A mechanism- and data-driven laser processing parameter optimization method

Abstract

The present invention discloses a method for optimizing laser processing parameters driven by a hybrid mechanism and data, comprising the following steps: establishing a neural network model embedded with a mechanism, and training it to obtain a proxy model; obtaining the current process parameters and the mechanism characteristics matching the current process parameters; inputting the current process parameters and the mechanism characteristics matching the current process parameters into the proxy model, and outputting the predicted values; calculating the adaptability function of the current process parameters, and performing selection, crossover, and mutation operations on the current process parameters to update the current process parameters; through continuous iteration, the adaptability function of the updated process parameters meets the optimization target conditions. The present invention solves the problem that the proxy model based on deep learning is used in the prior art to optimize the process parameters, but the proxy model based on deep learning lacks interpretability, which makes it difficult to understand the physical mechanism information of the temperature field during laser processing.

Description

Technical Field

[0001] The invention relates to the technical field of laser processing parameter optimization, and in particular to a laser processing parameter optimization method driven by a hybrid mechanism and data. Background Art

[0002] Laser processing is a high-precision and high-efficiency material processing method that uses a laser beam to locally heat the material and utilizes the material's melting, gasification or chemical reaction at high temperatures to achieve material processing operations. Compared with traditional processing methods, laser processing has many advantages such as no contact, no wear, high precision, and high speed, and is widely used in industry, medical care, scientific research and other fields.

[0003] In laser processing, process parameters refer to a series of key variables and settings that affect the quality, efficiency and effect of processing. The selection and adjustment of these parameters directly determine the accuracy, speed and surface quality of processing. Correct selection and optimization of process parameters are the key to achieving high-quality laser processing. At present, commonly used process parameter optimization methods include numerical simulation and theoretical models, Taguchi method, response surface method, multivariate analysis, etc. These methods can systematically study the influence of each parameter and find the best parameter combination. However, there are some defects in the above methods: although numerical simulation and theoretical models can intuitively present information such as temperature distribution and heat-affected zone during processing, help understand the processing mechanism and optimize parameters, they are usually based on a series of assumptions and simplified conditions, so the accuracy of the model is limited and may not fully reflect the complex situation in the actual processing process. The relationship between process parameters and processing quality presents nonlinear characteristics. As the dimension of parameters increases, this nonlinear relationship becomes more complex and difficult to determine. Optimization methods such as Taguchi method, multivariate analysis, response surface method, etc. rely on specific mathematical models, but in high-dimensional and nonlinear problems, these models are often oversimplified and lack sufficient flexibility to cope with the complexity of the system.

[0004] Deep learning has gradually emerged as an important branch of machine learning. It uses multi-layer neural networks to learn more complex feature representations from large-scale data, and can effectively solve the problem of modeling high-dimensional nonlinear laser processing systems. Deep learning technology can be used to establish a proxy model between process parameters and processing quality. By adjusting the input parameters of the proxy model, the output of the proxy model is close to the target value. Multi-objective optimization algorithms are commonly used in the industrial field. They can iteratively generate different combinations of process parameters, input them into the proxy model, and output processing performance indicators. This two-stage process of system modeling and process parameter optimization provides an objective and standardized method for optimizing process parameters. However, the proxy model based on deep learning lacks interpretability, which makes it difficult to understand the physical mechanism information of the temperature field during laser processing, thereby affecting its optimization of process parameters. Summary of the invention

[0005] In view of the above-mentioned defects, the present invention proposes a laser processing parameter optimization method driven by a hybrid of mechanism and data, aiming to solve the problem that the process parameters are optimized using a deep learning-based proxy model in the prior art, but the deep learning-based proxy model lacks interpretability, which makes it difficult to understand the physical mechanism information of the temperature field during laser processing, thereby affecting its optimization of process parameters.

[0006] To achieve this object, the present invention adopts the following technical solutions:

[0007] A laser processing parameter optimization method driven by a combination of mechanism and data, comprising the following steps:

[0008] Step S1: Establishing a mechanism model, a historical process parameter data set, and a mechanism characteristic data set under different process parameters;

[0009] Step S2: extracting several key temperature features from the mechanism model under different process parameters;

[0010] Step S3: establishing a mechanism-embedded neural network model according to different process parameters and several key temperature characteristics, and training the mechanism-embedded neural network model to obtain a proxy model;

[0011] Step S4: Determine the optimization variables, constraint parameter intervals of the optimization variables and optimization target conditions;

[0012] Step S5: randomly generating process parameters within the constraint parameter interval of the optimization variable;

[0013] Step S6: performing KNN clustering on a process parameter in the historical process parameter data set to obtain index values ​​of k process parameters, where k is a positive integer;

[0014] Step S7: according to the index values ​​of the k process parameters, the corresponding mechanism characteristics are retrieved from the mechanism characteristic data set, and all the mechanism characteristics are weighted and summed to generate the mechanism characteristics matching the current process parameters;

[0015] Step S8: input the current process parameters and the mechanism characteristics matching the current process parameters into the proxy model, and output the predicted values;

[0016] Step S9: Calculate the adaptability function of the current process parameters according to the predicted values, and perform selection, crossover, and mutation operations on the current process parameters to update the current process parameters;

[0017] Step S10: Repeat steps S6-S9 to make the adaptability function of the updated process parameters meet the optimization target conditions until a preset maximum number of iterations is reached to obtain the optimal process parameters.

[0018] Preferably, in step S1, establishing a mechanism model under different process parameters includes the following sub-steps:

[0019] Step S11: geometrically modeling the processed product to obtain a geometric model of the processed product;

[0020] Step S12: obtaining thermal characteristic data of the geometric model of the processed product under different working conditions and analyzing the data;

[0021] Step S13: network the geometric model of the processed product to establish a finite element model, and set product material properties and relevant process parameters of laser processing;

[0022] Step S14: setting the solution response to the thermal field of the laser processing process, and visualizing the transient temperature and thermal stress distribution during the laser processing process.

[0023] Preferably, in step S2, the following sub-steps are specifically included:

[0024] Step S21: acquiring a dynamic temperature field image from a mechanism model under different process parameters;

[0025] Step S22: parsing the dynamic temperature field image into multiple key frames;

[0026] Step S23: establishing a mapping relationship between color and temperature field on each key frame;

[0027] Step S24: Capture the color information of each key frame, and extract key temperature features according to the color information of each key frame.

[0028] Preferably, in step S4, the optimization variable is V=(x CS , x PO , x PW , x FR , x DE ) T , where V represents the process parameter variable group; x CS Parameter variable representing cutting speed; x PO The parameter variable representing power; x PW The parameter variable representing the pulse width; x FR The parameter variable representing the frequency; x DE represents the parameter variable of the defocus amount; T represents the transpose of the matrix V;

[0029] The constraint parameter interval of the optimization variable is

[0030] The optimization target condition is

[0031] Among them, f1 represents the objective function of process parameters and cross-section burrs; f2 represents the objective function of process parameters and cut surface burrs; f3 represents the objective function of process parameters and heat-affected zone width.

[0032] Preferably, in step S6, the following sub-steps are specifically included:

[0033] Step S61: Calculate the distance between a process parameter point and each process parameter point in the historical process parameter data set;

[0034] Step S62: sorting all process parameter points from small to large according to the distance between a process parameter point and each process parameter point in the historical process parameter data set;

[0035] Step S63: Select k process parameter points with the smallest distances and output corresponding index values.

[0036] Preferably, in step S3, the proxy model includes a process parameter input branch layer, a mechanism feature input branch layer, a feature fusion layer, and an output layer;

[0037] In step S8, the following sub-steps are specifically included:

[0038] Step S81: inputting the current process parameters and the mechanism features matching the current process parameters into the process parameter input branch layer and the mechanism feature input branch layer of the proxy model respectively, and outputting two low-dimensional features;

[0039] Step S82: two low-dimensional features are input into the feature fusion layer of the proxy model for splicing, and the spliced ​​features are output;

[0040] Step S83: The spliced ​​feature input and output layers output the predicted values ​​of the cross-section burrs, the cut surface burrs and the width of the heat-affected zone respectively.

[0041] Preferably, in step S9, the current process parameters are selected, crossed, and mutated, which specifically includes the following sub-steps:

[0042] Step S91: performing non-dominated sorting on the current process parameters and calculating the congestion degree of the individual current process parameters;

[0043] Step S92: selecting a parent individual according to the crowding degree of the current process parameter individual;

[0044] Step S93: Perform crossover and mutation operations on the selected parent individuals.

[0045] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0046] In this scheme, a proxy model is obtained by establishing and training a mechanism-embedded neural network model, and the proxy model is used to optimize the process parameters to achieve high-quality product laser processing. Compared with the traditional proxy model based on deep learning, the mechanism-embedded neural network model of this scheme not only takes into account the powerful learning ability of deep learning neural networks, but also combines the physical mechanism information of the temperature field, which improves the interpretability of the proxy model and enables it to understand the physical mechanism information of the temperature field during laser processing, thereby helping to optimize the process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of the steps of the laser processing parameter optimization method driven by a hybrid of mechanism and data. DETAILED DESCRIPTION

[0048] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0049] A laser processing parameter optimization method driven by a combination of mechanism and data, comprising the following steps:

[0050] Step S1: Establishing a mechanism model, a historical process parameter data set, and a mechanism characteristic data set under different process parameters;

[0051] Step S2: extracting several key temperature features from the mechanism model under different process parameters;

[0052] Step S3: establishing a mechanism-embedded neural network model according to different process parameters and several key temperature characteristics, and training the mechanism-embedded neural network model to obtain a proxy model;

[0053] Step S4: Determine the optimization variables, constraint parameter intervals of the optimization variables and optimization target conditions;

[0054] Step S5: randomly generating process parameters within the constraint parameter interval of the optimization variable;

[0055] Step S6: performing KNN clustering on a process parameter in the historical process parameter data set to obtain index values ​​of k process parameters, where k is a positive integer;

[0056] Step S7: according to the index values ​​of the k process parameters, the corresponding mechanism characteristics are retrieved from the mechanism characteristic data set, and all the mechanism characteristics are weighted and summed to generate the mechanism characteristics matching the current process parameters;

[0057] Step S8: input the current process parameters and the mechanism characteristics matching the current process parameters into the proxy model, and output the predicted values;

[0058] Step S9: Calculate the adaptability function of the current process parameters according to the predicted values, and perform selection, crossover, and mutation operations on the current process parameters to update the current process parameters;

[0059] Step S10: Repeat steps S6-S9 to make the adaptability function of the updated process parameters meet the optimization target conditions until a preset maximum number of iterations is reached to obtain the optimal process parameters.

[0060] This scheme is a mechanism and data hybrid driven laser processing parameter optimization method, such as Figure 1As shown, the first step is to establish a mechanism model, a historical process parameter data set and a mechanism feature data set under different process parameters. In this embodiment, by establishing a mechanism model under different process parameters, it is beneficial to obtain a dynamic temperature field. By establishing a historical process parameter data set, it is beneficial to store and manage process parameters. By establishing a mechanism feature data set, it is beneficial to store and manage the mechanism features in the mechanism model. The second step is to extract several key temperature features from the mechanism model under different process parameters. In this embodiment, since the burrs on the cross section and cut surface of the product after laser cutting are usually the result of the coupling of temperature and thermal stress, and the heat affected zone is mainly related to temperature, before subsequently establishing a mechanism embedded neural network model, it is necessary to extract key temperature features from the mechanism model, wherein the key temperature features include maximum temperature, minimum temperature, average temperature, temperature gradient, temperature kurtosis and temperature skewness. The third step is to establish a mechanism embedded neural network model according to different process parameters and several key temperature features, and train the mechanism embedded neural network model to obtain a proxy model. In this embodiment, by establishing a mechanism embedded neural network model, it is beneficial to optimize the process parameters based on the mechanism embedded neural network model in the future. The mechanism-embedded neural network model includes two input branch layers, namely, a process parameter input branch layer and a mechanism feature input branch layer. The process parameter input branch layer is composed of several fully connected layers, and the mechanism feature input branch layer is composed of a one-dimensional convolution layer, a tiling layer, and several fully connected layers. In addition, by training the mechanism-embedded neural network model, it is beneficial to improve the accuracy of the prediction of the mechanism-embedded neural network model. The fourth step is to determine the optimization variables, the constraint parameter interval of the optimization variables, and the optimization target conditions. In one embodiment, taking laser cutting as an example, the optimization variables are power, speed, defocus, frequency, and pulse width; the constraint parameter interval of the optimization variables is the constraint parameter interval of power, speed, defocus, frequency, and pulse width; the optimization target condition is to minimize cross-sectional burrs, cutting surface burrs, and heat-affected zones. The fifth step is to randomly generate process parameters within the constraint parameter interval of the optimization variables. In one embodiment, taking laser cutting as an example, a process parameter is randomly generated from the constraint parameter interval of power, speed, defocus, frequency, and pulse width, thereby forming a group of process parameters, and inputting the proxy model for iterative optimization. The sixth step is to perform KNN clustering on a process parameter in the historical process parameter data set to obtain index values ​​of k process parameters, where k is a positive integer. In this embodiment, the K nearest neighbor (KNN) algorithm is used to cluster a process parameter in the historical process parameter data set to obtain the index value of the process parameter that is most similar to the process parameter.The seventh step is to retrieve the corresponding mechanism features in the mechanism feature data set according to the index values ​​of k process parameters, and perform weighted summation on all mechanism features to generate mechanism features matching the current process parameters. In this embodiment, by searching the mechanism feature data set by the index value, it is possible to ensure that the mechanism features that accurately correspond to the process parameters are obtained. The comprehensive value of the mechanism features obtained by weighted summation can evaluate the comprehensive effect under different process parameter combinations. The eighth step is to input the current process parameters and the mechanism features matching the current process parameters into the proxy model and output the predicted values. In this embodiment, the current process parameters and the mechanism features matching the current process parameters are first input into the process parameter input branch layer and the mechanism feature input branch layer of the proxy model respectively, and can be extracted into two low-dimensional features, and then the two low-dimensional features are spliced ​​in the feature fusion layer of the proxy model, and finally the spliced ​​features are output in the output layer of the proxy model, and the output is the predicted value of the quality index of the product processed by the current process parameters. In one embodiment, taking laser cutting as an example, the quality index of the product processed by the current process parameters includes cross-section burrs, cutting surface burrs and heat-affected zone width. The ninth step is to calculate the adaptability function of the current process parameters according to the predicted values, and select, cross, and mutate the current process parameters to update the current process parameters. In this embodiment, by calculating the adaptability function of the current process parameters, it is beneficial to analyze whether the quality indicators of the products processed and formed under the current process parameters are in the optimal state. By selecting, crossing, and mutating the current process parameters, the process parameters are updated in the most optimized direction. The tenth step is to repeat steps S6-S9 so that the adaptability function of the updated process parameters meets the optimization target conditions until the preset maximum number of iterations is reached to obtain the optimal process parameters. In one embodiment, taking laser cutting as an example, the preset maximum number of iterations is 500 times. By continuously iterating and updating the process parameters composed of power, speed, defocus, frequency, and pulse width, when the set of process parameters is iterated to 500 times, the set of process parameters finally obtained is the optimal process parameters. Under the optimal process parameters, the cut product has the smallest cross-section burrs, cutting surface burrs, and heat-affected zone width.

[0061] In one embodiment, taking laser cutting as an example, by continuously adjusting the process parameters composed of power, speed, defocus, frequency and pulse width, a laser cutting experiment is carried out based on these process parameters to obtain different product quality indicators under different process parameters, and these data are recorded in the laser cutting experiment process parameter information table. The laser cutting experiment process parameter information table is shown in Table 1 below. The cutting object is a thin-walled plate with a thickness of 0.2 mm. The laser used is a pulsed laser. The process parameters related to the pulsed laser are: defocus, power, cutting speed, frequency and pulse width. Three levels are set in each process parameter, and there are 243 groups in the comprehensive experiment. After the product is cut, the molding quality of the slit position is particularly important. The main molding quality indicators are: cross-section burrs, cutting surface burrs and heat-affected zone width. The optimization goal of the process parameters is to minimize cross-section burrs, cutting surface burrs and heat-affected zone width.

[0062]

[0063] Table 1

[0064] In this scheme, a proxy model is obtained by establishing and training a mechanism-embedded neural network model, and the proxy model is used to optimize the process parameters to achieve high-quality product laser processing. Compared with the traditional proxy model based on deep learning, the mechanism-embedded neural network model of this scheme not only takes into account the powerful learning ability of deep learning neural networks, but also combines the physical mechanism information of the temperature field, which improves the interpretability of the proxy model and enables it to understand the physical mechanism information of the temperature field during laser processing, thereby helping to optimize the process parameters.

[0065] Preferably, in step S1, establishing a mechanism model under different process parameters includes the following sub-steps:

[0066] Step S11: geometrically modeling the processed product to obtain a geometric model of the processed product;

[0067] Step S12: obtaining thermal characteristic data of the geometric model of the processed product under different working conditions and analyzing the data;

[0068] Step S13: network the geometric model of the processed product to establish a finite element model, and set product material properties and relevant process parameters of laser processing;

[0069] Step S14: setting the solution response to the thermal field of the laser processing process, and visualizing the transient temperature and thermal stress distribution during the laser processing process.

[0070] In this embodiment, in step S11, geometric modeling describes the shape and size of the processed product, and the product is converted into a digital geometric model suitable for computer processing by computer-aided design (CAD) software. In step S12, by analyzing the thermal characteristic data, the physical quantities that need to be calculated, such as heat flux, etc., can be determined. In step S13, by meshing the geometric model, the key details in the geometric model can be captured, thereby improving the accuracy of the calculation. By establishing a finite element model, complex geometric shapes and material properties can be processed, making it possible to analyze complex products. By setting the material properties of the product, it is easy to ensure that the geometric model reflects the real physical behavior of the product during the simulation process. By setting the relevant process parameters of laser processing, such as power, pulse width, frequency, defocus, speed, etc., it is easy to ensure the quality of the processed product. In step S14, by visualizing the transient temperature and thermal stress distribution during the laser processing process, the staff can more intuitively understand the complex process of the interaction between the laser and the material, which helps to reveal the processing mechanism and provide theoretical support for process optimization.

[0071] Preferably, in step S2, the following sub-steps are specifically included:

[0072] Step S21: acquiring a dynamic temperature field image from a mechanism model under different process parameters;

[0073] Step S22: parsing the dynamic temperature field image into multiple key frames;

[0074] Step S23: establishing a mapping relationship between color and temperature field on each key frame;

[0075] Step S24: Capture the color information of each key frame, and extract key temperature features according to the color information of each key frame.

[0076] In this embodiment, a dynamic process multi-factor feature extraction method is proposed for the extraction of key temperature features. First, the dynamic temperature field image is parsed into multiple key frames, 11 in this embodiment. In each frame, the color value of each pixel in the image is obtained, usually the RGB value. Based on the "jet" standard color mapping in the "Colormap", a mapping of color and temperature is established, and the color value is mapped back to the temperature value. Usually, the RGB value needs to be converted into a scalar value first, and then the temperature value is reversely searched according to the known color mapping. Based on this, the global temperature data in each key frame can be obtained. Based on these temperature data, 13 key information including maximum temperature, minimum temperature, average temperature, temperature variance, median, kurtosis, skewness, hot spot, gradient, high temperature area, and entropy are further calculated and extracted. This information covers all the element characteristics of temperature. There is a strong correlation between the relevant features and burrs and heat-affected zones, which has a strong auxiliary significance for system modeling.

[0077] Preferably, in step S4, the optimization variable is V=(x CS , x PO , x PW , x FR , x DE ) T , where V represents the process parameter variable group; x CS Parameter variable representing cutting speed; x PO The parameter variable representing power; x PW The parameter variable representing the pulse width; x FR The parameter variable representing the frequency; x DE represents the parameter variable of the defocus amount; T represents the transpose of the matrix V;

[0078] The constraint parameter interval of the optimization variable is

[0079] The optimization target condition is

[0080] Among them, f1 represents the objective function of process parameters and cross-section burrs; f2 represents the objective function of process parameters and cut surface burrs; f3 represents the objective function of process parameters and heat-affected zone width.

[0081] In one embodiment, taking laser cutting as an example, by continuously iteratively optimizing the process parameters consisting of power, speed, defocus, frequency and pulse width, the ultimate optimization goal is to minimize cross-sectional burrs, cutting surface burrs and heat-affected zone width.

[0082] Preferably, in step S6, the following sub-steps are specifically included: step S61: calculate the distance between a process parameter point and each process parameter point in the historical process parameter data set; step S62: sort all process parameter points in ascending order according to the distance between a process parameter point and each process parameter point in the historical process parameter data set; step S63: select k process parameter points with the smallest distance, and output the corresponding index values.

[0083] In this embodiment, KNN clustering is performed on a process parameter in a historical process parameter data set, and Euclidean distance is selected as the distance measurement method. The distance between a new process parameter point and each process parameter point in the historical process parameter data set is calculated, and k process parameter points with the smallest distance are selected. The results output under these process parameter points can predict and estimate the possible output results under the new process parameter points.

[0084] Preferably, in step S3, the proxy model includes a process parameter input branch layer, a mechanism feature input branch layer, a feature fusion layer, and an output layer;

[0085] In step S8, the following sub-steps are specifically included: step S81: input the current process parameters and the mechanism characteristics matching the current process parameters into the process parameter input branch layer and the mechanism characteristics input branch layer of the proxy model respectively, and output two low-dimensional features; step S82: the two low-dimensional features are input into the feature fusion layer of the proxy model for splicing, and the spliced ​​features are output; step S83: the spliced ​​features are input into the output layer, and the predicted values ​​of the cross-section burrs, the cutting surface burrs and the width of the heat-affected zone are output respectively.

[0086] In one embodiment, taking laser cutting as an example, the output layer of the proxy model has three branches, which respectively predict the cross-section burr, the cutting surface burr and the width of the heat-affected zone. The proxy model of this solution integrates two modes of process parameters and mechanism characteristics to achieve hybrid drive of mechanism and data.

[0087] Preferably, in step S9, selection, crossover and mutation operations are performed on the current process parameters, which specifically include the following sub-steps: non-dominated sorting of the current process parameters and calculating the crowding of the current process parameter individuals; selecting the parent individual according to the crowding of the current process parameter individuals; and performing crossover and mutation operations on the selected parent individual.

[0088] In this embodiment, the NSGA-II algorithm is used to perform genetic iteration on the process parameters. In step S91, the current process parameters are process parameters of different combinations. The algorithm sorts the process parameters of different combinations according to non-dominated sorting. Through non-dominated sorting, the process parameters of different combinations are divided into multiple fronts, wherein the first front contains all non-dominated individuals, the second front contains non-dominated individuals among the individuals dominated by the first front individuals, and so on. After calculating the non-dominated sorting, for each individual in the front, calculating its crowding degree helps to maintain the diversity of the population in the subsequent selection process. In step S92, the algorithm will give priority to selecting individuals from the front with higher levels. If there are multiple individuals in the same front, individuals with higher crowding degree are given priority. This selection mechanism is intended to ensure that the selected individuals have both good fitness and maintain the diversity of the population. In step S93, the algorithm can continuously update the process parameter combination by performing crossover and mutation operations on the selected parent individuals, and gradually approach the optimal solution through fitness evaluation and selection mechanisms.

[0089] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0090] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A laser processing parameter optimization method driven by a combination of mechanism and data, characterized in that: The following steps are involved: Step S1: Establishing a mechanism model, a historical process parameter data set, and a mechanism characteristic data set under different process parameters; Step S2: extracting several key temperature features from the mechanism model under different process parameters; In step S2, the following sub-steps are specifically included: Step S21: acquiring a dynamic temperature field image from a mechanism model under different process parameters; Step S22: parsing the dynamic temperature field image into multiple key frames; Step S23: establishing a mapping relationship between color and temperature field on each key frame; Step S24: capturing the color information of each key frame, and extracting key temperature features according to the color information of each key frame; Among them, the key temperature characteristics include maximum temperature, minimum temperature, average temperature, temperature gradient, temperature kurtosis and temperature skewness; Step S3: establishing a mechanism-embedded neural network model according to different process parameters and several key temperature characteristics, and training the mechanism-embedded neural network model to obtain a proxy model; Step S4: Determine the optimization variables, constraint parameter intervals of the optimization variables and optimization target conditions; Step S5: randomly generating process parameters within the constraint parameter interval of the optimization variable; Step S6: performing KNN clustering on a process parameter in the historical process parameter data set to obtain index values ​​of k process parameters, where k is a positive integer; Step S7: according to the index values ​​of the k process parameters, the corresponding mechanism characteristics are retrieved from the mechanism characteristic data set, and all the mechanism characteristics are weighted and summed to generate the mechanism characteristics matching the current process parameters; Step S8: inputting the current process parameters and the mechanism characteristics matching the current process parameters into the proxy model, and outputting the predicted values, wherein the predicted values ​​are the predicted values ​​of the cross-section burr, the cutting surface burr and the width of the heat-affected zone respectively; Step S9: Calculate the adaptability function of the current process parameters according to the predicted values, and perform selection, crossover, and mutation operations on the current process parameters to update the current process parameters; Step S10: Repeat steps S6-S9 to make the adaptability function of the updated process parameters meet the optimization target conditions until a preset maximum number of iterations is reached to obtain the optimal process parameters.

2. The method for optimizing laser processing parameters driven by a combination of mechanism and data according to claim 1, characterized in that: In step S1, a mechanism model under different process parameters is established, including the following sub-steps: Step S11: geometrically modeling the processed product to obtain a geometric model of the processed product; Step S12: obtaining thermal characteristic data of the geometric model of the processed product under different working conditions and analyzing the data; Step S13: network the geometric model of the processed product to establish a finite element model, and set product material properties and relevant process parameters of laser processing; Step S14: setting the solution response to the thermal field of the laser processing process, and visualizing the transient temperature and thermal stress distribution during the laser processing process.

3. The method for optimizing laser processing parameters driven by a hybrid mechanism and data according to claim 1, characterized in that: In step S4, the optimization variable is V=(x CS , x PO , x PW , x FR , x DE ) T , where V represents the process parameter variable group; x CS Parameter variable representing cutting speed; x PO The parameter variable representing power; x PW The parameter variable representing the pulse width; x FR The parameter variable representing the frequency; x DE represents the parameter variable of the defocus amount; T represents the transpose of the matrix V; The constraint parameter interval of the optimization variable is The optimization target condition is Among them, f1 represents the objective function of process parameters and cross-section burrs; f2 represents the objective function of process parameters and cut surface burrs; f3 represents the objective function of process parameters and heat-affected zone width.

4. The method for optimizing laser processing parameters driven by a combination of mechanism and data according to claim 1, characterized in that: In step S6, the following sub-steps are specifically included: Step S61: Calculate the distance between a process parameter point and each process parameter point in the historical process parameter data set; Step S62: sorting all process parameter points from small to large according to the distance between a process parameter point and each process parameter point in the historical process parameter data set; Step S63: Select k process parameter points with the smallest distances and output corresponding index values.

5. The method for optimizing laser processing parameters driven by a hybrid mechanism and data according to claim 1, characterized in that: In step S3, the proxy model includes a process parameter input branch layer, a mechanism feature input branch layer, a feature fusion layer, and an output layer; In step S8, the following sub-steps are specifically included: Step S81: inputting the current process parameters and the mechanism features matching the current process parameters into the process parameter input branch layer and the mechanism feature input branch layer of the proxy model respectively, and outputting two low-dimensional features; Step S82: two low-dimensional features are input into the feature fusion layer of the proxy model for splicing, and the spliced ​​features are output; Step S83: The spliced ​​feature input and output layers output the predicted values ​​of the cross-section burrs, the cut surface burrs and the width of the heat-affected zone respectively.

6. The method for optimizing laser processing parameters driven by a combination of mechanism and data according to claim 1, characterized in that: In step S9, the current process parameters are selected, crossed, and mutated, which specifically includes the following sub-steps: Step S91: performing non-dominated sorting on the current process parameters and calculating the congestion degree of the individual current process parameters; Step S92: selecting a parent individual according to the crowding degree of the current process parameter individual; Step S93: Perform crossover and mutation operations on the selected parent individuals.

Images