A method for laser machining parameter optimization for tool surface texturing

Abstract

The present application belongs to the technical field of laser surface micro-texturing processing, and particularly relates to a laser processing parameter optimization method for tool surface texturing. The present application adopts a combined modeling method of Gaussian process regression and random forest weighted fusion, and aims at the problem of limited ability of depicting the strong nonlinear coupling relationship between the laser processing parameters and the processing results in the prior art. The present application fully gives play to the high-precision fitting advantage of Gaussian process regression to complex nonlinear relationship under small sample conditions, simultaneously utilizes the good generalization ability and anti-overfitting characteristics of the random forest model, realizes accurate representation of the multiple-input and multiple-output mapping relationship, and significantly improves the prediction accuracy and stability of the model.

Description

Technical Field

[0001] This invention belongs to the field of laser surface microtexturing technology, specifically relating to a method for optimizing laser processing parameters for tool surface texturing. Background Technology

[0002] Laser-based surface microtexturing is an important means to improve the cutting performance of cutting tools. Its processing quality is highly dependent on the proper matching of multiple parameters such as laser power, scanning speed, pulse frequency, and number of scans. However, there are complex and strongly nonlinear coupling relationships between these parameters and multiple output performance indicators such as microtexture depth, width, and surface roughness. Existing techniques for parameter optimization often rely on empirical parameter tuning, orthogonal experiments, or simple regression models, which have limited ability to characterize the inherent laws between multiple inputs and multiple outputs, resulting in low accuracy and poor stability of optimization results. To address this, Mao Y et al. used a response surface methodology (RSM) and a Grey Wolf Optimized Backpropagation (GWO-BP) neural network model for prediction and optimization. Their research shows that the intelligent algorithm based on neural networks improves the accuracy of parameter prediction to some extent and helps describe the complex nonlinear relationships in laser processing. However, this type of method still requires building a large-scale training sample set, resulting in numerous experiments, high costs, and its generalization ability under small sample conditions still needs improvement.

[0003] In terms of efficiency optimization, existing technologies typically rely on a large amount of experimental data for model building or parameter selection. Empirical methods or algorithms such as neural networks often require a large number of experiments to obtain sufficient training samples, resulting in long experimental cycles and high costs. In response, Hu et al. (Mechanical Engineer, 2018(12):51-53) optimized the laser etching process parameters for microtextures on ceramic cutting tool surfaces based on orthogonal experimental design. This method reduced experimental costs and improved optimization efficiency to some extent. However, orthogonal experimental design is essentially still a parameter selection method based on experimental design. It mainly relies on a limited number of experimental combinations for analysis, with limited parameter space coverage. It is difficult to fully characterize the complex nonlinear coupling relationship between multiple factors such as laser power, frequency, scanning speed, and number of scans. Under limited experimental sample conditions, it is difficult to quickly approximate the optimal parameter combination, which limits its application in engineering practice.

[0004] Furthermore, existing technologies often employ single-performance index optimization methods when optimizing parameters, focusing only on single surface quality indicators such as surface roughness. This makes it difficult to coordinately optimize multiple surface quality parameters, such as the arithmetic mean deviation of the profile and the maximum height of the profile, while ensuring that the microtexture geometry meets the target requirements. Consequently, it fails to achieve a synergistic improvement in geometric accuracy and surface integrity during tool surface texturing. In response, Zhou L et al. systematically studied the effects of parameters such as laser power, pulse frequency, scanning speed, and number of scans on the microtexture morphology, surface properties, and cutting performance of cemented carbide tools, achieving a certain degree of performance improvement. However, this study analyzed the microtexture geometry and its impact on cutting performance or surface roughness separately, focusing on the optimization of single or a few performance indicators, lacking a unified modeling and coordinated optimization mechanism for multiple indicators such as microtexture depth, width, and surface roughness.

[0005] In summary, existing technologies for laser processing parameter optimization still have the following shortcomings: First, they struggle to effectively handle the strong nonlinear coupling between multiple parameters and performance indicators; second, the optimization process is highly dependent on a large amount of experimental data, resulting in high experimental costs and low optimization efficiency; and third, there is a lack of a synergistic optimization method for multiple indicators, including geometric accuracy and surface quality. Therefore, there is an urgent need for a laser processing parameter optimization method that can simultaneously consider processing efficiency and optimization accuracy under limited experimental sample conditions, and achieve synergistic optimization of multiple objectives, including microtexture geometry and surface quality. Summary of the Invention

[0006] To overcome the problems in the prior art, this invention proposes a laser processing parameter optimization method for tool surface texturing.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This invention provides a method for optimizing laser processing parameters for tool surface texturing, comprising the following steps: Step 100: Obtain experimental data of laser processing for texturing tool surface, construct a sample dataset between laser processing parameters and processing results, and preprocess it. The processing results include microtexture depth, microtexture width, profile arithmetic mean deviation, and profile maximum height. Step 200: Based on the preprocessed sample dataset, for each processing result, with the preprocessed laser processing parameters as input and the corresponding preprocessed processing result as output, construct and train independent Gaussian process regression and random forest fusion prediction models respectively; Step 300: Generate a set of candidate laser processing parameters within the preset laser processing parameter range, and input each set of candidate laser processing parameters into the Gaussian process regression and random forest fusion prediction model to obtain the predicted microtexture depth, predicted microtexture width, predicted profile arithmetic mean deviation and predicted profile maximum height respectively. Step 400: Construct a laser processing parameter optimization model. The laser processing parameter optimization model uses candidate laser processing parameters as decision variables, predicts microtexture depth and microtexture width to meet target requirements as constraints, and predicts surface roughness index as optimization objective to construct a surface quality evaluation function. Step 500: Solve the laser processing parameter optimization model using the Bayesian optimization method to obtain the optimal combination of laser processing parameters.

[0008] Furthermore, in step 100, the laser processing parameters include laser power, pulse frequency, number of scans, and scanning speed.

[0009] Further, step 200 includes: Based on the preprocessed sample dataset, an independent Gaussian process regression prediction model is constructed and trained for each processing result. Each Gaussian process regression prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output. Based on the preprocessed sample dataset, independent random forest prediction models are constructed and trained for each type of processing result. Each random forest prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output. The prediction results of the Gaussian process regression prediction model and the random forest regression prediction model are weighted and fused to obtain the final prediction result.

[0010] Furthermore, in step 200, a composite kernel function consisting of a constant kernel function, a Matern kernel function, and a white noise kernel function is used for the microtexture width Gaussian process regression prediction model and the maximum profile height Gaussian process regression prediction model. For the Gaussian process regression prediction model of microtexture depth and the Gaussian process regression prediction model of profile arithmetic mean deviation, a composite kernel function consisting of constant kernel function, RBF kernel function, Mater kernel function and white noise kernel function is adopted.

[0011] Furthermore, the prediction results of the Gaussian process regression prediction model and the prediction results of the random forest regression prediction model are weighted and fused to obtain the final prediction result, including: ; in, To merge the predicted values, These are the predicted values ​​from the Gaussian process regression prediction model. These are the predicted values ​​from the random forest regression prediction model. and These are the weights for the Gaussian process regression prediction model and the random forest regression prediction model, respectively.

[0012] Furthermore, step 300 also includes: Within the preset range of laser processing parameters, multiple candidate laser processing parameter combinations are generated by random sampling. Duplicate candidate laser processing parameters are deduplicated, and only one set of laser processing parameter combinations is retained, thus obtaining a set of non-duplicate candidate laser processing parameters.

[0013] Further, in step 400, the constraint condition is that the predicted microtexture depth and predicted microtexture width meet the target requirements, including: Given a target microtexture depth and a target microtexture width, calculate the relative depth error between the predicted microtexture depth and the target microtexture depth, and the relative width error between the predicted microtexture width and the target microtexture depth; the relative depth error and the relative width error are respectively less than or equal to preset error thresholds.

[0014] Further, in step 400, a surface quality evaluation function is constructed with the prediction of surface roughness index as the optimization objective, including: ; In the above formula, This is the comprehensive evaluation value for surface quality; To predict the arithmetic mean deviation of the profile; To predict the maximum height of the profile; and These are the weighting coefficients.

[0015] Compared with the prior art, the present invention has the following technical effects: (1) This invention adopts a combined modeling method of weighted fusion of Gaussian process regression (GPR) and random forest (RF). In view of the problem that the existing technology has limited ability to characterize the strong nonlinear coupling relationship between laser processing parameters and processing results, it gives full play to the high-precision fitting advantage of Gaussian process regression under small sample conditions for complex nonlinear relationships. At the same time, it utilizes the good generalization ability and anti-overfitting characteristics of the random forest model to realize the accurate characterization of multi-input multi-output mapping relationship, which significantly improves the prediction accuracy and stability of the model.

[0016] (2) This invention proposes a collaborative optimization method of "geometric constraint priority + Bayesian optimization". Addressing the problems of difficulty in achieving coordinated optimization of microtexture geometric accuracy and surface quality, and low optimization efficiency in existing technologies, this method prioritizes meeting the target requirements for microtexture depth and width as constraints. Based on this, it optimizes for minimum surface roughness, combining the efficient search capability of the Bayesian optimization algorithm to achieve automatic optimization of processing parameters. This method ensures both the geometric dimensional accuracy of the microtexture and effectively reduces surface roughness, achieving a synergistic improvement in geometric accuracy and surface quality. Simultaneously, through the exploration and development of the expected value improvement acquisition function balance in Bayesian optimization, optimization efficiency is improved, enabling the rapid finding of the optimal combination of laser processing parameters that satisfies the constraints, thus meeting the needs of practical engineering applications. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solutions proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. Specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] In this embodiment, refer to Figure 1 A method for optimizing laser processing parameters for tool surface texturing is provided, comprising the following steps: Step 100: Obtain experimental data of laser processing for texturing tool surface, construct a sample dataset between laser processing parameters and processing results, and preprocess it. The processing results include microtexture depth, microtexture width, profile arithmetic mean deviation, and profile maximum height. Step 200: Based on the preprocessed sample dataset, for each processing result, with the preprocessed laser processing parameters as input and the corresponding preprocessed processing result as output, construct and train independent Gaussian process regression and random forest fusion prediction models respectively; Step 300: Generate a set of candidate laser processing parameters within the preset laser processing parameter range, and input each set of candidate laser processing parameters into the Gaussian process regression and random forest fusion prediction model to obtain the predicted microtexture depth, predicted microtexture width, predicted profile arithmetic mean deviation and predicted profile maximum height respectively. Step 400: Construct a laser processing parameter optimization model. The laser processing parameter optimization model uses candidate laser processing parameters as decision variables, predicts microtexture depth and microtexture width to meet target requirements as constraints, and predicts surface roughness index as optimization objective to construct a surface quality evaluation function. Step 500: Solve the laser processing parameter optimization model using the Bayesian optimization method to obtain the optimal combination of laser processing parameters.

[0021] The following is a detailed explanation of each of the above steps: Step 100: Obtain experimental data of laser processing for texturing tool surfaces, construct a sample dataset between laser processing parameters and processing results, and preprocess the dataset.

[0022] As an example, step 100 specifically includes: Step 110: Obtain experimental data of laser processing for texturing tool surfaces and construct a sample dataset between laser processing parameters and processing results.

[0023] The laser processing parameters include laser power P (W), pulse frequency f (kHz), number of scans N (times), and scanning speed v (mm / s). After processing, the microtexture morphology is measured using a white light interferometer or confocal microscope to obtain processing result data. The processing results include microtexture depth D (μm), microtexture width W (μm), profile arithmetic mean deviation Ra (μm), and maximum profile height Rz (μm). These four indicators serve as output variables of the model, representing the geometric dimensional accuracy and surface quality level of the microtexture, respectively. Thus, each combination of laser processing parameters and its corresponding four processing result indicators constitute a complete sample data pair. All experimental data are then compiled to form a sample dataset.

[0024] The sample dataset is divided into a training set and a test set for subsequent model training, cross-validation, and evaluation. Specifically, random partitioning or stratified sampling can be used to allocate 70%–85% of the sample dataset to the training set and the remaining 15%–30% to the test set. The training set is used to build and train the subsequent surrogate model, enabling it to fully learn the complex nonlinear mapping relationship between laser processing parameters and processing results. The test set is completely independent of the training process and is used to objectively evaluate the model's prediction accuracy and generalization ability after training, avoiding overfitting.

[0025] Step 120: Perform data preprocessing on the sample dataset. The data preprocessing includes laser processing parameter preprocessing and processing result preprocessing, which is used to improve the training stability and prediction accuracy of the subsequent prediction model.

[0026] For laser processing parameters, namely laser power, pulse frequency, number of scans, and scanning speed, since the dimensions and value ranges of each parameter are different, normalization is preferred to map each laser processing parameter to a uniform numerical range. The expression is: ; In the above formula, For the first One set of original laser processing parameters; These are the normalized laser processing parameters. and These are the minimum and maximum values ​​of the laser processing parameter in the sample dataset, respectively.

[0027] Preferably, to improve subsequent prediction capabilities, the depth of microtexture can be... And the arithmetic mean deviation of the profile The logarithmic transformation is performed, and its expression is: ; in, This is the original output value. This is the transformed output value.

[0028] After the prediction model completes its prediction, the original data is restored through inverse transformation, the expression of which is: .

[0029] Step 200: Based on the preprocessed sample dataset, for each processing result, with the preprocessed laser processing parameters as input and the corresponding preprocessed processing result as output, construct and train independent Gaussian process regression and random forest fusion prediction models respectively.

[0030] Different tool materials exhibit variations in laser absorptivity, thermal conductivity, melting point, and ablation characteristics, which can influence the dimensions and surface quality of microwoven structures. This embodiment establishes a mapping relationship between processing parameters and results using a fusion model of Gaussian process regression and random forest, enabling adaptive modeling for different tool materials. For different materials, only a small amount of corresponding experimental data is needed to complete model training under small sample conditions. Combined with Bayesian optimization, optimized processing parameters suitable for that material are obtained, ensuring good applicability and stability of this method under different tool material conditions.

[0031] Gaussian process regression (GPR), a nonparametric model based on Bayesian theory, excels at handling small-sample, highly nonlinear problems. It can flexibly capture the complex coupling relationship between laser processing parameters such as laser power, pulse frequency, number of scans, and scanning speed, and processing results such as microtexture depth D and width W through radial basis function (RBF) kernels, effectively addressing the insufficient ability of existing technologies to characterize highly nonlinear couplings. Random forest (RF), as an ensemble learning model composed of multiple decision trees, has advantages in resisting overfitting, handling high-dimensional data, and feature importance analysis, improving the predictive stability of processing results and compensating for the limitations of single models. A weighted fusion strategy is adopted to integrate the prediction results of Gaussian process regression and random forest, ensuring that the fused model combines the small-sample fitting advantage of Gaussian process regression with the high-dimensional anti-overfitting ability of random forest.

[0032] For four processing result indicators—microtexture depth, microtexture width, arithmetic mean deviation of the contour, and maximum height of the contour—independent prediction models are constructed. Each prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output. The parameters of each prediction model are optimized using training set data.

[0033] As an example, step 200 specifically includes: Step 210: Based on the preprocessed sample dataset, construct and train an independent Gaussian process regression prediction model for each processing result. Each Gaussian process regression prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output.

[0034] For four processing result indicators—microtexture depth, microtexture width, contour arithmetic mean deviation, and contour maximum height—Gaussian process regression models are constructed by selecting different kernel function combinations based on different processing results. These models include Gaussian process regression prediction models for microtexture depth, microtexture width, contour arithmetic mean deviation, and contour maximum height. Each prediction model takes preprocessed laser processing parameters as input and the corresponding processing result as output. The parameters of each prediction model are optimized using training set data.

[0035] The core of Gaussian process regression is the kernel function, which defines two normalized laser processing parameters in the input space. and Similarity, i.e., kernel function values The larger the value, the higher the output value. and The more relevant.

[0036] For the microtexture width and the maximum profile height Rz, a composite kernel function consisting of a constant kernel function, a Matern kernel function, and a white noise kernel function is preferred: ; In the above formula, Indicates constant kernel hyperparameters; This refers to the Matern kernel function, specifically its smoothing parameter. The preferred value is 2.5; This represents the white noise kernel function.

[0037] For the arithmetic mean deviation Ra of microtexture depth and profile, a composite kernel function consisting of a constant kernel function, an RBF kernel function, a Mater kernel function, and a white noise kernel function is preferred. ; In the above formula, Represents the RBF kernel function; for the arithmetic mean deviation Ra of microtexture depth from the profile, the Matern kernel function smoothing parameter of the microtexture depth. Preferably, 1.5 is the Matern kernel function smoothing parameter for the profile arithmetic mean deviation Ra. 2.5 is preferred.

[0038] During the training of the Gaussian process regression model, the kernel function hyperparameters are optimized by maximizing the logarithmic marginal likelihood function, and multiple random restarts are employed to reduce the risk of getting trapped in local optima. Preferably, for the Gaussian process regression prediction model of microtexture depth and the Gaussian process regression prediction model of profile arithmetic mean deviation, the number of random restarts is set to 6; for the Gaussian process regression prediction model of microtexture width and the Gaussian process regression prediction model of profile maximum height, the number of random restarts is set to 4. The initial value of the constant kernel coefficient is 1.0, and its range is [missing value]. The initial value for the length scale is 1.0, and its range is [missing value]. The initial value for white noise and nuclear noise levels is... or The range of values ​​is The kernel function hyperparameters were determined by maximizing the logarithmic marginal likelihood function and optimizing through multiple random restarts.

[0039] Step 220: Based on the preprocessed sample dataset, construct and train an independent random forest prediction model for each processing result. Each random forest prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output.

[0040] For four processing result indicators—microtexture depth, microtexture width, contour arithmetic mean deviation, and contour maximum height—random forest prediction models for microtexture depth, microtexture width, contour arithmetic mean deviation, and contour maximum height were constructed based on random forest. Each prediction model takes preprocessed laser processing parameters as input and the corresponding processing result as output. The parameters of each prediction model were optimized using training set data.

[0041] This embodiment sets the hyperparameters of the random forest regression model for different output results. The preferred number of decision trees is 220–350, and the preferred maximum tree depth is 8–10 layers.

[0042] Step 230: The prediction results of the Gaussian process regression prediction model and the prediction results of the random forest regression prediction model are weighted and fused to obtain the final prediction result.

[0043] The Gaussian process regression prediction model and the random forest regression prediction model are weighted and fused to obtain the final prediction result: ; in, To merge the predicted values, These are the predicted values ​​from the Gaussian process regression prediction model. These are the predicted values ​​from the random forest regression prediction model. and These are the weights for the Gaussian process regression prediction model and the random forest regression prediction model, respectively.

[0044] The fusion weights of the Gaussian process regression prediction model and the random forest regression prediction model are determined based on the prediction errors of each base model in the validation set or cross-validation process. For any response variable (processing result), the prediction errors of the Gaussian process regression prediction model and the random forest regression prediction model are calculated separately, and the model with the smaller prediction error is assigned a larger fusion weight.

[0045] Let the first The cross-validation error of the Gaussian process regression prediction model corresponding to each response variable (processing outcome) is: The cross-validation error of the random forest regression prediction model is The weights of the Gaussian process regression prediction model Weights of the random forest regression prediction model They are respectively: ; ; And it satisfies: ; Wherein, the cross-validation error and The root mean square error (RMSE), mean absolute error (MAE), or a weighted average of the two can be used.

[0046] Preferably, different fusion weights are used for different response variables. For the microtexture depth prediction model, , For the profile arithmetic mean deviation prediction model, , For the microtexture width prediction model, , For the maximum height prediction model of the profile, , .

[0047] Five-fold cross-validation was used to evaluate the predictive performance of the fusion model, and the coefficient of determination was employed. The mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation indicators.

[0048] Step 300: Generate a set of candidate laser processing parameters within the preset laser processing parameter range. Input each set of candidate laser processing parameters into the Gaussian process regression and random forest fusion prediction model to obtain the predicted microtexture depth, predicted microtexture width, predicted profile arithmetic mean deviation, and predicted profile maximum height.

[0049] In this embodiment, the ranges of laser power, pulse frequency, number of scans, and scanning speed are not arbitrarily set, but are determined comprehensively based on the rated operating range of the laser equipment, the laser processing response characteristics of the tool material, the results of pre-experimental forming, and experience in the micro-textured processing technology of the tool surface. For example, the laser power can be 2W to 6W, the pulse frequency can be 200kHz to 300kHz, the number of scans can be 2 to 5, and the scanning speed can be 100mm / s to 300mm / s.

[0050] Preferably, multiple candidate laser processing parameter combinations are generated using a random sampling method, and duplicate points are removed. The criterion for determining duplicate points is: if the laser power of two candidate laser processing parameter combinations exceeds the threshold value... Pulse frequency Number of scans and scanning speed If all parameters are identical, they are considered duplicates; during deduplication, each candidate laser processing parameter combination is represented as a quadruple. Using this quadruple as a unique identifier, duplicate quadruples are removed by hash deduplication, set deduplication, or unique value filtering of data tables, retaining only one set of laser processing parameter combinations, thus obtaining a set of candidate laser processing parameters without duplication.

[0051] Based on the GPR-RF fusion prediction model established in step 200, a laser processing parameter optimization model for tool surface texturing is constructed. Specifically, each combination of laser processing parameters in the candidate laser processing set is input into the GPR-RF fusion prediction model constructed in step 200 to obtain the corresponding predicted microtexture depth, predicted microtexture width, predicted profile arithmetic mean deviation, and predicted profile maximum height.

[0052] Step 400: Construct a laser processing parameter optimization model. The laser processing parameter optimization model uses candidate laser processing parameters as decision variables, the predicted microtexture depth and predicted microtexture width meeting the target requirements as constraints, and the predicted surface roughness index as the optimization objective. Construct a surface quality evaluation function and select the laser processing parameter combination with the smallest surface quality evaluation value. The surface roughness index includes the profile arithmetic mean deviation and the maximum profile height.

[0053] Surface roughness is chosen as the optimization target because microtexture depth and width are primarily used to ensure that the texture geometry meets design requirements, while surface roughness directly reflects the surface quality after laser processing. Excessive surface roughness can easily lead to defects such as melt buildup, burrs, microcracks, or expansion of the heat-affected zone, thus affecting the friction and wear performance, coating adhesion, and service life of the tool surface. Therefore, microtexture depth and width are first used as constraints to ensure that the microtexture dimensions meet the target accuracy requirements. Based on this, the comprehensive surface roughness evaluation value, composed of the arithmetic mean deviation Ra and the maximum profile height Rz, is used as the optimization target. The optimal parameter combination for surface quality is selected from candidate parameters that meet the dimensional constraints, thus balancing the functional dimensions of the texture and the surface machining quality of the tool.

[0054] Let the target microtexture depth be The target micro-texture width is Based on the predicted microtexture depth and predicted microtexture width, a depth relative error is defined. relative error with width They are respectively: ; ; In the above formula, Indicates the predicted microtexture depth; This indicates the predicted microtexture width.

[0055] Preferably, the geometric constraints are set as follows: ; The error threshold in the geometric constraints is set to 5%. This threshold is determined comprehensively based on the target dimensional accuracy requirements of the microtexture on the tool surface, the repeatability of the laser processing equipment, the measurement error of the detection equipment, and the results of preliminary experimental processing. The microtexture depth and width directly affect the oil storage, friction reduction, chip removal, and frictional contact state of the tool surface. If their deviations from the target values ​​are too large, the functional effect of the microtexture will be weakened. If the error threshold is set too small, the feasible parameter space will be significantly reduced, making it difficult to obtain candidate parameters that satisfy the constraints during the optimization process. Limiting both the relative depth error and the relative width error to within 5% ensures the dimensional accuracy of the microtexture while retaining a reasonable parameter search space, thus balancing processing quality, optimization efficiency, and engineering feasibility. In other words, only candidate parameter combinations that meet the target depth and width error requirements are considered feasible solutions.

[0056] Based on the above geometric constraints, a surface quality evaluation function is constructed: ; In the above formula, This is the comprehensive evaluation value for surface quality; To predict the arithmetic mean deviation of the profile; To predict the maximum height of the profile; and The weighting coefficients are the weighting factors. and The weighting can be determined based on machining quality evaluation requirements, tool service performance requirements, and preliminary experimental results. Specifically, multiple candidate weight combinations can be selected, for example... The predicted surface quality evaluation values ​​corresponding to the recommended parameters under different weight combinations were calculated, and the final weight combination was determined in combination with the actual processing verification results.

[0057] Preferably, when the surface texturing of the cutting tool focuses more on the overall surface smoothness, frictional contact stability, and average roughness, the following settings are provided: Weight greater than Weights, for example , When focusing on extreme defects such as local peak and valley heights, ablation pits, burrs, or molten deposits, settings can be configured. Weight greater than or equal to Weights.

[0058] In a preferred embodiment, taking into account... The characterization effect on the average surface roughness and To characterize local peak-valley defects, the following settings are used: ; The surface quality evaluation function is: ; Under the premise of satisfying the geometric constraints of microtexture depth and width, select The smallest candidate combination of laser processing parameters is taken as the optimization result.

[0059] Step 500: Solve the laser processing parameter optimization model using the Bayesian optimization method to obtain the optimal combination of laser processing parameters.

[0060] By using a predictive model as an approximation of the objective and constraint functions, the most promising sampling points are iteratively selected within the parameter space based on the acquisition function to balance global exploration and local exploitation. The acquisition function is constructed based on Expected Improvement (EI).

[0061] Let the current minimum surface quality rating in the training samples be... The expected improvement value of the candidate point is defined as: ; in: ; In the above formula, This represents the standardized boost value; it is an intermediate calculation variable. This represents the mean, the predicted value for a given candidate point. Standard deviation represents the uncertainty of the model regarding the predicted values; The cumulative distribution function of the standard normal distribution; The probability density function is the standard normal distribution. The hyperparameter for controlling the degree of exploration is preferably set between 0.01 and 0.1. Used to balance global exploration and local exploitation in the Bayesian optimization process. When When the value is small, the algorithm tends to select candidate parameters that are better than the current predicted target value in order to accelerate local convergence; when... When the value is large, the algorithm tends to select candidate parameters with higher prediction uncertainty to enhance the exploration of the unknown parameter region. Preferably, Adjustments are made based on the pre-optimization results, i.e., several candidates are selected first. The value is pre-optimized with a small number of iterations, and the final value is determined based on the number of feasible solutions, the improvement of the objective function, and the convergence trend of the optimization. In a preferred embodiment, .

[0062] Furthermore, during the Bayesian optimization process, priority is given to determining whether the candidate laser processing parameters satisfy the constraints of the microtexture target depth and target width. The surface quality evaluation function is only evaluated within the feasible region that satisfies the constraints. Optimization was performed to obtain the laser processing parameter combination that meets the requirements for geometric accuracy and has the best surface quality.

[0063] Referring to Table 1, to verify the effectiveness of the weighted fusion modeling strategy of Gaussian process regression (GPR) and random forest (RF), this experiment compared the prediction performance of single models (GPR, RF) and combined models (GPR-RF) on the geometric dimensions (microtexture depth, microtexture width) and surface roughness (profile arithmetic mean deviation Ra, profile maximum height Rz) of laser-processed microtextures. Although single models are simple to implement, GPR is easily limited by sample size and has insufficient generalization ability under the strong nonlinear coupling of multiple parameters in laser processing. While RF has strong robustness, its accuracy is limited, making it difficult to simultaneously achieve high-precision prediction of multiple indicators such as microtexture depth, width, and surface roughness. Therefore, the GPR-RF combined modeling strategy was finally adopted to improve prediction accuracy and stability.

[0064] Table 1 Comparative Experiment

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for optimizing laser processing parameters for tool surface texturing, characterized in that, Includes the following steps: Step 100: Obtain experimental data of laser processing for texturing tool surface, construct a sample dataset between laser processing parameters and processing results, and preprocess it. The processing results include microtexture depth, microtexture width, profile arithmetic mean deviation, and profile maximum height. Step 200: Based on the preprocessed sample dataset, for each processing result, with the preprocessed laser processing parameters as input and the corresponding preprocessed processing result as output, construct and train independent Gaussian process regression and random forest fusion prediction models respectively; Step 300: Generate a set of candidate laser processing parameters within the preset laser processing parameter range, and input each set of candidate laser processing parameters into the Gaussian process regression and random forest fusion prediction model to obtain the predicted microtexture depth, predicted microtexture width, predicted profile arithmetic mean deviation and predicted profile maximum height respectively. Step 400: Construct a laser processing parameter optimization model. The laser processing parameter optimization model uses candidate laser processing parameters as decision variables, predicts microtexture depth and microtexture width to meet target requirements as constraints, and predicts surface roughness index as optimization objective to construct a surface quality evaluation function. Step 500: Solve the laser processing parameter optimization model using the Bayesian optimization method to obtain the optimal combination of laser processing parameters; Step 200 includes: Based on the preprocessed sample dataset, an independent Gaussian process regression prediction model is constructed and trained for each processing result. Each Gaussian process regression prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output. Based on the preprocessed sample dataset, independent random forest prediction models are constructed and trained for each type of processing result. Each random forest prediction model takes the preprocessed laser processing parameters as input and the corresponding processing result as output. The prediction results from the Gaussian process regression model and the random forest regression model are weighted and fused to obtain the final prediction result, which includes: ； in, To merge the predicted values, These are the predicted values ​​from the Gaussian process regression prediction model. These are the predicted values ​​from the random forest regression prediction model. and These are the weights of the Gaussian process regression prediction model and the random forest regression prediction model, respectively. In step 400, the constraint condition is that the predicted microtexture depth and the predicted microtexture width meet the target requirements. This includes: setting the target microtexture depth and the target microtexture width, and calculating the relative depth error between the predicted microtexture depth and the target microtexture depth, and the relative width error between the predicted microtexture width and the target microtexture depth; the relative depth error and the relative width error are respectively less than or equal to preset error thresholds. In step 400, a surface quality evaluation function is constructed with the prediction of surface roughness index as the optimization objective, including: ； In the above formula, This is the comprehensive evaluation value for surface quality; To predict the arithmetic mean deviation of the profile; To predict the maximum height of the profile; and These are the weighting coefficients.

2. The laser processing parameter optimization method for tool surface texturing according to claim 1, characterized in that, In step 100, the laser processing parameters include laser power, pulse frequency, number of scans, and scanning speed.

3. The laser processing parameter optimization method for tool surface texturing according to claim 1, characterized in that, In step 200, a composite kernel function consisting of a constant kernel function, a Matern kernel function, and a white noise kernel function is used for the microtexture width Gaussian process regression prediction model and the profile maximum height Gaussian process regression prediction model. For the Gaussian process regression prediction model of microtexture depth and the Gaussian process regression prediction model of profile arithmetic mean deviation, a composite kernel function consisting of constant kernel function, RBF kernel function, Mater kernel function and white noise kernel function is adopted.

4. The laser processing parameter optimization method for tool surface texturing according to claim 1, characterized in that, Step 300 further includes: Within the preset range of laser processing parameters, multiple candidate laser processing parameter combinations are generated by random sampling. Duplicate candidate laser processing parameters are deduplicated, and only one set of laser processing parameter combinations is retained, thus obtaining a set of non-duplicate candidate laser processing parameters.

Images