized training data set, use the K-fold cross-validation method for training to optimize the hyperparameters of each basic machine learning model.

5. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 4, wherein During the training process, use the Optuna algorithm to automatically search for and optimize the hyperparameters of each basic machine learning model. In each cross-validation, use the optimized hyperparameters to train the basic machine learning model and evaluate the performance of the trained model on the validation set until the basic machine learning model with the minimized RMSE is obtained.

6. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 1, wherein Using the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model further includes: Take the weight combination of the basic machine learning model as the optimization object of the sand cat swarm algorithm, take the RMSE of the ensemble model on the validation set as the fitness function, and obtain the weight combination when the fitness function is minimized through the sand cat swarm algorithm.

7. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 6, wherein Randomly generate a preset number of sand cat individuals, and each sand cat individual represents a group of weight combinations. Calculate the fitness value of each individual using the fitness function. According to the search characteristics of the sand cat swarm algorithm, update the weight combination through the following formula: Among them, represents the weight vector of the current individual, represents the weight vector of the individual with the minimum fitness, represents the weight vector of a randomly selected individual, and represents a random coefficient with a value in [0, 1]; When the sand cat individual reaches the preset condition, further adjust the weight combination through the following formula to quickly converge to the optimal solution: Among them, represents the weight vector of a randomly selected individual, represents the weight vector of another randomly selected individual, and represents a random coefficient with a value in [0, 1], represents a random angle.

8. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 6, wherein, Based on the weight combination optimized by the sand cat swarm algorithm, construct an ensemble model and obtain the prediction formula of the ensemble model. Among them, represents the prediction result of the integrated model, represents the dynamic weight allocation of different basic machine learning models, represents the prediction result of the basic machine learning model on the validation set.

9. The method for predicting the forming quality of metal additive manufacturing based on ensemble learning according to claim 1, wherein After obtaining the integrated model, a model interpretation strategy based on game feature contribution quantification is used to interpret the integrated model, and the contribution values of different key process parameters affecting the forming quality of the WAAM process to the prediction results of the integrated model are calculated to improve the interpretability of the integrated model.

10. A forming quality prediction device for metal additive manufacturing based on ensemble learning, characterized in that It includes: A data acquisition module for collecting key process parameters affecting the forming quality of the WAAM process, including welding current, welding speed, swing amplitude, swing frequency, and cooling time, and constructing an original data set; A data augmentation module for augmenting the original data set based on a multi-modal data augmentation architecture fused with Gaussian distribution to obtain an augmented training data set; A model optimization module for selecting multiple basic machine learning models, training each basic machine learning model based on the augmented training data set, and optimizing the hyperparameters of each basic machine learning model; An ensemble learning module for constructing an initial ensemble model based on the optimized basic machine learning models; using the sand cat swarm algorithm to optimize the weights of each basic machine learning model in the ensemble model to obtain an optimized ensemble model; A prediction module for predicting the forming quality of metal additive manufacturing based on the optimized ensemble model to obtain prediction results.

Images