A laser texturing parameter recommendation method based on pulse lap joint and energy accumulation
By constructing a joint feature vector based on the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area, the problems of inaccurate laser texturing roughness prediction and insufficient cross-device applicability in the existing technology are solved, and higher accuracy and stable laser texturing parameter recommendation are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN XIANGMING LASER TECH CO LTD
- Filing Date
- 2026-05-15
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies have failed to construct a complete feature system that can simultaneously characterize the pulse spatial overlap state, the coverage of the scanning trajectory, and the energy accumulation effect of multiple scans, resulting in insufficient roughness control accuracy and cross-equipment versatility in the laser texturing industry.
By constructing a joint feature vector consisting of overlap rate along the scanning direction, inter-line overlap rate, energy per unit area, and cumulative energy per unit area, a laser texturing parameter recommendation method is established. Machine learning models are used to predict roughness, and process parameters are recommended under the constraints of equipment capability and process feasibility.
It improves the accuracy of laser texturing roughness prediction and the stability of cross-equipment process parameter recommendations, ensuring precise control under different equipment and process conditions.
Smart Images
Figure CN122436044A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of laser hairization technology, specifically to a method for recommending laser hairization parameters based on pulse overlap and energy accumulation. Background Technology
[0002] Laser texturing technology requires adjusting various parameters such as average laser power, frequency, scanning speed, scanning spacing, number of scans, and spot diameter to control surface roughness. Existing research has attempted to establish predictive models of parameters and roughness using data-driven methods. For example, CN120362730A proposes a parameter adjustment method based on a preset overlap rate target, and CN121506314A discloses a machine learning-guided design method for laser-formed ultra-high strength and toughness steel based on the integration of physicochemical features. This method also discloses a scheme that integrates engineering features such as energy density and energy per unit length into a neural network to predict surface roughness.
[0003] However, the aforementioned existing technologies either directly model using original equipment parameters or only use one-dimensional / two-dimensional energy features, failing to construct a complete feature system capable of simultaneously characterizing the pulse space overlap state, the coverage of the scanning trajectory, and the cumulative effect of energy from multiple scans. Furthermore, they have not demonstrated through rigorous ablation experiments the necessity for the inseparable joint use of multiple physical features, resulting in insufficient predictive stability and parameter transfer capability of the model under different equipment conditions and process windows, making it difficult to meet the actual needs of the laser texturing industry for roughness control accuracy and cross-equipment versatility. Summary of the Invention
[0004] This invention proposes a laser texturing parameter recommendation method based on pulse overlap and energy accumulation. It solves the technical problem of how to improve the prediction accuracy of laser texturing roughness and the stability of cross-equipment process parameter recommendation by constructing a physical feature vector composed of overlap rate along the scanning direction, inter-line overlap rate, energy per unit area, and cumulative energy per unit area, which has been proven by ablation experiments to be necessary for joint use.
[0005] The technical solution of this invention is implemented as follows: A method for recommending laser texturing parameters based on pulse overlap and energy accumulation includes the following steps: S1. Establish a laser texturing experimental database to record material type, initial surface roughness, laser processing parameters, and surface roughness results after processing; S2. Based on the laser processing parameters, calculate a joint feature vector consisting of the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area; wherein, The overlap rate along the scanning direction is used to characterize the degree of overlap between adjacent pulses on the same scanning trajectory. The line overlap rate is used to characterize the degree of coverage between adjacent scan trajectories. The energy per unit area is used to characterize the energy input level received per unit area during a single scan. The cumulative energy per unit area is used to characterize the cumulative effect of energy input per unit area under multiple scan conditions; S3. Based on the joint feature vector and experimental data, construct a training dataset and train a roughness prediction model to establish a mapping relationship between the joint feature vector and the surface roughness after processing. S4. When the target roughness is input, candidate process parameter combinations are generated within the constraints of equipment capability and process feasibility. The joint feature vector corresponding to each candidate process parameter combination is calculated and input into the roughness prediction model. Recommended process parameter combinations are selected and output based on the prediction results. Among them, the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area must all exist simultaneously in the joint feature vector to synergistically characterize the multi-scale action mechanism from pulse spatial overlap to energy accumulation during laser texturing.
[0006] Furthermore, the laser processing parameters include one or more of the following: average laser power, laser frequency, pulse width, scanning speed, scanning interval, number of scans, and spot diameter.
[0007] Furthermore, the overlap rate along the scanning direction The calculation formula is: ,in The pulse spacing along the scanning direction. For scanning speed, The laser frequency, denoted as the diameter of the light spot.
[0008] Furthermore, the inter-line overlap rate The calculation formula is: ,in For scanning spacing, denoted as the diameter of the light spot.
[0009] Furthermore, the energy per unit area The calculation formula is: ,in For single-pulse energy, The average power of the laser. The laser frequency, The pulse spacing along the scanning direction. This represents the scanning interval.
[0010] Furthermore, the cumulative energy per unit area The calculation formula is: ,in The number of scan passes. Energy per unit area.
[0011] Furthermore, the joint feature vector also includes single-pulse energy. and / or single-pulse energy normalization characteristics , as an auxiliary enhancement feature.
[0012] Furthermore, the single-pulse energy normalization feature The calculation formula is: ,in This is the maximum single pulse energy of the device.
[0013] Furthermore, the roughness prediction model is a tree model.
[0014] Furthermore, the selection of recommended process parameters is based on the deviation between the predicted roughness and the target roughness, and the recommended parameter combination is determined in combination with equipment capability constraints and process feasibility constraints.
[0015] The beneficial effects of the technical solution provided in this application are as follows: 1. In this application, a joint feature vector composed of overlap rate along the scanning direction, inter-line overlap rate, energy per unit area, and cumulative energy per unit area is used to synergistically characterize pulse overlap, trajectory coverage, and energy accumulation during the formation of laser-induced roughness. The roughness prediction model based on this joint feature vector achieves higher prediction accuracy with fewer input dimensions, and ablation experiments demonstrate that the absence of any feature leads to increased prediction error, thus proving the necessity of the indivisibility of the four features.
[0016] 2. The specific calculation formulas for the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area in this application ensure that each physical characteristic has clear calculation consistency under different laser processing equipment and different process conditions. Attached Figure Description
[0017] To more clearly illustrate the technical solutions 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 laser hairization experimental database structure of the present invention; Figure 2 This is a flowchart of the calculation of the combined characteristics of pulse overlap and energy accumulation in this invention; Figure 3 This is a flowchart of the machine learning model training process of the present invention; Figure 4 A flowchart is recommended for the process parameters of the invention. Detailed Implementation
[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0020] This embodiment provides a method for recommending laser texturing parameters based on pulse overlap and energy accumulation.
[0021] Establish a laser hairization experimental database
[0022] Reference Figure 1 The database stores all experimental data, providing a unified data foundation for subsequent feature calculations, model training, and validation. The database records at least the following three types of information: The first category is basic information about the workpiece and equipment, including material type, equipment type, laser type, scanning system type, focusing optical conditions, and initial surface roughness before processing. This information is used to identify the properties of workpieces and equipment under different experimental conditions, facilitating subsequent data analysis and model transfer across materials and equipment.
[0023] The second category is laser processing parameters, specifically including average laser power. Laser frequency Pulse width galvanometer scanning speed Scanning Spacing Number of scans Spot diameter and the device's maximum single pulse energy These parameters are raw input quantities that can be directly controlled and measured in the laser texturing process.
[0024] The third category is the processing result label, namely the surface roughness after processing. In addition, in some implementations, processing effect scores and apparent risk levels may also be recorded.
[0025] To provide a reliable data foundation for subsequent feature calculations and model training, this embodiment conducted multiple sets of experiments according to the parameter ranges shown in Table 1. Table 1 shows the processing results of three typical materials—aluminum alloy, carbon steel, and stainless steel—under different laser power, frequency, scanning speed, number of scans, spot diameter, scanning spacing, and initial roughness. The roughness after processing ranged from 0.93 to 11.7 μm. The data in Table 1 covers the commonly used process window for laser texturing, including processing conditions with low, medium, and high energy input levels and different degrees of overlap.
[0026] Table 1. Example of laser hair removal experimental data.
[0027] Table 1 is merely an example of experimental data; the present invention is not limited to the material types, parameter ranges, and data scales described above. Furthermore, under certain experimental conditions, the scanning interval... It can be larger than the diameter of the light spot. This indicates that there is a non-overlapping interval between adjacent scan trajectories. The scan spacing in this paper... This represents the actual spacing between adjacent scan trajectories. Through this experimental database, texturing results data under different materials, equipment, and process parameter combinations can be obtained, providing a foundation for subsequent physical feature calculations, machine learning model training, and process parameter recommendations.
[0028] Calculation of physical characteristics of laser energy
[0029] After obtaining the original laser processing parameters, this invention does not directly use all equipment setting parameters as model input, but rather follows... Figure 2 The illustrated process calculates a joint eigenvector characterizing the pulse space overlap state and energy accumulation effect. The purpose of this transformation is to map the parameters of the device control layer to variables adapted to the device hardware.
[0030] The specific calculation steps are as follows: According to the formula Calculate single pulse energy. The average power of the laser. The laser frequency is the unit of energy carried by a single laser pulse. This physical quantity characterizes the fundamental energy carried by a single laser pulse and forms the basis for subsequent calculations of other energy-related characteristics. The single-pulse energy directly affects the molten pool size and evaporation rate of the roughening pit, and is a significant driving factor in roughness formation.
[0031] According to the formula Calculate the spatial distance between adjacent pulses along the scanning trajectory. For scanning speed, The laser frequency is represented by this spacing. This spacing determines the sparsity of the pulses along the scanning direction. Smaller than the spot diameter When the time pulses overlap, Greater than There are gaps between the time pulses.
[0032] According to pulse spacing and spot diameter According to the formula Calculation. This feature is used to characterize the degree of overlap between adjacent pulses on the same scan trajectory. The larger the value, the more severe the pulse overlap and the denser the distribution of the roughening pits in the scanning direction, which is beneficial for forming a continuous and uniform roughness profile.
[0033] According to the scanning interval and spot diameter According to the formula Calculation. This feature is used to characterize the coverage between adjacent scan trajectories. When A positive value indicates overlap between trajectories, while a negative value indicates the presence of gaps. This feature directly affects the uniformity of the distribution of texturing pits perpendicular to the scanning direction.
[0034] Based on single pulse energy Pulse spacing and scan spacing According to the formula Calculation. This feature characterizes the energy input level received per unit area during a single scan. It integrates the effects of single-pulse energy, pulse spatial distribution, and scan line spacing, and is an important indicator reflecting the laser energy density.
[0035] Based on the number of scans According to the formula Calculation. This feature is used to characterize the cumulative effect of energy input per unit area under multiple scan conditions. As the number of scans increases, the material surface undergoes multiple heating and cooling cycles, and the cumulative energy input has a decisive influence on the final roughness.
[0036] The above overlap rate along the scanning direction Line overlap rate Energy per unit area and cumulative energy per unit area Four features synergistically constitute the joint feature set of this invention. This joint feature set is not a simple list of arbitrary physical quantities, but rather a joint physical characterization of the laser-induced roughness formation mechanism, starting from multiple scale processes including pulse spatial distribution, local overlap, trajectory coverage, and cumulative input. Among them, and Together they described the distribution density and continuity of the texturing pits on the two-dimensional surface; This reflects the energy input intensity of a single scan; This reflects the cumulative enhancement or attenuation effect of multiple scans. These four features decouple and characterize the masculinization process from different dimensions, complementing each other and being irreplaceable.
[0037] In some implementations, to further improve fitting accuracy, an auxiliary enhancement feature, such as single-pulse energy, may be introduced. and single-pulse energy normalization characteristics ,in . This is used to characterize the utilization level of the current single-pulse energy relative to the maximum single-pulse energy of the equipment, which helps the model learn the constraints of the equipment's capability ceiling on process feasibility. Additionally, for comparative verification, the duty cycle DC and single-pulse surface energy can also be calculated. However, they do not constitute the essential content of the core joint feature set of this invention.
[0038] Feature dataset construction
[0039] Reference Figure 3 After calculating the joint features of all samples, a roughened feature dataset is constructed for machine learning training. The input feature vector X of the training dataset includes: material type, initial surface roughness Ra0, and the aforementioned core joint feature set. Furthermore, in some implementations, some or all of the original laser processing parameters, as well as auxiliary enhancement features, may be selectively included. The output label Y represents the surface roughness after processing. .
[0040] In this embodiment, the LightGBM tree model is used as the roughness prediction model. The tree model was chosen because it can automatically handle nonlinear interactions between features, is insensitive to the dimensions of input features, and has fast training speed and relatively low overfitting risk. During training, five-fold cross-validation is used to evaluate model stability, and Bayesian optimization is used to fine-tune hyperparameters (such as maximum tree depth, learning rate, number of leaf nodes, etc.). After training, the mapping function is obtained: ,in, To predict roughness, For the input feature vector, This is the trained mapping model.
[0041] Recommended process parameters
[0042] like Figure 4 As shown, in practical applications, the user inputs the material type, initial surface roughness, and target surface roughness. Then, the system enters the process parameter recommendation process.
[0043] First, the system generates multiple sets of candidate process parameter combinations based on equipment capability constraints and process feasibility constraints. These constraints may include: single-pulse energy not exceeding a preset proportion of the equipment's maximum single-pulse energy; average laser power within the equipment's allowable range; laser frequency within the allowable frequency range; scanning speed, scanning interval, and number of scan passes within the available process range; and pulse width taken from the discrete pulse width set supported by the equipment.
[0044] Secondly, the system automatically calculates the corresponding joint feature vector for each set of candidate process parameter combinations and constructs candidate input feature vectors. Input it into the trained roughness prediction model to obtain the corresponding predicted roughness. .
[0045] Then, based on the deviation between the predicted roughness and the target roughness, and in conjunction with equipment capability constraints and process feasibility constraints, the system filters candidate parameter combinations and determines the recommended process parameter combination.
[0046] In this way, the process parameter recommendation of the present invention is not simply based on empirical settings of the original equipment parameters, but rather, under the constraints of equipment capability and process feasibility, it completes the roughening process parameter recommendation based on the joint feature vector of pulse overlap and energy accumulation, thereby improving the accuracy and stability of target roughness control.
[0047] Comparative experiments of different input feature schemes
[0048] To verify the effectiveness of the core joint feature set constructed in this invention for predicting laser-textured roughness, we set up six sets of comparative experiments with different input feature schemes under the same training and test set partitioning. In this embodiment, to examine the impact of different feature schemes on the model's prediction performance, the following six sets of input features were set: Group A is the original parameter group, which uses material type, initial surface roughness, average laser power, laser frequency, pulse width, scanning speed, scanning interval, number of scans, spot diameter, and maximum single pulse energy of the device as input features; Group B1 has a single feature. The group uses material type, initial surface roughness, and single-pulse energy as input features; Group B2 has a single characteristic. The group uses material type, initial surface roughness, and single-pulse surface energy as input features; Group B3 is the basic physical feature group, which uses material type, initial surface roughness, single pulse energy, duty cycle, single pulse surface energy, and single pulse energy normalization feature as input features. Group C is a joint feature group, which uses material type, initial surface roughness, overlap rate along the scanning direction, line overlap rate, energy per unit area, and cumulative energy per unit area as input features; Group D is the joint feature enhancement group, which further introduces single pulse energy and single pulse energy normalization features based on Group C to form an enhanced joint feature vector.
[0049] Under the same training and testing conditions, the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) were used to evaluate each model group. 2 The results are shown in Table 2. The experiments used the same training / test sample set (randomly partitioned, with 80% training and 20% test), and maintained consistent model structure and hyperparameter settings to eliminate interference from other variables in the performance comparison.
[0050] Table 2 Comparison of roughness prediction results for different input feature schemes
[0051] The following conclusions can be drawn from Table 2: (1) The prediction errors of groups B1 and B2, which only use a single energy feature, are very large, R 2 A value less than 0.57 indicates that single-pulse energy or single-pulse surface energy cannot describe the effects of pulse overlap and cumulative input during the roughening process alone, and is insufficient for accurate roughness prediction.
[0052] (2) The basic physical feature group B3 contains several common physical quantities, and its prediction accuracy (R) 2 =0.809) is higher than that of a single feature group but still lower than that of a combined feature group C. This indicates that even with the introduction of multiple physical quantities, the prediction performance remains limited if the correct feature combination is not selected for the formation mechanism of roughness.
[0053] (3) Although the core joint feature group C has significantly fewer input features (5, including material type and initial roughness) than group A (more than 10), its MAE decreases to 0.256 μm, RMSE decreases to 0.415 μm, and R 2 The score improved to 0.908, which is better than Group A's 0.797. This indicates that the performance improvement of Group C is not due to a simple increase in information content, but rather to the synergistic physical characterization of the roughness formation process by the four features.
[0054] (4) Group D, by further introducing single-pulse energy and its normalization characteristics based on Group C, achieved the best results, with a MAE of only 0.098 μm and R 2The accuracy is as high as 0.986. This indicates that the auxiliary enhancement features can further improve the model's fitting accuracy to the roughness results while maintaining the core joint physical representation logic, making it suitable for scenarios with extremely high prediction accuracy requirements.
[0055] Furthermore, to examine the generalization ability of different feature schemes under cross-device or cross-process window conditions, this embodiment uses the maximum single-pulse energy of the device. and spot diameter The samples were grouped (e.g., the data were divided into high-energy group and low-energy group, and large spot group and small spot group), and the extrapolation prediction error of the model was tested by leave-one-out cross-validation. The results are shown in Table 3.
[0056] Table 3 Comparison of extrapolation prediction errors for different feature schemes under different processing condition groups
[0057] As shown in Table 3, the average extrapolation MAE of the core joint feature group C is 0.794 μm, which is much lower than the 0.919 μm of the original parameter group A, and the extrapolation RMSE is also significantly reduced. This indicates that group C has a more stable characterization ability under different processing conditions and can better transfer to unseen combinations of equipment parameters. The extrapolation error of group D is further reduced, indicating that auxiliary enhancement features also help improve consistency across conditions.
[0058] ablation experiment
[0059] To verify that the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area must all exist simultaneously in the joint feature vector, this invention conducts ablation experiments. Based on the complete core joint feature group of group C, one feature was removed from each group, while all other conditions remained unchanged. The experimental results are shown in Table 4. The extrapolation prediction error is also given in the table for comparison.
[0060] Table 4 Comparison of ablation experimental results for core combined feature groups
[0061] As can be seen from Table 4: (1) After removing the overlap rate along the scanning direction (group C-1), the MAE increased from 0.256 μm to 0.355 μm (an increase of 39%), R² decreased from 0.908 to 0.847, and the extrapolated MAE increased from 0.794 to 0.808. This indicates that the overlap rate along the scanning direction is crucial for characterizing the degree of overlap of pulses in the scanning trajectory direction. The lack of this feature will cause the model to be unable to correctly distinguish the texturing effect under different overlap rates.
[0062] (2) After removing the inter-line overlap rate (group C-2), the MAE increased to 0.357 μm (an increase of 39%), the R² decreased to 0.835, and the extrapolated MAE increased to 0.897. This indicates that the inter-line overlap rate is a key parameter characterizing the coverage between adjacent scan trajectories, and the absence of this feature will significantly reduce the model's ability to predict the uniformity of the lateral distribution of pores.
[0063] (3) After removing the energy per unit area (C-3 group), the MAE increased to 0.298 μm (an increase of 16%), R² decreased slightly to 0.901, and the extrapolated MAE increased to 0.813. Although the decrease was relatively small, it still showed a downward trend. Moreover, energy per unit area and cumulative energy per unit area are not physically equivalent; the former represents the energy input of a single scan, while the latter represents the cumulative effect of multiple scans. Retaining the energy per unit area helps maintain the integrity of the physical relationship between single input and cumulative input, thereby ensuring the integrity of the core joint feature group in physical interpretation. Therefore, this feature should not be deleted due to data fluctuations.
[0064] (4) After removing the cumulative unit area energy (C-4 group), the performance deteriorated most drastically: MAE soared to 0.528 μm (an increase of 106%), R² dropped to 0.751, and extrapolated MAE rose to 1.049 μm (an increase of 32%). This fully demonstrates that the cumulative unit area energy plays a dominant role in characterizing the energy accumulation effect under multiple scanning conditions and is an indispensable core feature in the prediction of roughness.
[0065] Based on the comprehensive ablation experiment results, it can be concluded that each of the four features—overlap rate along the scanning direction, inter-line overlap rate, energy per unit area, and cumulative energy per unit area—contributes substantially to prediction accuracy, and the absence of any one feature will lead to a significant decrease in model performance. Therefore, the core joint feature set of this invention is not an arbitrary splicing of known physical quantities, but a synergistic feature system that has undergone rigorous ablation verification and is necessary for joint use.
[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for recommending laser texturing parameters based on pulse overlap and energy accumulation, characterized in that, Includes the following steps: S1. Establish a laser texturing experimental database to record material type, initial surface roughness, laser processing parameters, and surface roughness results after processing; S2. Based on the laser processing parameters, calculate a joint feature vector consisting of the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area; wherein, The overlap rate along the scanning direction is used to characterize the degree of overlap between adjacent pulses on the same scanning trajectory. The line overlap rate is used to characterize the degree of coverage between adjacent scan trajectories. The energy per unit area is used to characterize the energy input level received per unit area during a single scan. The cumulative energy per unit area is used to characterize the cumulative effect of energy input per unit area under multiple scan conditions; S3. Based on the joint feature vector and experimental data, construct a training dataset and train a roughness prediction model to establish a mapping relationship between the joint feature vector and the surface roughness after processing. S4. When the target roughness is input, candidate process parameter combinations are generated within the constraints of equipment capability and process feasibility. The joint feature vector corresponding to each candidate process parameter combination is calculated and input into the roughness prediction model. Recommended process parameter combinations are selected and output based on the prediction results. Among them, the overlap rate along the scanning direction, the inter-line overlap rate, the energy per unit area, and the cumulative energy per unit area must all exist simultaneously in the joint feature vector to synergistically characterize the multi-scale action mechanism from pulse spatial overlap to energy accumulation during laser texturing.
2. The laser texturing parameter recommendation method according to claim 1, characterized in that, The laser processing parameters include one or more of the following: average laser power, laser frequency, pulse width, scanning speed, scanning interval, number of scans, and spot diameter.
3. The laser texturing parameter recommendation method according to claim 1, characterized in that, The overlap rate along the scanning direction The calculation formula is: ,in The pulse spacing along the scanning direction. For scanning speed, The laser frequency, denoted as the diameter of the light spot.
4. The laser texturing parameter recommendation method according to claim 1, characterized in that, The line overlap rate The calculation formula is: ,in For scanning spacing, denoted as the diameter of the light spot.
5. The laser texturing parameter recommendation method according to claim 1, characterized in that, Energy per unit area The calculation formula is: ,in For single-pulse energy, The average power of the laser. The laser frequency, The pulse spacing along the scanning direction. This represents the scanning interval.
6. The laser texturing parameter recommendation method according to claim 1, characterized in that, The cumulative energy per unit area The calculation formula is: ,in The number of scan passes. Energy per unit area.
7. The laser texturing parameter recommendation method according to claim 1, characterized in that, The joint feature vector also includes single-pulse energy. and / or single-pulse energy normalization characteristics , as an auxiliary enhancement feature.
8. The laser texturing parameter recommendation method according to claim 7, characterized in that, The single-pulse energy normalization feature The calculation formula is: ,in This is the maximum single pulse energy of the device.
9. The laser texturing parameter recommendation method according to claim 1, characterized in that, The roughness prediction model is a tree model.
10. The laser texturing parameter recommendation method according to claim 1, characterized in that, The selection of recommended process parameters is based on the deviation between the predicted roughness and the target roughness, and the recommended parameter combination is determined in combination with equipment capability constraints and process feasibility constraints.