Method for optimizing residual stress of additive manufacturing part

Abstract

The invention discloses an additive manufacturing part residual stress optimization method, and relates to the technical field of stress optimization, and the method comprises the steps: checking the exposure energy fraction and volume expansion coefficient of a material as input parameters of a stress BP genetic optimization algorithm; the influence degree of different process parameters on stress is analyzed through analogue simulation, and equivalent stress is output; and designing a stress BP genetic optimization algorithm, establishing an anonymous function constraint prediction range of the analog input parameter and the output parameter, and outputting a corresponding process parameter combination. According to the method, the problem that a process adjustment method depends on experience in traditional stress optimization is solved, automatic optimization of process parameters is achieved, and the efficiency and precision of the optimization process are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of stress optimization, and specifically provides a method for optimizing the residual stress of additive manufacturing parts. Background Technique

[0002] The working principle of additive manufacturing technology is to use special software to slice the designed three-dimensional model to obtain the cross-sectional data of each layer. With the help of a horizontal scraper, metal powder is evenly laid on the substrate. According to the data information of each layer, a high-energy laser beam will accurately scan and melt the metal powder, and then layer by layer shape the object. This process will continue until the entire component is completely manufactured. After manufacturing, some post-processing operations may be required, such as removing the support structure and surface treatment, to ensure that the final manufactured part can obtain the required performance and appearance.

[0003] During the additive manufacturing process, the high-temperature laser heat source moves at high speed on the powder bed. The metal powder changes from a solid state to a liquid state and then cools to a solid state, experiencing a large temperature change, resulting in uneven stress distribution. When the stress is too large and the thermal stress accumulates continuously and cannot be released, residual stress will be formed inside the model. If the residual stress cannot be controlled within an appropriate range, it may cause printing defects in the parts. Therefore, domestic and foreign scholars have conducted a lot of research and done a lot of work on the stress field during selective laser melting. The research is mainly carried out in two aspects. One is the research on the stress itself, such as analyzing the numerical magnitude, type, distribution and law of the stress field of different parts. The other is that a large number of scholars are researching methods to optimize the stress field, reducing the adverse effects of residual stress on parts from aspects such as process parameters, temperature field and heat treatment.

[0004] Traditional methods for reducing residual stress, such as heat treatment and machining, usually cannot completely eliminate the residual stress generated during additive manufacturing, and have problems such as long time and high cost. Therefore, it is particularly important to develop a method for optimizing residual stress suitable for additive manufacturing that can effectively control or reduce residual stress. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for optimizing the residual stress of additive manufacturing parts to solve the existing problems mentioned in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for optimizing the residual stress of additive manufacturing parts, including the following steps:

[0007] S1. Check the exposure energy fraction and volume expansion coefficient of the material as the input parameters of the stress BP genetic optimization algorithm;

[0008] S2. Analyze the influence degree of different process parameters on stress through simulation, and output the equivalent stress;

[0009] S3. Design a stress BP genetic optimization algorithm, establish the anonymous function constraint prediction range of the simulation input parameters and output parameters, and output the corresponding process parameter combinations.

[0010] The further improvement of the present invention lies in that the specific steps of S2 include:

[0011] S21. Collect equivalent stress data under different process parameters through simulation analysis; the process parameters include laser power, scanning speed and powder layer thickness;

[0012] S22. Establish a process parameter - equivalent stress model to describe the influence of process parameters on equivalent stress, and obtain the process influence coefficient;

[0013] S23. Combine the process influence coefficient to obtain the equivalent stress.

[0014] The further improvement of the present invention lies in that the specific steps of the stress BP genetic optimization algorithm include:

[0015] S31. Construct the basic framework of the BP neural network and establish the initial network weight threshold length;

[0016] S32. Use the error obtained by training the BP neural network as the fitness function of the genetic algorithm;

[0017] S33. Take the process parameters selected by the genetic algorithm as the initial parameters of the BP neural network;

[0018] S34. Output the process parameter prediction result through the weight optimization strategy.

[0019] The further improvement of the present invention lies in that the specific steps of step S32 include:

[0020] S321. Randomly generate n groups of process parameters and their corresponding equivalent stresses, and perform binary encoding on each group of parameters, which consists of four parts: the connection weight between the input layer and the hidden layer, the hidden layer threshold, the connection weight between the hidden layer and the output layer, and the output layer threshold;

[0021] S322. Randomly generate a population PI(0) containing multiple process parameter schemes, and each scheme is a coded task representation;

[0022] S323. Extract the BP neural network error as the fitness function;

[0023] S323. Select the top 5 process parameter schemes with the best fitness in PI(0) as the first population PI(1);

[0024] S324. Randomly select two process parameter solutions in PI(1) as the parental task assignment solutions, select a position in the parental process parameter solutions for crossover operation to form new offspring process parameter solutions;

[0025] S325. Repeat steps S323 - S324, and calculate the fitness function of the new process parameter solutions;

[0026] S326. List the results of all fitness functions in the fitness sequence. When the standard deviation of the data in the fitness sequence is less than the set fitness fluctuation threshold, stop the iteration, and output the fitness corresponding to the process parameter solutions at this time as the weight threshold in step S33.

[0027] A further improvement of the present invention lies in that step S32 further includes using the trained BP neural network to predict the residual stress values under different process parameter combinations. When the deviation between the predicted residual stress value and the target value is greater than the set residual stress deviation threshold, the mutation probability is increased by 100 times. If it is less than or equal to the set residual stress deviation threshold, the mutation probability is reduced by combining the attenuation factor with the initial mutation probability.

[0028] A further improvement of the present invention lies in that the weight optimization strategy includes establishing an anonymous function constraint prediction range for each process parameter and the equivalent stress, setting the prediction performance of the neural network as the fitness standard of the genetic algorithm, and continuously adjusting the weights and biases in the neural network, so as to inversely deduce the optimal process parameter combination when the equivalent stress is the smallest.

[0029] A further improvement of the present invention lies in that the calculation formula of the process parameter - equivalent stress model is σ eq = a·P + b·v + c·h + d; where P represents the laser power, v represents the scanning speed, h represents the powder bed thickness, a represents the influence coefficient corresponding to the laser power, b represents the influence coefficient corresponding to the scanning speed, c represents the influence coefficient corresponding to the powder bed thickness, and d represents a constant; by fitting the data in step S21, the influence coefficients a, b, and c are obtained; then return to the process parameter - equivalent stress model calculation formula.

[0030] On the other hand, the present invention provides an additive manufacturing part residual stress optimization system, including:

[0031] An input parameter verification module, which verifies the material exposure energy fraction and the volume expansion coefficient as the input parameters of the stress BP genetic optimization algorithm;

[0032] An equivalent stress simulation module, which analyzes the influence degree of different process parameters on stress through simulation and outputs the equivalent stress;

[0033] The process parameter output module designs a BP genetic optimization algorithm for stress, establishes an anonymous function constraint prediction range for the simulated input parameters and output parameters, and outputs the corresponding process parameter combinations.

[0034] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned method for optimizing the residual stress of additive manufacturing parts.

[0035] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the above-mentioned method for optimizing the residual stress of additive manufacturing parts.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] 1. First, by combining the BP neural network and the genetic algorithm, the present invention can optimize the process parameter combinations according to the simulation analysis results in the complex additive manufacturing process, and precisely control the distribution of the equivalent stress. This optimization method can reduce the stress concentration caused by process defects while ensuring the quality of the parts, thereby reducing the occurrence of deformation and cracks, and improving the overall performance and reliability of the parts.

[0038] 2. Adjust the mutation probability of different deviations of the predicted residual stress in the algorithm to enhance the global exploration ability of the search, prompt the optimization algorithm to jump out of the local optimum, and improve the convergence quality.

[0039] 3. Through the stress BP genetic optimization algorithm, the process parameters can be automatically adjusted and optimized without relying on manual experience or experiments, saving a large amount of experimental time and costs. By automatically calculating the optimal process parameter combinations through the algorithm, not only the production efficiency is improved, but also the risk of parameter mismatch caused by human operation errors is effectively reduced. It can be dynamically adjusted in real time according to the input process parameters and simulation data, improving the manufacturing accuracy while enhancing the adaptability and flexibility of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flowchart of a method for optimizing the residual stress of additive manufacturing parts according to the present invention;

[0041] Figure 2 is a flowchart of the stress BP genetic optimization algorithm for a method for optimizing the residual stress of additive manufacturing parts according to the present invention;

[0042] Figure 3 shows a trend chart of the influence of the laser power of the present invention on the equivalent peak stress;

[0043] Figure 4Shows the influence trend diagram of the scanning speed of the present invention on the equivalent peak stress;

[0044] Figure 5 Shows the influence trend diagram of the powder laying layer thickness of the present invention on the equivalent peak stress;

[0045] Figure 6 Shows the error comparison diagram between the training set and the actual value in this embodiment;

[0046] Figure 7 Shows the error comparison diagram between the test set and the actual value in this embodiment;

[0047] Figure 8 Shows the regression diagram of the BP neural network in this embodiment;

[0048] Figure 9 Shows the regression diagram of the stress BP genetic optimization algorithm in this embodiment;

[0049] Figure 10 Shows the node coordinates and their corresponding stress values at the initial process parameters in this embodiment;

[0050] Figure 11 Shows the node coordinates and their corresponding stress values when predicting the process parameters in this embodiment;

[0051] Figure 12 Shows the printed radar shielding device model in this embodiment;

[0052] Figure 13 Shows the front side of the deviation scanning data of the initial process parameters of the radar shielding device in this embodiment;

[0053] Figure 14 Shows the side view of the deviation scanning data of the initial process parameters of the radar shielding device in this embodiment;

[0054] Figure 15 Shows the front side of the deviation scanning data of the predicted process parameters of the radar shielding device in this embodiment;

[0055] Figure 16 Shows the side view of the deviation scanning data of the predicted process parameters of the radar shielding device in this embodiment;

[0056] Figure 17 Is a framework diagram of an optimization system for residual stress of additive manufacturing parts of the present invention. Detailed implementation manners

[0057] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0058] The term "and / or" is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the front and rear associated objects.

[0059] Embodiment 1

[0060] Figure 1 A flowchart for optimizing the residual stress of an additive manufacturing part disclosed in this embodiment is shown as follows:

[0061] S1. In selective laser melting printing, subtle changes such as different metal powder materials, different printers, or even changes in the machine position may lead to different environments during the printing process. Therefore, it is necessary to understand the physical and chemical properties of the metal powder, the selection of the machine, and the working parameters. At the same time, it is necessary to check the material property parameters such as the exposure energy fraction and the volume expansion coefficient at a certain temperature to reduce or eliminate the influence of different materials and equipment. Therefore, checking the exposure energy fraction and the volume expansion coefficient of the material is used as the input parameter of the stress BP genetic optimization algorithm.

[0062] Before the thermo-mechanical coupling simulation analysis, it is necessary to first check the exposure energy fraction and the volume expansion coefficient of the material, and use the obtained results as the parameter input for the next simulation. The material property is selected as isotropic, that is, the physical and chemical properties of the object do not change with direction.

[0063] The exposure energy fraction (EEF for short) is a key parameter in the powder bed fusion additive manufacturing process. Specifically, it refers to the energy density transferred by a laser or other form of energy source when acting on the powder bed. This index is crucial for controlling the quality and performance of the printed part because it directly affects the melting and solidification process of the powder, thereby determining the microstructure and mechanical properties of the final product.

[0064] The volumetric expansion factor (VEF) is another important physical quantity in materials science. It describes the ratio of the relative change in the volume of an object to its original volume under a unit temperature change. In thermo-mechanical coupling analysis, this coefficient is widely used to calculate the internal stress and external deformation generated by thermal expansion when the temperature field of an object changes. Through this analysis, we can more accurately predict and evaluate the performance of materials in practical applications, especially under extreme temperature conditions.

[0065] In powder bed fusion additive manufacturing, the exposure energy fraction and the volumetric expansion factor are two factors that need to be considered comprehensively. Appropriate exposure energy can ensure that the powder material melts evenly and sufficiently, resulting in high-quality printed products. At the same time, understanding the volumetric expansion factor of the material helps us control the dimensional accuracy of the product during printing and prevent problems such as warping or cracking caused by temperature changes.

[0066] S2. Analyze the influence degree of different process parameters on stress through simulation, and output the equivalent stress;

[0067] During the selective laser melting process, the forming of the model is affected by many factors, and the representative one is the influence of process parameters, which determine the quality of part processing. Improper selection of process parameters will affect the processing accuracy, processing efficiency, and even cause deformation and cracking of the part, resulting in printing failure.

[0068] The present invention first makes an appropriate division of the network model according to the characteristics of the model, selects a set of process parameters to analyze the distribution changes of deformation, temperature, temperature gradient, and stress of the selected points and trajectories. Then, multiple sets of process parameters are set for simulation calculation, several control data points are selected to draw a line chart, and then the influence of changes in laser power, scanning speed, and powder bed thickness on the equivalent peak stress is analyzed.

[0069] S3. Design a stress BP genetic optimization algorithm, establish an anonymous function constraint prediction range for the simulated input parameters and output parameters, and output the corresponding process parameter combination.

[0070] The specific steps of S2 include:

[0071] S21. Collect equivalent stress data under different process parameters through simulation analysis; the process parameters include laser power, scanning speed, and powder bed thickness;

[0072] S22. Establish a process parameter - equivalent stress model to describe the influence of process parameters on the equivalent stress, and obtain the process influence coefficient; the calculation formula of the process parameter - equivalent stress model is σ eq= a·P + b·v + c·h + d; where P represents the laser power, v represents the scanning speed, h represents the powder layer thickness, a represents the influence coefficient corresponding to the laser power, b represents the influence coefficient corresponding to the scanning speed, c represents the influence coefficient corresponding to the powder layer thickness, and d represents a constant; by fitting the data in step S21, the influence coefficients a, b, and c are obtained; then return to the process parameter - equivalent stress model calculation formula;

[0073] S221. Influence analysis of laser power on the stress field simulation results

[0074] Analyze the laser power data in S21, Figure 3 The influence trend diagram of the laser power of the present invention on the equivalent peak stress is shown. It can be found that at a lower scanning speed, as the laser power gradually increases, the equivalent peak stress also gradually increases; this is because the increase in laser power increases the energy input per unit area of the powder bed surface per unit time; however, at a higher scanning speed, as the laser power gradually increases, the equivalent peak stress shows a decreasing trend; the change in the equivalent peak stress at 1000 mm / s and 1200 mm / s is relatively small, and the curve is relatively flat. Therefore, a laser - scanning speed threshold is set, and the experimental data on both sides of the laser - scanning speed threshold are respectively fitted to obtain the influence coefficients corresponding to the laser power in the two cases.

[0075] S222. Influence analysis of scanning speed on the stress field simulation results

[0076] Analyze the laser power data in S21, Figure 4 The influence trend diagram of the scanning speed of the present invention on the equivalent peak stress is shown. It can be found that as the scanning speed continuously increases, the equivalent peak stress as a whole shows a decreasing trend. This is because after the scanning speed increases, the energy input per unit area of the powder material per unit time decreases, the overall temperature of the powder bed decreases, the temperature gradient becomes smaller, and the stress value becomes smaller. Therefore, the influence coefficient corresponding to the scanning speed is obtained by linear fitting;

[0077] S223. Influence analysis of powder layer thickness on the stress field simulation results

[0078] Analyze the powder layer thickness data in S21, Figure 5 The influence trend diagram of the powder layer thickness of the present invention on the equivalent peak stress is shown. It can be found that when the powder layer thickness is 0.06 mm, the overall equivalent stress value is lower than when the powder layer thickness is 0.03 mm. This is because when the powder layer thickness increases, the energy input per unit area of the powder per unit time remains unchanged, but the amount of powder material increases, the energy required for melting increases, the overall temperature of the powder bed decreases, the temperature gradient becomes smaller, and the equivalent stress becomes smaller.

[0079] S23. Combine the process influence coefficients to obtain the equivalent stress.

[0080] Example 2

[0081] Figure 2 The flowchart of the stress BP genetic optimization algorithm for the residual stress optimization method of an additive manufacturing part according to the present invention is shown. Based on the inventive concept of Example 1, the present invention proposes a stress BP genetic optimization algorithm for the residual stress optimization method of an additive manufacturing part. The specific steps include:

[0082] S31. Construct the basic framework of the BP neural network and establish the initial network weight threshold length;

[0083] In this embodiment, the trainlm training algorithm is selected. 22 groups of simulation values are substituted into the BP neural network model. 17 groups of values are selected as the training set for the BP neural network to learn; the other 5 groups of parameters are used as the test set to detect the prediction accuracy of the BP neural network.

[0084] Figure 6 The error comparison diagram between the training set and the actual value in this embodiment is shown. It can be found that the data points have a relatively high degree of coincidence and the error values fluctuate slightly. Figure 7 The error comparison diagram between the test set and the actual value in this embodiment is shown. It can be found that the data points change greatly and the error fluctuations are more obvious.

[0085] Figure 8 The regression diagram of the BP neural network in this embodiment is shown. In the results of the command bar, the mean absolute error MAE is 71.9028 and the root mean square error RMSE is 94.8837. The BP neural network shows a large prediction error in multiple operations. For example, the root mean square error RMSE fluctuates between 60 and 150. This may be due to overfitting. One reason for overfitting is that the initial weights and thresholds are randomly large, which affects the learning effect and is prone to falling into local optima. Another reason is that the training data set is insufficient and it is difficult to meet the requirements of the BP neural network for a large amount of data. To improve the prediction accuracy, the genetic algorithm can be used to optimize the initial parameters of the BP neural network to achieve global optimization and avoid the local optimum problem.

[0086] S32. Based on the above results, the error obtained by training the BP neural network in the present invention is used as the fitness function of the genetic algorithm to improve the prediction accuracy.

[0087] S321. Randomly generate n groups of equivalent stresses corresponding to process parameters of the machine, and perform binary encoding on each group of parameters, which consists of four parts: the connection weight between the input layer and the hidden layer, the hidden layer threshold, the connection weight between the hidden layer and the output layer, and the output layer threshold;

[0088] S322. Randomly generate a population PI(0) containing multiple process parameter schemes, and each scheme is a coded task representation;

[0089] S323. Extract the error of the BP neural network as the fitness function;

[0090] Use the trained BP neural network to predict the residual stress values under different combinations of process parameters. When the deviation between the predicted residual stress value and the target value is large, and the set residual stress deviation threshold indicates that the currently searched combination of process parameters may deviate from the optimal solution. At this time, increase the mutation probability of the genetic algorithm to prompt the algorithm to jump out of the current local optimal solution and explore more solution spaces. Specifically, set a residual stress deviation threshold. When the deviation between the predicted residual stress value and the target value is greater than the set residual stress deviation threshold, the mutation probability is increased by 100 times. If it is less than or equal to the set residual stress deviation threshold, the mutation probability is reduced by combining the attenuation factor with the initial mutation probability to protect the current better solution and avoid destroying the existing excellent genes due to excessive mutation.

[0091] S323. Select the top 5 process parameter solutions with the highest fitness in PI(0) as the first population PI(1);

[0092] S324. Randomly select two process parameter solutions in PI(1) as the parent task assignment solutions, and select a position in the parent process parameter solutions for crossover operation to form new offspring process parameter solutions;

[0093] S325. Repeat steps S323 - S324, and calculate the fitness function of the new process parameter solutions;

[0094] S326. List the results of all fitness functions in the fitness sequence. When the standard deviation of the data in the fitness sequence is less than the set fitness fluctuation threshold, stop the iteration and output the fitness corresponding to the process parameter solution at this time as the weight threshold in step S33.

[0095] Figure 9 The regression graph of the stress BP genetic optimization algorithm in this embodiment is shown. Comparing the results predicted by the BP neural network, it can be seen that the test set is no longer overfitted, and the prediction error degree with the training set becomes smaller. The RMSE drops from more than 90 to the maximum value of 42 in the test set. The difference reflected in the regression graph is also getting smaller. This verifies that it is reasonable and effective to optimize the weight threshold of the neural network using the genetic algorithm, increases the prediction success rate, and it is relatively easy to calculate a good optimization model under multiple runs.

[0096] S33. Use the process parameters selected by the genetic algorithm as the initial parameters of the BP neural network;

[0097] S34. Output the process parameter prediction results through the weight optimization strategy. Table 1 shows the output process parameter prediction results;

[0098] Table 1 Prediction Results of Process Parameters

[0099]

[0100] It can be found that the results calculated twice do not vary much. To verify the beneficial effects of the present invention, simulation experiment verification and deviation experiment verification are carried out on the algorithm output.

[0101] In the simulation experiment verification, first, the initial process parameters: laser power 300W, scanning speed 900mm / s, powder laying layer thickness 0.03mm are used as the input of the simulation verification model, as Figure 10 , which shows several node coordinates and their corresponding stress values under the initial process parameters of the present invention; it can be observed that the stress values of all nodes are above 1000Mpa, belonging to a set of process parameters with relatively large stress in the data set. The points closest to the selected point coordinates are the points numbered 3 and 4, with an average value of 1026.5Mpa;

[0102] Subsequently, the parameters predicted by the neural network are substituted into the simulation software for simulation calculation, as Figure 11 , which shows several node coordinates and their corresponding stress values under the predicted process parameters of the present invention; it can be found that the stress values of several coordinate points in the figure are relatively small, basically around 800Mpa. The points closest to the selected point coordinates are the points numbered 1 and 2, with an average value of 833.9465Mpa. The predicted value of the stress BP genetic optimization algorithm is about 900Mpa, and the prediction error is 7.34%. In the data set of the algorithm, the average value of 22 groups of data is 941.58018Mpa, and the optimization degree is 11.43%. After simulation verification, the predicted value of the stress BP genetic optimization algorithm is relatively accurate, and the optimization result is also good, successfully reducing the stress value and optimizing the stress field.

[0103] In this embodiment, the deviation experiment verification is realized by printing SLM. In SLM printing, the temperature gradient caused by the rapid heating and cooling of the material will affect the residual stress value; and too large residual stress value will cause plastic deformation of the part, resulting in deformation, such as bending, twisting and warping, etc., and cracks and delamination may also occur; so the deviation size of the part is usually directly related to the residual stress level. Larger deviation often means higher residual stress. Therefore, the situation of residual stress can be indirectly evaluated by measuring and comparing the deviation between the actual size and the designed size of the printed part.

[0104] As Figure 12 , which shows the radar shield model after printing of the present invention. By scanning the model, the deviation distribution and deviation values of each point of the radar shield can be obtained to analyze and compare the deviation situations before and after optimization, so as to reflect the optimization degree of the residual stress.

[0105] Figure 13 andFigure 14 The deviation scan data report of the initial process parameters of the radar blocker of the present invention is shown. It can be found that the deviation distribution is uneven, irregular, changes relatively violently, and the deviation values in most areas are large, and the deformation effect is relatively serious, which can reflect that the residual stress value in the part is too high, the stress field distribution is uneven, and there is a need for optimization.

[0106] Therefore, input the predicted process parameters into the printer, print the physical object and scan the deviation data of the part, compare the changes with the initial process parameters, and analyze the magnitude of the deviation value, so as to indirectly reflect the optimization degree of the residual stress value; Figure 15 and Figure 16 The deviation scan data report of the predicted process parameters of the radar blocker of the present invention is shown. It can be found that in terms of the deviation value, the overall deviation amplitude decreases after optimization, and in terms of the deviation distribution, the change is relatively stable. Comparing with the scan data before optimization, the deformation situation has been improved, the deviation situation is better, and at the same time, it can also reflect that the overall level and distribution of the residual stress have been optimized, proving that the optimization method of the present invention is effective.

[0107] The weight optimization strategy includes establishing an anonymous function constraint prediction range of each process parameter and the equivalent stress, setting the prediction performance of the neural network as the fitness standard of the genetic algorithm, and continuously adjusting the weights and biases in the neural network, so as to inversely deduce the optimal process parameter combination when the equivalent stress is the smallest.

[0108] Example 3

[0109] Figure 17 The framework diagram of the residual stress optimization system for an additive manufacturing part of the present invention is shown. Based on the same inventive concept as in Example 1 and Example 2, the present invention provides a residual stress optimization system for an additive manufacturing part, including

[0110] An input parameter verification module, which verifies the material exposure energy fraction and the volume expansion coefficient as the input parameters of the stress BP genetic optimization algorithm;

[0111] An equivalent stress simulation module, which analyzes the influence degree of different process parameters on the stress through simulation and outputs the equivalent stress;

[0112] A process parameter output module, which designs a stress BP genetic optimization algorithm, establishes an anonymous function constraint prediction range of the simulated input parameters and output parameters, and outputs the corresponding process parameter combination.

[0113] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0114] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0115] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0117] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Those of ordinary skill in the art, under the inspiration of the present invention and without departing from the spirit and scope protected by the present invention and the claims, can still make many forms, and all of these fall within the protection scope of the present invention.

Claims

1. An optimization method for residual stress of additive manufacturing parts, characterized in that: It includes the following steps: S1. Check the exposure energy fraction of the material and the volume expansion coefficient, and use them as the input parameters of the stress BP genetic optimization algorithm; S2. Analyze the influence degree of different process parameters on the stress through simulation, and output the equivalent stress; S3. Design the stress BP genetic optimization algorithm, establish the anonymous function constraint prediction range of the simulated input parameters and output parameters, and output the corresponding process parameter combination.

2. The method for optimizing the residual stress of an additive manufacturing part according to claim 1, characterized in that: The specific steps of S2 include: S21. Collect the equivalent stress data under different process parameters through simulation analysis; the process parameters include laser power, scanning speed, and powder layer thickness; S22. Establish a process parameter - equivalent stress model to describe the influence of process parameters on the equivalent stress, and obtain the process influence coefficient; S23. Combine the process influence coefficient to obtain the equivalent stress.

3. The residual stress optimization method for an additive manufacturing part according to claim 1, characterized in that: The specific steps of the stress BP genetic optimization algorithm include: S31. Construct the basic framework of the BP neural network and establish the initial network weight threshold length; S32. Use the error obtained from the training of the BP neural network as the fitness function of the genetic algorithm; S33. Use the process parameters selected by the genetic algorithm as the initial parameters of the BP neural network; S34. Output the process parameter prediction result through the weight optimization strategy.

4. The method for optimizing the residual stress of an additive manufacturing part according to claim 3, characterized in that: The specific steps of step S32 include: S321. Randomly generate n groups of process parameters and their corresponding equivalent stresses, and perform binary encoding on each group of parameters, which consists of four parts: the connection weight between the input layer and the hidden layer, the hidden layer threshold, the connection weight between the hidden layer and the output layer, and the output layer threshold; S322. Randomly generate a population PI(0) containing multiple process parameter schemes, and each scheme is a coded task representation; S323. Extract the BP neural network error as the fitness function; S323. Select the top 5 process parameter schemes with the highest fitness in PI(0) as the first population PI(1); S324. Randomly select two process parameter schemes in PI(1) as the parent task assignment schemes, and perform a crossover operation at a selected position in the parent process parameter schemes to form new offspring process parameter schemes; S325. Repeat steps S323 - S324, and calculate the fitness function of the new process parameter schemes; S326. List the results of all fitness functions in the fitness sequence. When the standard deviation of the data in the fitness sequence is less than the set fitness fluctuation threshold, stop the iteration, and output the fitness corresponding to the process parameter scheme at this time as the weight threshold in step S33.

5. A method for optimizing the residual stress of an additive manufacturing part according to claim 3, characterized in that: Step S32 also includes using the trained BP neural network to predict the residual stress values under different process parameter combinations. When the deviation between the predicted residual stress value and the target value is greater than the set residual stress deviation threshold, the mutation probability increases by 100 times. If it is less than or equal to the set residual stress deviation threshold, the mutation probability is reduced by combining the attenuation factor with the initial mutation probability.

6. The method for optimizing the residual stress of an additive manufacturing part according to claim 4, characterized in that: The weight optimization strategy includes establishing an anonymous function constraint prediction range for each process parameter and the equivalent stress, setting the prediction performance of the neural network as the fitness criterion of the genetic algorithm, and continuously adjusting the weights and biases in the neural network to inversely deduce the optimal process parameter combination when the equivalent stress is minimized.

7. A method for optimizing residual stress of an additive manufacturing part according to claim 2, characterized in that: The calculation formula of the process parameter - equivalent stress model is σ eq = a·P + b·v + c·h + d; where P represents the laser power, v represents the scanning speed, h represents the powder layer thickness, a represents the influence coefficient corresponding to the laser power, b represents the influence coefficient corresponding to the scanning speed, c represents the influence coefficient corresponding to the powder layer thickness, and d represents a constant; by fitting the data in step S21, the influence coefficients a, b, and c are obtained; then return to the calculation formula of the process parameter - equivalent stress model.

8. An additive manufacturing part residual stress optimization system for implementing an additive manufacturing part residual stress optimization method according to any one of claims 1-7, characterized in that: It includes: An input parameter verification module that verifies the material exposure energy fraction and the volume expansion coefficient as the input parameters of the stress BP genetic optimization algorithm; An equivalent stress simulation module that analyzes the influence degree of different process parameters on the stress through simulation and outputs the equivalent stress; A process parameter output module that designs a stress BP genetic optimization algorithm, establishes an anonymous function constraint prediction range for the simulated input parameters and output parameters, and outputs the corresponding process parameter combination.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for optimizing the residual stress of an additive manufacturing part according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, it implements a method for optimizing the residual stress of an additive manufacturing part according to any one of claims 1-7.

Images