A method and system for predicting SLM forming performance and optimizing process parameters

Abstract

The present invention relates to the field of SLM forming, and particularly relates to a method and system for predicting SLM forming performance and optimizing process parameters. S1. Design a multi-step orthogonal experiment to obtain the process parameters and performance of SLM formed parts; S2. Randomly divide the data set; S3. Construct a Gaussian process regression model and a multiple stepwise regression model respectively. The Gaussian process regression model is constructed through a constant mean function and a rational quadratic covariance function, and the hyperparameters are adjusted by using a block bisection strategy; S4. Use three indicators, namely MAE, RMSE, and R<supgt;2< / supgt;, to evaluate the two models in S2; S5. Construct a Gaussian process regression - multiple stepwise regression prediction model to accurately predict the performance of formed parts from SLM process parameters; S6. Take the prediction model constructed in S5 as the fitness function, and obtain the recommended process parameters through optimization by the teaching-learning-based optimization algorithm; S7. Stepwise screen and verify the recommended process parameters. The present invention can not only accurately predict the performance according to the SLM process parameters, but also efficiently give the recommended process parameters.

Description

Technical Field

[0001] The present invention belongs to the field of selective laser melting (SLM) forming of metals, and more particularly, relates to a method and system for predicting SLM forming performance and optimizing process parameters. Background Art

[0002] The SLM technology is a process in which metal powder is completely melted under the action of high-energy laser, metallurgically welded to the base metal after heat dissipation and solidification, and then a three-dimensional entity is formed by layer-by-layer accumulation. It can directly form metal parts that are nearly fully dense and have good mechanical properties. Compared with traditional metal material forming, SLM forming can directly form parts with complex geometric shapes, and after SLM parts are formed, they can be used with only a small amount of processing or no further processing, and the surface quality is excellent. Therefore, SLM is widely used in the fields of aerospace, military equipment, etc.

[0003] The main process parameters of SLM forming include laser power, powder layer thickness, scanning speed, scanning spacing, etc., and these process parameters have a significant impact on the performance of the formed parts. The main performances of SLM forming include density, tensile strength, yield strength, elongation after fracture, surface roughness, etc. During the manufacturing process of SLM formed parts, not only there are interactions between various process parameters, but also the SLM forming process itself has very complex changes, and the performance of SLM formed parts is difficult to predict under the combined action of various factors.

[0004] Regarding the acquisition of performance data of SLM formed parts, the existing method is to obtain it through a series of performance detection tests after the formed parts are manufactured and cooled. This method takes a long time to obtain the performance of specific SLM process parameters. When it is necessary to obtain the performance corresponding to multiple groups of process parameters, the time and materials consumed will increase exponentially. The process parameters for producing high-performance SLM formed parts are obtained by the experimental method. A large number of process parameter groups are set, and each group is produced by an SLM device one by one, and then the performance of the obtained SLM formed parts is detected one by one. The process parameters corresponding to the SLM formed parts with better performance are recorded as the process parameters for subsequent industrial production. This method for obtaining high-performance SLM forming process parameters is extremely inefficient, consuming a large amount of raw materials, and the time consumed for printing and detection is also very long. For example, using the full factorial experiment method, for the four process parameters of SLM, if 4 experimental points are set for each group of process parameters, then 4 4 , that is, 256 groups of experiments are required; if 10 experimental points are set for each group of process parameters, then 10 4, that is, 10,000 groups of experiments. This approach is inefficient, consuming a large amount of raw materials, and the time spent on printing and performance testing is also very long. Moreover, these quantities cannot ensure that the obtained process parameters are optimal. In short, it is impossible to quickly obtain the performance of the formed parts under known SLM process parameters, and the existing methods for obtaining the process parameters required for high-performance SLM formed parts in industrial production are inefficient.

[0005] Therefore, it is highly necessary to accurately and quickly obtain the performance of the formed parts through other methods under known SLM process parameters, and to efficiently obtain the process parameters for manufacturing high-performance SLM formed parts for industrial production. Summary of the Invention

[0006] Aiming at the defects of the prior art, the purpose of the present invention is to provide a method and system for predicting SLM forming performance and optimizing process parameters, aiming to solve the problems of slow speed and long time-consuming in obtaining the performance of SLM formed parts under known SLM process parameters and the low efficiency in obtaining the process parameters for high-performance SLM formed parts.

[0007] To achieve the above object, in the first aspect, the present invention provides a method for predicting SLM forming performance and optimizing process parameters, including the following steps:

[0008] Design multiple groups of SLM process parameters through multi-step orthogonal experiments, and conduct actual manufacturing on the designed process parameters to determine the performance of the corresponding SLM formed parts. Summarize the designed SLM process parameters and the corresponding formed part performance into a data set;

[0009] Train a Gaussian process regression model based on the data set to obtain a first mapping relationship between SLM process parameters and SLM formed part performance; and train a multiple stepwise regression model based on the data set to obtain a second mapping relationship between SLM process parameters and SLM formed part performance; both the first mapping relationship and the second mapping relationship are used to predict the performance of SLM formed parts;

[0010] Combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model; wherein, the SLM performance prediction model fuses the two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal;

[0011] Optimize the SLM performance prediction model through the teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements of the performance of SLM formed parts;

[0012] According to the influence of SLM process parameters on the service life of the equipment, the multiple sets of recommended SLM process parameters are screened step by step to obtain multiple sets of SLM process parameters with low damage to the equipment service life and stable process parameters.

[0013] In an optional example, the method further includes the following steps:

[0014] For the multiple sets of SLM process parameters after step-by-step screening, actual manufacturing is carried out to verify the performance of the obtained SLM formed parts. The parameters whose performance meets the expected standards are listed in the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM formed parts; the parameters whose performance does not meet the expected standards are listed in the process parameter group with unsatisfactory verification results;

[0015] The process parameter group with unsatisfactory verification results is added to the data set to update the data set, and the Gaussian process regression model training, the multiple stepwise regression model training, the SLM process parameter recommendation and screening process are repeated to obtain multiple sets of newly screened SLM process parameters.

[0016] In an optional example, multiple sets of SLM process parameters are designed through multi-step orthogonal experiments. Specifically:

[0017] Design multi-step orthogonal experiments, design the SLM process parameter groups through orthogonal experiments globally, and carry out actual manufacturing on the designed process parameters to detect the performance of the preliminarily designed SLM formed parts;

[0018] Conduct a preliminary detection of the global performance of SLM, find out the suspicious areas with better performance of the preliminarily detected SLM formed parts, and redesign orthogonal experiments for the suspicious areas. After multiple designs of orthogonal experiments, multiple sets of SLM process parameters corresponding to the suspicious areas with better performance of the SLM formed parts are obtained.

[0019] In an optional example, the Gaussian process regression model is trained based on the data set, and the multiple stepwise regression model is trained based on the data set. Specifically:

[0020] Construct a Gaussian process regression model; train the Gaussian process regression model based on the designed multiple sets of SLM process parameters;

[0021] Construct a multiple stepwise regression model; train the multiple stepwise regression model based on the designed multiple sets of SLM process parameters; among them, considering the mutual influence between process parameters, high-order terms and cross terms are added to construct a complete set of alternative items; after obtaining the complete set of alternative items, through significance tests, alternative items are gradually introduced and eliminated.

[0022] In an alternative example, during the training process of the Gaussian process regression model and the multiple stepwise regression model, the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination R 2 are used to perform weighted evaluation on the two models through three indicators to evaluate the deviation between the model prediction results and the actual results, and to optimize the model parameters; the weights of the three indicators are: MAE:RMSE:R 2 = 40:40:20.

[0023] In an alternative example, the teaching and learning based optimization (TLBO) algorithm is used to optimize the SLM performance prediction model to obtain multiple sets of recommended SLM process parameters that meet the requirements of the SLM part performance. Specifically:

[0024] Set an appropriate number of initial populations, and use the SLM performance prediction model as the fitness function of the optimization algorithm;

[0025] By calculating the individual fitness, select the individual with the maximum fitness value as the teacher. The initial population is updated through the "teaching phase" and "learning phase". After the population update is completed, reselect the teacher individual. Repeat this process until the loop termination condition is reached to obtain the current optimal fitness individual, that is, the current optimal performance value and the corresponding process parameters, and use the process parameters corresponding to the optimal performance value as the recommended SLM process parameters.

[0026] In an alternative example, the SLM process parameters include: laser power, powder layer thickness, scanning speed, and scanning spacing; laser power and scanning speed have relatively greater impacts on the life of the equipment, and laser power has a relatively greater impact on the process stability of the equipment. According to the laser power and scanning speed of the actual equipment, set multiple thresholds for the two process parameters. When the laser power or scanning speed of the actual equipment is in different threshold intervals, the laser power or scanning speed is in different ratings, and different ratings result in different losses to the equipment;

[0027] According to the influence of the SLM process parameters on the service life of the equipment, perform step-by-step screening on the multiple sets of recommended SLM process parameters to obtain multiple sets of SLM process parameters with low damage to the service life of the equipment. Specifically:

[0028] Classify the multiple sets of recommended SLM process parameters into four categories according to: ① laser power level 1, scanning speed level 1; ② laser power level 1, scanning speed level 2; ③ laser power level 2, scanning speed level 1; ④ laser power level 2, scanning speed level 2. Among them, laser power level 1 causes the lowest damage to the service life of the equipment, and as the laser power level increases, the damage to the service life of the equipment increases; scanning speed level 1 causes the lowest damage to the service life of the equipment, and as the scanning speed level increases, the damage to the service life of the equipment increases;

[0029] After step-by-step screening to remove the parameters in multiple groups of recommended SLM process parameters where the laser power and scanning speed do not fall into the above four categories, they are displayed in order from ① to ④ from top to bottom, and multiple groups of process parameters after screening are obtained.

[0030] In a second aspect, the present invention provides an SLM forming performance prediction and process parameter optimization system, including:

[0031] A parameter data set design unit, which is used to design multiple groups of SLM process parameters through multi-step orthogonal experiments, actually manufacture the designed process parameters, determine the performance of the corresponding SLM formed parts, and summarize the designed SLM process parameters and the corresponding formed part performance into a data set;

[0032] A regression model training unit, which is used to train a Gaussian process regression model based on the data set to obtain a first mapping relationship between SLM process parameters and SLM formed part performance; and train a multiple stepwise regression model based on the data set to obtain a second mapping relationship between SLM process parameters and SLM formed part performance; both the first mapping relationship and the second mapping relationship are used to predict the performance of SLM formed parts;

[0033] A regression model combination unit, which is used to combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model; wherein, the SLM performance prediction model fuses the two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal;

[0034] A prediction model optimization unit, which is used to optimize the SLM performance prediction model through a teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements of the SLM formed part performance;

[0035] A process parameter screening unit, which is used to perform step-by-step screening on the multiple groups of recommended SLM process parameters according to the influence of SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment and stable process parameters.

[0036] In an optional example, the system further includes:

[0037] A process parameter grouping unit, which is used to actually manufacture the multiple groups of SLM process parameters after step-by-step screening to verify the performance of the obtained SLM formed parts, list the parameters with performance meeting the expected standard into the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM formed parts; list the parameters with performance not meeting the expected standard into the process parameter group with unsatisfactory verification results;

[0038] A parameter new recommendation screening unit is used to add the process parameter groups with unsatisfactory verification results to the dataset, update the dataset, repeat the Gaussian process regression model training, the multiple stepwise regression model training, the SLM process parameter recommendation and screening process, and obtain multiple groups of SLM process parameters after new screening.

[0039] In an optional example, the SLM process parameters include: laser power, powder layer thickness, scanning speed, and scanning spacing; the laser power and the scanning speed have a relatively large impact on the life of the device. According to the laser power and scanning speed of the actual device, multiple thresholds corresponding to the two process parameters are set. When the laser power or scanning speed of the actual device is in different threshold intervals, the laser power or scanning speed is in different ratings, and different ratings result in different losses to the device.

[0040] The process parameter screening unit classifies multiple groups of recommended SLM process parameters into four categories according to: ① laser power level one, scanning speed level one; ② laser power level one, scanning speed level two; ③ laser power level two, scanning speed level one; ④ laser power level two, scanning speed level two. Among them, laser power level one causes the lowest damage to the service life of the device, and as the laser power level increases, the damage to the service life of the device increases; scanning speed level one causes the lowest damage to the service life of the device, and as the scanning speed level increases, the damage to the service life of the device increases. After step-by-step screening to remove the parameters in multiple groups of recommended SLM process parameters where the laser power and scanning speed do not fall into the above four categories, they are displayed sequentially from top to bottom in the order from ① to ④ to obtain multiple groups of screened process parameters.

[0041] In a third aspect, the present invention provides an electronic device, including: a memory and a processor; the memory is used to store a computer program; the processor is used to implement the method provided in the first aspect above when executing the computer program.

[0042] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method provided in the first aspect above is implemented.

[0043] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects are achieved:

[0044] The present invention provides an SLM forming performance prediction and process parameter optimization method and system, constructs a Gaussian process regression - multiple stepwise regression model, can accurately predict the forming part performance through SLM process parameters; through the teaching - learning - based optimization algorithm to find the optimal solution, efficiently gives recommended process parameters; can provide recommended process parameters for high - performance SLM forming parts without consuming a large amount of material and time costs, which is of great significance to SLM industrial production. Description of the Drawings

[0045] Figure 1 This is the flowchart of the SLM forming performance prediction and process parameter optimization method provided by the embodiments of the present invention.

[0046] Figure 2 This is the flowchart of the SLM performance prediction and process parameter optimization method provided by the embodiments of the present invention.

[0047] Figure 3 This is the flowchart of the combination of the Gaussian process regression - multiple stepwise regression model and the teaching - learning - based optimization algorithm for the SLM performance prediction and process parameter optimization method provided by the embodiments of the present invention.

[0048] Figure 4 This is the architecture diagram of the SLM forming performance prediction and process parameter optimization system provided by the embodiments of the present invention. Detailed implementation manners

[0049] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

[0050] In the description of the present invention, the descriptions of reference terms such as "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0051] The present invention obtains SLM process parameters and performance data through designed experiments, constructs an SLM performance prediction model, accurately and quickly obtains the performance of formed parts through SLM process parameters, and searches for the optimal solution through an optimization algorithm to efficiently obtain recommended process parameters for high - performance SLM formed parts and apply them to industrial production.

[0052] Figure 1 This is the flowchart of the SLM forming performance prediction and process parameter optimization method provided by the embodiments of the present invention. As Figure 1 shown, it includes the following steps:

[0053] S101, design multiple groups of SLM process parameters through multi - step orthogonal experiments, perform actual manufacturing on the designed process parameters, determine the performance of the corresponding SLM formed parts, and summarize the designed SLM process parameters and the corresponding formed part performance as a data set;

[0054] S102. Train a Gaussian process regression model based on the dataset to obtain a first mapping relationship between SLM process parameters and SLM part performance; and train a multiple stepwise regression model based on the dataset to obtain a second mapping relationship between SLM process parameters and SLM part performance. Both the first mapping relationship and the second mapping relationship are used to predict the performance of SLM parts.

[0055] S103. Combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model. Among them, the SLM performance prediction model fuses the two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal.

[0056] S104. Optimize the SLM performance prediction model through a teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements for the performance of SLM parts.

[0057] S105. Stepwise screen the multiple groups of recommended SLM process parameters according to the influence of SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment and stable process parameters.

[0058] In an optional example, the method further includes the following steps:

[0059] For the multiple groups of SLM process parameters after stepwise screening, conduct actual manufacturing to verify the performance of the obtained SLM parts. List the parameters with performance meeting the expected standards in the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM parts; list the parameters with performance not meeting the expected standards in the process parameter group with unsatisfactory verification results.

[0060] Add the process parameter group with unsatisfactory verification results to the dataset, update the dataset, and repeat the processes of training the Gaussian process regression model, training the multiple stepwise regression model, recommending SLM process parameters, and screening to obtain multiple groups of newly screened SLM process parameters.

[0061] In a specific embodiment, the object of the present invention is achieved through the following technical solutions:

[0062] Obtain the performance of the formed parts corresponding to the process parameters through the design of multi-step orthogonal experiments. Design the main process parameters of SLM: laser power, powder layer thickness, scanning speed, and scanning spacing. Detect and obtain the performance of the formed parts: relative density, tensile strength, yield strength, elongation after fracture, surface roughness, etc. Respectively construct a Gaussian process regression model and a multiple stepwise regression model. Both models can predict the performance through the SLM process parameters. Then fuse the two models to construct a Gaussian process regression-multiple stepwise regression model, and predict the performance value of the SLM formed part from the SLM process parameters, realizing the accurate and rapid prediction of the performance of the SLM formed part from the SLM process parameters. And fuse the obtained Gaussian process regression-multiple stepwise regression model with the teaching and learning optimization algorithm. Take the obtained SLM performance prediction model as the fitness function of the teaching and learning algorithm. Through the optimization of the teaching and learning algorithm, obtain the individual with the optimal fitness value, that is, the optimal performance of the SLM formed part and the corresponding process parameters, and take the optimal performance and the corresponding process parameters as the recommended process parameters. Multiple runs can obtain multiple groups of recommended process parameters. Screen the multiple groups of recommended process parameters step by step according to process parameters such as laser power and scanning speed to obtain the screened recommended process parameters. Finally, conduct actual verification on the obtained screened process parameters.

[0063] Specifically, the SLM performance prediction and process parameter optimization method of the present invention is as Figure 2 shown, and includes the following steps:

[0064] S1. Design multi-step orthogonal experiments to obtain SLM forming process parameters and performance data, and divide the data set. The traditional orthogonal experiment design can be completed in one step. Although the orthogonal experiment designed in this way can cover the whole range, it has no focus. For the problem of obtaining the process parameters of high-performance SLM formed parts, more attention needs to be paid to the high-performance forming area of SLM than other areas, while the traditional orthogonal experiment cannot meet this requirement. Therefore, the present invention adopts multi-step orthogonal experiments for design. Design multi-step orthogonal experiments. First, design the process parameter groups through orthogonal experiments within the global range, and actually manufacture the designed process parameters, and detect the performance of the preliminarily designed SLM formed parts. Through the preliminary detection of the global performance of SLM, find out the suspicious areas with better preliminary detected performance, and design orthogonal experiments for these suspicious areas with better performance to obtain more information about the suspicious areas with better performance, providing more support for the subsequent SLM formed part performance prediction model to accurately predict high-performance formed parts. Finally, obtain a data set that covers the whole range and has reasonable density.

[0065] S2. Randomly divide the data set. Randomly divide the data into a training set and a prediction set for subsequent training and testing of the Gaussian process regression model and the multiple stepwise regression model.

[0066] S3. Respectively construct a Gaussian process regression model and a multiple stepwise regression model, asFigure 3 As shown below:

[0067] ① Construct a Gaussian process regression model: Gaussian process regression, as a type of machine learning, can achieve accurate prediction without a large amount of data compared to the common BP neural network. Dozens of groups of experimental data can be designed for SLM forming performance prediction. Select appropriate mean functions and covariance functions, and assign initial values to the hyperparameters. Within the normal SLM forming range, the performance values of SLM formed parts are generally stable. It is appropriate to use a constant mean function for the Gaussian process regression model predicting SLM performance; while the rational quadratic covariance function has strong generalization performance and better prediction ability for unknown data, so it is appropriate to use the rational quadratic covariance function for the Gaussian process regression model predicting SLM performance.

[0068] The form of the constant mean function is as follows:

[0069] m(x) = C, where C is a constant

[0070] The form of the rational quadratic covariance function is as follows: Let x and x′ be different input vectors, let r = ||x - x′||, and the expression of the rational quadratic kernel function is:

[0071] where α > 0, l > 0

[0072] Its hyperparameters are the mixing parameter α and the length scale parameter l, which are required to be positive numbers. Among them, α is mainly used to control the decay rate of this kernel function. The constructed Gaussian process regression model is f(x) ∼ GP(m(x), k(x)). GP represents the Gaussian process. Among them, the value of the constant mean function is the average of the performance values of the training set data, and the rational quadratic covariance function needs to adjust the hyperparameters. When constructing the Gaussian process regression model and adjusting the initial values of the hyperparameters, the block binary search strategy is adopted. For example, divide the hyperparameter α from 0 to 10 into 10 equally spaced regions, substitute the initial values of the hyperparameters at the nodes of each part into the model, and calculate the root mean square error for comparison. The hyperparameter model with a smaller root mean square error has better prediction ability, and intuitively grasp the influence of the overall hyperparameters on the prediction ability of the model. Then select an appropriate block region, and then adopt the binary search strategy. Each time the experiment takes the midpoint of the two adjacent nodes with the largest spacing as the hyperparameter test point, gradually refine the region, and conduct multiple experiments until the initial values of the hyperparameters with satisfactory scores are obtained. The scoring is carried out in the manner specified in S4. Before conducting the experiment using the binary search strategy, the overall change trend can be known through blocking, and some inappropriate regions can be excluded, reducing the binary search strategy experiment region and reducing the experimental workload.

[0073] ②Construct a multiple stepwise regression model: The stepwise regression method is used to sequentially select an independent variable with the most significant variance contribution from the candidate independent variables and add it to the regression model. When introducing a new variable, each of the previously introduced independent variables is tested one by one, and the insignificant ones are removed until no new independent variable can be introduced into the regression equation and no independent variable can be removed from the regression equation. Considering the mutual influence among the process parameters of SLM forming, in addition to the laser power, powder layer thickness, scanning speed, and scanning spacing, the candidate independent variables should also include all the terms formed by the permutation and combination of the four process parameters, and the order of the highest term is determined according to requirements. The variable selection rule needs to be set manually. For example, it is set that when the significance P value of the variable is less than 0.05, it is introduced into the regression equation; when the significance P value of the variable is greater than 0.10, it is removed from the regression equation. The calculation process of the multiple stepwise regression model is complex and can be solved with the help of statistical analysis software. For example, the SPSS software is used to solve the multiple stepwise regression model for the training set data. For the constructed multiple stepwise regression model, the scoring rule in S4 is used to evaluate the model. By adjusting the variable selection rule of the model, the multiple stepwise regression model is changed, optimized, and the prediction ability is improved.

[0074] S4. The test and evaluation system for the Gaussian process regression model and the multiple stepwise regression model. The prediction effects of the Gaussian process regression model and the multiple stepwise regression model are both evaluated by predicting the test set data and calculating the weighted evaluation of the three indicators of the mean absolute error MAE, root mean square error RMSE, and determination coefficient R. 2 The mean absolute error MAE represents the average of the absolute errors between the predicted values and the observed values. The smaller the MAE, the smaller the gap between the overall prediction and the actual value.

[0075] The mean square error MSE and the root mean square error RMSE are a measure reflecting the degree of difference between the estimator and the estimated quantity. The smaller the MSE, the smaller the difference between the predicted data and the actual data. The root mean square error RMSE is obtained by taking the square root of the mean square error. The determination coefficient R 2 reflects the fitting degree of the regression model to the actual value. R 2 is between 0 and 1. When R 2 = 0, it means that the model does not fit the data at all. When R 2 = 1, it means that the model fits the data perfectly. The larger the value of R 2 , the better the fitting degree of the model.

[0076] The above three evaluation data have their own focus. A single evaluation can only evaluate the prediction results from one aspect and cannot provide a more comprehensive evaluation of the prediction results. In order to make full use of the three evaluation indicators, the present invention sets the following evaluation criteria: if MAE is less than 0.01, the score is 100 points; if MAE is less than 0.02, the score is 90 points, and so on; if RMSE is less than 0.01, the score is 100 points; if RMSE is less than 0.02, the score is 90 points, and so on; if R 2 If the score is greater than 0.95, the indicator score is 100 points. 2 If the value is greater than 0.85, the indicator is scored as 90 points, and the score is evenly recursive. Because the forming process and data acquisition are affected by external factors and have fluctuation interference, slight deviations between the actual value and the predicted value of the performance data are allowed. Therefore, each evaluation indicator is scored as 100 points when the error is extremely small.

[0077] After the model prediction is completed, the actual focus is on the deviation between the model's prediction results and the actual results. Therefore, among the three evaluation indicators, MAE and RMSE need to be emphasized. When the three indicators are weighted to calculate the final score, the weight of each is: MAE: RMSE: R 2 =40:40:20, and the final score is calculated based on this. A final score of 90 points or above is considered accurate; a final score of 80-90 points is considered excellent; a final score of 70-80 points is considered effective; a final score of 60-70 points is considered poor; a final score below 60 points or a score below 60 points for any individual item is considered unpredictable.

[0078] S5. Construct a Gaussian process regression - multiple stepwise regression model. The Gaussian process regression model constructed in S3 and the multiple stepwise regression model are fused by weighting. For a set of input process parameters, the Gaussian process regression model and the multiple stepwise regression model respectively perform performance predictions on them, and the two obtained performance prediction values are fused by weighting. Appropriate weights are assigned to the prediction results of the two models, and the result obtained by weighted calculation is used as the prediction value of the Gaussian process regression - multiple stepwise regression model. Both the Gaussian process regression model and the multiple stepwise regression model can realize the prediction of performance from SLM process parameters, but each has its own advantages and disadvantages. The Gaussian process regression model is very accurate in predicting data "close" to the known points, but the prediction accuracy drops significantly when predicting "far from" the known points, that is, it is very accurate in predicting local data points, but there are some deficiencies in predicting the global region; while the multiple stepwise regression model is not as accurate as the Gaussian process regression model in predicting the local region, but the multiple stepwise regression model has stable prediction ability in the global range, and its prediction accuracy in the global range is also relatively good. The Gaussian process regression model and the multiple stepwise regression model are fused by weighting. In the region "close" to the known points, the prediction result of the multiple stepwise regression model is corrected by the prediction result of the Gaussian process regression model, and in the region "far from" the known points, the prediction result of the Gaussian process regression model is corrected by the prediction result of the multiple stepwise regression model, to obtain a Gaussian process regression - multiple stepwise regression model with more prominent prediction ability in the global range.

[0079] In the scoring system of S4, when the scores of both the Gaussian process regression model and the multiple stepwise regression model reach 90 points, it is considered that the two models can be fused to construct a Gaussian process regression - multiple stepwise regression model. The advantages and disadvantages of the weighted weights of the two separate models in the constructed Gaussian process regression - multiple stepwise regression model are evaluated by the score of the prediction test set data of the Gaussian process regression - multiple stepwise regression model, and the scoring method adopts the evaluation system in S4. The weighted weights are required to be accurate to 0.01. By using the traversal method and looping 99 times through the program, the optimal weight structure can be accurately solved. At this time, a Gaussian process regression - multiple stepwise regression model with more stable and accurate SLM performance prediction is obtained. The Gaussian process regression - multiple stepwise regression model constructed by fusing the two separate models absorbs the respective advantages of the two models. The prediction ability of the fused model in the global range is better than that of the separate Gaussian process regression model and multiple stepwise regression model, and it can realize more accurate prediction of the performance of SLM formed parts from SLM process parameters in the global range.

[0080] S6. Use the teaching and learning algorithm to optimize and obtain the recommended process parameters for the SLM process. By optimizing with the teaching and learning algorithm, the individual with the optimal fitness value can be obtained, and thus the process parameters with the optimal performance value can be obtained. Compared with traditional optimization algorithms such as genetic algorithms and particle swarm algorithms, the teaching and learning algorithm has a simple principle, is easy to implement, requires very few parameters to be tuned, and has a higher computational efficiency than traditional methods, making it suitable for optimizing the above SLM performance prediction model. First, set an appropriate number of initial populations, and use the Gaussian process regression-multiple stepwise regression prediction model obtained in S4 as the fitness function of the optimization algorithm. By calculating the individual fitness, select the individual with the largest fitness value as the teacher. The initial population is updated through the "teaching stage" and the "learning stage". After the population update is completed, reselect the teacher individual, and so on in a loop until the loop termination condition is reached, obtaining the current optimal fitness individual, that is, the current optimal performance value and the corresponding process parameters, and use the process parameters corresponding to the optimal performance value as the recommended process parameters.

[0081] S7. Stepwise screen and verify the SLM recommended process parameters. During the SLM forming process, if the process parameters are not selected appropriately, it is easy to cause damage to the equipment and reduce the service life of the equipment. The recommended process parameters do not guarantee that all process parameters are appropriate, and the process parameters need to be screened. Randomly divide the data set, run the Gaussian process regression-multiple stepwise regression model for SLM performance prediction and the teaching and learning algorithm for process parameter optimization 10 times, and 10 groups of different recommended process parameters can be obtained, and screen the 10 groups of process parameters. Among the four process parameters of laser power, powder layer thickness, scanning speed, and scanning spacing, laser power and scanning speed have a great impact on the life of the equipment. Specifically, the impact of laser power on the service life of the equipment is greater than that of scanning speed. It is necessary to set the thresholds of the two process parameters according to the laser power and scanning speed of the actual equipment.

[0082] The present invention sets the laser power at 40% - 60% of the maximum power of the laser device as level one; at 20% - 40% or 60% - 80% of the maximum power as level two; and at less than 20% or greater than 80% of the maximum power as level three. The lower the rating of the laser power, the smaller the damage to the equipment by the process parameters. The rating of the scanning speed is the same as that of the laser power, both being evaluated according to the percentage of the device maximum value, and the lower the rating, the smaller the damage. For the 10 groups of recommended process parameters with respect to laser power and scanning speed, first classify them according to the rating standard of laser power, and eliminate the process parameter groups with the laser power rated as level three. Then, further subdivide the process parameter groups with laser power at level one and level two according to the rating of the scanning speed, and eliminate the process parameter groups with the scanning speed rated as level three. Finally, classify them into four categories: ① laser power at level one, scanning speed at level one; ② laser power at level one, scanning speed at level two; ③ laser power at level two, scanning speed at level one; ④ laser power at level two, scanning speed at level two. After partial elimination of the 10 groups of process parameters through step-by-step screening, display them in order from ① to ④ from top to bottom to obtain the recommended process parameters after screening. The recommended process parameters after screening are then actually verified. Input the screened process parameters into the equipment for printing. On the one hand, it can verify the predicted performance, and on the other hand, it can expand the data set. For the process parameter groups after step-by-step screening, the process parameter groups with satisfactory verification results are the process parameter groups for industrial production of high-performance SLM formed parts and can be included in the industrial production list of high-performance SLM formed parts; for the process parameter groups with unsatisfactory verification, add the data of the process parameter groups with unsatisfactory verification to the data set, update the data set, and repeat steps S2 to S7 to obtain new recommended process parameters.

[0083] Further, before an individual calculates the fitness according to the Gaussian process regression - multiple stepwise regression model in S5, it is necessary to consider the accuracy of the process parameters that can be input into the equipment, such as the accuracy of the laser power, the accuracy of the scanning speed, etc. After the update of the individual is completed, the input of each individual needs to be converted into the corresponding process parameter value that can be input into the equipment according to the rounding method and then used as the input of the individual for subsequent operations.

[0084] Example 1

[0085] The SLM performance prediction and process parameter optimization method of this embodiment obtains process parameters and performance data through designed experiments, randomly divides the obtained data into a training set and a test set, constructs a Gaussian process regression model and a multiple stepwise regression model, and adjusts the initial values of the hyperparameters so that the two constructed models can accurately predict the performance parameters of the test set. Then, by fusing the above two models, a Gaussian process regression-multiple stepwise regression model is constructed, which is fused with the teaching and learning algorithm. The obtained fusion model is used as the fitness function, and through the optimization of the teaching and learning algorithm, the optimal performance and the process parameters under the optimal performance are obtained, that is, the process parameters are optimized. In this example, the input process parameters are laser power, powder layer thickness, scanning speed, and scanning spacing, and the output is density. The density, a single performance, is predicted. If other performance parameters need to be predicted, the operation is the same as predicting the density, and only the hyperparameters need to be readjusted and optimized.

[0086] First, data is obtained through designed experiments, and a multi-step orthogonal experiment is designed. First, an orthogonal experiment is designed globally, and the SLM formed parts with the initially designed process parameters are actually manufactured, and their density is detected; then, according to the density results of the SLM formed parts obtained from the initial design, an orthogonal experiment is redesigned for the suspicious areas with better detected performance, and the density is manufactured and detected again to obtain an experimental group data set that covers the whole and is reasonably dense and sparse.

[0087] In this embodiment, 60 groups of data are taken, and 50 groups are randomly selected as the training set data, and 10 groups are used as the test set data.

[0088] A Gaussian process regression model is constructed and trained with the training set data. A Gaussian process regression model is constructed using a constant mean function and a rational quadratic covariance function. The prediction ability of the Gaussian process regression model is evaluated by the results of the MAE, RMSE, and R 2 of the predicted values and the actual values of the density of the test set data. The specific value of the constant mean function is determined by the average value of the density of the training set data; for the determination of the hyperparameters in the rational quadratic covariance function, the hyperparameters α and l are adjusted through the block bisection strategy, continuously improving the score of the Gaussian process regression model, and raising the prediction result score of the model for the test set data to more than 90 points. Finally, an accurate SLM density prediction model is obtained. The calculation part of the Gaussian process regression model is solved by a computer through existing programs, but the part of the hyperparameter experiment and modification requires manual operation.

[0089] Construct a multiple stepwise regression model. To further improve the prediction accuracy of the multiple stepwise regression model, a multiple third-order stepwise regression model is constructed in this embodiment. Based on the existing four process parameters, the second-order square terms, second-order cross terms, third-order cubic terms, and third-order cross terms of each item are calculated, and a complete set of candidate independent variables including all first-order terms, second-order terms, and third-order terms is constructed. Subsequently, using the stepwise regression method, when introducing a new variable, a significance test is performed on each of the already introduced independent variables, and the insignificant ones are removed until no new independent variable can be introduced into the regression equation and no independent variable can be removed from the regression equation, thus completing the construction of the multiple third-order stepwise regression model. The prediction ability of the multiple third-order stepwise regression model is also evaluated by the results of the three indicators of MAE, RMSE, and R 2 for the predicted values and actual values of the density of the test set data. The multiple third-order stepwise regression model optimizes the model by adjusting the variable selection rules, so that the score of the predicted results of the multiple third-order stepwise regression model for the test set reaches more than 90 points. The calculation and stepwise regression analysis of the multiple third-order stepwise regression model are completed using SPSS statistical analysis software.

[0090] Construct a Gaussian process regression - multiple third-order stepwise regression model. The already constructed Gaussian process regression model and the multiple third-order stepwise regression model are fused by weighting. Through 100 times of program loop traversal, the weighted weight structure with an accuracy of 0.01 is accurately solved to obtain a Gaussian process regression - multiple third-order stepwise regression model with better and more stable prediction ability.

[0091] Integrate the Gaussian process regression - multivariate third - order stepwise regression model with the teaching - learning - based optimization algorithm, and optimize through the teaching - learning - based optimization algorithm. Set the initial population size to 100, loop 200 times, initialize the population, that is, assign initial values to the four inputs of each individual. Take the obtained Gaussian process regression - multivariate third - order stepwise regression model as the fitness function, input the process parameter group, and use the predicted density value output as the fitness value. Evaluate the quality of individuals based on the magnitude of the fitness value. The population continuously undergoes the "teaching" stage and the "learning" stage, and is continuously updated until the loop end condition is met, completing the optimization process of the teaching - learning - based optimization algorithm, obtaining the individual with the maximum fitness value, and then obtaining the process parameters under the optimal performance and the optimal performance. And take the process parameters with the optimal performance as the recommended process parameters. Run 10 times repeatedly, and 10 groups of different recommended process parameters will be obtained. For the 10 groups of recommended process parameters, by screening the laser power and scanning speed step by step, and grading the process parameter group according to the actual maximum parameters of the equipment, eliminate the inappropriate recommended process parameters to prevent damage to the equipment caused by inappropriate process parameters. The screened process parameter groups are divided into four categories: ① Laser power level 1, scanning speed level 1; ② Laser power level 1, scanning speed level 2; ③ Laser power level 2, scanning speed level 1; ④ Laser power level 2, scanning speed level 2. The recommended degrees of the four - category recommended process parameters gradually decrease from ① to ④, and the four - category process parameters are arranged from top to bottom according to the rule, as the screened process parameters.

[0092] Finally, conduct actual verification on the screened process parameters. The screened recommended process parameters are further verified actually. Input the screened process parameters into the equipment for printing and test their performance for verification. For the process parameter groups after step - by - step screening, the process parameter groups with satisfactory verification results are the process parameter groups for industrial production of high - performance SLM formed parts, and they are included in the industrial production list of high - performance SLM formed parts; for the process parameter groups with unsatisfactory verification results, save the data and add it to the data set to expand the data set. Repeat the above steps to obtain new recommended process parameters.

[0093] Figure 4 It is the system architecture diagram of the SLM forming performance prediction and process parameter optimization provided by the embodiment of the present invention, as Figure 4 shown, including:

[0094] The parameter data set design unit 410 is used to design multiple groups of SLM process parameters through multi - step orthogonal experiments, actually manufacture the designed process parameters, determine the performance of the corresponding SLM formed parts, and summarize the designed SLM process parameters and the corresponding formed part performance into a data set.

[0095] A regression model training unit 420 is configured to train a Gaussian process regression model based on the data set to obtain a first mapping relationship between SLM process parameters and the performance of SLM formed parts; and train a multiple stepwise regression model based on the data set to obtain a second mapping relationship between SLM process parameters and the performance of SLM formed parts; both the first mapping relationship and the second mapping relationship are used to predict the performance of SLM formed parts.

[0096] A regression model combination unit 430 is configured to combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model; wherein, the SLM performance prediction model fuses two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal.

[0097] A prediction model optimization unit 440 is configured to optimize the SLM performance prediction model through a teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements for the performance of SLM formed parts.

[0098] A process parameter screening unit 450 is configured to stepwise screen the multiple groups of recommended SLM process parameters according to the influence of SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment and stable process parameters.

[0099] A process parameter grouping unit 460 is configured to verify the performance of the obtained SLM formed parts through actual manufacturing for the multiple groups of SLM process parameters after stepwise screening, list the parameters with performance meeting the expected standards into the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM formed parts; and list the parameters with performance not meeting the expected standards into the process parameter group with unsatisfactory verification results.

[0100] A new parameter recommendation and screening unit 470 is configured to add the process parameter group with unsatisfactory verification results to the data set, update the data set, and repeat the processes of Gaussian process regression model training, multiple stepwise regression model training, SLM process parameter recommendation and screening to obtain multiple groups of newly screened SLM process parameters.

[0101] It can be understood that the detailed function implementations of the above units can be referred to the descriptions in the foregoing method embodiments and will not be elaborated here.

[0102] In addition, an embodiment of the present invention provides an electronic device, which includes: a memory and a processor;

[0103] The memory is configured to store a computer program;

[0104] The processor is used to implement the method in the above embodiments when executing the computer program.

[0105] In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method in the above embodiments is implemented.

[0106] Based on the method in the above embodiments, the present invention provides a computer program product. When the computer program product runs on a processor, the processor is caused to execute the method in the above embodiments.

[0107] Based on the method in the above embodiments, the present invention also provides a chip, including one or more processors and an interface circuit. Optionally, the chip may further include a bus. Wherein:

[0108] The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method may be completed by the integrated logic circuit in the hardware of the processor or by instructions in software form. The above-mentioned processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute each method and step disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0109] The interface circuit may be used for sending or receiving data, instructions or information. The processor may use the data, instructions or other information received by the interface circuit for processing and may send the processed information through the interface circuit. Optionally, the chip further includes a memory, and the memory may include a read-only memory and a random access memory and provide operation instructions and data to the processor. A part of the memory may further include a non-volatile random access memory (NVRAM).

[0110] Optionally, the memory stores an executable software module or a data structure, and the processor may execute corresponding operations by calling the operation instructions stored in the memory (the operation instructions may be stored in the operating system). Optionally, the interface circuit may be used to output the execution result of the processor. It should be noted that the respective functions of the processor and the interface circuit may be implemented through hardware design, may also be implemented through software design, or may be implemented through a combination of software and hardware, and are not limited here. It should be understood that each step of the above method embodiment may be completed by the logic circuit in the hardware form or by instructions in software form in the processor.

[0111] It should be understood that the sequence numbers of the steps in the above embodiments do not indicate the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. In addition, in some possible implementation manners, the steps in the above embodiments can be selectively executed according to actual situations, can be partially executed, or can be fully executed, and no limitation is imposed herein.

[0112] It should be understood that the processor in the embodiments of the present application can be a central processing unit (CPU), or can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor can be a microprocessor or any conventional processor.

[0113] The method steps in the embodiments of the present application can be implemented in a hardware manner or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in a random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable hard disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in the ASIC.

[0114] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium may be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0115] It is easy for those skilled in the art to understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting SLM forming performance and optimizing process parameters, characterized in that, Including the following steps: Design multiple groups of SLM process parameters through multi-step orthogonal experiments, perform actual manufacturing on the designed process parameters, determine the performance of the corresponding SLM formed parts, and summarize the designed SLM process parameters and the performance of the corresponding formed parts into a data set; Train a Gaussian process regression model based on the data set to obtain the first mapping relationship between SLM process parameters and SLM formed part performance; And train a multiple stepwise regression model based on the data set to obtain the second mapping relationship between SLM process parameters and SLM formed part performance; Both the first mapping relationship and the second mapping relationship are used to predict the performance of SLM formed parts; Combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model; wherein, the SLM performance prediction model fuses the two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal; Optimize the SLM performance prediction model through the teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements for the performance of SLM formed parts; Perform step-by-step screening on the multiple groups of recommended SLM process parameters according to the influence of SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment and stable process parameters; The Gaussian process regression model for predicting SLM performance uses a rational quadratic covariance function, and the form of the constant mean function is as follows: , where C is a constant; The rational quadratic covariance function is in the following form: Let x and be different input vectors. Let . The expression of the rational quadratic kernel function is: , where ; Its hyperparameters are the mixing parameter α and the length scale parameter l , which are required to be positive. Among them, α is mainly used to control the decay rate of the kernel function. The constructed Gaussian process regression model is f(x)~GP(m(x), k(x)), where GP represents the Gaussian process. The value of the constant mean function is the average of the performance values of the training set data. The rational quadratic covariance function requires adjusting hyperparameters. When adjusting the initial values of the hyperparameters for the constructed Gaussian process regression model, a block binary search strategy is adopted. A suitable block region is selected, and then the binary search strategy is used. In each trial, the midpoint of the two adjacent nodes with the largest distance is taken as the hyperparameter test point, and the region is gradually refined through multiple trials until the initial values of the hyperparameters with satisfactory scores are obtained.

2. The method according to claim 1, wherein It also includes the following steps: For the multiple groups of SLM process parameters after step-by-step screening, perform actual manufacturing and verify the performance of the obtained SLM formed parts. List the parameters with performance meeting the expected standards in the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM formed parts; List the parameters with performance not meeting the expected standards in the process parameter group with unsatisfactory verification results; Add the process parameter group with unsatisfactory verification results to the data set, update the data set, and repeat the processes of Gaussian process regression model training, multiple stepwise regression model training, SLM process parameter recommendation and screening to obtain multiple groups of newly screened SLM process parameters.

3. The method according to claim 1, characterized in that, Design multiple groups of SLM process parameters through multi-step orthogonal experiments, specifically: Design a multi-step orthogonal experiment, design SLM process parameter groups through orthogonal experiments within the global range, perform actual manufacturing on the designed process parameters, and detect the performance of the preliminarily designed SLM formed parts; Conduct a preliminary detection of the global performance of SLM, find out the suspicious areas with better performance of the preliminarily detected SLM formed parts, and redesign orthogonal experiments for the suspicious areas. After multiple design of orthogonal experiments, obtain multiple groups of SLM process parameters corresponding to the suspicious areas with better performance of SLM formed parts.

4. The method according to any one of claims 1 to 3, characterized in that, Train a Gaussian process regression model based on the data set, and train a multiple stepwise regression model based on the data set, specifically: Construct a Gaussian process regression model; train the Gaussian process regression model based on the designed multiple groups of SLM process parameters; Construct a multiple stepwise regression model; train the multiple stepwise regression model based on multiple groups of designed SLM process parameters; among them, considering the mutual influence among process parameters, add high-order terms and cross terms to construct a complete set of alternative items; After obtaining the complete set of alternative items, gradually introduce and eliminate the alternative items through significance tests.

5. The method according to claim 4, characterized in that, During the training process of the Gaussian process regression model and the multiple stepwise regression model, the two models are weighted and evaluated through three indicators: the mean absolute error (MAE), the root mean square error (RMSE), and the coefficient of determination R 2 to evaluate the deviation between the model prediction results and the actual results, and optimize the model parameters; the weights of the three indicators are: MAE: RMSE: R 2 = 40:40:

20.

6. The method according to claim 1, characterized in that, Optimize the SLM performance prediction model through the teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements for the performance of SLM formed parts, specifically: Set an appropriate number of initial populations and use the SLM performance prediction model as the fitness function of the optimization algorithm; By calculating the individual fitness, select the individual with the largest fitness value as the teacher. The initial population is updated through the "teaching stage" and "learning stage". After the population update is completed, re-select the teacher individual, and so on in a loop until the loop termination condition is reached, obtain the current optimal fitness individual, that is, the current optimal performance value and the corresponding process parameters, and use the process parameters corresponding to the optimal performance value as the recommended SLM process parameters.

7. The method according to claim 1, wherein The SLM process parameters include: laser power, powder layer thickness, scanning speed, and scanning spacing; laser power and scanning speed have relatively greater impacts on the life of the equipment, and laser power has relatively greater impacts on the process stability of the equipment. Set multiple thresholds corresponding to the two process parameters according to the laser power and scanning speed of the actual equipment. When the laser power or scanning speed of the actual equipment is in different threshold intervals, the laser power or scanning speed is in different ratings, and different ratings result in different losses to the equipment; Stepwise screen the multiple groups of recommended SLM process parameters according to the influence of the SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment, specifically: Classify the multiple groups of recommended SLM process parameters into four categories according to: ① laser power level one, scanning speed level one, ② laser power level one, scanning speed level two, ③ laser power level two, scanning speed level one, ④ laser power level two, scanning speed level two; among them, laser power level one has the lowest damage to the service life of the equipment, and as the laser power level increases, the damage to the service life of the equipment increases; scanning speed level one has the lowest damage to the service life of the equipment, and as the scanning speed level increases, the damage to the service life of the equipment increases; After stepwise screening and eliminating the parameters in the multiple groups of recommended SLM process parameters where the laser power and scanning speed are not in the above four categories, display them in order from ① to ④ from top to bottom to obtain the multiple groups of screened process parameters.

8. A system for predicting SLM forming performance and optimizing process parameters, characterized in that, Include: A parameter dataset design unit for designing multiple groups of SLM process parameters through multi-step orthogonal experiments, actually manufacturing the designed process parameters, determining the performance of the corresponding SLM formed parts, and summarizing the designed SLM process parameters and the corresponding formed part performance into a dataset; A regression model training unit for training a Gaussian process regression model based on the dataset to obtain the first mapping relationship between SLM process parameters and SLM formed part performance; And train a multiple stepwise regression model based on the dataset to obtain the second mapping relationship between the SLM process parameters and the performance of the SLM formed part; Both the first mapping relationship and the second mapping relationship are used to predict the performance of the SLM formed part; A regression model combination unit is used to combine the trained Gaussian process regression model and the trained multiple stepwise regression model to obtain an SLM performance prediction model; wherein, the SLM performance prediction model fuses the two performance prediction results of the Gaussian process regression model and the multiple stepwise regression model in a weighted manner, and the weights of the weighted manner are determined by traversal; A prediction model optimization unit is used to optimize the SLM performance prediction model through a teaching and learning algorithm to obtain multiple groups of recommended SLM process parameters that meet the requirements for the performance of the SLM formed part; A process parameter screening unit is used to stepwise screen the multiple groups of recommended SLM process parameters according to the influence of the SLM process parameters on the service life of the equipment to obtain multiple groups of SLM process parameters with low damage to the service life of the equipment and stable process parameters; The Gaussian process regression model for predicting SLM performance adopts a rational quadratic covariance function, and the form of the constant mean function is as follows: , where C is a constant; The rational quadratic covariance function is in the following form: Let x and be different input vectors, and let . The expression of the rational quadratic kernel function is: , where ; Its hyperparameters are the mixing parameter α and the length scale parameter l , which are required to be positive. Among them, α is mainly used to control the decay rate of the kernel function; the constructed Gaussian process regression model is f(x)~GP(m(x), k(x)), where GP represents Gaussian process; the value of the constant mean function is the average of the performance values of the training set data, and the rational quadratic covariance function requires adjusting hyperparameters; when adjusting the initial values of the hyperparameters of the constructed Gaussian process regression model, a block-by-block bisection strategy is adopted. First, a suitable block area is selected, and then the bisection strategy is used. In each trial, the midpoint of the two adjacent nodes with the largest spacing is taken as the hyperparameter test point, and the area is gradually refined through multiple trials until the initial values of the hyperparameters with satisfactory scores are obtained.

9. The system according to claim 8, wherein It also includes: A process parameter grouping unit is used to verify the performance of the obtained SLM formed part through actual manufacturing for the multiple groups of SLM process parameters after stepwise screening, list the parameters with performance meeting the expected standard into the process parameter group with satisfactory verification results as the process parameter group for high-performance SLM formed parts; List the parameters with performance not meeting the expected standard into the process parameter group with unsatisfactory verification results; A parameter new recommendation screening unit is used to add the process parameter group with unsatisfactory verification results to the dataset, update the dataset, and repeat the processes of Gaussian process regression model training, multiple stepwise regression model training, SLM process parameter recommendation and screening to obtain multiple groups of SLM process parameters after new screening.

10. The system according to claim 8 or 9, characterized in that The SLM process parameters include: laser power, powder layer thickness, scanning speed, and scanning spacing; the laser power and scanning speed have relatively large impacts on the service life of the equipment, and the laser power has relatively large impacts on the process stability of the equipment. Set multiple thresholds corresponding to the two process parameters according to the laser power and scanning speed of the actual equipment. When the laser power or scanning speed of the actual equipment is in different threshold intervals, the laser power or scanning speed is in different ratings, and different ratings result in different losses to the equipment; The process parameter screening unit classifies multiple groups of recommended SLM process parameters into four categories according to: ① laser power level one, scanning speed level one; ② laser power level one, scanning speed level two; ③ laser power level two, scanning speed level one; ④ laser power level two, scanning speed level two. Among them, laser power level one causes the lowest damage to the service life of the equipment, and as the laser power level increases, the damage to the service life of the equipment increases; scanning speed level one causes the lowest damage to the service life of the equipment, and as the scanning speed level increases, the damage to the service life of the equipment increases. After step-by-step screening to eliminate the parameters of laser power and scanning speed in multiple groups of recommended SLM process parameters that do not fall into the above four categories, they are displayed sequentially from top to bottom in the order from ① to ④ to obtain multiple groups of screened process parameters.

Images