Method for predicting mechanical property of selective laser melting technology molded part

Abstract

The invention discloses a method for predicting the mechanical property of a selective laser melting technology forming part. The method comprises the steps that SLM process data is obtained, and a data set is established; constructing a prediction model by adopting a random forest regression method, selecting laser power and scanning speed as input characteristics, introducing density of a formed part as an intermediate variable, and taking tensile strength, yield strength, elongation and comprehensive mechanical properties as output characteristics; training and optimizing the prediction model by using the data set; and applying the trained prediction model to a new sample, and predicting four output characteristics of the tensile property index of the molded part by inputting the laser power and the scanning speed of the new sample and combining a predicted value of the density of the molded part. According to the method, the double-layer multi-output prediction model is constructed through the random forest regression method, the density of the formed part is introduced to serve as an intermediate variable, the mechanical property of the SLM material can be predicted only through the laser power and the scanning speed in the SLM technology, the limitation of a traditional physical modeling and single-output prediction method is effectively overcome, and the prediction precision and adaptability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the properties of metal materials. Specifically, it relates to a method for predicting the mechanical properties of components formed by selective laser melting technology based on a random forest regression algorithm. Background Art

[0002] Selective laser melting technology is a technology for forming by selectively melting metal powder with a laser. The principle steps are as follows: First, a three-dimensional model is sliced layer by layer, and parameters such as laser output power, scanning speed, scanning path, scanning spacing, and layer thickness are planned to obtain a data file that can be executed by an SLM device; Second, the SLM device controls the laser beam through a scanning galvanometer according to the data read layer by layer to selectively melt the metal powder; After one layer is processed, the powder feeding cylinder rises, and the forming cylinder descends by the height of one layer thickness. The powder spreading mechanism transports the powder from the powder feeding cylinder to the forming platform, and the device again controls the scanning path of the laser beam to selectively melt the newly spread powder layer and fuse it with the previous layer. This process is repeated until the processing is completed. Metal additive manufacturing has now been widely used in the fields of aerospace, medical devices, and high-end equipment manufacturing. However, the mechanical properties of components formed by SLM technology highly depend on complex process parameters, including laser power, scanning speed, scanning interval, etc. These parameters have a significant impact on the density and mechanical properties of the formed components (such as tensile strength, yield strength, elongation, etc.). Therefore, the process control of advanced metal alloy SLM additive manufacturing technology has been a hot topic in selective laser melting technology in recent years and has become the research content of many experts and scholars. How to achieve the prediction of processing quality through process parameter optimization methods has great research significance for the additive manufacturing industry.

[0003] Traditional methods for predicting the mechanical properties of formed components are mostly based on physical models or empirical formulas. For example, heat conduction and fluid dynamics models are used to solve heat transfer and melt flow dynamics in the molten pool to predict the depth and width of the molten pool. The calculation complexity is high, it is difficult to obtain results quickly, and the mechanical parameters of the formed components cannot be directly predicted. Using machine learning methods to predict the mechanical properties of formed components is relatively more direct. By establishing the correlation between input features and output features, a non-linear model is directly established to achieve the prediction result. Common "parameter - performance" prediction methods include support vector machine models and Gaussian regression models. Usually, the output features of the laser such as laser power, scanning speed, scanning spacing, etc. are used as the input of the model, and the tensile strength, yield strength, or elongation of the formed component is used as a single output, which has certain limitations. Moreover, the model structures of these methods usually do not perform hierarchical optimization for the characteristics of the mechanical properties of formed components. Most of them are simple "input - output" mapping models, ignoring the important role of intermediate state variables. Summary of the Invention

[0004] Aiming at the problems existing in the above-mentioned prior art, the present invention provides a method for predicting the mechanical properties of components formed by selective laser melting technology. By introducing density as an intermediate variable, and ultimate tensile strength (UTS), yield strength, elongation, and the comprehensive performance index UTS*Elongation as multi-output features, combined with random forest hierarchical modeling, it overcomes the defects of low prediction accuracy, poor model adaptability, and insufficient physical interpretability in the prior art, and provides an effective solution for the rapid and accurate prediction of the mechanical properties of components formed by laser additive manufacturing and process optimization.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A method for predicting the mechanical properties of components formed by selective laser melting technology, comprising the following steps:

[0007] S100. Based on selective laser melting technology, use high-entropy alloy powder materials to print components through different process parameters, measure the density of the components and conduct tensile tests on the components, and use the ultimate tensile strength, yield strength, elongation, and comprehensive mechanical properties obtained from the tensile tests as the tensile performance indicators of the components to establish a training data set, where the process parameters include laser power and scanning speed, and the comprehensive mechanical properties are obtained by multiplying the ultimate tensile strength and elongation;

[0008] S200. Use the random forest regression method to construct a prediction model, select laser power and scanning speed as input features, and introduce the density of the component as an intermediate variable, and use the ultimate tensile strength, yield strength, elongation, and comprehensive mechanical properties as output features;

[0009] S300. Use the training data set established with the data obtained in step S100 to train the prediction model. When optimizing the model, select the point with the highest prediction error and add it to the next iteration to reduce the prediction error rate of the model for the untrained area;

[0010] S400. Define a prediction function based on the trained prediction model, input the laser power, scanning speed, and density of the component of the new sample, and output the predicted ultimate tensile strength, yield strength, elongation, and comprehensive mechanical properties to provide a prediction basis for the forming quality of the new sample;

[0011] S500. Apply the defined prediction function to the new sample, and predict the tensile performance indicators of the component by inputting the process parameters including laser power and scanning speed of the new sample, combined with the predicted value of the density of the component.

[0012] Specifically, in the step S100, molded parts with 30 sets of parameters based on selective laser melting technology are obtained, with 3 samples in each set of parameter samples, for a total of 90 sets of samples. The corresponding experimental data are collected, including process parameters, molded part density, and molded part tensile property indexes. These experimental data are organized into a training data set for machine learning, where the input features include laser power, scanning speed, and molded part density, and the output features include ultimate tensile strength (UTS), yield strength, elongation, and comprehensive mechanical property UTS * elongation. And the training data set is loaded by writing a program to ensure data integrity and correctness.

[0013] Specifically, the loaded training data set is also preprocessed in the step S100, including: checking data integrity, filling in missing values or removing abnormal data; normalizing the input features to ensure the scale consistency of the features; dividing the training set and the test set, where the data points with a scanning speed of not less than 1000 mm / s are used as the training set.

[0014] Specifically, when constructing the prediction model using the random forest regression method in the step S200, according to the characteristics of the random forest regression model, a global optimization algorithm is used to search for the best parameter combination in the random forest regression model, including setting the number of decision trees, the maximum tree depth, and the random seed value, and finding the best parameter combination with the minimum root mean square error within this range.

[0015] Specifically, the prediction model in the step S200 includes a two-layer model structure. The first-layer model structure takes laser power and scanning speed as input features and molded part density as the output feature. The second-layer model structure takes laser power, scanning speed, and molded part density as input features and ultimate tensile strength, yield strength, elongation, and comprehensive mechanical property as output features.

[0016] Specifically, in the step S300, the uncertainty sampling method of active learning is adopted during training. The data in the training set divided from the training data set are used to train the prediction model. The first-layer model uses laser power and scanning speed as input features to predict the molded part density. The second-layer model combines the molded part density output by the first-layer model, as well as laser power and scanning speed as input features to predict ultimate tensile strength, yield strength, elongation, and comprehensive mechanical property. When optimizing the model, the points with the highest prediction error in the test set divided from the training data set are selected and added to the training set in the next iteration.

[0017] Specifically, in step S300, after the prediction model is trained, the test set divided from the training data set is used for prediction. The predicted value of the density of the formed part is obtained by the first-layer model, and the predicted values of the tensile strength, yield strength, elongation rate, and comprehensive mechanical properties are obtained by the second-layer model. The predicted values are compared with the corresponding true values to evaluate the performance of the model.

[0018] Specifically, in step S300, the prediction ability of the prediction model is also quantified by evaluation indicators. The evaluation indicators used include the root mean square error and the mean absolute percentage error, and the prediction performance of the prediction model for the training data set is calculated respectively. At the same time, the model predicted value is compared with the true value to ensure that the prediction result meets the engineering accuracy requirements.

[0019] Specifically, in the evaluation indicators, the root mean square error RMSE is expressed as: The mean absolute percentage error MAPE is expressed as: In the formula, y i represents the true value, represents the predicted value, and n is the number of samples.

[0020] Furthermore, in step S300, it also includes displaying the prediction results and performance of the prediction model through visualization means, specifically including: drawing a correlation heat map to show the correlation between input features and output features; drawing a comparison graph of predicted values and true values to intuitively show the prediction ability of the prediction model; based on the prediction results of laser power, scanning speed, and the density of the formed part, generating a three-dimensional prediction surface graph to show the change of the tensile property index of the formed part with input features.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention constructs a double-layer multi-output prediction model through the random forest regression method, introduces the density of the formed part as an intermediate variable, and only needs the laser power and scanning speed in the SLM process to predict the mechanical properties of SLM materials, effectively overcoming the limitations of traditional physical modeling and single-output prediction methods, and improving the prediction accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a schematic flowchart of an embodiment of the present invention.

[0024] Figure 2 is a schematic structural diagram of the random forest regression model in an embodiment of the present invention.

[0025] Figure 3 is a predicted surface graph of the density of the formed part generated by the prediction model in an embodiment of the present invention.

[0026] Figure 4The predicted surface plot of the yield strength generated by the prediction model in an embodiment of the present invention.

[0027] Figure 5 The predicted surface plot of the tensile strength generated by the prediction model in an embodiment of the present invention.

[0028] Figure 6 The predicted surface plot of the elongation rate generated by the prediction model in an embodiment of the present invention.

[0029] Figure 7 The predicted surface plot of the comprehensive mechanical properties generated by the prediction model in an embodiment of the present invention.

[0030] Figure 8 The chart showing the change of the prediction error with the number of model training iterations in an embodiment of the present invention.

[0031] Figure 9 The comparison chart of the prediction performance of the density of the formed part in an embodiment of the present invention.

[0032] Figure 10 The comparison chart of the prediction performance of the yield strength in an embodiment of the present invention.

[0033] Figure 11 The comparison chart of the prediction performance of the tensile strength in an embodiment of the present invention.

[0034] Figure 12 The comparison chart of the prediction performance of the elongation rate in an embodiment of the present invention.

[0035] Figure 13 The comparison chart of the prediction performance of the comprehensive mechanical properties in an embodiment of the present invention.

[0036] Figure 14 The Pearson correlation coefficient heat map between the input process parameters and the output tensile properties index of the formed part in an embodiment of the present invention.

[0037] Figure 15 The comparison chart between the prediction result of the prediction model for the new sample on the density index of the formed part and the experimental verification in an embodiment of the present invention.

[0038] Figure 16 The comparison chart between the prediction result of the prediction model for the new sample on the yield strength index and the experimental verification in an embodiment of the present invention.

[0039] Figure 17 The comparison chart between the prediction result of the prediction model for the new sample on the tensile strength index and the experimental verification in an embodiment of the present invention.

[0040] Figure 18 The comparison chart between the prediction result of the prediction model for the new sample on the elongation rate index and the experimental verification in an embodiment of the present invention.

[0041] Figure 19 This is a comparison chart of the prediction results of the prediction model for new samples in terms of comprehensive mechanical property indexes and experimental verification in an embodiment of the present invention. Specific embodiments

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include but are not limited to the following embodiments.

[0043] Embodiment

[0044] As Figure 1 shown, the method for predicting the mechanical properties of a selective laser melting technology formed part includes the following steps:

[0045] S100. Based on the Selective Laser Melting (SLM) technology, use high-entropy alloy powder materials to print formed parts through different process parameters, measure the density ρ of the formed parts, and conduct tensile tests on the formed parts. Take the ultimate tensile strength UTS, yield strength Yield Strength, elongation Elongation, and comprehensive mechanical property UTS*Elongation obtained from the tensile tests as the tensile property indexes of the formed parts, and use them to establish a training data set. The process parameters include laser power and scanning speed, and the comprehensive mechanical property is obtained by multiplying the ultimate tensile strength and elongation.

[0046] Specifically, taking the high-entropy alloy powder material AlCoCrFeNi2.1 as an example, the powder particle size range is 15 to 53 μm. During the laser processing, the laser power range is selected to be 150 W to 450 W, the scanning speed range is 600 mm / s to 1200 mm / s, the laser scanning interval is 0.07 mm, the rotation angle of the laser scanning path is set to 90°, the powder layer thickness is 0.03 mm, and argon gas Ar is used as the shielding gas. The laser spot diameter is set to 20 μm, and the number of layers of the formed part is 70 layers. The formed parts are printed through the above thirty groups of different process parameters, with three samples for each process parameter, a total of ninety samples. Measure the density of the formed parts and conduct tensile tests. The obtained ultimate tensile strength UTS, yield strength Yield Strength, elongation Elongation, and comprehensive mechanical property UTS*Elongation are used as the evaluation indexes of the mechanical properties of the formed parts and as the output features of the subsequent machine learning data set.

[0047] Among them, the density measurement method is the Archimedes drainage method. Use an electronic scale to measure the mass m of the sample, and use the volume change of the graduated cylinder to represent the volume V of the sample. Therefore, the density ρ = m / V. The ultimate tensile strength UTS = F max / A 0, Yield Strength = F yield / A 0 , Elongation = (L - L 0 ) / L 0 * 100%, where A 0 represents the original cross-sectional area of the material, F max represents the maximum tensile force that the material can withstand before fracture, F yield represents the stress that the material begins to undergo permanent deformation (yield), L 0 represents the original length of the material, and L represents the length of the material after fracture.

[0048] Organize these experimental data into a training dataset for machine learning, where the input features include laser power, scanning speed, and part density, and the output features include ultimate tensile strength (UTS), yield strength, elongation, and comprehensive mechanical property UTS * elongation; and load the training dataset by writing a program to ensure data integrity and correctness.

[0049] Then preprocess the loaded training dataset, including: checking data integrity, filling in missing values or removing abnormal data; normalizing the input features to ensure the scale consistency of the features; dividing the training set and the test set, where the data points with a scanning speed of not less than 1000 mm / s are used as the training set to enhance the generalization ability of the model.

[0050] S200. Build a prediction model using the random forest regression method, select laser power and scanning speed as input features, and introduce part density as an intermediate variable, and use ultimate tensile strength, yield strength, elongation, and comprehensive mechanical property as output features.

[0051] Specifically, when building a prediction model using the random forest regression method, according to the characteristics of the random forest regression model, use a global optimization algorithm to search for the best parameter combination in the random forest regression model, including setting the number of decision trees, the maximum tree depth, and the random seed value, and find the best parameter combination with the minimum root mean square error within this range, which is beneficial to model training and reduces the prediction error.

[0052] The prediction model includes a two-layer model structure. The first-layer model structure uses laser power and scanning speed as input features and part density as the output feature; the second-layer model structure uses laser power, scanning speed, and part density as input features and ultimate tensile strength, yield strength, elongation, and comprehensive mechanical property as output features.

[0053] Figure 2The principle structure diagram of the random forest regression model is shown. Its structural principle is to construct multiple decision tree models, obtain the prediction results respectively, and then integrate and take the average to obtain the final prediction result. First, randomly sample from the original training dataset by the method of sampling with replacement (Bootstrap) to generate multiple sub-datasets, and each sub-dataset is used to train a decision tree; for each decision tree, randomly select some features when splitting nodes, and find the optimal splitting point based on the randomly selected features, thereby introducing the randomness of feature selection to increase the diversity of the model. For the test data, each decision tree in the random forest makes predictions independently and outputs a prediction value, and the prediction results of all decision trees are averaged to obtain the final regression prediction value. By integrating multiple decision trees, the random forest can effectively reduce the risk of overfitting of a single decision tree, thereby improving the generalization ability of the model. In addition, the random forest also uses the "out-of-bag data" (Out of Bag, OOB) to verify the performance of the model, further improving the robustness of the model. Generally speaking, the random forest regression algorithm, through randomizing samples and features and combining the integrated predictions of multiple decision trees, has strong generalization ability, high prediction accuracy and good stability, and at the same time retains strong interpretability of the model, which is an effective method to solve complex regression problems.

[0054] S300. Use the data in the training set divided from the training dataset to train the prediction model. Adopt the uncertainty sampling method of active learning, and select the points with the highest prediction error in the test set to be added to the training set in the next iteration to reduce the prediction error rate of the model for the un-trained area. The first-layer model uses the laser power and scanning speed as input features to predict the density of the formed part. The second-layer model combines the density of the formed part output by the first-layer model, as well as the laser power and scanning speed as input features to predict four output features such as tensile strength, yield strength, elongation rate and comprehensive mechanical properties.

[0055] Use the test set for prediction to obtain the predicted values of the density of the formed part and the four output features. Among them, the predicted value of the density of the formed part output by the first-layer model, and the predicted values of tensile strength, yield strength, elongation rate and comprehensive mechanical properties output by the second-layer model; compare each predicted value with the corresponding true value to evaluate the performance of the model.

[0056] Quantify the prediction ability of the prediction model through evaluation metrics. The evaluation metrics used include root mean square error and mean absolute percentage error, and calculate the prediction performance of the prediction model for the training dataset respectively; at the same time, compare the model predicted values with the true values to ensure that the prediction results meet the engineering accuracy requirements. Among them, the root mean square error RMSE is expressed as: The mean absolute percentage error MAPE is expressed as: In the formula, y irepresents the true value, represents the predicted value, and n is the number of samples.

[0057] The prediction results and performance of the prediction model are displayed through visualization means, specifically including: drawing a correlation heat map to show the correlation between input features and output features; drawing a comparison graph of predicted values and true values to visually show the prediction ability of the prediction model; based on the prediction results of laser power, scanning speed, and formed part density, generating a three-dimensional prediction surface graph to show the change of the tensile property index of the formed part with the change of input features.

[0058] S400. Define the prediction function based on the trained prediction model, input the laser power, scanning speed, and formed part density of a new sample, and output the predicted tensile strength, yield strength, elongation, and comprehensive mechanical properties to provide a prediction basis for the forming quality of the new sample.

[0059] S500. Apply the defined prediction function to the new sample. By inputting the process parameters of the new sample including laser power and scanning speed, combined with the predicted value of the formed part density, predict the tensile property index of the formed part. For example, the process parameters of three groups of new samples are configured in the experiment as follows: (1) laser power P = 300W, scanning speed v = 800mm / s; (2) laser power P = 390W, scanning speed v = 1000mm / s; (3) laser power P = 200W, scanning speed v = 1200mm / s.

[0060] Verify the accuracy of the prediction results through experiments, provide guidance for the optimization of the mechanical properties of the new sample, and finally realize the prediction function of the mechanical properties of the new sample based on the SLM technology.

[0061] As Figures 3 to 7 shown is the prediction result of the prediction model. The visualization of the prediction surface on the X and Y axes based on the laser output power and scanning speed is shown in the figure. The scatter points represent the true experimental values in the dataset. The experimental values basically coincide with the prediction surface, indicating that the model has a good prediction effect and can fully support the influence of input indicators on the output quantity. The five figures respectively show the influence of two input variables, laser power and scanning speed, on the density, yield strength, tensile strength, elongation, and comprehensive mechanical properties of the metal material. The density prediction surface has a good coincidence with the scatter points, indicating that the model has a good prediction effect on density and is more sensitive to the change of laser power, showing an obvious upward trend with the increase of laser power. The yield strength and tensile strength are more sensitive to the changes of laser power and scanning speed, showing peak characteristics. The elongation is relatively smooth with the change of process parameters, but shows good plasticity in the low parameter region, which may be related to the density of the material.

[0062] As Figure 8The figure shows a graph of the prediction error during the model optimization process changing with the number of iterations. By using the uncertainty sampling method to optimize the random forest regression model, data points with larger model prediction errors (i.e., samples that the model cannot accurately predict currently) are preferentially selected for annotation and added to the training set to continuously update the model. Through this method, the areas with poor model predictions can be focused on to optimize the overall performance. It can be seen that as the model is continuously iterated, the prediction error for the output target generally shows a downward trend.

[0063] As Figures 9 to 13 The figure shows a comparison graph of the predicted values and the true values of the dataset, which shows the differences between the true values and the predicted values, and intuitively demonstrates the prediction ability of the model. The dashed line represents the perfect prediction area. The closer the scatter points are to the dashed line, the stronger the prediction ability and the smaller the error. When deviating from the dashed line, it indicates that the difference between the predicted value and the true value is too large and the prediction is inaccurate. It can be seen from the figure that all the scatter points in the dataset, including the test set, are close to the dashed line area, indicating good prediction ability.

[0064] As Figure 14 It is a heatmap of Pearson correlation coefficients. All 7 features are applicable to model construction. The closer the numbers in the figure are to 1 and the darker the color, the higher the correlation between the two features. A positive value indicates a positive correlation, and vice versa. Understanding the correlations between features helps to understand the relationship between input and output and simplifies the prediction model. It can be seen from the figure that there is a strong positive correlation between density and power, and strong negative correlations between power (Laser Power) and yield strength (Yield Strength) and ultimate tensile strength (UTS). When the power increases, both the yield strength and the ultimate tensile strength will decrease; the strong negative correlations between density (Density) and yield strength (Yield Strength) and ultimate tensile strength (UTS) indicate that when the density increases, both the yield strength and the ultimate tensile strength will decrease.

[0065] As Figures 15 to 19 The figure shows a comparison graph of the prediction results of the prediction model for new samples and the experimental verification. The trained model is used to predict three groups of unprocessed parameters. The density predicted by the first-layer density prediction model is combined with the laser power and scanning speed to predict the UTS, Yield Strength, Elongation, and UTS*Elongation values of the new samples. Then, these unprocessed parameters are printed by SLM to test the true values, and the results shown in the figure are obtained. It can be seen that for various mechanical properties of the new samples, the prediction accuracy of the model is high and the prediction error is small, verifying the usability of the model.

[0066] The present invention constructs a double-layer multi-output prediction model based on the random forest regression method, which can predict the mechanical properties of SLM materials by only inputting the laser power and scanning speed in the selective laser melting (SLM) process. It effectively overcomes the limitations of traditional physical modeling and single-output prediction methods, improves the prediction accuracy and adaptability. According to the experimental results, the prediction error of the yield strength of the final model is 5.88%, the prediction error of the tensile strength is 1.17%, the prediction error of the elongation is 2.93%, and the prediction error of the comprehensive mechanical properties is 3.99%. The experimental results show that the present invention performs excellently in terms of prediction accuracy and generalization ability, and can provide effective support for the process optimization of SLM technology. It should be particularly noted that the present invention is not limited to the AlCoCrFeNi2.1 high-entropy alloy processed by SLM shown in the embodiments, and can also be applied to other types of alloy materials and different processing methods.

[0067] The above embodiments are only the preferred embodiments of the present invention, and do not limit the protection scope of the present invention. Any changes made by adopting the design principle of the present invention and non-creative labor on this basis shall fall within the protection scope of the present invention.

Claims

1. A method for predicting mechanical properties of parts formed by laser selective melting technology, characterized in that: The following steps are involved: S100, based on the selective laser melting technology, high entropy alloy powder material is used to print molded parts with different process parameters, the density of the molded parts is measured and the molded parts are subjected to tensile tests. The tensile strength, yield strength, elongation and comprehensive mechanical properties obtained from the tensile test are used as tensile performance indicators of the molded parts to establish a training data set, where the process parameters include laser power and scanning speed, and the comprehensive mechanical properties are obtained by multiplying the tensile strength and elongation; S200, using random forest regression method to build a prediction model, selecting laser power and scanning speed as input features, introducing molded part density as an intermediate variable, and taking tensile strength, yield strength, elongation and comprehensive mechanical properties as output features; S300, training the prediction model using the training data set established by the data obtained in step S100, and selecting the point with the highest prediction error to be added to the next iteration when optimizing the model, so as to reduce the prediction error rate of the model for the untrained area; S400, defining a prediction function based on the trained prediction model, inputting the laser power, scanning speed and molded part density of the new sample, outputting the predicted tensile strength, yield strength, elongation and comprehensive mechanical properties, and providing a prediction basis for the molding quality of the new sample; S500, applying the defined prediction function to the new sample, predicting the tensile performance index of the molded part by inputting the process parameters of the new sample including the laser power and the scanning speed, combined with the predicted value of the molded part density.

2. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 1 is characterized in that: In the step S100, 30 groups of parameters of molded parts based on the selective laser melting technology are obtained, with 3 parameter samples in each group, for a total of 90 groups of samples, and corresponding experimental data are collected, including process parameters, molded part density and molded part tensile performance indicators; these experimental data are organized into a training data set for machine learning, wherein the input features include laser power, scanning speed and molded part density, and the output features include tensile strength UTS, yield strength Yield Strength, elongation Elongation and comprehensive mechanical properties UTS*Elongation; and the training data set is loaded by writing a program to ensure data integrity and correctness.

3. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 2 is characterized in that: In step S100, the loaded training data set is also preprocessed, including: checking data integrity, filling missing values ​​or removing abnormal data; standardizing input features to ensure the scale consistency of features; dividing the training set and the test set, wherein the data points with a scanning speed of not less than 1000 mm / s are used as the training set.

4. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 1 is characterized in that: When the random forest regression method is used to construct the prediction model in step S200, according to the characteristics of the random forest regression model, a global optimization algorithm is used to search for the best parameter combination in the random forest regression model, including setting the number of decision trees, the maximum tree depth and the random seed value, and finding the best parameter combination with the smallest root mean square error within this range.

5. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 4 is characterized in that: The prediction model in step S200 includes a two-layer model structure. The first layer model structure uses laser power and scanning speed as input features, and uses molded part density as output features; the second layer model structure uses laser power, scanning speed and molded part density as input features, and uses tensile strength, yield strength, elongation and comprehensive mechanical properties as output features.

6. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 1 is characterized in that: In the step S300, an active learning uncertainty sampling method is used during training, and the prediction model is trained using the data in the training set divided by the training data set. The first layer model uses laser power and scanning speed as input features to predict the density of the molded part. The second layer model combines the density of the molded part output by the first layer model, as well as laser power and scanning speed as input features, to predict the tensile strength, yield strength, elongation and comprehensive mechanical properties. When optimizing the model, the point with the highest prediction error is selected in the test set divided by the training data set and added to the training set in the next iteration.

7. The method for predicting mechanical properties of a part formed by selective laser melting technology according to claim 6, characterized in that: In the step S300, after the prediction model is trained, a test set divided from the training data set is used for prediction, the predicted value of the output density of the molded part is obtained from the first layer model, and the predicted values ​​of the output tensile strength, yield strength, elongation and comprehensive mechanical properties are obtained from the second layer model; each predicted value is compared with the corresponding true value to evaluate the model performance.

8. The method for predicting mechanical properties of parts formed by selective laser melting technology according to claim 7, characterized in that: In step S300, the prediction ability of the prediction model is also quantified by evaluation indicators, including root mean square error and mean absolute percentage error, which respectively calculate the prediction performance of the prediction model for the training data set; at the same time, the model prediction value is compared with the true value to ensure that the prediction result meets the engineering accuracy requirements.

9. The method for predicting mechanical properties of a part formed by selective laser melting technology according to claim 8, characterized in that: Among the evaluation indicators, the root mean square error RMSE is expressed as: The mean absolute percentage error MAPE is expressed as: In the formula, y i represents the true value, represents the predicted value, and n is the number of samples.

10. The method for predicting mechanical properties of a part formed by selective laser melting technology according to claim 8, characterized in that: The step S300 also includes displaying the prediction results and performance of the prediction model by means of visualization, specifically including: drawing a correlation heat map to display the correlation between input features and output features; drawing a comparison chart of predicted values ​​and true values ​​to intuitively display the prediction ability of the prediction model; based on the prediction results of laser power, scanning speed and molded part density, generating a three-dimensional prediction surface map to display the changes in the tensile performance indicators of the molded parts with the input features.

Images