Processing method for laser metal deposition manufacturing process parameters

Abstract

The embodiment of the invention relates to a processing method for laser metal deposition manufacturing technological parameters. The processing method comprises the steps that multiple pieces of technological parameter information in a laser metal deposition manufacturing multi-technological-parameter deposition experiment are obtained; determining the influence of each piece of process parameter information on the forming quality; according to the influence, determining the influence range of each piece of process parameter information, and taking the influence range as a constraint condition; according to a second algorithm and the objective function, the number of each quality grade in each technological parameter combination is determined; constructing an objective function of process parameter optimization; according to a preset data driving prediction model and a target function, predicting the number of quality grades; calculating the fitness of the initial population; performing non-dominated sorting on the initial population to form a Pareto frontier; according to the fitness, generating a next generation of population through genetic manipulation, and repeatedly iterating until a target population is determined to form a Pareto frontier set; and determining a target process parameter combination according to the Pareto frontier set and the new search strategy.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a method for processing process parameters of laser metal deposition manufacturing. Background Art

[0002] Process parameter optimization is an important aspect of laser metal additive manufacturing (AM) process monitoring that has attracted widespread attention and is a key link in achieving "end-to-end" monitoring. As one of the representative technologies in the field of laser additive manufacturing, laser metal deposition manufacturing (LMD) has the advantages of short production cycle, good economy, and unlimited forming shapes. It is widely used in important fields such as aerospace, automobile manufacturing, medical treatment, and shipbuilding. However, the metal parts formed by it also have various forming quality problems and defects. In the manufacturing process monitoring, the change of process parameters is directly related to the forming quality of the component. Therefore, the process parameters must be optimized to determine the optimal process parameters.

[0003] With the development of machine learning, developing suitable machine learning algorithms to monitor the manufacturing process in real time can improve the quality of parts and achieve "shape control" of metal parts. Therefore, developing effective process parameter optimization algorithms and embedding them into manufacturing process monitoring has received widespread attention. For example, Patil et al. developed a process parameter optimization model based on image processing technology to determine the optimal combination of process parameters for laser additive manufacturing components. Kumar et al. developed a process parameter optimization model based on genetic algorithm to minimize voids and maximize material yield, and optimize process parameters such as wire feeding speed, torch speed, and gas flow rate. Although the above methods have many advantages, the defect-based process parameter optimization method in the manufacturing process requires a lot of experiments and engineering experience, there are multiple related optimization objectives, and there is no optimal solution. Summary of the invention

[0004] The purpose of the present invention is to provide a method for processing laser metal deposition manufacturing process parameters in view of the defects in the prior art, so as to solve the problems in the prior art.

[0005] To achieve the above object, the present invention provides a method for processing laser metal deposition manufacturing process parameters, comprising:

[0006] Obtain multiple process parameter information in laser metal deposition manufacturing multi-process parameter deposition experiment;

[0007] According to the first algorithm, respectively determine the influence of each process parameter information among the plurality of process parameter information on the forming quality; the influence is represented by a quality grade;

[0008] According to the influence, determining the influence range of each process parameter information, and taking the influence range as a constraint condition;

[0009] Combining the plurality of process parameter information to determine a plurality of groups of process parameter combinations;

[0010] Determining the number of each quality level in each set of process parameter combinations according to the second algorithm and the objective function;

[0011] According to the number of each quality grade and each set of process parameter combinations, the objective function of process parameter optimization is constructed;

[0012] Predicting the number of quality levels corresponding to each process parameter information according to a preset data-driven prediction model and the objective function;

[0013] A set of random process parameter combinations is used as an initial population, and the fitness of the initial population is calculated;

[0014] Performing non-dominated sorting on the initial population to form a Pareto front;

[0015] According to the fitness, the next generation population is generated through genetic operation, and the iteration is repeated until the target population is determined, thereby forming a Pareto frontier set;

[0016] According to the Pareto frontier set, a target process parameter combination is determined according to a new search strategy.

[0017] In a possible implementation, the process parameter information includes laser power, powder feeding rate and travel speed; and the determining, according to the first algorithm, respectively the influence of each process parameter information on the forming quality includes:

[0018] Through the variable control method, any one of the laser power, powder feeding rate and travel speed is adjusted, and the other two are fixed, and the influence on the forming quality when one process parameter information is changed and the other two process parameter information remain unchanged is determined, and multiple groups of quality grades are obtained; wherein each group of the quality grades includes multiple quality levels, and each quality level has a quantity; the first algorithm is the control variable method.

[0019] In a possible implementation, determining the influence range of each process parameter information according to the influence, and using the influence range as a constraint condition specifically includes:

[0020] According to the influence on the forming quality, the upper limit and the lower limit of the influence of each process parameter information are determined, and the upper limit and the lower limit of each process parameter information are used as a constraint condition of the process parameter information;

[0021] The constraints corresponding to each process parameter information constitute the final constraints.

[0022] In a possible implementation, determining the number of each quality level in each set of process parameter combinations according to the second algorithm and the objective function specifically includes:

[0023] training the second algorithm by the constraint conditions and the quantity of each quality level;

[0024] Input multiple sets of process parameter combinations into the second algorithm to predict the number of quality levels corresponding to each set of process parameter combinations; wherein the second algorithm is a label propagation algorithm.

[0025] In a possible implementation, the objective function of optimizing the process parameters is constructed by taking the minimum number of each quality level as a sub-objective of the objective function and according to the number of each process parameter information and each set of process parameter combinations. Specifically, the objective function includes:

[0026] Findoptimal process parameters=(P,T,V)

[0027] according to Construct the objective function;

[0028] Where P is the laser power, T is the powder feeding rate, V is the travel speed, and f Level-A 、f Level-B 、f Level-C and f Level-D Represents the number of quality grades A, B, C, and D respectively.

[0029] In a possible implementation, determining the number of quality levels corresponding to each process parameter information according to the preset data-driven prediction model and the objective function specifically includes:

[0030] The historical laser power, powder feeding rate, and travel speed are used as input features, and the number of quality grades is used as output. The experimental data is divided into a training set and a test set using a random forest regression model, and the model is trained to predict the number of quality grades under different process parameters.

[0031] By using any one of cross validation, mean square error and R2 score as a model evaluation indicator, the performance of the random forest regression model is evaluated, and when the evaluation passes, it is determined that the training of the random forest regression model is completed;

[0032] A set of process parameter combinations in the objective function are input into the trained random forest regression model to determine the number of quality levels corresponding to each process parameter information.

[0033] In a possible implementation, the step of taking a set of random process parameter combinations as an initial population and calculating the fitness of the initial population specifically includes:

[0034] A set of random process parameter combinations is created as the initial population, each individual includes three parameters: laser power P, powder feeding rate T, and travel speed V;

[0035] determining the size of the initial population;

[0036] Sort the individuals in the initial population to form different non-dominated hierarchies;

[0037] The crowding distance is calculated between individuals in the same non-dominated hierarchy; wherein the fitness is represented by the non-dominated hierarchy and the crowding distance.

[0038] In a possible implementation, performing non-dominated sorting on the initial population according to the fitness to form a Pareto front specifically includes:

[0039] If the number of individuals in the last level exceeds the limit, selection is performed based on the crowding distance, and individuals with larger crowding distance are retained;

[0040] The individuals in the initial population are divided into multiple non-dominated levels, the first level includes all individuals that are not dominated by other individuals, and the second level includes individuals that are only dominated by individuals in the first level; wherein the first level in the initial population is the Pareto front.

[0041] In a possible implementation, generating the next generation population through genetic operation according to the fitness, repeating the iteration until the target population is determined, and forming the Pareto frontier set specifically includes:

[0042] When selecting the next generation of population, select individuals with lower non-dominated levels. If two individuals are at the same level, select the individual with larger crowding distance to form the next generation of population.

[0043] Determine the Pareto front from the next generation population;

[0044] The Pareto fronts of multiple populations form a Pareto front set.

[0045] In a possible implementation, determining the target process parameter combination according to the Pareto front set and the new search strategy specifically includes:

[0046] Determine a new search strategy based on minimizing defects, maximizing quality, comprehensive scoring, user preference, or diversity;

[0047] According to the new search strategy, the target process parameter combination is determined.

[0048] By applying the processing method of laser metal deposition manufacturing process parameters provided by the embodiment of the present invention, the data-driven model is combined with the multi-objective optimization algorithm, expensive and time-consuming multiple test experiments are eliminated, an optimal solution search strategy is established, the optimal process parameters are automatically searched from the Pareto solution set, the coupling between multiple quality objectives is balanced, and multi-objective optimization is achieved. This method can improve the accuracy and efficiency of process parameter optimization in the manufacturing process, and solve the problem that a large number of experiments are required in manufacturing process monitoring and multiple process parameter objectives are difficult to optimize. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic flow chart of a method for processing laser metal deposition manufacturing process parameters provided by an embodiment of the present invention;

[0050] Figure 2 It is a statistical diagram of the process parameter deposition experiment results;

[0051] Figure 3 Flowchart for data-driven prediction model;

[0052] Figure 4 Flowchart of the multi-objective optimization algorithm based on the new search strategy. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.

[0055] Figure 1 A schematic flow chart of a method for processing laser metal deposition manufacturing process parameters provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method for processing the laser metal deposition manufacturing process parameters includes the following steps:

[0056] Step 101, obtaining multiple process parameter information in a laser metal deposition manufacturing multi-process parameter deposition experiment;

[0057] Specifically, 17 typical laser metal deposition manufacturing multi-process parameter deposition experiments were carried out to ensure the accuracy and completeness of the experimental data. The multi-process parameter deposition experiment has three key process parameter information: laser power P, powder feeding rate T and travel speed V.

[0058] Step 102, according to the first algorithm, respectively determine the influence of each process parameter information among the plurality of process parameter information on the forming quality; the influence is represented by a quality grade;

[0059] Specifically, the first algorithm is a control variable method. Through the variable control method, any one of the laser power, powder feeding rate and travel speed is adjusted, and the other two are fixed. The influence on the forming quality when one process parameter information is changed and the other two process parameter information remain unchanged is determined, and multiple groups of quality grades are obtained; wherein each group of the quality grades includes multiple quality levels, and each quality level has a quantity.

[0060] Among them, the quality grade can include four grades AD, A represents the best quality, B represents average quality and requires further testing, C represents poor quality and is not suitable for industrial applications, and D represents the worst quality and is directly discarded.

[0061] Step 103, determining the influence range of each process parameter information according to the influence, and using the influence range as a constraint condition;

[0062] Specifically, according to the influence on the forming quality, the upper limit and the lower limit of the influence of each process parameter information are determined, and the upper limit and the lower limit of each process parameter information are used as a constraint condition of the process parameter information;

[0063] The constraints corresponding to each process parameter information constitute the final constraints. The constraints can be as follows:

[0064]

[0065] Among them, P min and P max Respectively represent the lower and upper limits of laser power, T min and T max Respectively represent the lower and upper limits of the powder feeding rate, V min and V max Respectively represent the lower and upper limits of the travel speed.

[0066] Figure 2It is a statistical diagram of the results of multi-process parameter deposition experiments. 17 typical combinations of laser metal deposition manufacturing process parameters were used for multi-process parameter deposition experiments, as shown in Table 1. In Table 1, groups A1 to A6 only change the laser power P, groups B1 to B5 only change the powder feeding speed T, and groups C1 to C6 only change the travel speed V. Then, as shown in the figure, based on four quality levels (grade A indicates the best quality; grade B indicates general quality and requires further testing; grade C indicates poor quality and is not suitable for industrial applications; grade D is the worst quality and is directly abandoned), the number of each quality level of each group of process parameters is counted. It can be found from the figure that the forming quality of A4 in group A is the best, so the parameter optimization range of laser power should be controlled between 1400-1800W; the forming quality of B3 in group B is the best, so the parameter optimization range of powder feeding speed should be controlled between 7-9g / min; the forming quality of C2 in group C is the best, so the parameter optimization range of travel speed should be controlled between 1000-1400mm / min. Finally, the process parameter constraint values ​​are obtained:

[0067]

[0068]

[0069]

[0070] Table 1

[0071] Step 104, combining the plurality of process parameter information to determine a plurality of sets of process parameter combinations;

[0072] Specifically, the three process parameter information may be arbitrarily combined to generate multiple groups of process parameter combinations. For example, at least two of them may be combined to determine four groups of process parameter combinations.

[0073] Step 105, determining the number of each quality level in each set of process parameter combinations according to the second algorithm;

[0074] Specifically, the second algorithm is trained by the constraint condition and the number of each quality level;

[0075] Input multiple sets of process parameter combinations into the second algorithm to predict the number of quality levels corresponding to each set of process parameter combinations; wherein the second algorithm is a label propagation algorithm.

[0076] Among them, when training the second algorithm, the data before step 105 can be used as input for training. After training the second algorithm, the trained label propagation algorithm can be used to input the process parameter combination into the trained label propagation algorithm to predict the quality level of the unlabeled process parameter combination, thereby realizing the determination of the number of quality levels of the unlabeled process parameter combinations and greatly expanding the data set, that is, adding the process parameter combination to the data set to improve the generalization ability of the model.

[0077] Step 106, constructing an objective function for optimizing process parameters according to the number of each quality grade and each set of process parameter combinations;

[0078] Findoptimal process parameters=(P,T,V)

[0079] Specifically, according to Construct the objective function;

[0080] Where P is the laser power, T is the powder feeding rate, V is the travel speed, and f Level-A 、f Level-B 、f Level-C and f Level-D Represents the number of quality grades A, B, C, and D respectively.

[0081] Specifically, the objective function may include four sub-objectives, namely, minimizing the number of quality D grades, minimizing the number of quality C grades, maximizing the number of quality B grades, and maximizing the number of quality A grades.

[0082] Step 107, predicting the number of quality levels corresponding to each process parameter information according to the preset data-driven prediction model and the objective function;

[0083] Among them, the data-driven prediction model can utilize the random forest algorithm. Specifically, the process parameter combination predicted in step 106 and the laser power, powder feeding rate, and travel speed corresponding to the original process parameter information are used as input features, and the number of quality grades is used as output. The random forest regression model is used to divide the experimental data into a training set and a test set, and the model is trained to predict the number of quality grades under different process parameters; thereby further improving the prediction accuracy of the number of quality grades.

[0084] Through cross validation, mean square error and R 2 Any one of the scores is used as a model evaluation indicator to perform performance evaluation on the random forest regression model, and when the evaluation passes, it is determined that the training of the random forest regression model is completed;

[0085] A set of process parameter combinations in the objective function are input into the trained random forest regression model to determine the number of quality grades corresponding to each process parameter information. Thus, when a set of process parameter combinations are input, the trained random forest model can be used to predict the corresponding number of quality grades.

[0086] like Figure 3 As shown in the figure, RF consists of multiple decision trees, each of which is generated independently and is not affected by other decision trees. Each decision tree is trained based on a randomly selected feature subset and sample subset. For prediction problems, RF uses the averaging method to average the regression prediction results of each decision tree as the final modeling result. The model process is as follows: 1) From the initial data set, select N samples as training sets by assuming return sampling, and retain unselected samples as outer packaging samples; 2) For each training set, randomly select K feature subsets as input features of the decision tree; 3) Based on the selected features and sample subsets, train the decision tree model; 4) Repeat steps 1-3 to build a decision tree model; 5) Take the average of the prediction results of N decision trees as the final prediction result of process parameters and the number of four quality grades.

[0087] Step 108, taking a set of random process parameter combinations as an initial population, and calculating the fitness of the initial population;

[0088] Specifically, a set of random process parameter combinations is created as the initial population, and each individual in the initial population includes three parameters: laser power P, powder feeding rate T, and travel speed V; the size of the initial population is determined; the individuals in the initial population are sorted to form different non-dominated hierarchies; the crowding distance is calculated between individuals in the same non-dominated hierarchy; wherein the fitness is represented by the non-dominated hierarchy and the crowding distance. Thus, the initial population is evaluated by the fitness. The population size can be set as needed, such as 100 individuals, so as to ensure the diversity of individuals in the population.

[0089] Among them, non-dominated sorting is based on the concept of Pareto optimality, which stratifies individuals in the population according to their target values. Specifically, the dominance relationship means that if an individual is not inferior to another individual in all goals and is superior to the individual in at least one goal, the former dominates the latter. Non-dominated hierarchy means that individuals in the population are divided into multiple non-dominated hierarchies, the first level contains all individuals that are not dominated by other individuals, the second level contains those that are only dominated by individuals in the first level, and so on.

[0090] The calculation steps of non-dominated sorting are as follows:

[0091] Initialize and create an empty non-dominated hierarchy list fronts. Traverse the population and for each individual, check whether it is dominated by other individuals. Assign levels, put individuals that are not dominated by any individual into the first level, then remove these individuals from the population, and continue to process the remaining individuals until all individuals are assigned to a level.

[0092] Crowding distance is used to maintain diversity among individuals in the same non-dominated level. Specifically: Crowding distance means that the crowding distance of each individual is the sum of the distances between its "neighbors" on each target. The larger the distance, the sparser the individuals around the individual, and thus the more likely they are to be retained. Boundary individuals refer to individuals at the boundary of the target value, and their crowding distance is set to infinity to ensure that they are not easily eliminated.

[0093] The calculation steps of crowding distance are as follows:

[0094] Initialize, set the initial crowding distance of each individual to 0. Sort by target, for each target, sort the individuals according to the target value. Calculate the crowding distance, for each target, calculate the distance between adjacent individuals and add it to the crowding distance. Non-dominated sorting and crowding distance calculation are often used in combination to select the next generation of population.

[0095] Step 109, performing non-dominated sorting on the initial population to form a Pareto front;

[0096] Specifically, if the number of individuals in the last level exceeds the limit, selection is performed based on the crowding distance, and individuals with larger crowding distance are retained;

[0097] The individuals in the initial population are divided into multiple non-dominated levels, the first level includes all individuals that are not dominated by other individuals, and the second level includes individuals that are only dominated by individuals in the first level; wherein the first level in the initial population is the Pareto front.

[0098] Step 110, generating the next generation population through genetic operation according to the fitness, repeating the iteration until the target population is determined, and forming a Pareto frontier set;

[0099] The specific steps to generate the next generation population are as follows:

[0100] Select individuals from the initial population to participate in crossover and mutation, generate new individuals through crossover, introduce new individuals through mutation, use the new individuals as the offspring population, use the individuals that have been previously screened for fitness as the parent population, merge the parent and offspring populations, and merge the parent and offspring populations into a new population. Perform non-dominated sorting on the individuals in the new population, and perform non-dominated sorting on the merged population to form multiple non-dominated hierarchies. Select individuals, starting from the first level, and select individuals in sequence until the population size limit is reached. If the number of individuals in the last level exceeds the limit, select according to the crowding distance, and give priority to retaining individuals with larger crowding distances.

[0101] Specifically, when selecting the next generation population, individuals with lower non-dominated levels are selected. If two individuals are at the same level, the individual with a larger crowding distance is selected to form the next generation population; the Pareto frontier is determined from the next generation population; the Pareto frontiers of multiple populations form a Pareto frontier set.

[0102] Step 111, determining a target process parameter combination according to the Pareto front set and a new search strategy.

[0103] Specifically, a new search strategy is determined based on a defect minimization strategy, a quality maximization strategy, a comprehensive scoring strategy, a user preference-based strategy, or a diversity-based strategy; and a target process parameter combination is determined based on the new search strategy.

[0104] Specifically, in order to determine the final target process parameter combination from the Pareto front set, a new search strategy can be introduced. For example, one of the following strategies can be selected: Minimize defect strategy: select individuals with the least number of quality grades D and C. Maximize quality strategy: select individuals with the largest number of quality grades A and B. Comprehensive scoring strategy: combine multiple objectives and use weighted sum method or other methods to calculate the comprehensive score.

[0105] Therefore, the target process parameter combination can be determined through a new search strategy, which can not only improve the optimization efficiency, but also ensure that the optimal solution set that meets multiple objectives is found, and the most appropriate process parameter combination is selected according to actual needs.

[0106] See also Figure 4 , Steps 108-110 utilize the NSGA-II algorithm. In step 111, based on the NSGA-II algorithm, the search strategy first searches for the sample subset of level D, then searches for the sample subset of level A, and then searches for the sample subsets of level C and level B in turn according to the importance of each quality level, so as to achieve iterative optimization of the algorithm and give the optimal process parameter combination. The search strategy is based on the generated Pareto solution set, thereby avoiding the algorithm from falling into the local optimal solution.

[0107] By applying the processing method of laser metal deposition manufacturing process parameters provided by the embodiment of the present invention, the data-driven model is combined with the multi-objective optimization algorithm, expensive and time-consuming multiple test experiments are eliminated, an optimal solution search strategy is established, the optimal process parameters are automatically searched from the Pareto solution set, the coupling between multiple quality objectives is balanced, and multi-objective optimization is achieved. This method can improve the accuracy and efficiency of process parameter optimization in the manufacturing process, and solve the problem that a large number of experiments are required in manufacturing process monitoring and multiple process parameter objectives are difficult to optimize.

[0108] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0109] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0110] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for processing process parameters of laser metal deposition manufacturing, characterized in that: The method for processing the laser metal deposition manufacturing process parameters includes: Obtain multiple process parameter information in laser metal deposition manufacturing multi-process parameter deposition experiment; According to the first algorithm, respectively determine the influence of each process parameter information among the plurality of process parameter information on the forming quality; the influence is represented by a quality grade; According to the influence, determining the influence range of each process parameter information, and taking the influence range as a constraint condition; Combining the plurality of process parameter information to determine a plurality of groups of process parameter combinations; Determining the number of each quality level in each set of process parameter combinations according to the second algorithm and the objective function; According to the number of each quality grade and each set of process parameter combinations, the objective function of process parameter optimization is constructed; Predicting the number of quality levels corresponding to each process parameter information according to a preset data-driven prediction model and the objective function; A set of random process parameter combinations is used as an initial population, and the fitness of the initial population is calculated; Performing non-dominated sorting on the initial population to form a Pareto front; According to the fitness, the next generation population is generated through genetic operation, and the iteration is repeated until the target population is determined, thereby forming a Pareto frontier set; According to the Pareto frontier set, a target process parameter combination is determined according to a new search strategy.

2. The method according to claim 1, characterized in that The process parameter information includes laser power, powder feeding rate and travel speed; the influence of each process parameter information on the forming quality is specifically determined according to the first algorithm: Through the variable control method, any one of the laser power, powder feeding rate and travel speed is adjusted, and the other two are fixed, and the influence on the forming quality when one process parameter information is changed and the other two process parameter information remain unchanged is determined, and multiple groups of quality grades are obtained; wherein each group of the quality grades includes multiple quality levels, and each quality level has a quantity; the first algorithm is the control variable method.

3. The method according to claim 1, characterized in that Determining the influence range of each process parameter information according to the influence, and taking the influence range as a constraint condition specifically includes: According to the influence on the forming quality, the upper limit and the lower limit of the influence of each process parameter information are determined, and the upper limit and the lower limit of each process parameter information are used as a constraint condition of the process parameter information; The constraints corresponding to each process parameter information constitute the final constraints.

4. The method according to claim 1, characterized in that: Determining the number of each quality level in each set of process parameter combinations according to the second algorithm and the objective function specifically includes: training the second algorithm by the constraint conditions and the quantity of each quality level; Input multiple sets of process parameter combinations into the second algorithm to predict the number of quality levels corresponding to each set of process parameter combinations; wherein the second algorithm is a label propagation algorithm.

5. The method according to claim 1, characterized in that The objective function of optimizing the process parameters is constructed by taking the minimum number of each quality grade as the sub-objective of the objective function and according to the number of each process parameter information and each set of process parameter combinations. Specifically, the objective function includes: According to constructing the objective function; Where P is the laser power, T is the powder feeding rate, V is the travel speed, and f Level-A 、f Level-B 、f Level-C and f Level-D Represents the number of quality grades A, B, C, and D respectively.

6. The method according to claim 1, characterized in that Determining the number of quality levels corresponding to each process parameter information according to the preset data-driven prediction model and the objective function specifically includes: The historical laser power, powder feeding rate, and travel speed are used as input features, and the number of quality grades is used as output. The experimental data is divided into a training set and a test set using a random forest regression model, and the model is trained to predict the number of quality grades under different process parameters. Through cross validation, mean square error and R 2 Any one of the scores is used as a model evaluation indicator to perform performance evaluation on the random forest regression model, and when the evaluation passes, it is determined that the training of the random forest regression model is completed; A set of process parameter combinations in the objective function are input into the trained random forest regression model to determine the number of quality levels corresponding to each process parameter information.

7. The method according to claim 1, characterized in that The step of taking a set of random process parameter combinations as an initial population and calculating the fitness of the initial population specifically includes: A set of random process parameter combinations is created as the initial population, each individual includes three parameters: laser power P, powder feeding rate T, and travel speed V; determining the size of the initial population; Sort the individuals in the initial population to form different non-dominated hierarchies; The crowding distance is calculated between individuals in the same non-dominated hierarchy; wherein the fitness is represented by the non-dominated hierarchy and the crowding distance.

8. The method according to claim 7, characterized in that The non-dominated sorting of the initial population according to the fitness to form the Pareto front specifically includes: If the number of individuals in the last level exceeds the limit, selection is performed based on the crowding distance, and individuals with larger crowding distance are retained; The individuals in the initial population are divided into multiple non-dominated levels, the first level includes all individuals that are not dominated by other individuals, and the second level includes individuals that are only dominated by individuals in the first level; wherein the first level in the initial population is the Pareto front.

9. The method according to claim 1, characterized in that: The generation of the next generation population through genetic operation according to the fitness, and repeated iteration until the target population is determined, thereby forming a Pareto frontier set specifically includes: When selecting the next generation of population, select individuals with lower non-dominated levels. If two individuals are at the same level, select the individual with larger crowding distance to form the next generation of population. Determine the Pareto front from the next generation population; The Pareto fronts of multiple populations form a Pareto front set.

10. The method according to claim 1, characterized in that Determining the target process parameter combination according to the Pareto frontier set and the new search strategy specifically includes: Determine a new search strategy based on minimizing defects, maximizing quality, comprehensive scoring, user preference, or diversity; According to the new search strategy, the target process parameter combination is determined.

Images