Multi-target process parameter optimization method for laser powder bed melting process based on Q learning improved CMOCSO algorithm

Abstract

The invention relates to a multi-target process parameter optimization method for a laser powder bed melting process based on a Q learning improved CMOCSO algorithm, and belongs to the technical field of laser manufacturing and intelligent manufacturing. The method mainly comprises the following steps: S1, constructing a laser powder bed melting process multi-target process parameter optimization model comprehensively considering carbon emission, processing quality and processing efficiency; s2, adopting a Q learning-based improved CMOCSO algorithm to carry out multi-objective optimization model solving; and S3, selecting a Pareto solution set based on the entropy weight-TOPSIS. According to the method, carbon emission in the melting process of the laser powder bed is effectively reduced while the quality and the efficiency are guaranteed, and powerful support is provided for quality improvement, efficiency improvement and carbon reduction in the laser additive manufacturing process.

Description

Technical Field

[0001] This invention relates to a multi-objective process parameter optimization method for laser powder bed melting process based on Q-learning improved CMOCSO algorithm, belonging to the field of laser manufacturing and intelligent manufacturing technology. Background Technology

[0002] Laser powder bed melting (LBD) is currently one of the representative technologies in additive manufacturing. It uses a high-power laser beam to melt metal powder material layer by layer to construct complex three-dimensional objects, and has been widely used in aerospace, automotive, and medical fields. However, there is efficiency loss in the process of converting electrical energy into laser energy, with a large amount of energy dissipated as heat, generating significant carbon emissions. Therefore, proposing a multi-objective process parameter optimization method that considers carbon emissions is of great significance for the green development of additive manufacturing.

[0003] Existing research on LPBF (Laser Bed Fusion) technology has limited multi-objective optimization models that simultaneously consider carbon emissions, production efficiency, and quality. Therefore, incorporating carbon emissions, production efficiency, and quality as the core of multi-objective optimization not only better meets the actual production needs of enterprises but also reduces carbon emissions and costs while ensuring part quality. This multi-objective optimization method provides enterprises with more comprehensive decision support in the additive manufacturing field, promoting the development of low-carbon, efficient, and sustainable manufacturing models. Summary of the Invention

[0004] To address the aforementioned problems, this invention designs a multi-objective process parameter optimization method for laser powder bed melting processes based on an improved CMOCSO algorithm (Constrained Multi-Objective Optimization Competitive Cooperative Group Optimization Algorithm) using Q-learning. The characteristics of the laser powder bed melting process are analyzed to establish a multi-objective process parameter optimization model with carbon emissions, printing efficiency, and processing quality as optimization objectives. Based on this, an improved CMOCSO multi-objective solution algorithm based on Q-learning is proposed to improve solution quality and efficiency. Finally, an entropy-weighted TOPSIS (Approximation of Ideal Solution Ranking Method) Pareto solution set selection method is used to obtain the optimal combination of process parameters. Experiments verify the effectiveness and feasibility of the model and optimization results.

[0005] This invention relates to a multi-objective process parameter optimization method for laser powder bed melting process based on Q-learning improved CMOCSO algorithm, comprising the following steps:

[0006] S1: Construct a multi-objective process parameter optimization model for the laser powder bed melting process that comprehensively considers carbon emissions, processing quality, and processing efficiency;

[0007] S2: The multi-objective process parameter optimization model is solved by using the Q-learning-based improved CMOCSO algorithm;

[0008] S3: Pareto solution set selection based on entropy weight-TOPSIS method.

[0009] Preferably, step S1 includes the following sub-steps:

[0010] S11: Power measurements and fitting analyses were performed on each subsystem of the LPBF. Based on the carbon emission generation mechanism of each LPBF subsystem, the carbon emission objective function for the LPBF manufacturing process was constructed as follows:

[0011]

[0012] In the formula, T s T represents the total time the scraper moves. m The time required for laser printing one layer, T z h represents the platform printing (Z-axis rise) descent time, and h represents the part height. s For a single layer thickness, S m V represents the distance the scraper travels. s V is the scraper movement speed. s1 V is the moving speed of the brush for spreading powder. s2 V is the return travel speed of the tool. z To reduce the printing platform descent speed, V p The vertical speed of the powder feeding platform is m, the length of the workpiece is n, the width of the workpiece is D, the laser scanning distance is V, the laser moving speed is C poc It is a carbon emission factor for electricity.

[0013] S12: A function of the build efficiency S and the print time t*, with powder layer thickness h. s Setting the value to 0.03mm, the objective function for LPBF manufacturing efficiency can be obtained as follows:

[0014] S = 0.03 * D * V * 3600

[0015] In the formula, V is the laser scanning speed in mm / s, and D is the scanning distance in mm.

[0016] S13: For 6061 aluminum alloy raw material powder, the number of pores with a diameter greater than 10μm is used as the quality standard for LPBF formed parts. Data fitting is performed on the pore size to obtain the LPBF quality objective function:

[0017] T=exp(-110.716493932921+0.364971644657P+0.129157357531V

[0018] +1.358114995374D-0.000679200093P 2-0.000080097064V 2

[0019] -0.003148605630D 2 -0.003063481883PD-0.001424403143VD

[0020] +0.000000444463PPP-0.000000251339PPV

[0021] +0.000004599196PPD+0.000000111856PVV

[0022] +0.000000447720VVD+0.000003751448VDD)

[0023] In the formula, P is the laser power, D is the scanning distance, and V is the laser scanning speed.

[0024] S14: Apply corresponding constraints to the laser power, scanning speed, and scanning spacing process parameters, and simultaneously construct microhardness constraint conditions through data fitting:

[0025] P min ≤P≤P max

[0026] V min ≤V≤V max

[0027] D min ≤D≤D max

[0028] 60HV≤μ≤120HV

[0029] μ=173.8313737-1.361576699P+0.00413014479P 2 +0.005173554075D 2

[0030] +0.000114294343PV-0.00000462336935P 3 +0.000000352796V 2 D

[0031] -0.000007371381VD 2

[0032] In the formula, P min P is the lower limit of laser input power. max V is the upper limit of laser input power. minSet a lower limit for the scan speed, V max To set an upper limit for the scanning speed, D min Set a lower limit for the scan spacing, D max Set an upper limit for the scanning spacing, where P is the laser power, D is the scanning spacing, and V is the laser scanning speed.

[0033] S15: Establish a multi-objective optimization model with carbon emissions, manufacturing efficiency, and processing quality as objectives.

[0034] F(P,V,D)={min(C A ),max(S),min(T)}

[0035]

[0036] Preferably, step S2 includes the following sub-steps:

[0037] S21: Build the original CMOCSO algorithm framework to solve multi-objective optimization models;

[0038] S22: Proposes a faster convergence update strategy CCSO:

[0039]

[0040] In the formula, m is the number of targets, T is the current algebra, and T c This represents the maximum number of generations. σ increases with the generation number t, from m... 2 As the value of σ decreases from +1 to 1, the interest rate decreases in the early stages of evolution and decreases slightly in the later stages. This results in a larger expansion in the early stages to accelerate convergence to the CPF, and a smaller expansion in the later stages to ensure the utilization of the CPF.

[0041] The proposed cooperative swarm optimizer is based on the arithmetic crossover operator, particle v w and x w The strategy is:

[0042]

[0043] In the formula, a i For each variable, randomly generate 0 or 1, l = 1, 2, ..., n, in this case, x w and x l It's a random swap.

[0044] S23: Q-Learning (QL) is introduced into the CMOCSO algorithm framework to assist in selecting the most suitable method for the task and avoid getting trapped in local optima. In QL, states, actions, and rewards are set so that the agent can learn the Q-Table and make corresponding decisions. The current position and velocity of each particle are used as the state; actions are defined as the particle moving towards a better solution (such as the optimal solution in the current iteration or a neighboring solution); a positive reward is given when the particle moves to a solution with a better objective function value and satisfies the constraints, and a negative reward is given otherwise; in each iteration, the particle can choose the optimal action (i.e., the direction and distance of movement) based on the current Q value, thereby gradually approaching the optimal solution.

[0045] The operation is performing an auxiliary task:

[0046] A = {a|a∈{T1,T2,...,T} i}}

[0047] In the formula, T1 to T i This represents the execution of the corresponding task, where i is the number of auxiliary tasks, and the state includes the currently used auxiliary tasks, which include:

[0048] S={s|s∈{(T,T1),(T,T2),...,(T,T i )}}

[0049] In the formula, (T, T1) indicates that the current algorithm uses T1 as an auxiliary task to execute the j-th task T. j The reward is from T j The transfer ratio to T is defined as follows:

[0050]

[0051] In the formula, P j It is T j population, P t j It is the set of solutions that are transferred to T (i.e., the next generation is selected to T), and then the Q-table is updated using the Bellman Function:

[0052] Q(s,a)=Q(s,a)+α[r+γmax a' Q(s',a')-Q(s,a)]

[0053] In the formula, a is the learning rate, γ is the discount factor, and the Bellman function is the core idea of ​​QL. It considers both the real-time reward of performing an action in a given state and the expected future reward of taking the next action. This Q-Table can estimate the Q value of performing any auxiliary task in any state.

[0054] Preferably, step S3 includes the following sub-steps:

[0055] S31: Standardize the maximum and minimum values ​​of the original data;

[0056] S32: The weights are determined using the entropy weight method;

[0057] S33: Apply the TOPSIS method for comprehensive evaluation.

[0058] This invention effectively reduces carbon emissions during the laser powder bed melting process while ensuring quality and efficiency, providing strong support for improving quality, efficiency, and reducing carbon emissions in laser additive manufacturing. Attached Figure Description

[0059] Figure 1 This is a schematic diagram illustrating the division of the laser powder bed melting stage in this invention;

[0060] Figure 2 This is a diagram of the real-time power monitoring platform for the laser powder bed melting process of the present invention;

[0061] Figure 3 This is a power variation curve of the printing platform module of the present invention;

[0062] Figure 4 This is a power variation curve of the powder feeding platform module of the present invention;

[0063] Figure 5 This is a graph showing the power variation of the laser module of the present invention.

[0064] Figure 6 This is a flowchart of the improved CMOCSO algorithm based on Q-learning in this invention;

[0065] Figure 7 This is the Pareto solution set graph after solving the problem using the improved algorithm of this invention;

[0066] Figure 8 This is a printed image showing the result before optimization of the laser powder bed melting parameters of this invention;

[0067] Figure 9 This is a printing result diagram after optimizing the laser powder bed melting parameters of the present invention. Detailed Implementation

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the embodiments are used to explain the present invention and are not intended to limit the present invention.

[0069] This invention discloses a multi-objective process parameter optimization method for laser powder bed melting process based on Q-learning improved CMOCSO algorithm. Figure 1 A schematic diagram showing the division of the laser powder bed melting stage; Figure 2Diagram of a real-time power monitoring platform for the laser powder bed melting process; Figure 3 A graph showing the power variation of the printing platform module; Figure 4 The graph shows the power variation of the powder delivery platform module. Figure 5 This is a graph showing the power variation of the laser module. Figure 6 The flowchart shows the improved CMOCSO algorithm based on Q-learning. Figure 7 To improve the Pareto solution set graph obtained after solving the algorithm; Figure 8 Printing results before laser powder bed melting parameters were optimized; Figure 9 The printing result is shown after optimizing the laser powder bed melting parameters.

[0070] The overall technical solution of this invention is a multi-objective process parameter optimization method for laser powder bed melting process based on Q-learning improved CMOCSO algorithm, comprising the following steps:

[0071] S1: Construct a multi-objective process parameter optimization model for the laser powder bed melting process that comprehensively considers carbon emissions, processing quality, and processing efficiency;

[0072] S2: The multi-objective process parameter optimization model is solved by using the Q-learning-based improved CMOCSO algorithm;

[0073] S3: Selection of Pareto solution set based on entropy weight-TOPSIS method;

[0074] S4: Case Analysis.

[0075] Step S1 includes the following sub-steps:

[0076] S11: Perform power measurement and fitting analysis on each subsystem of LPBF, and construct the carbon emission objective function of LPBF manufacturing process in combination with the carbon emission generation mechanism of each subsystem of LPBF.

[0077] S111: The LPBF manufacturing process is divided into the following time nodes t1, t2, t3, and t4. The time before t1 is the preparation stage; t1-t2 is the start-up standby stage; t2-t3 is the actual additive manufacturing stage, in which powder spreading and laser printing processes are repeatedly performed. t5-t6 is set as the time for one layer of powder spreading, and t7-t8 is set as the time for one layer of laser printing. t3-t4 is the end standby stage.

[0078] Establish the powder spreading time function T S (t5-t6) is:

[0079]

[0080] In the formula, S m V represents the distance the scraper travels. s This refers to the scraper's moving speed;

[0081] Laser printing time function T M (t7-t8) is:

[0082]

[0083] In the formula, m is the length of the workpiece, n is the width of the workpiece, d is the diameter of the laser spot, α is the overlap ratio, and v is the laser moving speed;

[0084] Platform printing (Z-axis rise) descent time function T Z for:

[0085]

[0086] In the formula, h s For a single layer thickness, V Z To reduce the printing platform descent speed;

[0087] The additive manufacturing time function is T A (t2-t3):

[0088]

[0089] S112: Constructing an energy consumption model for the laser powder bed melting process:

[0090] E = E F +E A +E E

[0091]

[0092] E A =E S +E P +E Z +E I +E C +E L +E G +E H +E W

[0093]

[0094] In the formula, E F To reduce power consumption during standby, E A For energy consumption in the additive manufacturing stage, E E To end the standby phase energy consumption, E S For the energy consumption of the powder spreading module, E P For the energy consumption of the powder supply module, E H For the energy consumption of the high-pressure blower module, E Z For the power consumption of the printing platform module, EI For the energy consumption of the heating plate module, E W For the energy consumption of the water chiller module, E C For the control panel module energy consumption, E L For the energy consumption of the laser module, E G This refers to the energy consumption of the galvanometer module.

[0095] S113: Establish an LPBF equipment energy consumption testing and monitoring platform, such as... Figure 2 ;

[0096] S114: The power variation function of the printing platform module is obtained by fitting the data in Table 1:

[0097] Y Z =2.069X Z +1.963

[0098] In the formula, Y Z For printing platform power, X Z For printing platform movement speed;

[0099] Table 1. Power Variation of Printing Platform

[0100]

[0101] S115: The power variation function of the powder delivery platform module is obtained by fitting the data in Table 2:

[0102] Y P =1.346X P +0.75

[0103] In the formula, Y P X is the output power of the powder delivery platform. P For the movement speed of the powder delivery platform;

[0104] Table 2 Power Variation of Powder Feeding Platform

[0105]

[0106] S116: The power variation function of the laser module is obtained by fitting the data in Table 3:

[0107] Y L =2.198X L +1612

[0108] In the formula, Y L X is the output power of the laser module. L The output power of the laser module; Table 3 shows the power variation of the laser module.

[0109]

[0110]

[0111] S117: Obtain the average operating power value of other modules;

[0112] S118: Constructing a carbon emission model for the laser powder bed melting process based on Tables 4 and 5:

[0113]

[0114] In the formula, T s T represents the total time the scraper moves. m h represents the time required to laser print one layer, and h represents the height of the part. s For a single layer thickness, V s1 V is the moving speed of the brush for spreading powder. s2 V is the return travel speed of the tool. z To reduce the printing platform descent speed, V p To accelerate the growth of the fan-giving platform, C poc It is a carbon emission factor for electricity.

[0115] Table 4. Laser Powder Bed Melting Power Parameters

[0116]

[0117] Table 5 Other Test Parameters for Equipment

[0118]

[0119] S12: Establish a mapping model between production efficiency and process parameters:

[0120]

[0121] In the formula, S represents the production efficiency during the manufacturing process, V* represents the volume of the printed part, t* represents the time consumed to print one part, m represents the part length, n represents the part width, D represents the laser scanning spacing, V represents the laser scanning speed, and h represents the part height. s For the thickness of the printed part layers;

[0122] Table 6. Efficiency Quantification Table of Orthogonal Experiment for Process Parameters

[0123]

[0124]

[0125] Based on the efficiency S of the construction and the printing time t*, the powder layer thickness h s Setting the value to 0.03mm, the objective function for LPBF manufacturing efficiency can be obtained as follows:

[0126] S = 0.03 * D * V * 3600

[0127] In the formula, D is the laser scanning speed in mm / s, and V is the scanning distance in mm.

[0128] S13: The mapping model between process parameters and pore size is obtained by fitting the data in Table 7:

[0129] T=exp(-110.716493932921+0.364971644657P

[0130] +0.129157357531V+1.358114995374D

[0131] -0.000679200093P 2 -0.000080097064V 2

[0132] -0.003148605630D 2 -0.003063481883PD

[0133] -0.001424403143VD+0.000000444463PPP

[0134] -0.000000251339PPV+0.000004599196PPD

[0135] +0.000000111856PVV+0.000000447720VVD

[0136] +0.000003751448VDD

[0137] Table 7. Statistics on Pore Size in Experiments with Multiple Process Parameters

[0138]

[0139]

[0140] S14: The mapping model between process parameters and microhardness obtained by fitting the data in Table 8 is: μ=173.8313737-1.361576699P+0.00413014479P 2 +0.005173554075D 2 +0.000114294343PV-0.00000462336935P 3 +0.000000352796V 2 D-0.000007371381VD 2

[0141] Table 8. Statistical Table of Hardness Measurement of Parts with Multiple Process Parameter Groups

[0142]

[0143] S15: Establish a multi-objective optimization model with carbon emissions, printing efficiency, and processing quality as objectives.

[0144] F(P,V,D)={min(C A ),max(S),min(T)}

[0145]

[0146] Step S2 includes the following sub-steps:

[0147] S21: Design of an improved CMOCSO solution algorithm based on Q-learning, the algorithm flowchart is as follows. Figure 6 As shown.

[0148] S22: Proposes a faster convergence update strategy CCSO:

[0149]

[0150] In the formula, m is the number of targets, T is the current algebra, and T c This represents the maximum number of generations. σ increases with the generation number t, from m... 2 As the value of σ decreases from +1 to 1, the interest rate decreases in the early stages of evolution and decreases slightly in the later stages. This results in a larger expansion in the early stages to accelerate convergence to the CPF, and a smaller expansion in the later stages to ensure the utilization of the CPF.

[0151] The proposed cooperative swarm optimizer is based on the arithmetic crossover operator, particle v w and x w The strategy is:

[0152]

[0153] In the formula, a i For each variable, randomly generate 0 or 1, l = 1, 2, ..., n, in this case, x w and x l It's a random swap.

[0154] S23: Q-Learning (QL) is introduced into the CMOCSO algorithm framework to assist in selecting the most suitable method for the task and avoid getting trapped in local optima. In QL, states, actions, and rewards are set so that the agent can learn the Q-Table and make corresponding decisions. The current position and velocity of each particle are used as the state; actions are defined as the particle moving towards a better solution (such as the optimal solution in the current iteration or a neighboring solution); a positive reward is given when the particle moves to a solution with a better objective function value and satisfies the constraints, and a negative reward is given otherwise; in each iteration, the particle can choose the optimal action (i.e., the direction and distance of movement) based on the current Q value, thereby gradually approaching the optimal solution.

[0155] The operation is performing an auxiliary task:

[0156] A = {a|a∈{T1,T2,...,T} i}}

[0157] In the formula, T1 to T i This represents the execution of the corresponding task, where i is the number of auxiliary tasks, and the state includes the currently used auxiliary tasks, which include:

[0158] S={s|s∈{(T,T1),(T,T2),...,(T,T i )}}

[0159] In the formula, (T, T1) indicates that the current algorithm uses T1 as an auxiliary task to execute the j-th task T. j The reward is from T j The transfer ratio to T is defined as follows:

[0160]

[0161] In the formula, P j It is T j population, P t j It is the set of solutions that are transferred to T (i.e., the next generation is selected to T), and then the Q-table is updated using the Bellman Function:

[0162] Q(s,a)=Q(s,a)+α[r+γmax a' Q(s',a')-Q(s,a)]

[0163] In the formula, a is the learning rate, γ is the discount factor, and the Bellman function is the core idea of ​​QL. It considers both the real-time reward of performing an action in a given state and the expected future reward of taking the next action. This Q-Table can estimate the Q value of performing any auxiliary task in any state.

[0164] Step S3 includes the following sub-steps:

[0165] S31: Transform the raw data into dimensionless numerical values ​​and standardize the maximum and minimum values.

[0166] Positive indicators:

[0167] Y ij =(x ij -min(x ij )) / (max(x ij )-min(x ij ))

[0168] Negative indicators:

[0169] Y ij =(max(x) ij )-x ij ) / (max(x ij )-min(x ij ))

[0170] In the formula, x ij Y represents the original value of the i-th evaluation object on the j-th indicator. ij This represents the standardized numerical value.

[0171] S32: Determining weights using the entropy weight method:

[0172] Calculate the proportion values ​​of each indicator:

[0173]

[0174] Calculate the information entropy of each indicator:

[0175]

[0176] In the formula, when P ij When = 0, it is stipulated that Pij*ln(Pij) = 0. Calculate the difference coefficients for each indicator:

[0177] G j =1-E j

[0178] In the formula, G j E is the difference coefficient of the j-th indicator. j It is the information entropy of the j-th indicator, which measures the degree of disorder or uncertainty of the information contained in the indicator.

[0179] Calculate the weight of each indicator:

[0180]

[0181] In the formula, wj Let be the weight of the j-th indicator, and n be the total number of indicators. The weight of each indicator is proportional to its difference coefficient, meaning that the higher the discrimination of an indicator, the greater its weight in the comprehensive evaluation.

[0182] S33: Apply the TOPSIS method for comprehensive evaluation:

[0183] Constructing a weighted normalized decision matrix:

[0184] Z ij =W j +Y ij

[0185] Determine the ideal solution (Taking the maximum value for all evaluation objects i) and negative ideal solution (Minimum value for all evaluated object i):

[0186]

[0187] Calculate the distance between each evaluation object and the ideal solution. Distance to the negative ideal solution

[0188]

[0189] Calculate the relative proximity C of each evaluation object. i The relative proximity value ranges from [0,1], where 1 indicates that the evaluated object completely coincides with the ideal solution, and 0 indicates that the evaluated object completely coincides with the negative ideal solution. In practical applications, the evaluated object with a higher relative proximity is usually considered to be better.

[0190]

[0191] S34: Using 70mm*70mm*10mm parts as test pieces and a scanning layer thickness of 0.03mm, Matlab was used as the algorithm platform. An improved COMCSO algorithm was applied to solve the multi-objective model. The population size was set to 100, and the number of iterations was 10000. The objectives were to minimize carbon emissions, maximize efficiency, and minimize the number of pores, with hardness and pore diameter as constraints. The optimal Pareto solution set was selected based on entropy-weighted TOPSIS. Figure 7 As shown in Table 9, selecting the LPBF process parameters (P,V,D) = (200,1000,0.108) yields the best optimization results.

[0192] Table 9 Comparison of Optimization Results

[0193]

[0194] Step S4 includes the following sub-steps:

[0195] S41: Experiments were conducted using 6061 aluminum alloy powder with the process parameters shown in Table 10. Two groups of printed parts were set up: one group with optimized parameters and the other with empirical parameters. Both groups of test parts were 70mm*70mm*10mm, with a scanning layer thickness of 0.03mm. All other conditions were the same. The experimental results are shown in Tables 11, 12, and 13. The comparative results show that after optimizing the process parameters, carbon emissions decreased by 13.3%, pore diameter decreased by 23.03%, and manufacturing efficiency increased by 8%. Figure 8 Figure 9 Printing results before and after optimization of laser powder bed melting parameters.

[0196] Table 10 Verification Control Parameter Combination Table

[0197]

[0198] Table 11 Comparison of experimental carbon emission values ​​for the two sets of parameters

[0199]

[0200] Table 12. Statistics on porosity of molded parts with two sets of parameters.

[0201]

[0202] Table 13 Efficiency statistics of molded parts with two sets of parameters

[0203]

[0204] The above description represents preferred embodiments of the present invention and is not intended to limit the invention. Those skilled in the art can still modify the above technical solutions or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing multi-objective process parameters in laser powder bed melting process based on Q-learning improved CMOCSO algorithm, characterized in that, Includes the following steps: S1: Construct a multi-objective process parameter optimization model for the laser powder bed melting process that comprehensively considers carbon emissions, processing quality, and processing efficiency; S2: The multi-objective process parameter optimization model is solved by using the Q-learning-based improved CMOCSO algorithm; S3: Pareto solution set selection based on entropy weight-TOPSIS method.

2. The method for optimizing multi-objective process parameters of laser powder bed melting process based on Q-learning improved CMOCSO algorithm according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11: Power measurements and fitting analyses were performed on each subsystem of the LPBF. Based on the carbon emission generation mechanism of each LPBF subsystem, the carbon emission objective function for the LPBF manufacturing process was constructed as follows: In the formula, T s T represents the total time the scraper moves. m The time required for laser printing one layer, T z h represents the platform printing (Z-axis rise) descent time, and h represents the part height. s For a single layer thickness, S m V represents the distance the scraper travels. s V is the scraper movement speed. s1 V is the moving speed of the brush for spreading powder. s2 V is the return travel speed of the tool. z To reduce the printing platform descent speed, V p The vertical speed of the powder feeding platform is m, the length of the workpiece is n, the width of the workpiece is D, the laser scanning distance is V, the laser moving speed is C poc It is a carbon emission factor for electricity. S12: A function of the build efficiency S and the print time t*, with powder layer thickness h. s Setting the value to 0.03mm, the objective function for LPBF manufacturing efficiency can be obtained as follows: S = 0.03 * D * V * 3600 In the formula, V is the laser scanning speed in mm / s, and D is the scanning distance in mm. S13: For 6061 aluminum alloy raw material powder, the number of pores with a diameter greater than 10μm is used as the quality standard for LPBF formed parts. Data fitting is performed on the pore size to obtain the LPBF quality objective function: T=exp(-110.716493932921+0.364971644657P+0.129157357531V +1.358114995374D-0.000679200093P 2 -0.000080097064V 2 -0.003148605630D 2 -0.003063481883PD-0.001424403143VD +0.000000444463PPP-0.000000251339PPV +0.000004599196PPD+0.000000111856PVV +0.000000447720VVD+0.000003751448VDD) In the formula, P is the laser power, D is the scanning distance, and V is the laser scanning speed. S14: Apply corresponding constraints to the laser power, scanning speed, and scanning spacing process parameters, and simultaneously construct microhardness constraint conditions through data fitting: P min ≤P≤P max In min ≤V≤V max D min ≤D≤D max 60HV≤μ≤120HV μ=173.8313737-1.361576699P+0.00413014479P 2 +0.005173554075D 2 +0.000114294343PV-0.00000462336935P 3 +0.000000352796V 2 D -0.000007371381VD 2 In the formula, P min P is the lower limit of laser input power. max V is the upper limit of laser input power. min Set a lower limit for the scan speed, V max To set an upper limit for the scanning speed, D min Set a lower limit for the scan spacing, D max Set an upper limit for the scanning spacing, where P is the laser power, D is the scanning spacing, and V is the laser scanning speed. S15: Establish a multi-objective optimization model with carbon emissions, manufacturing efficiency, and processing quality as objectives. F(P,V,D)={min(C A ),max(S),min(T)} 3. The method for optimizing multi-objective process parameters of laser powder bed melting process based on Q-learning improved CMOCSO algorithm according to claim 1 or 2, characterized in that, Step S2 includes the following sub-steps: S21: Build the original CMOCSO algorithm framework to solve multi-objective optimization models; S22: Proposes a faster convergence update strategy CCSO: In the formula, m is the number of targets, T is the current algebra, and T c To maximize the number of generations, σ increases from m... 2 As the value of σ decreases from +1 to 1, the interest calculation decreases in the early stages of evolution and then decreases slightly in the later stages. This results in a larger expansion in the early stages to accelerate convergence to the CPF, and a smaller expansion in the later stages to ensure the utilization of the CPF. The proposed cooperative swarm optimizer is based on the arithmetic crossover operator, particle v w and x w The strategy is: In the formula, a i For each variable, randomly generate 0 or 1, l = 1, 2, ..., n, in this case, x w and x l It is a random exchange; S23: Q-Learning (QL) is introduced into the CMOCSO algorithm framework to assist in selecting the most suitable method for the task and avoid getting trapped in local optima. In QL, states, actions, and rewards are set so that the agent can learn the Q-Table and make corresponding decisions. The current position and velocity of each particle are used as the state; the action is defined as the particle moving towards a better solution (such as the optimal solution in the current iteration or a neighboring solution); when the particle moves to a solution with a better objective function value and meets the constraints, a positive reward is given, otherwise a negative reward is given; in each iteration, the particle can choose the optimal action (i.e., the direction and distance of movement) based on the current Q value, thereby gradually approaching the optimal solution. The operation is performing an auxiliary task: A={a|a∈{T1,T2,...,T i }} In the formula, T1 to T i This represents the execution of the corresponding task, where i is the number of auxiliary tasks, and the state includes the currently used auxiliary tasks, which include: S={s|s∈{(T,T1),(T,T2),...,(T,T i )}} In the formula, (T, T1) indicates that the current algorithm uses T1 as an auxiliary task to execute the j-th task T. j The reward is from T j The transfer ratio to T is defined as follows: In the formula, P j It is T j population, P t j It is the set of solutions that transition to T (i.e., the next generation is selected to T), and then the Q-table is updated using the Bellman function: Q(s,a)=Q(s,a)+α[r+γmax a' Q(s',a')-Q(s,a)] In the formula, a is the learning rate, γ is the discount factor, and the Bellman function is the core idea of ​​QL. It considers both the real-time reward of performing an action in a given state and the expected future reward of taking the next action. The Q-Table can estimate the Q value of performing any auxiliary task in any state.

4. The method for optimizing multi-objective process parameters of laser powder bed melting process based on Q-learning improved CMOCSO algorithm according to claim 3, characterized in that, Step S3 includes the following sub-steps: S31: Standardize the maximum and minimum values ​​of the original data; S32: The weights are determined using the entropy weight method; S33: Apply the TOPSIS method for comprehensive evaluation.

Images