Intelligent control system and method of crystal glass laser internal engraving machine

Abstract

The invention relates to the technical field of intelligent control, and discloses an intelligent control system and method for a crystal glass laser internal engraving machine. The method comprises the following steps: acquiring three-dimensional point cloud data of a crystal glass workpiece, extracting a material basic feature vector, and inputting the material basic feature vector into an axial self-crossover attention module for three-dimensional space feature association to obtain a three-dimensional space state map; fusing the acoustic emission signal, the light scattering signal and the temperature signal based on the three-dimensional space state map to perform to-be-excited state discrimination, and determining a to-be-excited area; and adjusting the laser power, the pulse frequency and the focusing depth according to the to-be-excited region to obtain a laser parameter combination, and performing position trajectory and control input optimization on the plurality of laser heads to generate an execution instruction of each laser head. The laser power, the pulse frequency and the focusing depth can be adaptively adjusted according to the recognition result of the to-be-excited area, and then the machining precision and stability of the crystal glass laser internal engraving machine are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to an intelligent control system and method for a crystal glass laser engraving machine. Background Technology

[0002] Traditional laser engraving technology employs a fixed processing mode with preset parameters, lacking real-time sensing capabilities of the material's state. This makes it difficult to address the inherent characteristics of crystal glass, such as uneven refractive index and complex thermal stress distribution, leading to frequent unexcited areas during processing and severely impacting engraving quality and efficiency. When handling complex 3D patterns, the lack of an effective coordination mechanism between laser heads prevents globally optimal processing path planning. Furthermore, the inability to accurately predict and identify unexcited areas during processing results in delayed parameter adjustments, further affecting the quality of laser engraving. Summary of the Invention

[0003] The main objective of this invention is to provide an intelligent control system and method for a crystal glass laser engraving machine. This invention can adaptively adjust the laser power, pulse frequency, and focusing depth according to the identification results of the area to be excited, thereby improving the processing accuracy and stability of the crystal glass laser engraving machine.

[0004] To achieve the above objectives, the present invention provides an intelligent control method for a crystal glass laser engraving machine, comprising the following steps: Three-dimensional point cloud data of crystal glass workpieces are collected, material basic feature vectors are extracted, and the material basic feature vectors are input into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. Based on the three-dimensional spatial state map, acoustic emission signals, light scattering signals, and temperature signals are fused to determine the state to be excited and identify the region to be excited. The laser power, pulse frequency, and focusing depth are adjusted according to the region to be excited to obtain a combination of laser parameters. Based on the combination of laser parameters, the position trajectory and control input of multiple laser heads are optimized to generate execution instructions for each laser head.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of acquiring three-dimensional point cloud data of a crystal glass workpiece, extracting material basic feature vectors, and inputting the material basic feature vectors into an axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map includes: The three-dimensional point cloud data of the crystal glass workpiece is obtained through the CCD vision positioning module, and the three-dimensional spatial coordinate matrix of the three-dimensional point cloud data is extracted. Based on the three-dimensional spatial coordinate matrix, the optical properties of the material are analyzed, and a basic feature vector of the material containing information on the workpiece geometry, refractive index distribution and material density is constructed. The material's basic feature vector is input into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of inputting the material basic feature vector into an axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map includes: The material's basic feature vector is input into the axial self-crossing attention module, where it is mapped to X-axis query matrix, X-axis key matrix, X-axis value matrix, Y-axis query matrix, Y-axis key matrix, Y-axis value matrix, Z-axis query matrix, Z-axis key matrix, and Z-axis value matrix, respectively. Calculate the X-axis attention weight matrix based on the X-axis query matrix, the X-axis key matrix, and the X-axis value matrix; calculate the Y-axis attention weight matrix based on the Y-axis query matrix, the Y-axis key matrix, and the Y-axis value matrix; calculate the Z-axis attention weight matrix based on the Z-axis query matrix, the Z-axis key matrix, and the Z-axis value matrix; The X-axis attention weight matrix, the Y-axis attention weight matrix, and the Z-axis attention weight matrix are weighted and fused to obtain a comprehensive attention feature; Based on the comprehensive attention feature encoding, the physical relationships between points in the three-dimensional space and the feature relationships at different depths are obtained to obtain a three-dimensional spatial state map.

[0007] Optionally, in a third implementation of the first aspect of the present invention, obtaining a three-dimensional spatial state map based on the physical relationships between points in the three-dimensional space and the feature relationships at different depths, using the integrated attention feature encoding, includes: Based on the comprehensive attention features, attention weight values ​​are assigned to the spatial coordinate points corresponding to the three-dimensional point cloud data to obtain associated weight distribution data; Based on the associated weight distribution data, the heat conduction associated paths and material continuity changes between each depth layer are calculated to obtain the characteristic associated data of each depth layer; Based on the feature correlation data of each depth layer, the heat-affected zone during the carving process is predicted, and a three-dimensional spatial state map is generated based on the heat-affected zone.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of determining the region to be excited by fusing acoustic emission signals, light scattering signals, and temperature signals based on the three-dimensional spatial state map includes: Based on the three-dimensional spatial state map, material excitation characteristic parameters at each spatial location are obtained, and a correspondence table between spatial location and material excitation characteristic parameters is established. Acoustic emission signals are collected to calculate acoustic emission power spectral density; light scattering signals are collected and normalized to obtain scattered light intensity; and temperature signals are collected and three-dimensionally interpolated to obtain local temperature values. The acoustic emission power spectral density, scattered light intensity, and local temperature values ​​are compared with the material excitation characteristic parameters in the corresponding relationship table to calculate the excitation state deviation value at each spatial location. The objective function value of the discrimination function for the state to be excited is calculated based on the excitation state deviation value. When the objective function value is less than the set state threshold, it is determined to be a region to be excited.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the step of adjusting the laser power, pulse frequency, and focusing depth according to the region to be excited to obtain a laser parameter combination, and optimizing the position trajectory and control input of multiple laser heads based on the laser parameter combination to generate execution instructions for each laser head, includes: Calculate the excitation state deviation value based on the spatial location of the region to be excited; The laser power adjustment coefficient is determined based on the deviation value of the state to be excited, and the standard laser power is multiplied by the laser power adjustment coefficient to obtain the adjusted laser power. The pulse frequency increment is calculated based on the deviation value of the state to be excited, and the standard pulse frequency is added to the pulse frequency increment to obtain the adjusted pulse frequency; The focusing depth offset is calculated based on the material refractive index deviation value corresponding to the region to be excited, and the standard focusing depth is added to the focusing depth offset for depth correction to obtain the adjusted focusing depth. The adjusted laser power, the adjusted pulse frequency, and the adjusted focus depth are combined to form a laser parameter combination; Execution instructions for multiple laser heads are generated based on the laser parameter combination.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of generating execution instructions for multiple laser heads based on the laser parameter combination includes: The laser parameter combination is used as the control input to construct a global optimization objective function for multiple laser heads; Based on the global optimization objective function, constraints such as non-overlapping thermally affected regions, maximum temperature rise, and material stress are set to obtain a multi-constraint optimization problem. Based on the aforementioned multi-constraint optimization problem, the state variables of each laser head node are iteratively updated and global variables are coordinated to generate execution instructions for each laser head.

[0011] Optionally, in the seventh implementation of the first aspect of the present invention, the step of iteratively updating the state variables of each laser head node and coordinating global variables based on the multi-constraint optimization problem to generate execution instructions for each laser head includes: Initialize the state variables, global variables, and Lagrange multipliers of each laser head node to obtain the initial configuration parameters; Based on the initial configuration parameters, the state variables of each laser head node are optimized using the ADMM algorithm to obtain the updated state variables of each laser head node. Based on the multi-constraint optimization problem, the update state variables of each laser head node are globally coordinated and updated to obtain a global coordination result. The iteration error is calculated for the updated state variables of each laser head node. When the iteration error is less than the convergence threshold, the iteration is terminated and the execution instructions for each laser head are generated.

[0012] Optionally, in the eighth implementation of the first aspect of the present invention, the step of optimizing the state variables of each laser head node based on the initial configuration parameters using the ADMM algorithm to obtain the updated state variables of each laser head node includes: A state variable vector is constructed based on the position coordinates, velocity components, and acceleration components of each laser head node, and a local objective function for each laser head node is established based on the state tracking error and control input energy consumption. The ADMM algorithm is used to construct a local optimization problem by combining the local objective function with the initial configuration parameters, thereby obtaining the complete optimization problem for each laser head node; The complete optimization problem of each laser head node is numerically solved to obtain the optimal state variable value of each laser head node, and the updated state variable of each laser head node is generated based on the optimal state variable value.

[0013] This invention also provides an intelligent control system for a crystal glass laser engraving machine, comprising: The extraction module is used to collect three-dimensional point cloud data of crystal glass workpieces, extract material basic feature vectors, and input the material basic feature vectors into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. The excitation state discrimination module is used to determine the excitation state based on the three-dimensional spatial state map by fusing acoustic emission signals, light scattering signals and temperature signals, and to determine the excitation region. The adjustment module is used to adjust the laser power, pulse frequency and focusing depth according to the region to be excited to obtain a combination of laser parameters, and optimize the position trajectory and control input of multiple laser heads based on the combination of laser parameters to generate execution instructions for each laser head.

[0014] In summary, this invention, by constructing an axial self-crossing attention module with parallel processing along the X, Y, and Z axes, can accurately capture the physical relationships between points in three-dimensional space. Compared to traditional single-dimensional analysis methods, it can comprehensively understand the three-dimensional spatial structural characteristics of materials. It also constructs a multi-sensor fusion-based excitation state discrimination mechanism, achieving real-time and accurate identification of the material's excitation state through weighted fusion of acoustic emission power spectral density, scattered light intensity, and local temperature values. This mechanism can proactively sense and predict the emergence of the excitation region. The ADMM algorithm is used to achieve collaborative optimization control of multiple laser heads. By setting multiple constraints such as non-overlapping thermally affected areas, maximum temperature rise constraints, and material stress constraints, the coordinated operation between laser heads is ensured. Based on the identification results of the excitation region, the laser power, pulse frequency, and focusing depth are adjusted accordingly. The laser power adjustment is based on dynamic calculation of the deviation coefficient, the pulse frequency is increased by a corresponding increment based on the standard value, and the focusing depth is precisely corrected according to the material's refractive index deviation, thereby improving the processing accuracy and stability of the crystal glass laser engraving machine. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the intelligent control method steps of a crystal glass laser engraving machine in one embodiment of the present invention; Figure 2 This is a block diagram of the intelligent control system of a crystal glass laser engraving machine according to an embodiment of the present invention.

[0016] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Reference Figure 1 This embodiment provides an intelligent control method for a crystal glass laser engraving machine, including the following steps: S1. Collect three-dimensional point cloud data of crystal glass workpiece, extract material basic feature vector, and input the material basic feature vector into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. Specifically, the CCD vision positioning module performs precise imaging and 3D modeling sampling on the crystal glass workpiece to be carved. The collected raw data is represented as a point cloud, a dense dataset containing the 3D coordinates of each spatial point, from which a 3D spatial coordinate matrix is ​​constructed. Based on the 3D spatial coordinate matrix, material optical properties are analyzed. The MOCA-Net material optical property attention network is used to perform convolution extraction on the local neighborhood of each point, extracting data feature terms reflecting the internal optical behavior of the workpiece, including local refractive index gradient changes, material density distribution trends, and microstructural continuity, generating a set of 256-dimensional material basic feature vectors. These material basic feature vectors are input into an axial self-crossing attention module, which consists of attention branches operating independently along the X, Y, and Z axes. Each branch constructs a query, key, and value matrix and performs directionally constrained correlation calculations to extract the correlation strength between features in each direction. After completing the attention calculation in each direction, an adaptive fusion mechanism is used to weighted integrate the attention outputs from the three directions, generating a unified state map representing the 3D physical state relationships.

[0019] S2, based on the three-dimensional spatial state map, fuses acoustic emission signals, light scattering signals and temperature signals to determine the state to be excited and identify the region to be excited; Specifically, the three-dimensional spatial state map encodes the physical state distribution characteristics of each spatial location point, including parameters such as refractive index gradient, density distribution, and thermal stress trend. Based on the three-dimensional spatial state map, material excitation characteristic parameters for each coordinate point are extracted, and a mapping relationship between spatial location and excitation characteristic parameters is established, forming a standardized space-characteristic reference table. Acoustic emission signals, light scattering signals, and temperature distribution data are simultaneously acquired during the on-site processing using a multimodal physical sensing device. The acoustic emission signal is sampled at high frequency and then subjected to Fast Fourier Transform (FFT) for spectral expansion to calculate its acoustic emission power spectral density. The light scattering signal is acquired by a photodiode and a high-gain transimpedance amplifier, and then normalized to obtain the scattered light intensity value. Temperature data is acquired in real-time by an infrared thermal imager at a frequency of 100Hz, and three-dimensional interpolation is performed using spatial coordinates to obtain the local temperature value at a specified location. The acoustic emission power spectral density, scattered light intensity, and local temperature value are compared with the material excitation characteristic parameters recorded in the state map according to their spatial location to obtain the excitation state deviation value corresponding to each spatial point, reflecting the degree of deviation between the current actual processing state and the expected material response. The excitation state deviation value is input into the excitation state discrimination function. The excitation state discrimination function constructs a multi-dimensional weighted target model, and calculates the target function value by linearly combining the deviations of acoustic emission power spectral density, normalized scattered light intensity and local temperature value according to the set physical contribution coefficient. When the target function value is lower than the preset state discrimination threshold, the corresponding position is determined to be in an excitation failure or insufficient excitation state, and the corresponding position is marked as the region to be excited.

[0020] S3. Adjust the laser power, pulse frequency and focusing depth according to the region to be excited to obtain the laser parameter combination. Based on the laser parameter combination, optimize the position trajectory and control input of multiple laser heads to generate the execution instructions of each laser head.

[0021] Specifically, based on the three-dimensional spatial coordinates of the region to be excited, the corresponding deviation value of the state to be excited is extracted by combining the fusion discrimination results. For dynamic correction of laser power, a mapping relationship is constructed based on the deviation value of the state to be excited to determine the laser power adjustment coefficient. The standard laser power value is multiplied by the laser power adjustment coefficient to obtain the laser power correction value required for the current region to be excited, achieving targeted control of laser energy density. Regarding adaptive pulse frequency processing, a frequency increment value proportional to the deviation value is calculated and added to the default standard pulse frequency to obtain an adjusted pulse frequency that meets the energy injection requirements, effectively compensating for insufficient energy caused by changes in material thermophysical properties or abnormal laser transmission. During the adjustment of focusing depth, the refractive index deviation value of the region to be excited is analyzed, and the corresponding focusing depth offset is calculated using a nonlinear response model between refractive index and the focusing plane. The focusing depth offset is superimposed on the standard focusing depth to achieve adaptive correction of the laser focus for specific spatial microstructures, improving the spatial deposition efficiency of laser energy. The adjusted laser power, pulse frequency, and focusing depth are integrated into a unified laser parameter combination. Based on this laser parameter combination and considering the current processing segment, task priority, and spatial interference relationship of each laser head in the multi-laser head system, a task scheduling and path optimization algorithm is used to generate multi-dimensional control execution commands for each laser head, including processing position, scanning speed, laser emission command, and focusing curve.

[0022] In one example, 3D point cloud data of a crystal glass workpiece is acquired, a material fundamental feature vector is extracted, and this material fundamental feature vector is input into an axial self-crossing attention module for 3D spatial feature association to obtain a 3D spatial state map, including: The three-dimensional point cloud data of the crystal glass workpiece is obtained through the CCD vision positioning module, and the three-dimensional spatial coordinate matrix of the three-dimensional point cloud data is extracted. Based on the three-dimensional spatial coordinate matrix, the optical properties of materials are analyzed, and a basic feature vector of materials containing information on workpiece geometry, refractive index distribution and material density is constructed. The material's basic feature vector is input into the axial self-crossing attention module to perform three-dimensional spatial feature association, resulting in a three-dimensional spatial state map.

[0023] In this example, a CCD vision positioning module deployed on a laser engraving platform performs a 3D scan of the crystal glass workpiece to be processed. The CCD module possesses sub-pixel-level image resolution capabilities and can acquire high-density point cloud information in conjunction with structured light or laser projection. Its output data records all visible surfaces and their inherent microstructural features in spatial coordinates, forming a point cloud set containing X, Y, and Z 3D coordinate data. Geometric reconstruction and noise removal are performed on the 3D point cloud data, organizing the spatial positions of all points into a standardized 3D spatial coordinate matrix. Based on the material optical property modeling network MOCA-Net, the 3D spatial coordinate matrix is ​​input into a multi-layer 3D convolutional module for local spatial context modeling. Simultaneously, feature fusion is performed using refractive index variation curves and material density distribution maps to construct a unified material-based feature vector describing the geometric structure, material density gradient, and optical response characteristics of the crystal glass workpiece. The material's basic feature vector is input into the axial self-crossing attention module. The axial self-crossing attention module processes the attention mechanism in the X, Y, and Z axes respectively, constructing an independent query, key, and value matrix in each dimension. By calculating the attention weights between vectors, it extracts the associated features and anomaly distribution information in different spatial directions. Based on the output results of each axis, a cross-axis fusion mechanism is introduced to weight and combine the attention outputs of the three directions according to preset weight coefficients to generate a set of three-dimensional spatial state maps that express global spatial information.

[0024] In one example, the material's fundamental feature vector is input into an axial self-crossing attention module for three-dimensional spatial feature association, resulting in a three-dimensional spatial state map, including: The material's basic feature vector is input into the axial self-crossing attention module, where it is mapped to the X-axis query matrix, X-axis key matrix, X-axis value matrix, Y-axis query matrix, Y-axis key matrix, Y-axis value matrix, Z-axis query matrix, Z-axis key matrix, and Z-axis value matrix, respectively. Calculate the X-axis attention weight matrix based on the X-axis query matrix, X-axis key matrix, and X-axis value matrix; calculate the Y-axis attention weight matrix based on the Y-axis query matrix, Y-axis key matrix, and Y-axis value matrix; calculate the Z-axis attention weight matrix based on the Z-axis query matrix, Z-axis key matrix, and Z-axis value matrix; The X-axis attention weight matrix, Y-axis attention weight matrix, and Z-axis attention weight matrix are weighted and fused to obtain the comprehensive attention features. Based on the comprehensive attention feature encoding of the physical relationships between points in the three-dimensional space and the feature relationships of different depth layers, a three-dimensional spatial state map is obtained.

[0025] In this example, the material's fundamental feature vector is input into the axial self-crossing attention module. Within the module, the input vector is projected through independent linear transformation channels into attention calculation matrices for three spatial axes: query, key, and value matrices along the X-axis, Y-axis, and Z-axis. These nine matrices together form the basis of the attention operation in the three-axis space. Each matrix is ​​generated by multiplying it with the 256-dimensional material fundamental feature vector, and learnable linear transformation weights are used to achieve vector dimensionality compression and directional projection, making the attention calculation process for each axial branch directionally sensitive and region-selective. Attention calculations are performed on three independent paths along the X, Y, and Z axes. This involves performing a dot product operation between the query matrix and the corresponding key matrix for each axis, dividing by a scaling factor, and then normalizing using a softmax function to obtain the attention weight distribution matrix for each axis. The attention weight matrix is ​​then multiplied and mapped to the axis value matrix to generate attention output features in the corresponding directions: X-axis attention features, Y-axis attention features, and Z-axis attention features. Each captures the associated structural and physical coupling characteristics of the material's fundamental features in the horizontal, vertical, and depth directions. To obtain a globally consistent spatial state representation, the attention outputs from the three axes are fused using an adaptive weighting mechanism. The fusion process is weighted by the learning parameters set in the network, integrating the information from the three directions into a unified attention encoding feature. This integrated attention feature possesses a sensitive ability to express changes in the state of heterogeneous materials, refractive index gradient distribution, and stress concentration trends in multidimensional space, effectively capturing the interaction relationships and local structural coupling mechanisms between different depth layers along the Z-axis. Using comprehensive attention features as basic data and combining them with the three-dimensional coordinate information of each spatial point, a three-dimensional spatial state map of the encoded material microstructure state and spatial topological physical relationship is generated.

[0026] In one example, a 3D spatial state map is obtained by encoding the physical relationships between points in 3D space and the feature relationships at different depths based on comprehensive attention feature encoding, including: Based on the comprehensive attention characteristics, attention weight values ​​are assigned to the spatial coordinate points corresponding to the 3D point cloud data to obtain the associated weight distribution data; Based on the correlation weight distribution data, the heat conduction correlation path and material continuity change between each depth layer are calculated to obtain the characteristic correlation data of each depth layer; Based on the feature correlation data of each depth layer, the heat-affected zone during the carving process is predicted, and a three-dimensional spatial state map is generated based on the heat-affected zone.

[0027] In this example, the attention weights in the comprehensive attention features are spatially mapped, and each weight is redistributed to its corresponding coordinates in the 3D point cloud data, forming a weighted point cloud weight distribution. An index mapping table is established to establish the correspondence between feature channels and point cloud coordinates, generating structured associated weight distribution data. Based on this associated weight distribution data, a deep hierarchical model is constructed, dividing the entire 3D point cloud into multiple hierarchical units along the Z-axis. Local thermal properties are extracted within each depth layer, including weight gradient distribution, inter-point connectivity density, and material optical continuity indices. Cross-layer analysis is used to calculate the thermal conduction paths between adjacent depth layers. The thermal conduction paths use the weight values ​​as a reference for heat flux potential. A physical continuity function between depths is constructed by combining material density and refractive index changes to reflect the resistance and deformation encountered when heat propagates from upper to lower layers. This yields comprehensive feature data for each depth layer under laser-induced thermal diffusion, stress concentration, and abrupt property changes. After obtaining the feature correlation data of each depth layer, the potential heat-affected zone during laser engraving is predicted based on the physical simulation model and the constitutive relationship of thermal transfer. The prediction process considers factors such as laser power density, action time, scanning trajectory, and material thermal conductivity, and identifies highly thermally sensitive areas as potential heat accumulation risk zones or stress triggering zones. The spatial distribution information of the heat-affected zone is fused into the original three-dimensional point cloud coordinate system, and combined with the original material basic characteristics and comprehensive attention characteristics, a three-dimensional spatial state map expressing the dynamics of heat conduction, spatial thermal coupling relationships, and physical continuity changes is encoded.

[0028] In one example, the state to be excited is determined by fusing acoustic emission signals, light scattering signals, and temperature signals based on a three-dimensional spatial state map, thus identifying the region to be excited, including: Based on the three-dimensional spatial state map, the material excitation characteristic parameters at each spatial location are obtained, and a correspondence table between spatial location and material excitation characteristic parameters is established. Acoustic emission signals are collected to calculate acoustic emission power spectral density; light scattering signals are collected and normalized to obtain scattered light intensity; and temperature signals are collected and three-dimensionally interpolated to obtain local temperature values. The acoustic emission power spectral density, scattered light intensity, and local temperature values ​​are compared with the material excitation characteristic parameters in the corresponding relationship table to calculate the excitation state deviation value at each spatial location. The objective function value of the discrimination function for the state to be excited is calculated based on the excitation state deviation value. When the objective function value is less than the set state threshold, it is determined to be the region to be excited.

[0029] In this example, excitation characteristic parameters for each spatial location are extracted based on the three-dimensional spatial state map, including the acoustic emission power response threshold under standard excitation, the typical excitation scattered light intensity range, and the steady-state excitation temperature window. The coordinates of all spatial points are mapped one-to-one with the material excitation reference characteristic values, constructing a correspondence table between spatial location and material excitation characteristic parameters. Simultaneously, a multimodal sensing link is activated to acquire physical response signals from the laser processing site in real time. The acoustic emission signal is acquired by a distributed ultrasonic sensor array and the acoustic emission power spectral density is calculated using a fast Fourier transform, reflecting whether micro-cracks, bubbles, or bursting responses occur during the laser-material interaction. The light scattering signal is acquired by a broadband silicon photodetector and normalized to obtain the unit intensity scattering response, used to quantify whether laser energy is effectively absorbed within the material and induces nonlinear optical effects. The temperature signal is acquired by a high-resolution infrared thermal imager, capturing a two-dimensional thermal image, and the thermal field is mapped to a three-dimensional coordinate system using a spatial interpolation function to obtain the local temperature value at any location. For each spatial point, the acoustic emission power spectral density, normalized scattered light intensity, and local temperature values ​​are compared with the standard material excitation parameters stored in the spatial correspondence table. These three types of differences are then normalized, quantized, and weighted to calculate the excitation state deviation value for the current spatial point. This deviation reflects the degree of deviation between the actual excitation behavior and the ideal state. Using the excitation state deviation value as the input variable, a discriminant function for the desired excitation state is constructed. This discriminant function uses a linear weighted superposition model or a weighted nonlinear mapping function to generate a target function value, which characterizes the probability level of whether the current spatial point is in an excited or unexcited state. When the target function value is lower than a preset state threshold (e.g., 0.15), the current spatial location is determined to be an area to be excited, indicating that effective excitation or insufficient excitation intensity has not been formed at this location during the current laser processing, suggesting problems such as abnormal local material structure, optical refraction interference, or thermal diffusion shielding.

[0030] In one example, the laser power, pulse frequency, and focus depth are adjusted according to the region to be excited to obtain a combination of laser parameters. Based on this combination, the position trajectories and control inputs of multiple laser heads are optimized to generate execution instructions for each laser head, including: Calculate the excitation state deviation value based on the spatial location of the region to be excited; The laser power adjustment coefficient is determined based on the deviation value of the state to be excited, and the standard laser power is multiplied by the laser power adjustment coefficient to obtain the adjusted laser power. The pulse frequency increment is calculated based on the deviation value of the state to be excited, and the standard pulse frequency is added to the pulse frequency increment to obtain the adjusted pulse frequency; The focusing depth offset is calculated based on the material refractive index deviation value corresponding to the region to be excited, and the standard focusing depth is added to the focusing depth offset for depth correction to obtain the adjusted focusing depth. The adjusted laser power, adjusted pulse frequency, and adjusted focus depth are combined to form the laser parameter combination; Execution instructions for multiple laser heads are generated based on combinations of laser parameters.

[0031] In this example, the standard excitation characteristic parameters and real-time monitoring signal response are located in the three-dimensional spatial state map based on the spatial location of the region to be excited. By comparing the real-time acoustic emission power spectral density, scattered light intensity, and local temperature values ​​with the expected excitation state parameters at the corresponding points, a comprehensive deviation is calculated to obtain the current region's excitation state deviation value. This deviation represents the degree of inadequacy of the material's response under laser irradiation and is the fundamental driving factor of the entire parameter adjustment model. A power adjustment coefficient calculation model is introduced based on the excitation state deviation value. For example, a linear or exponential function is used to map the deviation value to a set of laser power adjustment coefficients. The adjustment range is set between 1.0 and 1.25 to avoid the risk of thermal damage to the material. The laser power adjustment coefficient is multiplied by the standard laser power value to obtain the enhanced laser power output, enabling laser energy to be effectively deposited in the material and form an excitation point. Simultaneously, an adaptive correction amount for the pulse frequency is calculated. A frequency increment function is constructed based on the deviation value. The larger the deviation value, the higher the corresponding frequency increment. However, limited by the system's modulation capability, the maximum increment does not exceed a preset upper limit (e.g., 5kHz). The pulse frequency required for the current processing point is obtained by superimposing it onto a standard frequency, ensuring that the laser pulse can compensate for the response lag problem due to insufficient energy within the microscale thermal accumulation time window. The refractive index change information of the material in the region to be excited is retrieved, and the refractive index deviation value is obtained by differential calculation with the standard refractive index. Then, the corresponding focus depth offset is calculated using a focus offset model. The focus depth offset is calculated by multiplying the refractive index deviation value by a factor of k. The focus depth offset is superimposed with the standard focus depth to form a real-time depth correction value, ensuring that the laser focus is always aligned with the optimal excitation layer inside the material, avoiding laser diffusion and energy misalignment problems caused by focus drift. The adjusted laser power, pulse frequency, and focus depth are encapsulated into a set of structured laser parameter combinations. Combined with the current processing trajectory position of each laser head in the multi-laser head system, the distribution characteristics of the corresponding region to be excited, and the thermally affected boundary conditions, the parameter combinations are allocated to each laser head using a scheduling optimization algorithm to form independent execution instructions. These instructions include processing point coordinates, power level, pulse frequency setpoint, focusing curve coefficient, and dynamic response coefficient, and are synchronously sent to each laser head controller via a high-speed communication interface.

[0032] In one example, execution instructions for multiple laser heads are generated based on combinations of laser parameters, including: A global optimization objective function for multiple laser heads is constructed by using the combination of laser parameters as control input; Based on the global optimization objective function, constraints such as non-overlapping heat-affected regions, maximum temperature rise, and material stress are set to obtain a multi-constraint optimization problem. Based on the multi-constraint optimization problem, the state variables of each laser head node are iteratively updated and the global variables are coordinated to generate the execution instructions for each laser head.

[0033] In this example, laser parameters, including adjusted laser power, pulse frequency, and focusing depth, are input into a multi-laser head collaborative control framework. Based on this, a global optimization objective function is constructed. The main optimization indices in the objective function are the tracking accuracy of the processing trajectory, the control cost of the laser parameters, and the uniformity of the thermal field distribution. Specifically, the objective function is the sum of the weighted integral of the squared position deviation of each laser head over time and the weighted integral of the squared control input. At the same time, a temperature gradient penalty term is added to limit the local heat accumulation caused by uneven heat diffusion, forming a nonlinear weighted multi-objective control model. Based on the objective function, several rigid constraints are introduced. Among them, the non-overlapping heat-affected zone constraint limits the Euclidean distance between the action areas of any two laser heads in the same processing cycle to be greater than or equal to twice the heat-affected radius, thus avoiding local damage to the material structure caused by the cross-superposition of heat in space. The maximum temperature rise constraint is set according to the thermal stability threshold of the crystal glass material, for example, the upper limit of temperature rise is 180°C. The temperature rise rate and peak value of each laser point in a short time are limited by real-time thermal field simulation or infrared monitoring feedback. The stress constraint uses the condition that the von Mises equivalent stress does not exceed 0.7 times the yield strength of the material as the limit. The stress response of each laser point during the processing is calculated by a thermo-mechanical field coupled finite element model, and the scanning speed and focusing depth are adjusted accordingly. After formulating a multi-objective optimization problem with three types of constraints, a distributed iterative optimization algorithm is used to solve it. Each laser head node acts as an independent control subsystem, maintaining its own local state variables, including spatial position, velocity vector, laser power, and focusing depth. In each iteration, the state variables are locally optimized and updated according to the local objective function and constraints. The updated results are transmitted to the coordination module for global variable coordination through a shared communication mechanism. The coordination module uses the alternating direction multiplier method or the Lagrange multiplier method to transform the mutual constraints caused by thermal field coupling among multiple laser heads into decomposable optimization subproblems. After each global coordination, the global reference trajectory and laser parameter distribution direction are adjusted. After several rounds of iterative updates and global coordination, when the system meets the convergence criterion (i.e., the change in state variables between two adjacent rounds is less than a set threshold), the control instructions required for each laser head in the current cycle are output. These instructions include the target position coordinates, real-time scanning speed, dynamic laser power curve, and focusing depth trajectory. All instructions are synchronously sent to the laser head controller through a high-bandwidth bus interface, realizing collaborative processing among multiple laser heads under the coupling constraints of thermal-mechanical-optical multi-physics domains.

[0034] In one example, based on a multi-constraint optimization problem, the state variables of each laser head node are iteratively updated and global variables are coordinated to generate execution instructions for each laser head, including: Initialize the state variables, global variables, and Lagrange multipliers of each laser head node to obtain the initial configuration parameters; Based on the initial configuration parameters, the state variables of each laser head node are optimized using the ADMM algorithm to obtain the updated state variables of each laser head node. Based on the multi-constraint optimization problem, the updated state variables of each laser head node are globally coordinated and updated to obtain the global coordination result; The iteration error is calculated for the updated state variables of each laser head node. When the iteration error is less than the convergence threshold, the iteration is terminated and the execution instructions for each laser head are generated.

[0035] In this example, based on the carving task distribution, processing area division, and the 3D coordinate information of the area to be excited, initial state variables are configured for each laser head node, including position coordinate vector, velocity vector, laser power level, pulse frequency parameter, and focus depth value. Simultaneously, system-wide shared initial global variables are set. These initial global variables represent common reference variables used in the multi-laser head system to coordinate the thermal influence range, parameter distribution, and processing load balancing. Combining the coupling constraint terms in the multi-objective optimization problem, a set of initialized Lagrange multipliers is assigned to each laser head node to represent the deviation constraint mapping between local state variables and global variables, thus obtaining the initial configuration parameters. Based on the initial configuration parameters and following the parallel structure of the alternating direction multiplier method, a local state variable optimization process is performed for each laser head node. Each laser head node minimizes its local objective function, i.e., minimizes the deviation of its state variables (position, velocity, laser control parameters) from the desired trajectory and local energy consumption target. Lagrange multiplier terms and penalty terms are introduced to reflect the consistency constraint between the current node's state variables and global variables. The optimization solution is a convex quadratic programming problem with a squared penalty term. After local optimization, the updated state variables of all nodes are fed back to the central coordination module for global variable updates. The goal is to achieve global consistency by averaging the state variables of all nodes, while minimizing the difference between the overall node state and the global variables, resulting in a new global coordination result used as a reference for the next round of state variable updates. Simultaneously, the Lagrange multipliers of each node are updated, adjusting the penalty expression for constraint violations based on the difference between the current state variables and the coordinated global variables. The nodes are gradually adjusted towards convergence. After each complete iteration cycle, an iteration error assessment is performed on the changes in state variables of all laser head nodes. This is achieved by calculating the Euclidean distance between the state vectors of each node in two consecutive iterations and comparing it to a convergence threshold. When the changes in state variables of all nodes are less than the set convergence threshold (e.g., 10), the node is considered convergent. -4When the alternating direction multiplier algorithm has converged, the iteration process is terminated. The state variables of each laser head node are extracted as the final optimization result. A structured execution control instruction set is generated based on the state variables. The instruction content includes the optimized scanning trajectory path, speed setting sequence, laser power modulation curve, pulse frequency adjustment scheme and dynamic focusing control parameters. The instructions are synchronously distributed to the low-level control unit of each laser head through the real-time control interface.

[0036] In one example, based on the initial configuration parameters, the ADMM algorithm is used to optimize the state variables of each laser head node, resulting in the updated state variables of each laser head node, including: A state variable vector is constructed based on the position coordinates, velocity components, and acceleration components of each laser head node, and a local objective function for each laser head node is established based on the state tracking error and control input energy consumption. The ADMM algorithm is used to construct a local optimization problem by combining the local objective function with the initial configuration parameters, thereby obtaining the complete optimization problem for each laser head node; The complete optimization problem of each laser head node is numerically solved to obtain the optimal state variable values ​​of each laser head node, and the updated state variables of each laser head node are generated based on the optimal state variable values.

[0037] In this example, the kinematic information of each laser head node in the current control cycle is extracted in the task scheduling module. This includes the real-time position coordinates, instantaneous velocity vector, and current acceleration component of the laser head in three-dimensional space. A standardized state variable vector is constructed based on the physical state data. The state vector contains nine dimensions: three position components, three velocity components, and three acceleration components, to describe the dynamic behavior of the laser head. A state tracking error vector is constructed based on the reference trajectory data generated by the carving task. The sum of the squares of the state tracking error vector is used as a precision penalty term. At the same time, the variation amplitude of the control input variables, namely laser power, scanning speed, and focusing depth, is included as an energy consumption index in the penalty function. A weight matrix R is assigned to adjust the proportion of control input energy consumption in the overall optimization objective, forming the local objective function of each laser head node. The local objective function focuses on minimizing trajectory deviation and control energy consumption. After the local objective function is constructed, based on the configuration parameters provided during the initialization phase, including initial state variables, shared global variables, and Lagrange multiplier vectors, the local objective function is embedded into the solution structure of the ADMM algorithm. A complete optimization problem for each laser head node is constructed using a "local update-global coordination-multiplier update" strategy. While maintaining local optimality, the optimization problem introduces global consistency constraints and inter-node thermal coupling limitations to ensure no conflicts arise between multiple laser heads in terms of thermal field, spatial distribution, and trajectory coordination. In each ADMM iteration step, the laser head node independently solves its complete optimization problem. The optimization objective function is a constrained convex quadratic programming problem, which includes state error terms, control energy consumption terms, Lagrange multiplier penalty terms, and global consistency penalty terms. Shared global variables are used as external coupling inputs, and numerical solutions are performed in conjunction with local state prediction results. The solution process achieves high-precision iteration through numerical optimization algorithms such as the interior point method, gradient projection method, or conjugate gradient method. After the solution is obtained, the state variable values ​​corresponding to the optimal solution are extracted and used as the updated state variables of the laser head node in this iteration, including the new round of position commands, velocity scheduling strategy and acceleration transition curve.

[0038] Reference Figure 2 This embodiment provides an intelligent control system for a crystal glass laser engraving machine, including: Extraction module 1 is used to collect three-dimensional point cloud data of crystal glass workpieces, extract material basic feature vectors, and input the material basic feature vectors into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. The excited state discrimination module 2 is used to discriminate the excited state based on the fusion of acoustic emission signal, light scattering signal and temperature signal according to the three-dimensional spatial state map, and to determine the region to be excited. The adjustment module 3 is used to adjust the laser power, pulse frequency and focusing depth according to the region to be excited to obtain the laser parameter combination, and optimize the position trajectory and control input of multiple laser heads based on the laser parameter combination to generate the execution instructions of each laser head.

[0039] In this embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0040] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, system, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or method that includes that element.

[0041] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An intelligent control method for a crystal glass laser engraving machine, characterized in that, include: Three-dimensional point cloud data of crystal glass workpieces are collected, material basic feature vectors are extracted, and the material basic feature vectors are input into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. Based on the three-dimensional spatial state map, acoustic emission signals, light scattering signals, and temperature signals are fused to determine the state to be excited and identify the region to be excited. The laser power, pulse frequency, and focusing depth are adjusted according to the region to be excited to obtain a combination of laser parameters. Based on the combination of laser parameters, the position trajectory and control input of multiple laser heads are optimized to generate execution instructions for each laser head.

2. The intelligent control method for a crystal glass laser engraving machine according to claim 1, characterized in that, The process involves acquiring 3D point cloud data of the crystal glass workpiece, extracting the material's fundamental feature vector, and inputting this vector into an axial self-crossing attention module for 3D spatial feature association to obtain a 3D spatial state map, including: The three-dimensional point cloud data of the crystal glass workpiece is obtained through the CCD vision positioning module, and the three-dimensional spatial coordinate matrix of the three-dimensional point cloud data is extracted. Based on the three-dimensional spatial coordinate matrix, the optical properties of the material are analyzed, and a basic feature vector of the material containing information on the workpiece geometry, refractive index distribution and material density is constructed. The material's basic feature vector is input into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map.

3. The intelligent control method for a crystal glass laser engraving machine according to claim 2, characterized in that, The step of inputting the material's basic feature vector into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map includes: The material's basic feature vector is input into the axial self-crossing attention module, where it is mapped to X-axis query matrix, X-axis key matrix, X-axis value matrix, Y-axis query matrix, Y-axis key matrix, Y-axis value matrix, Z-axis query matrix, Z-axis key matrix, and Z-axis value matrix, respectively. Calculate the X-axis attention weight matrix based on the X-axis query matrix, the X-axis key matrix, and the X-axis value matrix; calculate the Y-axis attention weight matrix based on the Y-axis query matrix, the Y-axis key matrix, and the Y-axis value matrix; calculate the Z-axis attention weight matrix based on the Z-axis query matrix, the Z-axis key matrix, and the Z-axis value matrix; The X-axis attention weight matrix, the Y-axis attention weight matrix, and the Z-axis attention weight matrix are weighted and fused to obtain a comprehensive attention feature; Based on the comprehensive attention feature encoding, the physical relationships between points in the three-dimensional space and the feature relationships at different depths are obtained to obtain a three-dimensional spatial state map.

4. The intelligent control method for a crystal glass laser engraving machine according to claim 3, characterized in that, The process of encoding the physical relationships between points in the three-dimensional space and the feature relationships at different depths based on the comprehensive attention feature encoding yields a three-dimensional spatial state map, including: Based on the comprehensive attention features, attention weight values ​​are assigned to the spatial coordinate points corresponding to the three-dimensional point cloud data to obtain associated weight distribution data; Based on the associated weight distribution data, the heat conduction associated paths and material continuity changes between each depth layer are calculated to obtain the characteristic associated data of each depth layer; Based on the feature correlation data of each depth layer, the heat-affected zone during the carving process is predicted, and a three-dimensional spatial state map is generated based on the heat-affected zone.

5. The intelligent control method for a crystal glass laser engraving machine according to claim 1, characterized in that, The step of determining the region to be excited by fusing acoustic emission signals, light scattering signals, and temperature signals based on the three-dimensional spatial state map includes: Based on the three-dimensional spatial state map, material excitation characteristic parameters at each spatial location are obtained, and a correspondence table between spatial location and material excitation characteristic parameters is established. Acoustic emission signals are collected to calculate acoustic emission power spectral density; light scattering signals are collected and normalized to obtain scattered light intensity; and temperature signals are collected and three-dimensionally interpolated to obtain local temperature values. The acoustic emission power spectral density, scattered light intensity, and local temperature values ​​are compared with the material excitation characteristic parameters in the corresponding relationship table to calculate the excitation state deviation value at each spatial location. The objective function value of the discrimination function for the state to be excited is calculated based on the excitation state deviation value. When the objective function value is less than the set state threshold, it is determined to be a region to be excited.

6. The intelligent control method for a crystal glass laser engraving machine according to claim 1, characterized in that, The laser power, pulse frequency, and focusing depth are adjusted according to the region to be excited to obtain a laser parameter combination. Based on the laser parameter combination, the position trajectory and control input of multiple laser heads are optimized to generate execution instructions for each laser head, including: Calculate the excitation state deviation value based on the spatial location of the region to be excited; The laser power adjustment coefficient is determined based on the deviation value of the state to be excited, and the standard laser power is multiplied by the laser power adjustment coefficient to obtain the adjusted laser power. The pulse frequency increment is calculated based on the deviation value of the state to be excited, and the standard pulse frequency is added to the pulse frequency increment to obtain the adjusted pulse frequency; The focusing depth offset is calculated based on the material refractive index deviation value corresponding to the region to be excited, and the standard focusing depth is added to the focusing depth offset for depth correction to obtain the adjusted focusing depth. The adjusted laser power, the adjusted pulse frequency, and the adjusted focus depth are combined to form a laser parameter combination; Execution instructions for multiple laser heads are generated based on the laser parameter combination.

7. The intelligent control method for a crystal glass laser engraving machine according to claim 6, characterized in that, The execution instructions for generating multiple laser heads based on the laser parameter combination include: The laser parameter combination is used as the control input to construct a global optimization objective function for multiple laser heads; Based on the global optimization objective function, constraints such as non-overlapping thermally affected regions, maximum temperature rise, and material stress are set to obtain a multi-constraint optimization problem. Based on the aforementioned multi-constraint optimization problem, the state variables of each laser head node are iteratively updated and global variables are coordinated to generate execution instructions for each laser head.

8. The intelligent control method for a crystal glass laser engraving machine according to claim 7, characterized in that, The process, based on the multi-constraint optimization problem, involves iteratively updating the state variables of each laser head node and coordinating global variables to generate execution instructions for each laser head, including: Initialize the state variables, global variables, and Lagrange multipliers of each laser head node to obtain the initial configuration parameters; Based on the initial configuration parameters, the state variables of each laser head node are optimized using the ADMM algorithm to obtain the updated state variables of each laser head node. Based on the multi-constraint optimization problem, the update state variables of each laser head node are globally coordinated and updated to obtain a global coordination result. The iteration error is calculated for the updated state variables of each laser head node. When the iteration error is less than the convergence threshold, the iteration is terminated and the execution instructions for each laser head are generated.

9. The intelligent control method for a crystal glass laser engraving machine according to claim 8, characterized in that, Based on the initial configuration parameters, the ADMM algorithm is used to optimize the state variables of each laser head node to obtain the updated state variables of each laser head node, including: A state variable vector is constructed based on the position coordinates, velocity components, and acceleration components of each laser head node, and a local objective function for each laser head node is established based on the state tracking error and control input energy consumption. The ADMM algorithm is used to construct a local optimization problem by combining the local objective function with the initial configuration parameters, thereby obtaining the complete optimization problem for each laser head node; The complete optimization problem of each laser head node is numerically solved to obtain the optimal state variable value of each laser head node, and the updated state variable of each laser head node is generated based on the optimal state variable value.

10. An intelligent control system for a crystal glass laser engraving machine, characterized in that, The steps for implementing the intelligent control method for the crystal glass laser engraving machine according to any one of claims 1 to 9 include: The extraction module is used to collect three-dimensional point cloud data of crystal glass workpieces, extract material basic feature vectors, and input the material basic feature vectors into the axial self-crossing attention module for three-dimensional spatial feature association to obtain a three-dimensional spatial state map. The excitation state discrimination module is used to determine the excitation state based on the three-dimensional spatial state map by fusing acoustic emission signals, light scattering signals and temperature signals, and to determine the excitation region. The adjustment module is used to adjust the laser power, pulse frequency and focusing depth according to the region to be excited to obtain a combination of laser parameters, and optimize the position trajectory and control input of multiple laser heads based on the combination of laser parameters to generate execution instructions for each laser head.

Images