Artificial intelligence-based 3D printing powder proportioning optimization method and system

Abstract

The application relates to the technical field of intelligent optimization, and discloses a 3D printing powder proportioning optimization method and system based on artificial intelligence. The method comprises the following steps: collecting historical data corresponding to powder materials based on the printing process of a 3D printing product, performing multivariate analysis on the historical data, and generating a plurality of powder compositions; a material simulation model simulates the properties of any powder composition, outputs material property data, and extracts a powder proportioning scheme meeting printing requirements; a product prediction model predicts the product of any powder proportioning scheme, outputs a product prediction result, matches the product prediction result with ideal product data, and selects the best powder proportioning scheme and the corresponding best printing process; and the performance of an actual printing product is tested, and if the actual printing product does not meet the requirements, the step is repeatedly executed, and a new best powder proportioning scheme and a new best printing process are selected again. The application improves the precision of 3D printing powder proportioning optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent optimization, in particular to a 3D printing powder proportioning optimization method and system based on artificial intelligence. BACKGROUND

[0002] In the field of 3D printing, the proportioning of powder materials plays a crucial role in the quality, performance and efficiency of the printed products. Traditional powder proportioning methods mainly rely on empirical formulas and extensive experimental verification. This method not only consumes time and effort, but also is difficult to adapt to complex and variable printing requirements and material characteristics. With the rapid development of artificial intelligence technology, its application in materials science has gradually attracted attention. By using artificial intelligence algorithms, especially machine learning and deep learning technologies, a large amount of experimental data can be analyzed and mined, thereby achieving precise optimization of 3D printing powder proportioning. However, at present, the application of artificial intelligence technology in 3D printing powder proportioning optimization is relatively small, and lacks systematicness and efficiency. Therefore, developing a 3D printing powder proportioning optimization method and system based on artificial intelligence has important practical significance for improving 3D printing efficiency, reducing cost and improving product quality.

[0003] Similar prior art is Chinese patent application CN119294270A, which discloses a parameter optimization method for powder laser 3D printing forming process, the method comprising: determining the solution space of the particle swarm algorithm, performing finite element analysis on each process parameter, determining the objective function of the particle swarm algorithm, obtaining simulated particles and measured particles in the solution space, during the iterative training process of the pre-constructed support vector machine using the measured particles, calculating the credibility of the simulated particles at any time of training, when the product of the credibility and the normalized simulated fitness value is greater than a preset value, and the measured fitness value of the simulated particle is greater than a conversion threshold, the simulated particle is used as a measured particle in the subsequent training process, the training process is repeated, and the target process parameter is obtained after training. The application can improve the accuracy of the determined target process parameter while reducing the amount of calculation and reducing the time cost. Chinese patent application CN119720647A discloses a 3D printing parameter optimization method and system based on machine learning, relating to the field of 3D printing technology, comprising: obtaining a data set; preprocessing the data set; constructing a machine learning-based training model according to the data set; training the training model using the data set, and using the trained training model as a 3D printing parameter optimization model; and optimizing the 3D printing parameters to be optimized using the 3D printing parameter optimization model. The application constructs and trains a machine learning-based training model to obtain a 3D printing parameter optimization model, and optimizes the 3D printing parameters to be optimized using the 3D printing parameter optimization model, solving the problem of requiring a large amount of computing resources and time to determine 3D printing parameters in the prior art, and improving the calculation effect and accuracy of 3D printing parameters.

[0004] The deficiencies of the prior art mainly lie in that they mainly consider the optimization of process parameters, do not introduce the optimization of powder material ratio, and do not consider the influence of the characteristics of different materials on the performance of the printed product under the same process parameters. In actual situations, the printing process parameters and the printing powder ratio need to be combined to improve the performance of the printed product. SUMMARY

[0005] The present application provides a 3D printing powder ratio optimization method and system based on artificial intelligence, which is used to improve the accuracy of 3D printing powder ratio optimization based on artificial intelligence.

[0006] In a first aspect, the present application provides a 3D printing powder ratio optimization method based on artificial intelligence, which comprises:

[0007] Collecting historical data corresponding to the powder material based on the printing process of the 3D printed product, the historical data including material data and product performance data, performing multivariate analysis on the historical data based on the printing process and the product performance data to generate a plurality of powder compositions;

[0008] Building a material simulation model simulating the properties of any of the powder compositions, outputting material property data, and extracting a powder ratio meeting the printing requirements from all the powder compositions based on the material property data;

[0009] Building a product prediction model predicting the product based on the printing process for any of the powder ratios, outputting a product prediction result, and matching and optimizing the product prediction result with ideal product data to screen out the best powder ratio and the corresponding best printing process;

[0010] The best powder ratio is used for 3D printing using the best printing process, and the performance of the actual printed product is tested. If the actual printed product does not meet the requirements, repeat this step to re-screen a new best powder ratio and a new best printing process.

[0011] In combination with the first aspect, the generating a plurality of powder compositions comprises:

[0012] After normalizing the historical data, factor analysis is performed to obtain the explanatory factors between any variables in the historical data, and the influence value of the explanatory factors on the product performance data is obtained based on interaction analysis, and a plurality of key variables are extracted from all variables based on the influence value.

[0013] Based on the key variables, the proportion range of any material in the material data is obtained, and a plurality of powder compositions are randomly sampled based on the proportion range.

[0014] In combination with the first aspect, the building a material simulation model comprises:

[0015] The characteristic quantity of the material data is set as the explanatory variable, and the characteristic quantity of the product performance data is set as the target variable, and the explanatory variables and the target variables are collected and summarized to generate a first data set;

[0016] Based on the number of explanatory variables, the number of nodes of the input layer is set, based on the type of target variables, the number of nodes of the output layer is set, a multilayer perceptron model is built from the input layer, hidden layer and output layer, the first data set is trained using the multilayer perceptron model, and a first simulation value is output based on the target variable;

[0017] Adjust model parameters of the multilayer perceptron model through a comparison result of the product performance data and the first simulation value to generate a first model;

[0018] Obtain simulation variables of the explanatory variables using a theoretical calculation method, and collect and aggregate the explanatory variables, the simulation variables, and the target variables to generate a second data set;

[0019] Set a number of nodes of a new input layer based on a number of the explanatory variables and the simulation variables, build a new multilayer perceptron model from the new input layer, a hidden layer, and the output layer, train the second data set using the new multilayer perceptron model, and output a second simulation value based on the target variable;

[0020] Adjust model parameters of the new multilayer perceptron model through a comparison result of the product performance data and the second simulation value to generate a second model;

[0021] Combine the first model and the second model to set the material simulation model.

[0022] With reference to the first aspect, the output material property data includes:

[0023] Extract a first characteristic quantity of the powder composition, input the first characteristic quantity into the first model after preprocessing, and output a first estimation result;

[0024] Perform cluster analysis on the first estimation result and all the first simulation values, extract a cluster containing the first estimation result, and set the explanatory variable corresponding to the cluster as a reference variable;

[0025] Obtain a first simulation variable corresponding to the first characteristic quantity using a theoretical calculation method, input the first characteristic quantity, the first simulation variable, and the reference variable into the second model after combination, output a second estimation result, and set the second estimation result as the material property data.

[0026] With reference to the first aspect, the construction of the product prediction model includes:

[0027] Obtain printing parameters of different powder materials in the printing process, filter out a printing process parameter based on a correlation between the product performance data and the printing parameters, and set remaining printing parameters as auxiliary parameters;

[0028] Generate a mathematical formula between the printing process parameter and the product performance data based on a regression analysis method, calculate a first prediction value of a target performance based on the mathematical formula, and calculate a difference between the first prediction value and the product performance data based on a type of the target performance.

[0029] constructing a supplementary prediction model based on a neural network, setting the auxiliary parameters and the difference as input variables, setting the product performance data as output variables, and setting the combination of the input variables and the output variables as a training data set;

[0030] training the supplementary prediction model based on the input variables using the training data set, outputting a predicted correction value based on the output variables, and setting the trained prediction model as the product prediction model.

[0031] In combination with the first aspect, inputting the printing parameters and the powder ratio into the product prediction model, outputting a second prediction value based on the target performance, adding the second prediction value and the predicted correction value, and setting as the product prediction result.

[0032] In combination with the first aspect, the matching optimization of the product prediction result and the ideal product data includes:

[0033] comparing the product prediction result of the same category with the ideal product data, calculating an evaluation value, and storing the combination of the evaluation value, the powder ratio, and the product prediction result;

[0034] extracting a low-dimensional feature quantity associated with the evaluation value from the product prediction result based on principal component analysis, storing the combination of the low-dimensional feature quantity, the powder ratio, and the evaluation value;

[0035] generating a transformation parameter based on the powder ratio and the low-dimensional feature quantity, using a gradient descent algorithm to screen out the transformation parameter corresponding to the minimum evaluation value and setting it as the best transformation parameter, converting the best transformation parameter into a new input parameter based on inverse transformation, generating a new powder ratio based on the new input parameter, and storing the combination of the new powder ratio, the best transformation parameter, and the minimum evaluation value;

[0036] repeating this step based on the storage record to complete the matching optimization.

[0037] In combination with the first aspect, screening out the best powder ratio and the corresponding best printing process includes:

[0038] setting the new powder ratio generated after matching optimization as the best powder ratio;

[0039] extracting the product prediction result corresponding to the best powder ratio from the storage record, outputting the printing parameters corresponding to the product prediction result based on the product prediction model, and setting the printing parameters as the best printing process.

[0040] In combination with the first aspect, the generating the transformation parameter based on the powder ratio and the low-dimensional feature quantity comprises:

[0041] The parameter prediction model is constructed based on polynomial regression, the parameter prediction model is trained based on the feature quantity of all the powder ratios and the low-dimensional feature quantity, the feature quantity of any powder ratio is input into the trained parameter prediction model, a predicted feature quantity is output, and the predicted feature quantity is set as the transformation parameter.

[0042] In a second aspect, the present application provides a 3D printing powder ratio optimization system based on artificial intelligence, comprising:

[0043] A collection module is configured to collect historical data corresponding to powder materials according to a printing process of a 3D printing product, the historical data including material data and product performance data, perform multivariate analysis on the historical data based on the printing process and the product performance data, and generate a plurality of powder compositions.

[0044] A simulation module is configured to construct a material simulation model, the material simulation model simulating the properties of any of the powder compositions and outputting material property data, and extracting a powder ratio meeting printing requirements from all the powder compositions based on the material property data.

[0045] A prediction module is configured to construct a product prediction model, the product prediction model predicting a product based on the printing process for any of the powder ratios, outputting a product prediction result, and matching and optimizing the product prediction result with ideal product data to screen out an optimal powder ratio and a corresponding optimal printing process.

[0046] A printing module is configured to use the optimal printing process to perform 3D printing on the optimal powder ratio, test the performance of an actual printing product, and if the actual printing product does not meet the requirements, repeat the step to re-screen a new optimal powder ratio and a new optimal printing process.

[0047] The technical scheme provided in the application first, through collecting the historical data of the printing process and the product performance of the 3D printing finished product, performing multivariate analysis, identifying the key factors affecting the product performance, and generating multiple potential powder compositions, the limitations of relying on experience or trial and error method in the traditional method are avoided. Then, a material simulation model is constructed, the properties of the powder composition are simulated, the material property data are output, and the powder ratio meeting the printing requirements is screened based on the data, the powder ratio meeting the printing requirements can be accurately screened, and the accuracy of the ratio optimization is improved. Then, a finished product prediction model is constructed, the finished product performance is predicted based on the printing process, and the best powder ratio and the best printing process are screened through matching optimization with ideal finished product data, through the multi-model cooperation and correction mechanism, the accurate prediction of the finished product performance is realized, which provides a reliable basis for optimization and improves the intelligent and automatic level of optimization. Finally, through the iteration optimization and feedback mechanism, the continuous improvement of the ratio and process is realized, the performance of the finished product is continuously improved, the ratio and process parameters are adaptively adjusted according to the actual printing result, and the adaptability and robustness of the system are improved.

[0048] The application also realizes the fine adjustment of the powder ratio and the printing parameters through parameter generation, further improves the finished product performance, stores the powder ratio, the printing parameters, the finished product performance data and the intermediate results in the optimization process, forms a knowledge base, and provides a reference for subsequent optimization. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical scheme of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0050] Figure 1 An embodiment schematic diagram of the 3D printing powder ratio optimization method based on artificial intelligence in the embodiments of the application;

[0051] Figure 2 An embodiment schematic diagram of the multivariate analysis and powder composition generation process in the embodiments of the application;

[0052] Figure 3 An embodiment schematic diagram of the finished product prediction model construction and optimization process in the embodiments of the application;

[0053] Figure 4 An embodiment schematic diagram of the 3D printing powder ratio optimization system based on artificial intelligence in the embodiments of the application. DETAILED DESCRIPTION

[0054] The embodiments of the present application provide a 3D printing powder proportioning optimization method and system based on artificial intelligence. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0055] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 One embodiment of the 3D printing powder proportioning optimization method based on artificial intelligence in the embodiments of the present application includes:

[0056] Step S101, collect historical data corresponding to the powder material based on the printing process of the 3D printing finished product, the historical data including material data and finished product performance data, perform multivariate analysis on the historical data based on the printing process and the finished product performance data, and generate a plurality of powder compositions.

[0057] It can be understood that the execution subject of the present application can be a 3D printing powder proportioning optimization device based on artificial intelligence, and can also be a terminal or a server, and the specific embodiments are not limited herein. The embodiments of the present application take the server as the execution subject for example.

[0058] Specifically, the 3D printed product refers to an article manufactured using 3D printing technology, the printing process refers to the process data when using 3D printing technology, such as printing temperature, printing speed, layer thickness, laser power, etc. The powder material refers to the powder material data used by different 3D printed products, including but not limited to aluminum alloy powder, titanium alloy powder and various metal powder materials. The historical data refers to the data set generated by summarizing the 3D printing related data collected historically. Among them, the material data includes but is not limited to the proportion, particle size, purity, material properties and other parameters of various powder materials. The product performance data refers to the performance data of the 3D printed product, including but not limited to mechanical properties (such as tensile strength, hardness, toughness), physical properties (such as density, porosity) and chemical properties (such as corrosion resistance, heat resistance) and other data. Through multivariate analysis, the key factors affecting the product performance data and the mutual relationship between the variables can be identified, so as to recombine various powders, generate multiple powder compositions, and preliminarily screen the generated powder compositions to exclude those compositions that obviously do not meet the physical and chemical principles or actual production conditions.

[0059] Step S102, constructing a substance simulation model, the substance simulation model simulates the performance data of the product printed by any powder composition, outputs substance property data, and extracts powder ratio of the powder composition that meets the printing requirements from all powder compositions based on the substance property data.

[0060] Specifically, the substance simulation model refers to a model that simulates the performance data of the product printed by each powder composition without considering the printing process using historical data. The substance property data refers to the performance data of the product printed by the powder composition, such as density, hardness, melting point, thermal expansion coefficient and other physical and chemical properties. The printing requirement refers to the requirement of the product to be printed, and the powder ratio of the powder composition that may meet the printing requirement is extracted.

[0061] Step S103, constructing a product prediction model, the product prediction model predicts the product based on the printing process for any powder ratio, outputs the product prediction result, matches and optimizes the product prediction result with the ideal product data, and selects the best powder ratio and the corresponding best printing process.

[0062] Specifically, the product prediction model refers to a model that predicts the product printed by the powder ratio using different printing processes, and the product prediction result refers to the performance data of the product generated by the powder ratio under different printing processes. The ideal product data refers to the performance data of the product corresponding to the printing requirement. According to the ideal product data, the printing process can be adjusted, and then the product prediction result can be adjusted and optimized, so that the product prediction result meets the ideal product data, that is, the powder ratio and the printing process that best meet the printing requirement, that is, the best powder ratio and the best printing process, can be selected.

[0063] Step S104, the 3D printing of the optimal powder ratio is performed using the optimal printing process, and performance testing is performed on the actual printed product. If the actual printed product does not meet the requirements, the step is repeatedly executed, and a new optimal powder ratio and a new optimal printing process are reselected.

[0064] Specifically, the actual printed product refers to the real-time product after the 3D printing of the optimal powder ratio by the optimal printing process, and the performance testing refers to the performance detection of the actual printed product, including but not limited to mechanical properties, physical properties, and chemical properties, etc. If the actual printed product does not meet the requirements, it means that the performance test result of the actual printed product is different from the ideal product data, and the parameters of the optimal powder ratio and the optimal printing process need to be adjusted. Through the above steps, a new optimal powder ratio and a new optimal printing process can be reselected until the actual printed product meets the requirements.

[0065] In a specific embodiment, a plurality of powder compositions are generated, including:

[0066] (1) After the historical data is normalized, factor analysis is performed to obtain the explanatory factors between any variables in the historical data. Based on the interaction analysis, the influence value of the explanatory factors on the product performance data is obtained, and based on the influence value, a plurality of key variables are extracted from all variables.

[0067] (2) Based on the key variables, the proportion range of any material in the material data is obtained, and a plurality of powder compositions are randomly sampled based on the proportion range.

[0068] Specifically, Figure 2 A flowchart for multivariate analysis and powder composition generation is provided. Normalization refers to converting historical data of different dimensions to the same scale. Factor analysis refers to a process of identifying common factors between latent variables and explaining the correlation between variables. The explanatory factor refers to the common factor between variables. Regression analysis using interaction terms is used as the method of interaction analysis, and the influence of the explanatory factor on the product performance data is calculated, i.e., the influence value is output. For example, the printing speed in the historical data has a significant influence on the tensile strength, corresponding to a larger influence value. The variables corresponding to the explanatory factors with an influence value greater than a preset threshold are set as key variables.

[0069] The proportion range refers to determining the optimal proportion of different materials using an optimization algorithm (such as linear programming, genetic algorithm) according to the key variables and interaction analysis. Random sampling or grid sampling is used to generate a plurality of powder compositions, ensuring that each powder composition is within the proportion range.

[0070] In a specific embodiment, a substance simulation model is constructed, including:

[0071] (1) Set the characteristic quantity of the material data as the explanatory variable, and set the characteristic quantity of the finished product performance data as the target variable, collect and summarize the explanatory variable and the target variable to generate a first data set.

[0072] (2) Set the number of nodes of the input layer based on the number of explanatory variables, set the number of nodes of the output layer based on the type of target variable, build a multi-layer perceptron model from the input layer, hidden layer and output layer, train the first data set using the multi-layer perceptron model, and output the first simulation value based on the target variable.

[0073] (3) Adjust the model parameters of the multi-layer perceptron model through the comparison results of the finished product performance data and the first simulation value, and generate a first model.

[0074] (4) Obtain the simulation variable of the explanatory variable using a theoretical calculation method, collect and summarize the explanatory variable, simulation variable and target variable to generate a second data set.

[0075] (5) Set the number of nodes of the new input layer based on the number of explanatory variables and simulation variables, build a new multi-layer perceptron model from the new input layer, hidden layer and output layer, train the second data set using the new multi-layer perceptron model, and output the second simulation value based on the target variable.

[0076] (6) Adjust the model parameters of the new multi-layer perceptron model through the comparison results of the finished product performance data and the second simulation value, and generate a second model.

[0077] (7) Set the first model and the second model as a material simulation model.

[0078] Specifically, the characteristic quantity of the material data includes, but is not limited to, material composition ratio, material physical and chemical properties, and other related parameters. The characteristic quantity of the finished product performance data includes, but is not limited to, mechanical properties (such as tensile strength, hardness, toughness), physical properties (such as density, porosity), and chemical properties (such as corrosion resistance, heat resistance).

[0079] The hidden layer in the multi-layer perceptron (MLP) model contains multiple neurons. The explanatory variable is taken as the input layer, and the output layer will output the predicted value based on the type of target variable, i.e. the first simulation value. In the training process of the multi-layer perceptron model, set the loss function and optimization algorithm, repeat the forward propagation and back propagation process until the model converges or reaches the predetermined training number of rounds, and adjust the model parameters. Among them, the same target variable type corresponding to the finished product performance data and the first simulation value in the test set are compared for deviation, and the comparison result is output. Thus, the model parameters are adjusted to generate a first model. The main function of the first model is to preliminarily simulate and predict the first simulation value corresponding to various material data, i.e. to preliminarily simulate the properties of the material.

[0080] Theoretical calculation refers to calculating the material properties corresponding to the explanatory variables using quantum chemical calculation, molecular dynamics simulation and other methods as simulation variables.

[0081] The first simulation value output by the first model can be corrected and verified through the simulation variables, and the first data set can be supplemented to generate a second data set. By retraining the second model with the second data set, the generalization ability of the second model can be improved to adapt to different material data, improving the accuracy and reliability of the simulation of material properties. Therefore, the first model and the second model are combined to set up a material simulation model.

[0082] In a specific embodiment, the output material property data includes:

[0083] (1) Extracting the first characteristic quantity of the powder composition, inputting the first characteristic quantity after preprocessing into the first model, and outputting the first estimated result.

[0084] (2) Cluster analysis is performed on the first estimated result and all first simulation values, and a cluster containing the first estimated result is extracted, and the explanatory variable corresponding to the cluster is set as the reference variable.

[0085] (3) The first simulation variable corresponding to the first characteristic quantity is obtained using a theoretical calculation method, and the first characteristic quantity, the first simulation variable and the reference variable are combined and input into the second model to output the second estimated result. The second estimated result is set as the material property data.

[0086] Specifically, the first characteristic quantity refers to the characteristic quantity corresponding to the powder composition based on the type of material data characteristic quantity. The first estimated result refers to the first simulation value generated by the first model according to the first characteristic quantity.

[0087] After summarizing and clustering the first estimated result and all first simulation values, K-means clustering method can be used to divide into multiple clusters, and further extract the cluster containing the first estimated result. Through the cluster, the reference variable similar to the first characteristic quantity can be found among all explanatory variables, indicating that the powder material corresponding to the reference variable has relevance with the powder composition.

[0088] The first simulation variable is used to simulate the material properties of the powder composition. Combining the first characteristic quantity, the first simulation variable and the reference variable can expand the prediction accuracy of the characteristic quantity, and then output the corresponding second estimated result from the second model. For example, the input characteristic quantity is simulated tensile strength (480 MPa), hardness (190 HV), Fe: 60%, Ni: 20%, Cr: 20%, particle size: 15-53 μm, laser power: 200 W, scanning speed: 1000 mm / s. The corresponding output second estimated result is tensile strength (510 MPa), hardness (210 HV), etc.

[0089] In a specific embodiment, a finished product prediction model is constructed, comprising:

[0090] (1) Obtain printing parameters of different powder materials in the printing process, filter out printing process parameters based on the correlation between finished product performance data and printing parameters, and set the remaining printing parameters as auxiliary parameters.

[0091] (2) Generate a mathematical formula between the printing process parameters and the finished product performance data based on a regression analysis method, calculate a first prediction value of the target performance based on the mathematical formula, and calculate the difference between the first prediction value and the finished product performance data based on the type of the target performance.

[0092] (3) Construct a supplementary prediction model based on a neural network, set the auxiliary parameters and the difference as input variables, set the finished product performance data as output variables, and set the combination of the input variables and the output variables as a training data set.

[0093] (4) The supplementary prediction model is trained based on the input variables using the training data set, outputs a prediction correction value based on the output variables, and sets the trained prediction model as the finished product prediction model.

[0094] Specifically, Figure 3 A flowchart is constructed for the finished product prediction model. The construction of the finished product prediction model requires analysis and prediction based on the printing process. The printing parameters refer to the parameter data included in the printing process, such as printing temperature, printing speed, layer thickness, laser power, etc. The correlation is used to select the printing parameters that significantly affect the performance of the finished product, i.e. the printing process parameters. The remaining printing parameters are used as auxiliary features, i.e. auxiliary parameters.

[0095] The mathematical formula refers to the calculation formula corresponding to the multiple regression equation generated by the regression analysis method, which is used to describe the mathematical relationship between the printing process parameters and the finished product performance data. The target performance refers to the performance type corresponding to any finished product performance data. The first prediction value can be estimated for any target performance based on the mathematical formula. The difference is used to describe the deviation of the regression prediction by the mathematical formula.

[0096] The supplementary prediction model is used to learn the relationship between the auxiliary parameters and the difference, and introduces a correction mechanism, thereby improving the accuracy of the final prediction. The auxiliary parameters are the printing parameters with lower correlation that are filtered out, and the training data set can be constructed by input variables and output variables to achieve accurate prediction of the finished product performance.

[0097] The supplementary prediction model is trained according to the training data set to generate the finished product prediction model, which can achieve accurate prediction of the finished product performance data in combination with the key degree of the printing process.

[0098] In a specific embodiment, the printing parameters and the powder ratio are input into the product prediction model, a second prediction value is output based on the target performance, the second prediction value is added to the prediction correction value, and is set as the product prediction result.

[0099] Specifically, since the printing parameters and the powder ratio are multiple, any powder ratio and any printing parameter are input into the product prediction model, and the second prediction value corresponding to the target performance is output. Since the product prediction model introduces a correction mechanism, the second prediction value is added to the prediction correction value to generate the product prediction result.

[0100] In a specific embodiment, the product prediction result is matched and optimized with the ideal product data, including:

[0101] (1) The product prediction result of the same category is compared with the ideal product data, the evaluation value is calculated, and the evaluation value is combined with the powder ratio and the product prediction result for storage.

[0102] (2) Based on principal component analysis, low-dimensional feature quantities associated with the evaluation value are extracted from the product prediction result, and the low-dimensional feature quantities are combined with the powder ratio and the evaluation value for storage.

[0103] (3) Transform parameters are generated based on the powder ratio and the low-dimensional feature quantities, the gradient descent algorithm is used to select the transform parameters corresponding to the minimum evaluation value, and the best transform parameters are set. The best transform parameters are converted into new input parameters based on inverse transformation, new powder ratios are generated based on the new input parameters, and the new powder ratios are combined with the best transform parameters and the minimum evaluation value for storage.

[0104] (4) This step is repeated based on the storage records to complete the matching and optimization.

[0105] Specifically, the evaluation value represents the comparison result between the product prediction result and the ideal product data, and the weight coefficient can be set according to the category of the product prediction result. The evaluation value is calculated by summing the differences between the product prediction results and the ideal product data of all categories according to the weight coefficient.

[0106] Principal component analysis is suitable for linear dimension reduction, and the dimension with the largest variance can be extracted as a feature quantity. Since the categories of the product prediction results are multiple, the feature quantities of the product prediction results are high-dimensional, and low-dimensional feature quantities more directly related to the evaluation value, i.e., low-dimensional feature quantities, need to be extracted from the feature quantities of the high-dimensional product performance data. The correlation can be represented according to the Pearson correlation coefficient.

[0107] The transformation parameter can transform the high-dimensional input parameter space into a low-dimensional transformation parameter space, thereby reducing the complexity of the optimization problem. The gradient descent algorithm can be used as an optimization algorithm to find the transformation parameter that minimizes or maximizes the evaluation value. In this application, the smaller the evaluation value, the closer the finished product prediction result is to the ideal finished product data, and therefore the best transformation parameter can be selected. Since the transformation parameter is obtained by principal component analysis transformation, the best transformation parameter can be converted into a new input parameter using inverse transformation. The category of the new input parameter corresponds to the category of the powder ratio, and therefore a new powder ratio can be generated.

[0108] Storing all the above optimization records can realize the process of matching and optimizing the finished product prediction result with the ideal finished product data.

[0109] In a specific embodiment, the best powder ratio and the corresponding best printing process are selected, including:

[0110] (1) The new powder ratio generated after matching and optimization is set as the best powder ratio.

[0111] (2) The finished product prediction result corresponding to the best powder ratio is extracted from the storage record, and the printing parameters corresponding to the finished product prediction result are output based on the finished product prediction model, and the printing parameters are set as the best printing process.

[0112] Specifically, the new powder ratio generated from the matching and optimization process is obtained. These new powder ratios are converted from the best transformation parameter with the smallest evaluation value selected by the gradient descent algorithm.

[0113] In the finished product prediction model, the reverse mapping function of the finished product prediction model is used to input the best powder ratio and the target performance index, and the corresponding printing parameters, i.e. the best printing process, are output.

[0114] In a specific embodiment, the transformation parameter is generated based on the powder ratio and the low-dimensional feature quantity, including:

[0115] The parameter prediction model is constructed based on polynomial regression, the parameter prediction model is trained based on the feature quantity of all powder ratios and all low-dimensional feature quantities, the feature quantity of any powder ratio is input into the trained parameter prediction model, the predicted feature quantity is output, and the predicted feature quantity is set as the transformation parameter.

[0116] Specifically, the polynomial regression can capture the nonlinear relationship between variables, is suitable for complex data relationships, and is used to build a parameter prediction model. The characteristic quantity corresponding to the powder ratio and the low-dimensional characteristic quantity are set as input variables, and the predicted characteristic quantity is set as an output variable. The output variable can be a transformed characteristic quantity or other indicators that need to be predicted. A suitable polynomial order is selected to balance the complexity and generalization ability of the model. The characteristic quantity of all powder ratios and all low-dimensional characteristic quantities are used as a training set to train the parameter prediction model. The model performs prediction calculation based on the input data and outputs the predicted characteristic quantity.

[0117] The above describes the 3D printing powder ratio optimization method based on artificial intelligence in the embodiments of the present application. The 3D printing powder ratio optimization system based on artificial intelligence in the embodiments of the present application is described below. Please refer to Figure 4 An embodiment of the 3D printing powder ratio optimization system based on artificial intelligence in the embodiments of the present application includes:

[0118] The collection module 201 is configured to collect historical data corresponding to the powder material according to the printing process of the 3D printing finished product. The historical data includes material data and finished product performance data. The historical data is subjected to multivariate analysis based on the printing process and the finished product performance data, and a plurality of powder compositions are generated.

[0119] The simulation module 202 is configured to build a material simulation model. The material simulation model simulates the properties of any powder composition and outputs material property data. Based on the material property data, a powder ratio meeting the printing requirements is extracted from all powder compositions.

[0120] The prediction module 203 is configured to build a finished product prediction model. The finished product prediction model predicts the finished product based on the printing process for any powder ratio, and outputs a finished product prediction result. The finished product prediction result is matched and optimized with ideal finished product data, and the best powder ratio and the corresponding best printing process are selected.

[0121] The printing module 204 is configured to use the best printing process to perform 3D printing on the best powder ratio. The performance of the actual printed finished product is tested. If the actual printed finished product does not meet the requirements, this step is repeated to reselect a new best powder ratio and a new best printing process.

[0122] Through the cooperation of the above-mentioned components, first, by collecting the historical data of the printing process and the performance of the 3D printing finished product, a multivariate analysis is performed to identify the key factors affecting the performance of the finished product, and multiple potential powder compositions can be generated, avoiding the limitations of relying on experience or trial and error in traditional methods. Then, a material simulation model is constructed to simulate the properties of the powder composition, output material property data, and select the powder ratio that meets the printing requirements based on these data, which can accurately select the powder ratio that meets the printing requirements and improve the accuracy of the ratio optimization. Next, a finished product prediction model is constructed to predict the performance of the finished product based on the printing process, and through matching optimization with ideal finished product data, the best powder ratio and the best printing process are selected. Through the cooperation of multiple models and the correction mechanism, the performance of the finished product is accurately predicted, providing a reliable basis for optimization and improving the intelligent and automated level of optimization. Finally, through the iterative optimization and feedback mechanism, the ratio and process are continuously improved to ensure that the performance of the finished product is continuously improved, and the adaptability and robustness of the system are improved by adaptively adjusting the ratio and process parameters according to the actual printing results.

[0123] The present application also realizes fine adjustment of powder ratio and printing parameters through parameter generation, further improves the performance of the finished product, and stores the powder ratio, printing parameters, finished product performance data and intermediate results in the optimization process to form a knowledge base for subsequent optimization.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0125] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0126] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An artificial intelligence-based 3D printing powder proportioning optimization method, characterized in that, The AI-based 3D printing powder ratio optimization method comprises: Collecting historical data corresponding to the powder material based on the printing process of the 3D printing finished product, the historical data including material data and finished product performance data, performing multivariate analysis on the historical data based on the printing process and the finished product performance data, and generating a plurality of powder compositions; Building a material simulation model, which simulates the properties of any of the powder compositions, outputs material property data, and extracts a powder ratio meeting the printing requirements from all the powder compositions based on the material property data; Building a finished product prediction model, which predicts the finished product based on the printing process for any of the powder ratios, outputs the finished product prediction result, and optimizes the matching of the finished product prediction result and the ideal finished product data to screen out the best powder ratio and the corresponding best printing process; Wherein, the finished product prediction result of the same category is compared with the ideal finished product data, the evaluation value is calculated, and the evaluation value is combined with the powder ratio and the finished product prediction result for storage; the low-dimensional feature quantity associated with the evaluation value is extracted from the finished product prediction result based on principal component analysis, and the low-dimensional feature quantity is combined with the powder ratio and the evaluation value for storage; transformation parameters are generated based on the powder ratio and the low-dimensional feature quantity, the transformation parameters corresponding to the smallest evaluation value are screened out using the gradient descent algorithm, and are set as the best transformation parameters; the best transformation parameters are converted into new input parameters based on inverse transformation, new powder ratios are generated based on new input parameters, and the new powder ratios are combined with the best transformation parameters and the smallest evaluation value for storage; this step is repeated based on the storage record to complete the matching optimization; The new powder ratio generated after matching optimization is set as the best powder ratio; the finished product prediction result corresponding to the best powder ratio is extracted from the storage record, and the printing parameters corresponding to the finished product prediction result are output based on the finished product prediction model, and the printing parameters are set as the best printing process; The best powder ratio is used for 3D printing using the best printing process, and the performance of the actual printed finished product is tested. If the actual printed finished product does not meet the requirements, this step is repeated to re-screen new best powder ratios and new best printing processes. 2.The AI-based 3D printing powder proportioning optimization method according to claim 1, characterized in that, The generation of a plurality of powder compositions comprises: After normalizing the historical data, factor analysis is performed to obtain the explanatory factors between any variables in the historical data, and the influence value of the explanatory factors on the finished product performance data is obtained based on interaction analysis, and a plurality of key variables are extracted from all variables based on the influence value; Based on the key variables, the proportion range of any material in the material data is obtained, and a plurality of powder compositions are randomly sampled based on the proportion range. 3.The AI-based 3D printing powder formulation optimization method of claim 1, wherein, The construction of the material simulation model comprises: The characteristic quantity of the material data is set as an explanatory variable, and the characteristic quantity of the product performance data is set as a target variable. A first data set is collected by summarizing the explanatory variable and the target variable; The number of nodes of an input layer is set based on the number of the explanatory variables, and the number of nodes of an output layer is set based on the type of the target variable. A multilayer perceptron model is built by the input layer, a hidden layer and the output layer. The first data set is trained by using the multilayer perceptron model, and a first simulation value is output based on the target variable; A first model is generated by adjusting the model parameters of the multilayer perceptron model through the comparison result of the product performance data and the first simulation value; A simulation variable of the explanatory variable is obtained by using a theoretical calculation method. A second data set is generated by collecting and summarizing the explanatory variable, the simulation variable and the target variable; The number of nodes of a new input layer is set based on the number of the explanatory variable and the simulation variable. A new multilayer perceptron model is built by the new input layer, a hidden layer and the output layer. The second data set is trained by using the new multilayer perceptron model, and a second simulation value is output based on the target variable; A second model is generated by adjusting the model parameters of the new multilayer perceptron model through the comparison result of the product performance data and the second simulation value; The first model and the second model are combined to set the substance simulation model.

4. The AI-based 3D printing powder recipe optimization method of claim 3, wherein, The output substance property data includes: A first characteristic quantity of the powder composition is extracted. The first characteristic quantity is input into the first model after pretreatment, and a first estimation result is output; The first estimation result and all the first simulation values are subjected to cluster analysis. A cluster cluster containing the first estimation result is extracted. The explanatory variable corresponding to the cluster cluster is set as a reference variable; A first simulation variable corresponding to the first characteristic quantity is obtained by using a theoretical calculation method. The first characteristic quantity, the first simulation variable and the reference variable are combined and input into the second model. A second estimation result is output. The second estimation result is set as the substance property data. 5.The AI-based 3D printing powder proportioning optimization method of claim 1, wherein, The construction of the product prediction model includes: Printing parameters of different powder materials are obtained in the printing process. Based on the correlation between the product performance data and the printing parameters, a printing process parameter is screened out from all the printing parameters based on the correlation. The remaining printing parameters are set as auxiliary parameters; A mathematical formula between the printing process parameter and the product performance data is generated based on a regression analysis method. A first prediction value of a target performance is calculated based on the mathematical formula. A difference value between the first prediction value and the product performance data is calculated based on the type of the target performance; A supplementary prediction model is constructed based on a neural network. The auxiliary parameters and the difference value are set as input variables. The product performance data is set as an output variable. The input variables and the output variables are combined to set a training data set; The supplementary prediction model is trained based on the input variables using the training data set, outputs a predicted correction value based on the output variable, and sets the trained prediction model as the finished product prediction model.

6. The AI-based 3D printing powder recipe optimization method of claim 5, wherein, The printing parameters and the powder formula are input into the finished product prediction model, a second predicted value is output based on the target performance, the second predicted value is added to the predicted correction value, and the result is set as the finished product prediction result.

7. The AI-based 3D printing powder recipe optimization method of claim 1, wherein, The transformation parameters are generated based on the powder formula and the low-dimensional feature quantity, including: A parameter prediction model is constructed based on polynomial regression, the parameter prediction model is trained based on the feature quantity of all the powder formulas and all the low-dimensional feature quantities, the feature quantity of any powder formula is input into the trained parameter prediction model, a predicted feature quantity is output, and the predicted feature quantity is set as the transformation parameter.

8. An artificial intelligence-based 3D printing powder proportioning optimization system for implementing the 3D printing powder proportioning optimization method according to any one of claims 1-7, characterized in that, The 3D printing powder formula optimization system based on artificial intelligence includes: A collection module is configured to collect historical data of powder materials according to a printing process of a 3D printing finished product, the historical data including material data and finished product performance data, perform multivariate analysis on the historical data based on the printing process and the finished product performance data, and generate a plurality of powder compositions; An analog module is configured to construct a material simulation model, the material simulation model simulating the properties of any of the powder compositions and outputting material property data, and extracting a powder formula meeting printing requirements from all the powder compositions based on the material property data; A prediction module is configured to construct a finished product prediction model, the finished product prediction model predicting a finished product based on the printing process for any of the powder formulas and outputting a finished product prediction result, matching and optimizing the finished product prediction result with ideal finished product data, and screening out an optimal powder formula and a corresponding optimal printing process; A printing module is configured to use the optimal printing process to perform 3D printing on the optimal powder formula, test the performance of an actual printed finished product, and repeat the step if the actual printed finished product does not meet the requirements, and re-screen a new optimal powder formula and a new optimal printing process.

Images