CN-121988905-A - Intelligent process-matched galvanometer laser cutting text creation product customization generation method

Abstract

The invention discloses an intelligent process-matched vibrating mirror laser cutting text-created product customization generation method which comprises the steps of obtaining text-created vector data and material thermophysical parameters, constructing a heterogeneous graph data structure containing a node-side relationship, inputting the heterogeneous graph data structure into a pre-trained graph annotation force network model, extracting high-dimensional features representing geometric density and topological structure of a graph, predicting a transient space-time thermal field when a laser spot moves by utilizing a thermodynamic physical information neural network in combination with a thermal conduction physical law, and adaptively solving an optimal process parameter combination by taking a heat affected zone threshold as a boundary to generate a numerical control code to finish cutting. The invention solves the problem that personalized text creation processing relies on manual trial and error and heat accumulation to cause damage through cooperation of a physical mechanism and deep learning, and remarkably improves the processing yield and the custom response speed of complex patterns.

Inventors

• ZHANG TIANCHENG
• WANG CHENHUI
• LIU HUIYING

Assignees

• 安徽机电职业技术学院

Dates

Publication Date

20260508

Application Date

20260310

Priority Date

20251231

Claims (6)

1. 1. The intelligent process-matched galvanometer laser cutting text creation product customization generation method is characterized by comprising the following steps of: S1, acquiring original vector graphic data of a to-be-processed text-created product and thermophysical attribute parameters of a target processing material, analyzing geometric primitive types, coordinate positions and topological connection relations in the vector graphic data, and constructing a heterogeneous graph data structure containing various node types and side relations; S2, inputting the heterogeneous graph data structure into a pre-trained heterogeneous graph attention network model, calculating the aggregation weights of different types of neighborhood nodes to a central node by utilizing a multi-head attention mechanism, and extracting high-dimensional feature vectors representing the local geometric density degree and the global topological structure of the graph; S3, taking the high-dimensional feature vector as geometric condition constraint, combining the thermophysical attribute parameters, inputting the geometric condition constraint into a thermodynamic physical information neural network, and predicting transient space-time thermal field distribution data when a laser spot moves along a processing path by utilizing a neuron layer constrained by a network embedded by a thermal conduction physical law; S4, according to the transient space-time thermal field distribution data, taking a preset material heat affected zone critical threshold as a constraint boundary, and adaptively solving the optimal laser process parameter combination meeting the thermal control requirement aiming at different processing primitive areas; S5, generating a numerical control machining code comprising a vibrating mirror deflection coordinate sequence and a laser energy modulation signal according to the optimal laser process parameter combination, and transmitting the numerical control machining code to a laser execution terminal to finish cutting operation.
2. 2. The method according to claim 1, wherein the step S1 of constructing a heterogeneous graph data structure including a plurality of node types and side relationships includes the following sub-steps: S11, analyzing the original vector graphic data, and discretizing the graphic into a graphic represented by the original vector graphic data Primitive set composed of geometric line segments And extracting all endpoints and intersections of the geometric line segments to form a connection point set ; S12, defining a heterogeneous map Wherein the node sets By corresponding to the primitive set Is of the first type of node(s) And corresponding to the set of connection points Is of the second type Composition; S13, constructing a structure edge set based on the topological connection relation If and only if a geometric line segment Is connected with the end point of (C) At the corresponding node And (3) with A connecting edge is established between the two parts; s14, constructing a heat-associated edge set based on geometric space adjacency For any two nodes of the first type that do not share a connection point Calculate the corresponding geometric line segment And (3) with Shortest euclidean distance between If the formula is satisfied Then at And (3) with Establishing a thermal association edge between them, wherein Is a preset thermal influence coefficient, and the temperature of the heat-sensitive material is controlled to be the preset thermal influence coefficient, For thermal diffusivity in the thermophysical property parameter, Is a preset laser single-point action time constant.
3. 3. The method according to claim 1, wherein said step S2 comprises the following sub-steps: s21, for each node in the heterogeneous graph data structure Initializing original feature vectors And mapping different types of nodes to the feature space of the same dimension to obtain projection vectors The mapping formula is Wherein Representing nodes Type of (f) A learner projection matrix of a corresponding type; s22, calculating central node by utilizing multi-head attention mechanism With its neighborhood nodes Between the first two Correlation coefficient of individual attention heads The calculation formula is Wherein The number representing the attention header is given, For the total number of heads to be the same, Is the first The weight vectors of the individual attention headers, Is the first The feature transformation matrix of the individual attention heads, The vector concatenation operation is represented by a vector, Is an activation function; S23, carrying out normalization processing on the correlation coefficient to obtain normalized attention weight The calculation formula is Wherein Is a node Is a neighborhood node set; s24, carrying out weighted aggregation on neighborhood node characteristics based on the normalized attention weight, and splicing the outputs of all attention heads to obtain the high-dimensional characteristic vector The calculation formula is Wherein As a function of the non-linear activation, Representing a stitching operation of the multi-headed feature.
4. 4. A method according to claim 3, characterized in that said step S3 comprises the following sub-steps: S31, determining the current processing path point Associated primitive node Obtaining the corresponding high-dimensional feature vector Constructing an input vector comprising spatial coordinates, temporal and geometric features ; S32, inputting the input vector Inputting to the thermodynamic physical information neural network, outputting a predicted temperature value ; S33, the training and predicting process of the thermodynamic physical information neural network is constrained by a transient heat conduction partial differential equation, the equation describes the relation between the heat flux density generated by laser spot movement and a material temperature field, and the mathematical expression is as follows: Wherein, the For the temperature field distribution, For the thermal diffusivity as defined in claim 2, As the laser absorptivity of the material, For the moment of time Is used for the laser power of the (c), In order to achieve a material density of the material, For the specific heat capacity of the material, In order to be a material thickness, Is the radius of the laser spot, For the moment of time Instantaneous coordinates of the laser spot center, said 、 、 、 Belongs to the thermophysical property parameter of claim 1.
5. 5. The method according to claim 4, wherein said step S4 comprises the following sub-steps: s41, defining a laser processing process parameter set Wherein For the laser power to be high, For the frequency of the laser pulses, Is the laser scanning speed, which is defined as ; S42, constructing an optimization function aiming at maximizing processing efficiency And in meeting the heat affected zone critical threshold Solving the optimal parameters under the constraint condition, wherein the mathematical model is as follows: Wherein, the In order to achieve a total processing time, the total processing time, To be centered on the light spot As the center of a circle Is a localized heat affected zone of radius, And The minimum and maximum power limits of the laser respectively, Maximum linear energy density allowed for the material; s43, carrying out iterative solution on the mathematical model by adopting a gradient descent method to obtain an optimal laser process parameter combination at each moment 。
6. 6. An intelligent process-matched galvanometer laser cut literature custom generation system for performing the method of any one of claims 1 to 5, comprising: The heterogeneous composition construction module is used for acquiring original vector graphic data of a to-be-processed text-created product and thermophysical attribute parameters of a target processing material, analyzing geometric primitive types, coordinate positions and topological connection relations in the vector graphic data, and constructing a heterogeneous graph data structure containing various node types and side relations; The feature extraction module is used for inputting the heterogeneous graph data structure into a pre-trained heterogeneous graph annotation force network model, calculating the aggregation weights of different types of neighborhood nodes to the center node by utilizing a multi-head attention mechanism, and extracting high-dimensional feature vectors representing the local geometric density degree and the global topological structure of the graph; The thermal field prediction module is used for taking the high-dimensional feature vector as geometric condition constraint, combining the thermal physical attribute parameters, inputting the high-dimensional feature vector into a thermodynamic physical information neural network, and predicting transient space-time thermal field distribution data when a laser spot moves along a processing path by utilizing a neuron layer constrained by a network embedded by a thermal conduction physical law; The parameter optimization module is used for adaptively solving the optimal laser process parameter combination meeting the thermal control requirement aiming at different processing primitive areas by taking a preset material heat affected zone critical threshold as a constraint boundary according to the transient space-time thermal field distribution data; And the control generation module is used for generating a numerical control processing code comprising a galvanometer deflection coordinate sequence and a laser energy modulation signal according to the optimal laser process parameter combination, and transmitting the numerical control processing code to a laser execution terminal to finish cutting operation.

Description

Intelligent process-matched galvanometer laser cutting text creation product customization generation method Technical Field The invention relates to the technical field of laser precision machining and artificial intelligence auxiliary manufacturing, in particular to an intelligent process matching galvanometer laser cutting text product customization generation system. Background With the deep fusion of the "industry 4.0" and "C2M (Consumer to Manufacturer, consumer direct manufacturing)" modes, the literature product customization market presents a explosive growth situation. The laser hollowed-out paper carving lamp, the wooden assembly model, the acrylic art ornament and other literature products become hot spots in the current market because of strong design sense, high individuation degree and quick updating iteration. The core of the manufacture of such products is the laser fine machining technology, in particular the galvanometer laser cutting system (Galvanometer Laser Cutting System). Compared with the traditional gantry laser cutting machine, the vibrating mirror system realizes beam deflection by utilizing the X-Y axis high-speed scanning vibrating mirror and the F-theta field lens, has extremely high processing speed and dynamic response capability, and meets the requirement of the natural literature product on the rapid forming of complex patterns. In the existing galvanometer laser processing technology system, the data processing flow generally follows a linear mode of CAD design-CAM path planning-process parameter setting-processing execution. A user or designer firstly uses vector drawing software (such as CorelDRAW, illuster, autoCAD) to generate a two-dimensional vector graphic in DXF, PLT or SVG format, then an operator introduces the file into laser marking or cutting control software (such as EzCad, scanMaster and the like), key process parameters such as laser power, pulse frequency, scanning speed, jump Delay (Jump Delay), switching light Delay and the like are set in the software, and finally the system drives a galvanometer motor to finish machining according to the generated instructions. Although this technology route is relatively mature in standardized, mass production, the prior art still exposes significant limitations in the face of the production scenario of "multi-class, small-lot, highly customized" literature products, mainly in terms of both insufficient degree of intelligence in process matching and hysteresis in thermal effect control. The prior art lacks efficient resolution of deep mapping mechanisms between pattern geometry and process parameters. The current laser CAM software mainly generates a motion instruction based on the geometric locus (such as a straight line, an arc and a spline curve) of a graph, and part of high-end software has a simple corner automatic speed reduction function or a small circle speed limiting function, but the essence of the software is still to locally adjust based on a simple geometric rule, so that the overall topological structure and the density distribution of the graph cannot be understood. Graphics of the literature often have extremely high complexity, such as a paper-cut style vector diagram, which includes both large-area physical reserved areas and extremely dense hollowed-out texture areas. If uniform process parameters are adopted, large-area areas are possibly cut out and not transparent due to insufficient energy, and dense texture areas can cause serious heat accumulation (Heat Accumulation) due to high path overlapping degree and short heat dissipation time, so that the burning, carbonization and even firing of materials are caused, and the ornamental value and the yield of finished products are seriously affected. The prior art is highly dependent on manual experience and "trial and Error" in process parameter settings. Because of the wide variety of processing materials (special paper with different gram weights, basswood with different densities, leather and the like) of the literature product, the thermal physical properties (heat conductivity coefficient, specific heat capacity and thermal diffusivity) of each material are greatly different. In practice, operators often need to perform multiple trial cuts in the face of a new design pattern or a new material, and repeatedly adjust the focal length, power and speed by visually observing the slit width (KERF WIDTH) and the degree of edge charring. The mode relying on the experience of the master is low in efficiency, cannot meet the instant response requirement of the user ordering, namely production, in the e-commerce mode, and causes expensive sample material waste. For thermal field control during laser machining, existing solutions rely mainly on Finite Element Analysis (FEA) software (such as ANSYS, COMSOL) for offline simulation. Although the FEA method can accurately calculate the temperature field distribution inside the material based on the thermal co