Laser cutting parameter optimization method and system, electronic device and medium

Abstract

The present disclosure provides a laser cutting parameter optimization method, comprising: determining process parameter initial value information of a laser cutting machining process according to task description information of a current laser cutting task, the process parameter initial value information comprising: a dimensionless initial value of at least one process parameter; splicing the task description information of the current laser cutting task and the corresponding process parameter initial value information to obtain task process integration information of the current laser cutting task; inputting the task process integration information of the current laser cutting task into a trained residual compensation neural network, so that the residual compensation neural network outputs corresponding process parameter compensation information according to the input task process integration information, the process parameter compensation information comprising: a dimensionless compensation value of each process parameter; for each process parameter, determining a dimensionless final value of the process parameter according to the dimensionless initial value and the compensation value of the process parameter.

Description

Technical Field

[0001] This disclosure relates to the field of laser cutting technology, and in particular to a laser cutting parameter optimization method and system, electronic equipment and medium thereof. Background Technology

[0002] With the widespread adoption of high-power lasers and precision control technology, laser cutting has become an indispensable material processing method in modern manufacturing. In laser cutting, the setting of process parameters directly determines the cutting quality (such as surface roughness, taper, and slag adhesion) and processing efficiency. Traditional parameter setting methods mainly rely on manual experience or consulting fixed parameter manuals. This approach often proves inefficient and unsuitable when faced with batch variations in materials, aging of equipment optical components, or complex and variable processing tasks. Summary of the Invention

[0003] This disclosure aims to solve at least one of the technical problems existing in the prior art, and proposes a laser cutting parameter optimization method and its system, electronic equipment and medium.

[0004] In a first aspect, embodiments of this disclosure provide a method for optimizing laser cutting parameters, including:

[0005] Based on the task description information of the current laser cutting task, the initial value information of the process parameters of the laser cutting process is determined. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter.

[0006] By combining the task description information of the current laser cutting task with the corresponding initial value information of the process parameters, the task process integration information of the current laser cutting task is obtained.

[0007] The task process integration information of the current laser cutting task is input into the trained residual compensation neural network, so that the residual compensation neural network can output the corresponding process parameter compensation information according to the input task process integration information. The process parameter compensation information includes: the dimensionless compensation value of each process parameter.

[0008] For each of the process parameters, the dimensionless final value of the process parameter is determined based on the dimensionless initial value of the process parameter and the compensation value.

[0009] Optionally, before the step of inputting the task process integration information of the current laser cutting task into a preset residual compensation neural network, the method further includes:

[0010] Multiple historical laser cutting task samples and their corresponding tags are obtained. The historical laser cutting task samples record the task process integration information of the historical laser cutting tasks. The task process integration information is obtained by splicing the task description information of the historical laser cutting tasks and the initial value information of the laser cutting process parameters determined based on the task description information of the historical laser cutting tasks. The tags are the process parameter compensation information corresponding to the historical laser cutting tasks. The dimensionless compensation value of each process parameter in the process parameter compensation information corresponding to the historical laser cutting tasks is the difference between the dimensionless actual value and the dimensionless initial value of the process parameter.

[0011] Using the integrated information of the task process as the model input and the corresponding label as the model output, the preset residual compensation neural network is trained using multiple historical laser cutting task samples and their corresponding labels.

[0012] Optionally, during the training of a preset residual compensation neural network using multiple historical laser cutting task samples and their corresponding labels, the total loss of the residual compensation neural network is... :

[0013]

[0014] in, For mean square error loss, Loss due to physical constraints;

[0015] Mean square error loss :

[0016]

[0017] in, For the network prediction residual of the i-th sample, Let be the difference between the experimentally measured optimal parameter and the empirical parameter for the i-th sample, where n is the total number of samples, and i∈[1,n].

[0018] Physical constraint loss :

[0019]

[0020]

[0021]

[0022]

[0023] in, The penalty coefficient is... The value of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample. and Based on and preset floating coefficient The lower and upper bounds of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample are determined. This represents the value of the j-th dimension in the process parameter compensation information output by the trained model for the i-th sample. express and superposition, .

[0024] Optionally, the task description information includes at least one of the following: physical property information of the material to be cut, equipment property information of the laser cutting equipment, and gas type information of the auxiliary gas used in the laser cutting task.

[0025] Optionally, the physical property information of the material to be cut includes: the dimensionless value of the thickness of the material to be cut. dimensionless value of the material to be cut ;

[0026] The equipment attribute information of the laser cutting equipment includes: the dimensionless value of the rated maximum power of the laser cutting equipment. ;

[0027] The gas type information of the auxiliary gas includes: the dimensionless value of the gas type of the auxiliary gas. ;

[0028] At least one of the following process parameters is included: the set cutting power of the laser cutting equipment, the laser cutting speed, the laser cutting focal point position, and the auxiliary gas pressure;

[0029] The steps for determining the initial values ​​of the laser cutting process parameters based on the task description information of the current laser cutting task include:

[0030] The dimensionless initial value of the set cutting power of the laser cutting equipment in the current laser cutting task is determined based on the following formula (1). :

[0031] …Formula (1)

[0032] in, The dimensionless value of the optimal empirical power preset based on historical data fitting results. The preset thickness sensitivity index, The laser absorption coefficient corresponding to the material to be cut;

[0033] And / or,

[0034] The dimensionless initial value of the laser cutting speed in the current laser cutting task is determined based on the following equation (2). :

[0035] …Formula (2)

[0036] in, The dimensionless value of the optimal empirical velocity preset based on the fitting results of historical data. This is the preset attenuation index of the cutting speed due to the thickness. The preset power gain coefficient, The coefficient representing the impact of current laser cutting quality requirements on the reduction of cutting speed;

[0037] And / or,

[0038] The dimensionless initial value of the laser cutting focal position in the current laser cutting task is determined based on the following equation (3) or equation (4). :

[0039] …Formula (3)

[0040] in, The preset focus shift slope and , The intercept constant is preset and ;

[0041] …Formula (4)

[0042] in, The preset material thickness reference threshold, and The preset downward slope of the focal point is for different thickness segments. and These are preset intercept constants for different thickness ranges. and ;

[0043] And / or,

[0044] The dimensionless initial value of the auxiliary gas pressure in the current laser cutting task is determined based on the following equation (5). :

[0045]

[0046] Where interp() represents the set of historical experience points. linear interpolation function, Let be the dimensionless value of the material thickness within the i-th historical experience point in the historical experience point set. Let be the dimensionless value of the auxiliary gas pressure within the i-th historical experience point in the historical experience point set. This is the pressure correction factor corresponding to the type of auxiliary gas used in the current laser cutting task;

[0047] Optionally, the residual compensation neural network includes:

[0048] The input layer is configured to receive the input task process integration information and generate a corresponding vector;

[0049] The channel attention layer is configured to perform feature enhancement processing on the vector provided by the input layer based on the channel attention mechanism;

[0050] The residual mapping layer includes: multiple stacked residual blocks configured to extract features from the vectors provided by the channel attention layer based on residual operations;

[0051] The fully connected output layer is configured to perform a linear transformation on the feature vector output by the residual mapping layer to obtain the corresponding process parameter compensation information.

[0052] Optionally, for each of the process parameters, the step of determining the dimensionless final value of the process parameter based on the dimensionless initial value of the process parameter and the compensation value includes:

[0053] For each of the process parameters, the dimensionless initial value of the process parameter and the compensation value are summed, and the result is used as the dimensionless final value of the process parameter.

[0054] In a second aspect, embodiments of this disclosure provide a laser cutting parameter optimization system configured to implement the method described in the first aspect, the system comprising:

[0055] The determination module is configured to determine the initial value information of the process parameters of the laser cutting process based on the task description information of the current laser cutting task. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter.

[0056] The splicing module is configured to splice the task description information and the corresponding initial value information of the process parameters of the current laser cutting task to obtain the integrated task process information of the current laser cutting task.

[0057] The processing module is configured to input the task process integration information of the current laser cutting task into the trained residual compensation neural network, so that the residual compensation neural network can output corresponding process parameter compensation information according to the input task process integration information. The process parameter compensation information includes: dimensionless compensation values ​​of each process parameter.

[0058] The calculation module is configured to determine the dimensionless final value of each process parameter based on its dimensionless initial value and the compensation value.

[0059] Thirdly, this disclosure provides an electronic device, characterized in that it includes:

[0060] One or more processors;

[0061] Memory, used to store one or more programs;

[0062] When the one or more programs are executed by the one or more processors, the one or more processors implement the methods as described in any of the first aspects.

[0063] Fourthly, this disclosure provides a computer-readable medium storing a computer program, characterized in that, when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect.

[0064] The technical solution disclosed herein integrates basic operating conditions and initial process parameters. Instead of using the final parameters as the fitting target, the network model learns the correction residuals between the two. Based on the residual mapping logic, a mapping relationship is established from operating conditions to parameter adjustment amounts. Finally, the final values ​​of the process parameters are determined based on the initial process parameters and compensation values. This technical solution has the following advantages: 1) Comprehensive information integration, taking into account both on-site operating conditions and baseline process conditions, ensuring the model closely matches actual processing scenarios; 2) Simplified fitting target, making small residuals easier to learn, reducing the difficulty of model training, thereby significantly reducing dependence on massive training data and improving the convergence speed and optimization efficiency of the model under data-scarce conditions; 3) Correction based on initial parameters, making parameter fluctuations controllable, ensuring logical transparency in the parameter adjustment process, facilitating engineers' understanding, intervention, and trust in system decisions, and achieving human-machine collaborative evolution. Attached Figure Description

[0065] Figure 1 A flowchart illustrating a laser cutting parameter optimization method provided in this embodiment of the disclosure;

[0066] Figure 2 A flowchart illustrating a laser cutting parameter optimization method provided in this embodiment of the disclosure;

[0067] Figure 3 This is a structural block diagram of a laser cutting parameter optimization system provided in this disclosure;

[0068] Figure 4 This is a structural block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0069] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0070] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0071] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0072] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the said feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0073] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0074] In related technologies, to solve the problem of optimizing process parameters in laser cutting, there are mainly two technical approaches:

[0075] One approach is the existing expert system and lookup table method, whose core idea is to establish a mapping database of "material thickness-power-speed-air pressure". Operators select the corresponding parameter group from the menu on the control panel based on the type and thickness of the material to be processed. The equipment controller directly calls these preset values ​​for cutting. The parameters in the database are usually standard values ​​set at the factory, and cannot perceive actual environmental fluctuations on-site. When the actual working conditions deviate from the standard values, cutting defects are easily generated, requiring repeated manual trial cuts and adjustments, resulting in low automation.

[0076] The second type is the pure data-driven AI optimization solution, which aims to overcome the rigidity of expert systems by employing artificial intelligence-based optimization methods. This type of solution typically uses neural networks or deep learning models to fit a large amount of historical cutting data with the final cutting quality score, training an end-to-end predictive model. In practical applications, the AI ​​model directly searches and outputs a set of optimal parameters based on current material information. However, the AI ​​search space is multi-dimensional, easily generating parameter combinations that violate physical principles or exceed equipment limits. Furthermore, pure AI models require massive amounts of precisely labeled historical data to achieve ideal optimization results. In the "cold start" phase of actual production environments or when facing newly introduced material types, insufficient training samples lead to poor model prediction accuracy and slow convergence of the optimization process, making it difficult to meet the real-time requirements of industrial sites. In addition, existing AI solutions often lack transparent decision-making logic; engineers cannot intuitively understand the reasons behind the AI's recommendations of specific parameters, leading to a lack of trust in the system's recommendations by on-site operators, often requiring manual intervention and reducing the level of automation.

[0077] To address at least one of the technical problems existing in the aforementioned related technologies, this disclosure provides a method for optimizing laser cutting parameters. Figure 1 This is a flowchart illustrating a laser cutting parameter optimization method provided in an embodiment of this disclosure. Figure 1 As shown, the laser cutting parameter optimization method provided in this disclosure includes:

[0078] Step S1: Determine the initial value information of the process parameters of the laser cutting process based on the task description information of the current laser cutting task. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter.

[0079] In this disclosure, task description information refers to known information (which can also be understood as task details) used to describe the current laser cutting task. Examples include the physical properties of the material to be cut, the equipment properties of the laser cutting equipment, and the type of auxiliary gas used in the laser cutting task.

[0080] In some embodiments, the physical property information of the material to be cut includes: a dimensionless value of the thickness of the material to be cut. dimensionless value of the material to be cut The equipment attribute information for laser cutting equipment includes: the dimensionless value of the rated maximum power of the laser cutting equipment. Dimensionless value of lens cleanliness The gas type information for the auxiliary gas includes: the dimensionless value of the gas type for the auxiliary gas. .

[0081] Specifically, firstly, raw data describing the known information for the current laser cutting task can be acquired. This includes raw data on the thickness of the material to be cut (e.g., 5mm or 10mm), the material composition (e.g., stainless steel or carbon steel), the rated maximum power of the laser cutting equipment (e.g., 60kW), the cleanliness of the lens (e.g., 80%), and the type of auxiliary gas (e.g., air, oxygen, or nitrogen). Then, the acquired raw data is dimensionless to obtain the corresponding dimensionless values.

[0082] It should be noted that for some non-numerical raw data (such as the material to be cut, the type of auxiliary gas), a preset encoding method can be used to convert it into numerical expression.

[0083] Optionally, the material of the material to be cut is converted into a corresponding digital code (which can be represented by a one-hot vector) using a preset material type coding rule; for example, if the material to be cut is stainless steel, the dimensionless value after digital coding is... =0.8; The material to be cut is carbon steel, and the dimensionless value is obtained after digital encoding. =0.9.

[0084] Optionally, the gas type of the auxiliary gas is converted into a corresponding digital code (which can be represented by a one-hot vector) using a preset auxiliary gas coding rule; for example, if the auxiliary gas is air, the dimensionless value after digital coding is... =0.7; the auxiliary gas is nitrogen, and the dimensionless value is obtained after digital encoding. =0.8; the auxiliary gas is oxygen, and the dimensionless value is obtained after digital encoding. =0.9.

[0085] In this disclosure, normalization can be used to achieve dimensionless data, as follows:

[0086]

[0087] in, The original data values, and These are the mean and standard deviation from the historical database, respectively. This is the dimensionless value after normalization.

[0088] In step S1, based on the task description information of the current laser cutting task, the dimensionless initial value (i.e., initial value information of the process parameter) of at least one process parameter in the laser cutting process can be determined manually or automatically by an algorithm. This process of determining the initial value information of the process parameter in step S1 can be understood as the process of initially determining the laser cutting parameters.

[0089] Step S2: Combine the task description information and the corresponding initial value information of the process parameters of the current laser cutting task to obtain the integrated task process information of the current laser cutting task.

[0090] Step S3: Input the task process integration information of the current laser cutting task into the trained residual compensation neural network so that the residual compensation neural network can output the corresponding process parameter compensation information according to the input task process integration information. The process parameter compensation information includes: dimensionless compensation value of each process parameter.

[0091] The residual compensation neural network trained in this disclosure takes task process integration information (information obtained by splicing task description information and initial value information of process parameters) as input and outputs process parameter compensation information (dimensionless compensation value of each process parameter) as output.

[0092] Step S4: For each process parameter, determine the dimensionless final value of the process parameter based on the dimensionless initial value and compensation value of the process parameter.

[0093] In step S4, optionally, for each process parameter, the dimensionless initial value and the compensation value of the process parameter are summed, and the result is used as the dimensionless final value of the process parameter.

[0094] In this disclosure, the basic characteristics of the operating conditions and the initial process parameters are integrated. The network model does not use the final parameters as the fitting target, but instead learns the correction residuals between the two, and establishes a mapping relationship from operating conditions to parameter adjustment amounts based on the residual mapping logic. The technical solution of this disclosure has the following advantages: 1) Comprehensive information integration, taking into account both on-site operating conditions and benchmark process conditions, and modeling that fits the actual processing scenario; 2) Simplified fitting target, small residuals are easier to learn, reducing the difficulty of model training, thereby significantly reducing the dependence on massive training data and improving the convergence speed and optimization efficiency of the model under data-scarce conditions; 3) Correction based on initial parameters, parameter fluctuations are controllable, making the parameter adjustment process logically transparent, which is convenient for engineers to understand, intervene in and trust the system's decisions, and realize human-machine collaborative evolution.

[0095] In some embodiments, during step S1, the task description information also includes the dimensionless value of the thickness of the material to be cut. Dimensionless value of the material to be cut Dimensionless value of the rated maximum power of laser cutting equipment Dimensionless value of lens cleanliness Dimensionless value of the type of auxiliary gas Taking five-dimensional data as an example, the task description information can be represented by vector X as follows:

[0096]

[0097] In some embodiments, at least one of the following process parameters is included: the set cutting power of the laser cutting equipment, the laser cutting speed, the laser cutting focal position, and the auxiliary gas pressure.

[0098] The dimensionless initial values ​​of each process parameter can be determined based on the following steps Sa to Sd:

[0099] Step Sa: Determine the dimensionless initial value of the set cutting power of the laser cutting equipment in the current laser cutting task based on the following formula (1). :

[0100] …Formula (1)

[0101] in, The dimensionless value of the optimal empirical power preset based on historical data fitting results. The preset thickness sensitivity index (usually) ), This is the laser absorption coefficient corresponding to the material to be cut.

[0102] It should be noted that the laser absorptivity coefficient This is a dimensionless correction factor for a material's ability to absorb laser energy of a specific wavelength. It characterizes the difference in the efficiency of a material in converting laser energy into heat energy. The larger the value, the higher the utilization rate of the laser. For example, carbon steel is 1.0, stainless steel is 0.9, and aluminum alloy is 0.7. It can characterize the difficulty of laser cutting of materials. For example, the laser cutting difficulty of aluminum alloy is greater than that of stainless steel, and the laser cutting difficulty of stainless steel is greater than that of carbon steel.

[0103] Step Sb: Determine the dimensionless initial value of the laser cutting speed in the current laser cutting task based on the following formula (2). :

[0104] …Formula (2)

[0105] in, The dimensionless value of the optimal empirical velocity preset based on the fitting results of historical data. The attenuation index of cutting speed with respect to the preset thickness (usually greater than 1, indicating that the thicker the material, the faster the speed decreases). This is a preset power gain coefficient (usually less than 1, for example, 0.8). The value represents the impact of the current laser cutting task's cutting quality requirements on the reduction of cutting speed (ranging from 0 to 1; generally, the higher the cutting quality requirements, the smaller the value, for example, 0.8 for fine cutting, 0.9 for ordinary cutting, and 1.0 for coarse cutting).

[0106] In this disclosure, the laser cutting speed is inversely proportional to the thickness and directly proportional to the power. The above equation (2) in the disclosure adopts a hyperbolic attenuation model.

[0107] Step Sc: Determine the dimensionless initial value of the laser cutting focal position in the current laser cutting task based on the following equation (3) or equation (4). :

[0108] …Formula (3)

[0109] in, The preset focus downward slope and , The intercept constant is preset and ;

[0110] …Formula (4)

[0111] in, The preset material thickness reference threshold, and The preset downward slope of the focal point is for different thickness segments. and These are preset intercept constants for different thickness ranges. and .

[0112] In this disclosure, the laser cutting focal point is the position of the focal point relative to the upper surface of the material to be cut (the side facing the laser cutting equipment). The upper surface of the material to be cut is taken as point "0". When the value is positive, it indicates that the laser cutting focal point is above the upper surface of the material to be cut. When the value is negative, it indicates that the laser cutting focal point is inside the material to be cut. The focal point usually shifts linearly with the thickness (as shown in equation (3) above) or piecewise linearly (as shown in equation (4) above).

[0113] It should be noted that when “segmented linear shift” (i.e., as shown in the above formula (4)) is used, in order to ensure a smooth transition of the focal position, the values ​​of the two segments at T=Tth are equal. At this time, the following equation relationship exists: δ1*Tth+C1=δ2*Tth+C2. By rearranging, the relationship between C2 and C1 is: C2=C1+(δ1-δ2)*Tth.

[0114] Step Sd: Determine the dimensionless initial value of the auxiliary gas pressure in the current laser cutting task based on the following equation (5). :

[0115]

[0116] Where interp() represents the set of historical experience points. linear interpolation function, Let be the dimensionless value of the material thickness within the i-th historical experience point in the historical experience point set. Let be the dimensionless value of the auxiliary gas pressure within the i-th historical experience point in the historical experience point set. This is the pressure correction factor corresponding to the type of auxiliary gas used in the current laser cutting task;

[0117] It should be noted that different gases have different cutting characteristics and require different gas pressures. This disclosure introduces a gas pressure correction coefficient, which allows for the use of the same set of interpolation points to adapt to the gas pressure requirements of different gases. For example, oxygen: commonly used for carbon steel cutting, mainly relies on the exothermic oxidation reaction, requiring low auxiliary gas pressure; the gas pressure correction coefficient is taken as a baseline value of 1.0. Nitrogen: commonly used for stainless steel / aluminum alloy cutting, mainly relies on high pressure to blow away molten metal, requiring higher gas pressure; the gas pressure correction coefficient is taken as 1.2 to 1.5. Air: its cutting effect is between that of oxygen and nitrogen; the gas pressure correction coefficient is taken as 1.1.

[0118] In this disclosure, taking as an example at least one process parameter that simultaneously includes: the set cutting power of the laser cutting equipment, the laser cutting speed, the laser cutting focal position, and the auxiliary gas pressure, in four-dimensional data, the initial value information of the process parameters can be represented by a vector. Represented as:

[0119]

[0120] Accordingly, in step S2, the task-process integration information obtained by concatenating the task description information with the initial value information of the process parameters can be represented by a vector. Represented as:

[0121]

[0122] Accordingly, in step S3, the process parameter compensation information output by the residual compensation neural network can be represented by vectors. (Essentially a residual correction vector) is represented as:

[0123]

[0124] That is, the model input of the residual compensation neural network is The model output is Accordingly, in step S4, the dimensionless final values ​​of each process parameter can be obtained based on the following formula:

[0125]

[0126] in, The vector representation of the dimensionless final values ​​of all process parameters is shown below, and the calculation expression for each process parameter is as follows:

[0127]

[0128] It should be noted that the above task description information includes 5-dimensional data, while the initial value information of process parameters and the compensation information of process parameters both include 4-dimensional data. This is only one optional implementation scheme in this disclosure and will not limit the technical solution of this disclosure.

[0129] It should be noted that step S4 above yields the dimensionless final values ​​of the process parameters. The actual data values ​​of each process parameter can then be obtained through the inverse operation of the aforementioned dimensionless processing. The process of deriving the actual data values ​​from the dimensionless values ​​is a conventional technique in this field and will not be elaborated upon here.

[0130] Figure 2 This is a flowchart illustrating a laser cutting parameter optimization method provided in an embodiment of this disclosure. Figure 2 As shown, the laser cutting parameter optimization method provided in this disclosure not only includes steps S1 to S4 in the previous embodiments, but also includes steps S01 and S02 before step S1; only steps S01 and S02 will be described in detail below.

[0131] Step S01: Obtain multiple historical laser cutting task samples and their corresponding labels.

[0132] Among them, the historical laser cutting task samples record the task process integration information of historical laser cutting tasks. The task process integration information is obtained by splicing the task description information of historical laser cutting tasks and the initial value information of the laser cutting process parameters determined based on the task description information of historical laser cutting tasks. The tag is the process parameter compensation information corresponding to the historical laser cutting task. The dimensionless compensation value of each process parameter in the process parameter compensation information corresponding to the historical laser cutting task is the difference between the dimensionless actual value and the dimensionless initial value of the process parameter.

[0133] It should be noted that the implementation methods for obtaining the task description information of historical laser cutting tasks, the initial value information of the laser cutting process parameters determined based on the task description information of historical laser cutting tasks, and the method of splicing the task description information and the initial value information of the process parameters to obtain the task process integration information can refer to the relevant content described in steps S1 and S2 above, and will not be repeated here.

[0134] It should be noted that, for each historical laser cutting task sample, after determining the initial values ​​of the laser cutting process parameters based on the task description information, to prevent the AI ​​from searching for parameters that violate physical principles, boundary constraints can be generated based on the initial values ​​of the process parameters for that historical laser cutting task sample. :

[0135]

[0136] That is, during the training process, dimensionless values ​​(using vectors) of different process parameters for the historical laser cutting task sample were defined. The boundary of (representation) is the vector corresponding to the initial value information of the process parameters. Set the floating ratio around the center. (For example, The value is 0.2).

[0137] Step S02: Using task process integration information as model input and corresponding labels as model output, train the preset residual compensation neural network using multiple historical laser cutting task samples and their corresponding labels.

[0138] In some embodiments, during the training of a preset residual compensation neural network using multiple historical laser cutting task samples and their corresponding labels, the total loss of the residual compensation neural network is... :

[0139]

[0140] in, For mean square error loss, Loss due to physical constraints;

[0141] Mean square error loss :

[0142]

[0143] in, For the network prediction residual of the i-th sample, Let be the difference between the experimentally measured optimal parameter and the empirical parameter for the i-th sample, where n is the total number of samples, and i∈[1,n].

[0144] Physical constraint loss :

[0145]

[0146]

[0147]

[0148]

[0149] in, The penalty coefficient is... Let j be the value of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample. and Based on and preset floating coefficient The lower and upper bounds of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample are determined. This represents the value of the j-th dimension in the process parameter compensation information output by the trained model for the i-th sample. express and superposition, .

[0150] In this disclosure, by introducing a hard constraint mechanism based on physical experience during the training phase, the search space of the AI ​​is limited, ensuring that the generated parameters are always within a safe and feasible physical range, and preventing dangerous working conditions from occurring.

[0151] In some embodiments, the residual compensation neural network includes: an input layer, a channel attention layer, a residual mapping layer, and a fully connected output layer. The input layer is configured to receive input task process integration information and generate a corresponding vector (i.e., the aforementioned vector). The channel attention layer is configured to perform feature enhancement processing on the vectors provided by the input layer based on the channel attention mechanism; the residual mapping layer includes multiple stacked residual blocks, and the residual mapping layer is configured to extract features from the vectors provided by the channel attention layer based on residual operations; the fully connected output layer is configured to perform linear transformation on the feature vectors output by the residual mapping layer to obtain the corresponding process parameter compensation information.

[0152] In this disclosure, in order to automatically identify the importance of each feature under different working conditions (such as ultra-thin plate vs. ultra-thick plate), a channel attention layer is introduced before the residual mapping layer to perform feature weighting. The working process of this channel attention layer mainly includes the following three steps:

[0153] Step ①: Process the input tensor Global max pooling and global average pooling are performed separately to extract feature statistics. Global max pooling is used to capture the most significant "strong features" in the input features (e.g., power is often the dominant factor in thick plate cutting), and its calculation formula is as follows:

[0154]

[0155] Global average pooling is used to capture the "overall distribution trend" of input features (such as the combined influence of background parameters like ambient temperature and air pressure). Its calculation formula is as follows:

[0156]

[0157] in, Indicates the index of the feature channel (corresponding to different input parameter dimensions). Represents the spatial domain of the feature map. For spatial coordinate index, This represents the total number of pixels in the feature map (i.e., the size of the spatial domain). Indicates the input feature map at the th Channel, coordinates The value at that location.

[0158] Step 2: Calculate the attention weight vector using a multilayer perceptron shared network. :

[0159]

[0160] in, is the activation function, MLP() is the multilayer perceptron, AvgPool() is the average pooling process, and MaxPool() is the max pooling process.

[0161] Step 3: Weight the original features to obtain enhanced features. :

[0162]

[0163] in, This indicates element-wise multiplication. This step ensures that the network can adaptively focus on key features such as power or speed based on the material thickness.

[0164] In the residual mapping layer, let Traditional networks directly fit the target mapping (i.e., the true parameter corrections) that the network needs to fit. This scheme fits the residual function .

[0165] No. The output of each residual block The calculation formula is:

[0166]

[0167] in, This is the output of the previous residual block. The residual mapping learned for the current residual block (usually consisting of two fully connected layers and the nonlinear activation function ReLU). ( ) represents an identity mapping or linear projection.

[0168] By stacking N residual blocks, the network can deeply explore the weak correlation between thickness, power and cutting quality, and effectively avoid the gradient vanishing problem.

[0169] The technical solution disclosed herein first utilizes prior physical rules regarding material thickness and equipment power to construct an initial solution space and safety constraint boundary for process parameters. Then, through a deep neural network model, using the initial parameters as input conditions, it mines the nonlinear coupling characteristics between multiple process parameters, performing residual compensation and refined optimization for key variables such as focal position and gas dynamics parameters. Finally, based on real-time cutting quality feedback, it dynamically updates the parameter correlation graph. This solution effectively addresses the problems of poor adaptability of traditional empirical methods and the difficulties in cold start and weak interpretability of pure AI methods, truly achieving efficient and robust optimization of the laser cutting process.

[0170] Figure 3 This is a structural block diagram of a laser cutting parameter optimization system provided in this disclosure. Figure 3 As shown, the laser cutting parameter optimization system can implement the laser cutting parameter optimization method provided in any of the previous embodiments. The laser cutting parameter optimization system includes: a determination module, a splicing module, a processing module, and a calculation module.

[0171] The determination module is configured to determine the initial value information of the laser cutting process parameters based on the task description information of the current laser cutting task. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter.

[0172] The splicing module is configured to splice the task description information and the corresponding initial value information of the process parameters of the current laser cutting task to obtain the integrated task process information of the current laser cutting task.

[0173] The processing module is configured to input the task process integration information of the current laser cutting task into the trained residual compensation neural network, so that the residual compensation neural network can output the corresponding process parameter compensation information based on the input task process integration information. The process parameter compensation information includes the dimensionless compensation value of each process parameter.

[0174] The calculation module is configured to determine the dimensionless final value of each process parameter based on its dimensionless initial value and compensation value.

[0175] For a description of each of the above functional modules, please refer to the corresponding content in the previous method embodiments, which will not be repeated here.

[0176] Based on the same inventive concept, this disclosure also provides an electronic device. Figure 4 This is a structural block diagram of an electronic device provided in an embodiment of this disclosure. Figure 4 As shown, this disclosure provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement any of the laser cutting parameter optimization methods described in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, configured to enable information interaction between the processor and the memory.

[0177] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus).

[0178] In some embodiments, the processor 101, memory 102, and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0179] In some embodiments, the one or more processors 101 include a field-programmable gate array.

[0180] This disclosure also provides a computer-readable medium. The computer-readable medium stores a computer program, which, when executed by a processor, implements the steps of any of the laser cutting parameter optimization methods described in the above embodiments. The computer-readable storage medium may be volatile or non-volatile.

[0181] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described laser cutting parameter optimization method.

[0182] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0183] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0184] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0185] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0186] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0187] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0188] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0189] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0190] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0191] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for optimizing laser cutting parameters, characterized in that, include: Based on the task description information of the current laser cutting task, the initial value information of the process parameters of the laser cutting process is determined. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter. By combining the task description information of the current laser cutting task with the corresponding initial value information of the process parameters, the task process integration information of the current laser cutting task is obtained. The task process integration information of the current laser cutting task is input into the trained residual compensation neural network, so that the residual compensation neural network can output the corresponding process parameter compensation information according to the input task process integration information. The process parameter compensation information includes: the dimensionless compensation value of each process parameter. For each of the process parameters, the dimensionless final value of the process parameter is determined based on the dimensionless initial value of the process parameter and the compensation value.

2. The method according to claim 1, characterized in that, Before the step of inputting the task process integration information of the current laser cutting task into the preset residual compensation neural network, the following steps are also included: Multiple historical laser cutting task samples and their corresponding tags are obtained. The historical laser cutting task samples record the task process integration information of the historical laser cutting tasks. The task process integration information is obtained by splicing the task description information of the historical laser cutting tasks and the initial value information of the laser cutting process parameters determined based on the task description information of the historical laser cutting tasks. The tags are the process parameter compensation information corresponding to the historical laser cutting tasks. The dimensionless compensation value of each process parameter in the process parameter compensation information corresponding to the historical laser cutting tasks is the difference between the dimensionless actual value and the dimensionless initial value of the process parameter. Using the integrated information of the task process as the model input and the corresponding label as the model output, the preset residual compensation neural network is trained using multiple historical laser cutting task samples and their corresponding labels.

3. The method according to claim 2, characterized in that, During the training of a preset residual compensation neural network using multiple historical laser cutting task samples and their corresponding labels, the total loss of the residual compensation neural network is... : ； in, For mean square error loss, Loss due to physical constraints; Mean square error loss : ； in, For the network prediction residual of the i-th sample, Let be the difference between the experimentally measured optimal parameter and the empirical parameter for the i-th sample, where n is the total number of samples, and i∈[1,n]. Physical constraint loss : ； ； ； ； in, The penalty coefficient is... The value of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample. and Based on and preset floating coefficient The lower and upper bounds of the j-th dimension in the initial value information of the process parameters corresponding to the i-th sample are determined. This represents the value of the j-th dimension in the process parameter compensation information output by the trained model for the i-th sample. express and superposition, .

4. The method according to any one of claims 1 to 3, wherein, The task description information includes at least one of the following: physical property information of the material to be cut, equipment property information of the laser cutting equipment, and gas type information of the auxiliary gas used in the laser cutting task.

5. The method according to claim 4, wherein, The physical property information of the material to be cut includes: the dimensionless value of the thickness of the material to be cut. dimensionless value of the material to be cut ; The equipment attribute information of the laser cutting equipment includes: the dimensionless value of the rated maximum power of the laser cutting equipment. ; The gas type information of the auxiliary gas includes: the dimensionless value of the gas type of the auxiliary gas. ; At least one of the following process parameters is included: the set cutting power of the laser cutting equipment, the laser cutting speed, the laser cutting focal point position, and the auxiliary gas pressure; The steps for determining the initial values ​​of the laser cutting process parameters based on the task description information of the current laser cutting task include: The dimensionless initial value of the set cutting power of the laser cutting equipment in the current laser cutting task is determined based on the following formula (1). : …Formula (1) in, The dimensionless value of the optimal empirical power preset based on historical data fitting results. The preset thickness sensitivity index, The laser absorption coefficient corresponding to the material to be cut; And / or, The dimensionless initial value of the laser cutting speed in the current laser cutting task is determined based on the following equation (2). : …Formula (2) in, The dimensionless value of the optimal empirical velocity preset based on the fitting results of historical data. This is the preset attenuation index of the cutting speed due to the thickness. The preset power gain coefficient, The coefficient representing the impact of current laser cutting quality requirements on the reduction of cutting speed; And / or, The dimensionless initial value of the laser cutting focal position in the current laser cutting task is determined based on the following equation (3) or equation (4). : …Formula (3) in, The preset focus downward slope and , The intercept constant is preset and ; …Formula (4) in, The preset material thickness reference threshold, and The preset downward slope of the focal point is for different thickness segments. and These are preset intercept constants for different thickness ranges. and ; And / or, The dimensionless initial value of the auxiliary gas pressure in the current laser cutting task is determined based on the following equation (5). : ； Where interp() represents the set of historical experience points. linear interpolation function, Let be the dimensionless value of the material thickness within the i-th historical experience point in the historical experience point set. Let be the dimensionless value of the auxiliary gas pressure within the i-th historical experience point in the historical experience point set. This is the pressure correction factor corresponding to the type of auxiliary gas used in the current laser cutting task.

6. The method according to claim 1, wherein, The residual compensation neural network includes: The input layer is configured to receive the input task process integration information and generate a corresponding vector; The channel attention layer is configured to perform feature enhancement processing on the vector provided by the input layer based on the channel attention mechanism; The residual mapping layer includes: multiple stacked residual blocks configured to extract features from the vectors provided by the channel attention layer based on residual operations; The fully connected output layer is configured to perform a linear transformation on the feature vector output by the residual mapping layer to obtain the corresponding process parameter compensation information.

7. The method according to claim 1, wherein, For each of the process parameters, the step of determining the dimensionless final value of the process parameter based on the dimensionless initial value and the compensation value includes: For each of the process parameters, the dimensionless initial value of the process parameter and the compensation value are summed, and the result is used as the dimensionless final value of the process parameter.

8. A laser cutting parameter optimization system, characterized in that, The system, configured to implement the method as described in any one of claims 1 to 7, comprises: The determination module is configured to determine the initial value information of the process parameters of the laser cutting process based on the task description information of the current laser cutting task. The initial value information of the process parameters includes: at least one dimensionless initial value of the process parameter. The splicing module is configured to splice the task description information and the corresponding initial value information of the process parameters of the current laser cutting task to obtain the integrated task process information of the current laser cutting task. The processing module is configured to input the task process integration information of the current laser cutting task into the trained residual compensation neural network, so that the residual compensation neural network can output corresponding process parameter compensation information according to the input task process integration information. The process parameter compensation information includes: dimensionless compensation values ​​of each process parameter. The calculation module is configured to determine the dimensionless final value of each process parameter based on its dimensionless initial value and the compensation value.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images