Rapid optimization method for laser cladding process based on NSGA-II (Non-dominated Sorting Genetic Algorithm-II) algorithm

Abstract

The invention provides a laser cladding process dynamic optimization method and system based on NSGA-II and PSO-BPNN cooperation, and belongs to the field of laser additive manufacturing process optimization. The method is characterized in that key process parameters are preliminarily screened through an orthogonal experiment; a BP neural network (PSO-BPNN) optimized by a PSO algorithm is utilized to establish a high-precision nonlinear mapping model between the process parameters and cladding layer quality indexes (aspect ratio, dilution rate and hardness); adopting an NSGA-II algorithm to carry out multi-objective optimization in an initial parameter range; a dynamic range adjustment mechanism is innovatively introduced, when an optimal solution does not meet requirements, the most sensitive technological parameter range of substandard indexes is directionally adjusted based on the range analysis result of an orthogonal experiment, and optimization is carried out again until a satisfactory solution is obtained. According to the method, the problem of repeated experiments caused by improper setting of the initial parameter range in a traditional method is solved, rapid and accurate optimization under a small number of experiments is achieved, and the method is particularly suitable for on-site rapid remanufacturing of large equipment.

Description

Technical Field

[0001] This invention belongs to the field of laser additive manufacturing and remanufacturing technology, and specifically relates to a method for optimizing laser cladding process parameters, and in particular a dynamic optimization method and system that integrates orthogonal experiments, machine learning models and multi-objective evolutionary algorithms. Background Technology

[0002] Laser cladding technology, as a surface strengthening and remanufacturing technology, has advantages such as high metallurgical bonding strength, small heat-affected zone, and precise repair, and is widely used in mechanical parts repair and surface strengthening. However, there is a complex nonlinear relationship between laser cladding process parameters (such as laser power, scanning speed, powder feeding rate, etc.) and cladding layer properties (such as aspect ratio, dilution rate, hardness), making process parameter optimization a challenge.

[0003] Existing optimization methods mainly include orthogonal experimental design and response surface methodology. Orthogonal experiments, by selecting the best within a preset parameter combination, can screen for major influencing factors, but the optimization effect is limited by the experimental range and it is difficult to reflect the nonlinear relationship between parameters and performance. Response surface methodology relies on quadratic polynomial model fitting, which has insufficient prediction accuracy for highly nonlinear problems. If an inappropriate selection of the experimental range leads to poor optimization results, the experimental range needs to be adjusted and repeated experiments conducted. However, each experiment requires cumbersome procedures such as wire cutting and metallographic polishing, which is time-consuming and cannot meet the timeliness requirements of on-site repair.

[0004] In recent years, intelligent algorithms have been introduced into laser cladding optimization. For example, some studies have used response surface methodology combined with particle swarm optimization to optimize process parameters, or used the NSGA-II algorithm for multi-objective optimization of the cladding layer morphology. However, these methods still have shortcomings in fitting nonlinear relationships and global optimization capabilities. To address these issues, this invention proposes a rapid optimization method for laser cladding processes based on orthogonal experiments, BPNN, and the NSGA-II algorithm. This method can dynamically adjust the process range to achieve efficient optimization of the overall performance of the cladding layer. Summary of the Invention

[0005] To address the problems of insufficient nonlinear relationship fitting ability, low optimization efficiency, and difficulty in balancing multi-objective conflicts in existing laser cladding process optimization methods, this invention proposes a rapid optimization method for laser cladding process parameters based on the NSGA-II multi-objective optimization algorithm. The aim is to improve the aspect ratio of the cladding layer, reduce the dilution rate, increase the hardness, and significantly reduce the number of experiments and time costs.

[0006] To address the aforementioned technical problems, the present invention employs a rapid optimization method for laser cladding processes based on the NSGA-II algorithm, comprising the following steps: S1: Design experiments within a preset range of process parameters to obtain multiple combinations of process parameters and their corresponding cladding layer quality index data; S2: Based on the data, construct and train a machine learning model that can map the complex nonlinear relationship between process parameters and cladding layer quality indicators; S3: With the goal of optimizing one or more cladding layer quality indicators, a multi-objective evolutionary algorithm is used to find the best solution within the preset process parameters to obtain the first Pareto optimal solution set; S4: Determine whether the first Pareto optimal solution set meets the preset cladding layer quality requirements; if it does, select the final process parameter combination from it; if it does not, proceed to S5. S5: Dynamic parameter range adjustment step: Identify the quality indicators that do not meet the requirements, and based on the significance ranking of the influence of each process parameter on the indicator in the experimental design, adjust the value range of at least one process parameter with the most significant influence to form a new process parameter space. S6: Within the new process parameter space adjusted in S5, repeat S3 until the obtained Pareto optimal solution set meets the quality requirements.

[0007] Furthermore, in S1, the experimental design is an orthogonal experimental design, the process parameters include at least laser power, scanning speed and powder feeding rate, and the cladding layer quality indicators include at least aspect ratio, dilution rate and hardness.

[0008] Furthermore, in S5, the "ranking of the significance of the influence of each process parameter on the index according to the experimental design" is determined by range analysis of the orthogonal experimental results.

[0009] Furthermore, in S2, the machine learning model is a BP neural network model, and the initial connection weights and thresholds of the BP neural network model are optimized using the particle swarm optimization (PSO) algorithm to form a PSO-BPNN model.

[0010] Furthermore, the determination coefficient R² of the PSO-BPNN model is greater than 0.95.

[0011] Furthermore, in S3, the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm with an elitist strategy (NSGA-II), and its optimization objective function is: Minimize dilution rate: ; Maximize aspect ratio: ; Maximize hardness: ; in, P is the laser power; Vs represents the scan speed; Vr is the powder delivery rate; W / H refers to the aspect ratio; η is the dilution rate; H represents hardness.

[0012] Furthermore, the optimized process parameter combination obtained by the method is: laser power 934W±5%, scanning speed 352mm / min±5%, powder feeding rate 0.64r / min±5%. Applying this parameter combination, an Fe60 alloy cladding layer with an aspect ratio ≥3.0, a dilution rate ≤0.34, and a hardness ≥610HV can be prepared on a Q345 steel substrate.

[0013] This invention can be extended to a laser cladding process optimization system, which includes: The data acquisition module is used to perform S1 in claim 1; The model building and training module is used to perform S2 in claim 1; A multi-objective optimization module is used to execute S3 in claim 1; The dynamic adjustment and decision-making module is used to execute S4, S5 and S6 in claim 1 and output the final optimized combination of process parameters.

[0014] The present invention can be extended to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0015] This invention applies to a Q345 steel substrate part, the surface of which has an Fe60 alloy laser cladding layer prepared by the optimized process parameters. The microstructure of the cladding layer from the substrate interface to the surface consists of: a planar crystal zone with a thickness of 3-5 μm, a columnar crystal zone, and a cladding layer body mainly composed of equiaxed crystals and dendrites. The cladding layer is metallurgically bonded to the substrate and is free of pores and incomplete fusion defects.

[0016] Compared with the prior art, the present invention has the following beneficial effects.

[0017] 1. This invention proposes a dynamic range adjustment mechanism; this is an essential feature that distinguishes it from existing static optimization methods. When the initial optimization result is unsatisfactory, the system can automatically and directionally adjust the parameter range, fundamentally avoiding the predicament of repeated experimentation-adjustment-re-experimentation due to improper initial range settings, and achieving adaptive optimization.

[0018] 2. This invention constructs a high-precision collaborative model; the PSO-BPNN model integrates the powerful mapping capability of the BP neural network and the global optimization capability of the PSO algorithm. Its prediction accuracy (R²>0.95) is significantly higher than that of the traditional response surface model or ordinary BPNN, laying a solid foundation for the reliable optimization of NSGA-II.

[0019] 3. This invention achieves full-chain optimization coverage; from experimental design, model establishment, global optimization to dynamic adjustment, it forms a complete solution, and its results are reflected in the optimized process parameters, the system that implements the method, and the high-performance products prepared using the parameters, thus forming a tight intellectual property protection network.

[0020] 4. This invention has significant industrial application value; it can quickly lock the globally optimal or suboptimal process window through a small number of experiments, which greatly saves time and economic costs, and provides key technical support for the promotion of laser cladding technology in on-site emergency repair and personalized small-batch manufacturing scenarios with extremely high time requirements.

[0021] In summary, this invention addresses the challenges of existing orthogonal experiments and response surface methodology in handling the nonlinear relationship between process parameters and cladding layer performance, as well as the problem of inadequate optimization results due to improper selection of experimental parameter ranges, necessitating repeated experiments. It employs orthogonal experiments to screen key process parameters, uses PSO-BPNN to establish a nonlinear mapping model between process parameters (laser power, scanning speed, powder feeding rate) and cladding layer quality indicators (aspect ratio, dilution rate, hardness), and utilizes the NSGA-II multi-objective optimization algorithm for global optimization of process parameters. By adjusting the original experimental parameter ranges, the optimal combination of process parameters with comprehensive performance is ultimately obtained. This invention achieves rapid optimization of process parameters with a limited number of experiments and is applicable to fields such as on-site repair of large parts and rapid repair of small batches of parts. Attached Figure Description

[0022] The present invention will now be further described with reference to the accompanying drawings.

[0023] Figure 1 This is a schematic diagram of the frame structure of the present invention.

[0024] Figure 2 This is a schematic diagram of the process of the present invention.

[0025] Figure 3 This is a schematic diagram of the microstructure point / surface scanning structure of the present invention.

[0026] Figure 4 This is a schematic diagram of the microstructure of the coating of the present invention. Detailed Implementation

[0027] like Figure 1 , Figure 2As shown, this invention screens key factors through orthogonal experiments; fits the complex relationship between process and performance based on PSO-BPNN; and, with the goal of achieving a large aspect ratio, low dilution rate, and high hardness, employs the NSGA-II algorithm for multi-objective optimization to find the optimal solution and analyzes the microstructure. If the optimal solution does not meet the cladding layer quality requirements, the quality index items that do not meet the requirements are first selected. Based on the orthogonal experimental screening results, the range of process parameters that have the greatest impact on these indexes is adjusted. Then, the NSGA-II algorithm is used again for multi-objective optimization to find the optimal solution until the cladding layer performance meets the requirements.

[0028] The present invention will be further described below with reference to the embodiments.

[0029] Example 1: Preparation of high-performance Fe60 cladding layer based on dynamic optimization method 1. Initial Experimental Design and Data Acquisition Substrate and cladding material: This embodiment uses Q345 steel as the base material and Fe60 alloy as the cladding material to optimize the laser cladding process.

[0030] The elemental composition of the cladding material is shown in Table 1.

[0031]

[0032] Orthogonal experiment: A three-factor, five-level orthogonal array, specifically L25(5³), was used, and the factor levels are shown in Table 2. A total of 25 experiments were conducted, and the process parameters, aspect ratio, dilution rate, and hardness of each experiment were recorded.

[0033] The experimental parameter ranges are as follows:

[0034] 2. High-precision construction of the PSO-BPNN surrogate model: To improve the model's generalization ability, 25 more experiments were conducted on the basis of the orthogonal experiments (as shown in Table 3). 35 groups were randomly selected for training, 10 groups for validation, and 5 groups for testing.

[0035] Data preprocessing: All input (process parameters) and output (quality indicators) data are normalized and scaled to the [0,1] interval to accelerate network convergence.

[0036] Network structure determined: Through trial and error or pruning algorithms, the optimal number of hidden layer nodes was determined to be 10. Therefore, the network structure is 3-10-3.

[0037] Detailed explanation of the PSO optimization process for BPNN: Particle encoding: Encodes all connection weights and thresholds of the BPNN into a single particle, the dimension D of which is determined by the network structure (in this example). ).

[0038] Fitness function: defined as the reciprocal of the root mean square error (RMSE) of the BPNN on the training set. The smaller the RMSE, the higher the fitness.

[0039] PSO parameter settings: Particle swarm size M=40, maximum number of iterations T=200, learning factor c1=c2=2, inertia weight w linearly decreases from 0.9 to 0.4.

[0040] Optimization process: PSO population initialization → calculate the fitness of each particle (i.e., a set of weight thresholds) → update individual optimal and global optimal → update particle velocity and position → iterate until the termination condition is met → output the global optimal particle (i.e., the optimal weight threshold) → assign it to BPNN to form the PSO-BPNN model.

[0041] Model validation: Evaluation was conducted using multiple metrics including the coefficient of determination (R²) and mean absolute percentage error (MAPE). After PSO optimization, the average R² on the test set reached over 0.96, significantly better than the 0.87 of the unoptimized BPNN.

[0042]

[0043] 3. Coordinated optimization of NSGA-II and dynamic range adjustment: Initialization: Set NSGA-II parameters, such as population size N=100, crossover probability pc=0.9, mutation probability pm=1 / D, and maximum number of generations Gen_max=200.

[0044] First NSGA-II optimization: In the initial parameter space ( Within the range of P=[700,1000], Vs=[200,400], Vr=[0.5,1.0], optimization is performed using the three objective functions mentioned above. After Gen_max generation evolution, the first Pareto front is obtained. .

[0045] Decision-making and judgment: From The optimal solution is selected from the options: With P=1000W, Vs=372mm / min, Vr=0.60r / min, the predicted performance is W / H=3.17, η=0.347, HD=607HV. The set quality thresholds are: W / H≥3.0, η≤0.33, HD≥610HV. It was determined that the dilution rate η=0.347>0.33, which does not meet the requirements.

[0046] Perform dynamic range adjustment (S5): Identify non-compliant metrics: dilution rate (η).

[0047] Query significance ranking: According to the range analysis table of the orthogonal experiment (Table 4), the factor with the most significant impact on the dilution rate is laser power (P).

[0048] Adjustment strategy: Due to the excessive dilution rate (over-melting) of the current solution, the energy input needs to be reduced. Therefore, the upper limit of the laser power P is lowered from 1000W to 950W.

[0049] New parameter space It becomes: P=[700,950],Vs=[200,400],Vr=[0.5,1.0].

[0050] Second NSGA-II optimization: in a new parameter space Inside, the NSGA-II algorithm is rerun. After evolution, the second Pareto front is obtained. .

[0051] Final decision: From The optimal solution was reselected, yielding P=934W, Vs=352mm / min, Vr=0.64r / min, with predictive performance of W / H=3.06, η=0.332, and HD=613HV. All indicators met the threshold requirements, and the optimization process ended.

[0052]

[0053] Range analysis showed that laser power had the most significant impact on dilution rate and hardness, while powder feed rate had a significant impact on aspect ratio.

[0054] The optimal combination selected from the solution set is: laser power 934W, scanning speed 352mm / min, and powder feed rate 0.64r / min. The predicted values ​​corresponding to this combination are: aspect ratio 3.06, dilution rate 0.33, and hardness 613HV.

[0055] Verification experiments show that the error between the predicted and experimental values ​​is less than 6.5% (dilution rate error 3.01%, aspect ratio error 1.31%, hardness error 3.59%), the hardness of the cladding layer is about 3 times higher than that of the substrate, and there are no defects such as pores or lack of fusion.

[0056] The optimized cladding layer microstructure is dominated by equiaxed and dendritic grains, with fine grains at the top and slightly larger grains in the middle and bottom. The bottom microstructure of the coating is columnar, while the heat-affected zone is acicular martensite, forming a planar crystalline transition layer with a thickness of approximately 3-5 μm between the two. Elemental line scan analysis from the substrate to the coating shows that the contents of Fe and C gradually decrease, while the contents of Cr and Si gradually increase. Figure 3 This indicates that a good metallurgical bond has been formed.

[0057] Embodiment of the present invention: The cladding layer prepared using the final optimized process exhibits a uniform and continuous macroscopic morphology, free of defects. Microstructure ( Figure 4 The results show that it has a dense equiaxed and dendritic structure with good interfacial bonding. The performance fully meets the predicted level, and the error is within the acceptable range.

[0058] Example 2: System Implementation The laser cladding process optimization system can be integrated into the control computer of the laser cladding equipment in software form, or it can be provided through a cloud platform. The system comprises four main modules: data acquisition, model building and training, multi-objective optimization, and dynamic adjustment and decision-making, providing users with automated services from data input to parameter output.

[0059] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A rapid optimization method for laser cladding process based on the NSGA-II algorithm, characterized in that, Includes the following steps: S1: Design experiments within a preset range of process parameters to obtain multiple combinations of process parameters and their corresponding cladding layer quality index data; S2: Based on the data, construct and train a machine learning model that can map the complex nonlinear relationship between process parameters and cladding layer quality indicators; S3: With the goal of optimizing one or more cladding layer quality indicators, a multi-objective evolutionary algorithm is used to find the best solution within the preset process parameters to obtain the first Pareto optimal solution set; S4: Determine whether the first Pareto optimal solution set meets the preset cladding layer quality requirements; if it does, select the final process parameter combination from it; If not satisfied, then execute S5; S5: Dynamic parameter range adjustment step: Identify the quality indicators that do not meet the requirements, and based on the significance ranking of the influence of each process parameter on the indicator in the experimental design, adjust the value range of at least one process parameter with the most significant influence to form a new process parameter space. S6: Within the new process parameter space adjusted in S5, repeat S3 until the obtained Pareto optimal solution set meets the quality requirements.

2. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 1, characterized in that, In S1, the experimental design is an orthogonal experimental design, the process parameters include at least laser power, scanning speed and powder feeding rate, and the cladding layer quality indicators include at least aspect ratio, dilution rate and hardness.

3. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 2, characterized in that, In S5, the "ranking of the significance of the influence of each process parameter on the index according to the experimental design" is determined by range analysis of the orthogonal experimental results.

4. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 1, characterized in that, In S2, the machine learning model is a BP neural network model, and the initial connection weights and thresholds of the BP neural network model are optimized using the particle swarm optimization algorithm to form a PSO-BPNN model.

5. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 4, characterized in that, The determination coefficient R² of the PSO-BPNN model is greater than 0.

95.

6. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 1, characterized in that, In S3, the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm with an elitist strategy, and its optimization objective function is: Minimize dilution rate: ; Maximize aspect ratio: ; Maximize hardness: ; in, P is the laser power; Vs represents the scan speed; Vr is the powder delivery rate; W / H is the aspect ratio; η is the dilution rate; H represents hardness.

7. The rapid optimization method for laser cladding process based on the NSGA-II algorithm according to claim 1, characterized in that, The optimized process parameter combination obtained by the method is: laser power 934W±5%, scanning speed 352mm / min±5%, powder feeding rate 0.64r / min±5%. Using this parameter combination, an Fe60 alloy cladding layer with an aspect ratio ≥3.0, a dilution rate ≤0.34, and a hardness ≥610HV can be prepared on a Q345 steel substrate.

8. A laser cladding process optimization system, characterized in that, It includes: The data acquisition module is used to perform S1 in claim 1; The model building and training module is used to perform S2 in claim 1; A multi-objective optimization module is used to execute S3 in claim 1; The dynamic adjustment and decision-making module is used to execute S4, S5 and S6 in claim 1 and output the final optimized combination of process parameters.

9. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

10. A Q345 steel base part, characterized in that, Its surface has a Fe60 alloy laser cladding layer prepared by the optimized process parameters described in claim 7. The microstructure of the cladding layer from the substrate interface to the surface layer consists of: a planar crystal zone with a thickness of 3-5 μm, a columnar crystal zone, and a cladding layer body mainly composed of equiaxed crystals and dendrites. The cladding layer is metallurgically bonded to the substrate and has no pores or fusion defects.

Images