Monitoring method and system applied to laser cutting die machining

Abstract

The invention belongs to the technical field of intelligent manufacturing, and provides a monitoring method and system applied to laser cutting die machining. The method comprises the steps that when it is judged that the stability of laser cutting die machining does not meet a preset condition based on a plurality of judgment factors, material attribute information of a machined material and set power, transmitted by a power sensor, of a laser head are called, and surface features of the machined material in the laser cutting die machining process are obtained; the analysis and prediction model is used for processing, and the speculated real power of the laser head is obtained; and when the deviation between the speculated real power of the laser head and the set power is too large, outputting alarm information. According to the method, the acquired material attributes, the set power and the surface characteristic data are deeply processed by analyzing and predicting the depth of the model, so that the real laser power of the laser head can be accurately speculated, power abnormity can be early warned in time, the cutting quality problem is reduced, and the machining quality and stability of the laser cutting die are guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and more particularly, to a monitoring method and system applied to laser die processing. Background Art

[0002] In the field of laser die processing, the stability of laser power is one of the key factors determining the quality of die processing. Precise laser power can ensure smooth cutting edges, accurate dimensions of the die, and reduce the rejection rate.

[0003] Currently, in the industry, the monitoring of laser power during the laser die processing generally adopts the method of directly obtaining the output power value of the laser head. This traditional monitoring method relies on the power sensor of the laser device itself. However, due to factors such as the aging of the power sensor, measurement errors, and changes in the optical path loss inside the device, the laser head power data directly obtained may have large deviations and cannot truly reflect the actual power of the laser acting on the material surface.

[0004] When there is a deviation between the actual laser power and the set power, if it is not discovered and adjusted in time, it will lead to problems such as inconsistent cutting depths of the die, residual slag at the cut, and material deformation, seriously affecting the processing accuracy and service performance of the die.

[0005] In summary, how to more accurately monitor the laser power during the laser die processing to give an alarm in time for the unstable situation of the laser power is a technical problem that needs to be solved currently. Summary of the Invention

[0006] In view of this, the present invention provides a monitoring method, system, electronic device, computer storage medium, and computer program product applied to laser die processing to solve at least one of the above technical problems.

[0007] The present invention provides a monitoring method applied to laser die processing, including the following method steps:

[0008] S10. Determine whether the stability of the laser die processing meets a preset condition based on several determination factors. If not, execute S20;

[0009] S20. Retrieve the material property information of the material to be processed, the set power of the laser head transmitted by the power sensor, and obtain the surface characteristics of the material to be processed during the laser die processing, where the surface characteristics include thermal characteristics, optical characteristics, and surface roughness characteristics;

[0010] S30. Process according to the material property information, the set power, and the surface characteristics using an analysis and prediction model to obtain the estimated actual power of the laser head.

[0011] S40. When the deviation between the inferred actual power of the laser head and the set power is too large, an alarm message is output.

[0012] The present invention also provides a monitoring system applied to laser die processing. The system includes a controller and a storage medium. A computer program is stored in the storage medium. By calling and executing the computer program by the controller, the following steps are implemented:

[0013] S10. Based on several determination factors, determine whether the stability of laser die processing meets a preset condition. If not, execute S20;

[0014] S20. Retrieve the material property information of the material to be processed, the set power of the laser head transmitted by the power sensor, and obtain the surface characteristics of the material to be processed during laser die processing. The surface characteristics include thermal characteristics, optical characteristics, and surface roughness characteristics;

[0015] S30. According to the material property information, the set power, and the surface characteristics, use an analysis and prediction model for processing to obtain the inferred actual power of the laser head;

[0016] S40. When the deviation between the inferred actual power of the laser head and the set power is too large, an alarm message is output.

[0017] The present invention also provides an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the computer program is executed by the processor, the method described in any one of the foregoing is implemented.

[0018] The present invention also provides a computer storage medium, which stores a computer program that can be executed by a processor to implement the method described in any one of the foregoing.

[0019] The present invention also provides a computer program product, which includes a computer program that can be executed by a processor to implement the method described in any one of the foregoing.

[0020] The beneficial effects of the present invention are as follows: The monitoring method of the present invention can accurately infer the actual laser power of the laser head by deeply processing the collected material properties, set power, and surface feature data using the analysis and prediction model. Furthermore, it can timely warn of power abnormalities, reduce cutting quality problems, and ensure the quality and stability of laser die processing. Description of the Drawings

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0022] Figure 1 is a schematic flow chart of a monitoring method applied to laser die processing disclosed in an embodiment of the present invention;

[0023] Figure 2 is a schematic structural diagram of a feature extraction model disclosed in an embodiment of the present invention;

[0024] Figure 3 is a schematic structural diagram of an analysis and prediction model disclosed in an embodiment of the present invention;

[0025] Figure 4 is a schematic structural diagram of a monitoring system applied to laser die processing disclosed in an embodiment of the present invention. Specific Embodiments

[0026] The following specific embodiments illustrate the implementation manners of the present application. Those familiar with this technology can easily understand other advantages and effects of the present application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0027] In addition, the technical features involved in different implementation manners of the present application described below can be combined with each other as long as they do not conflict with each other.

[0028] As Figure 1 shown, an embodiment of the present invention discloses a monitoring method applied to laser die processing, including the following method steps:

[0029] S10. Judge whether the stability of laser die processing meets the preset conditions based on several determination factors. If not, execute S20.

[0030] During the laser die processing, the laser power of the laser head is the core factor determining the processing quality, and its stability directly affects the cutting accuracy and production efficiency of the die. In this step, the stability of the processing process is evaluated in real time through a set of multiple determination factors.

[0031] Among them, the fluctuation of the power supply voltage will directly lead to the instability of the output power of the laser head. For example, when the power supply voltage suddenly drops, the actual output power of the laser head may not reach the set value, thus affecting the cutting effect. By monitoring parameters such as the voltage and current of the power supply and comparing them with the preset stable range, it can be judged whether it is in a normal state.

[0032] In addition, the operating temperature of the laser generator is also closely related to the laser power. If the heat dissipation of the laser generator is poor after long-term operation and the temperature is too high, it will cause the performance of the internal optical components to decline, resulting in fluctuations in the laser power. Therefore, by monitoring the temperature of the laser generator in real time and setting a reasonable temperature threshold, when the temperature exceeds the threshold, it can be determined that the stability of the laser power may be affected.

[0033] The degree of melting or vaporization of the material during the processing can also intuitively reflect whether the laser power is stable. For example, when cutting materials of the same material and the same thickness, if the melting trace at the material cut is significantly wider or narrower during a certain period, it indicates that the laser power may have fluctuated. An industrial camera or other visual detection equipment can be used to take real-time images of the processing area, analyze the melting or vaporization characteristics of the material surface, and compare them with the images in the normal processing state, which can be used as a determination factor.

[0034] In addition, the smoke concentration generated during the processing is also related to the laser power. The higher the laser power, the greater the amount of material vaporization and the more smoke generated. By installing a smoke sensor to monitor the change of the smoke concentration in the processing area, if the smoke concentration shows abnormal fluctuations, it can also be inferred that there is a problem with the stability of the laser power.

[0035] Therefore, the determination factors can include at least one of the fluctuation of the power supply voltage, the operating temperature of the laser generator, the degree of melting or vaporization, and the smoke concentration.

[0036] At the same time, preset conditions are also set. The preset conditions are a series of thresholds or ranges set for the above-mentioned determination factors related to the laser power according to the quality standards and production requirements of laser die-cutting processing. For example, it is stipulated that the power supply voltage fluctuation range should be controlled within ±5%, the temperature of the laser generator should not exceed 60°C, the width change of the melting trace at the material cut should not exceed ±0.1 mm, and the smoke concentration fluctuation range should be within ±10%, etc.

[0037] When one or more determination factors exceed the preset condition range, it is determined that the processing stability of the laser die does not meet the requirements. At this time, the subsequent process is triggered to execute S20, so as to take measures in time to adjust the laser power and avoid the occurrence of processing quality problems.

[0038] S20, retrieve the material property information of the processed material, the set power of the laser head transmitted by the power sensor, and obtain the surface characteristics of the processed material during the laser die cutting process, wherein the surface characteristics include thermal characteristics, optical characteristics and surface roughness characteristics.

[0039] When the processing stability is not up to standard, the key data related to the processing process is comprehensively collected, including the material property information of the processed material, the set power of the laser head transmitted by the power sensor, and the surface characteristics of the processed material during the laser die cutting process.

[0040] Among them, the material property information of the processed material includes the material's thermal conductivity, specific heat capacity, melting point, reflectivity, density and other physical and chemical properties. These properties determine the material's absorption, conduction and response characteristics to laser energy.

[0041] The laser head set power transmitted by the power sensor is the power parameter of the device's theoretical output. However, due to factors such as sensor error and optical path loss, the set power is not the actual power acting on the material surface.

[0042] The surface characteristics of the processed material are an important basis for reflecting the actual laser power: 1) Thermal characteristics use infrared temperature measuring equipment, thermal imagers, etc. to obtain information such as the material surface temperature distribution and temperature change rate, which directly reflects the thermal response of the material after absorbing laser energy; 2) Optical characteristics use spectrometers, reflectivity meters, etc. to detect data such as the intensity, spectral distribution and reflectivity of reflected light on the material surface, which can analyze the energy distribution of the laser on the material surface; 3) Surface roughness characteristics use non-contact surface roughness measuring instruments to measure the surface microscopic morphology parameters of the material after processing, such as the arithmetic mean deviation of the profile Ra (the average value of the absolute value on the reference length), the ten-point height of the microscopic roughness Rz (the sum of the average value of the maximum profile peak height within the sampling length and the average value of the five largest profile valley depths), etc., which intuitively reflect the effect of laser processing on the material surface.

[0043] S30, using an analysis and prediction model to perform processing according to the material property information, the set power and the surface characteristics, to obtain an inferred real power of the laser head.

[0044] The present invention pre-constructs an analysis and prediction model, which can be constructed based on algorithms such as regression analysis and neural networks.

[0045] During the model training stage, processing experiments were carried out on a variety of materials under different laser power settings, and material properties, set power, surface characteristics, and real laser power data were simultaneously collected through high-precision measurement equipment. The data was learned using regression analysis, neural network and other algorithms, allowing the model to accurately capture the complex nonlinear relationship between each parameter and the real laser power.

[0046] After obtaining the above key data, the material property information, set power, and surface feature data among them are input into the analysis and prediction model. Through internal calculations and inferences, the analysis and prediction model outputs the speculation result of the actual power of the laser head.

[0047] S40. When the deviation between the speculated actual power of the laser head and the set power is too large, an alarm message is output.

[0048] It can be understood that this speculation result can more accurately reflect the power of the laser actually acting on the material surface, thus providing a reliable basis for subsequent adjustment of processing parameters and ensuring processing quality. For example, when the deviation between the speculated actual power of the laser head and the set power transmitted by the power sensor is too large, it can be determined that the laser head is affected by factors such as power sensor aging, measurement error, and changes in the optical path loss inside the equipment. At this time, an alarm message is output to prompt the staff to replace or recalibrate the laser head.

[0049] The monitoring method of the present invention can accurately speculate the actual laser power of the laser head by deeply processing the collected material properties, set power, and surface feature data using the analysis and prediction model. Furthermore, it can timely warn of power abnormalities, reduce cutting quality problems, and ensure the processing quality and stability of laser knife molds.

[0050] As an example, the obtaining of the surface features of the material to be processed during the processing of the laser knife mold includes:

[0051] Detecting the temperature change information, laser reflectivity, and luminescence information of the processing area of the material to be processed during the processing. The temperature change information includes surface temperature distribution and temperature change rate; and, detecting the surface roughness information and processed surface shape image of the processing area of the material to be processed after processing.

[0052] Based on the temperature change information and thermal deformation information, the thermal characteristics are constructed. Based on the laser reflectivity and luminescence information, the optical characteristics are constructed. The thermal deformation information is obtained by comparing the processed surface shape image with the surface shape image before processing; and, based on the surface roughness information, the surface roughness characteristics are constructed.

[0053] The thermal characteristics, the optical characteristics, and the surface roughness characteristics constitute the surface characteristics of the material to be processed.

[0054] In this embodiment, during the processing of the laser knife mold, the interaction between the material surface and the laser energy will generate various physical phenomena. These phenomena are related to the laser power and can be used to speculate the actual power of the laser head. Specifically as follows:

[0055] 1) Thermal characteristics:

[0056] Temperature change information: When laser energy acts on the surface of the material to be processed, most of the energy is absorbed by the material and converted into heat energy, causing the temperature of the material to rise. The surface temperature distribution reflects the spatial distribution state of laser energy on the material surface, and the temperature change rate reflects the speed at which the material absorbs laser energy.

[0057] Laser power is closely related to temperature change. The higher the power, the more energy the material absorbs per unit time, the higher the surface temperature, and the faster the heating rate; conversely, the lower the power, the less obvious the change in the surface temperature of the material. By monitoring the temperature change information in real time, the transfer of laser energy on the material surface can be intuitively reflected, and then the true power of the laser head can be inferred.

[0058] Thermal deformation information: After the material to be processed absorbs laser energy, it will deform due to thermal expansion and contraction and internal stress changes. The degree of thermal deformation is closely related to the thermophysical properties of the material itself and the input of laser energy. When the laser power is too high, the material absorbs too much energy, and excessive thermal expansion will cause obvious thermal deformation, such as bending and twisting; when the power is too low, the material is not heated sufficiently, and the degree of thermal deformation is weak.

[0059] By comparing the surface shape information of the material before and after processing to obtain thermal deformation information, the effect of laser power on the thermal action of the material can be evaluated, providing a key basis for analyzing the true laser power.

[0060] 2) Optical characteristics:

[0061] Laser reflectivity: When laser irradiates the surface of the material to be processed, part of the energy is reflected, and the reflectivity represents the ratio of the reflected energy to the incident laser energy. Factors such as the surface state and chemical composition of the material will affect its laser reflection ability, and changes in laser power will change the microstructure and physical and chemical properties of the material surface, thus affecting the reflectivity.

[0062] When the laser power is low, there is no significant change on the material surface, and the reflectivity is relatively stable; as the laser power increases, the material surface is melted and vaporized, the surface roughness increases, and the reflectivity will change.

[0063] By measuring the laser reflectivity, the reflection of laser energy on the material surface can be analyzed, and then the energy distribution during the interaction between the laser and the material can be inferred, assisting in inferring the true laser power.

[0064] Luminescence information: After the material to be processed absorbs laser energy, some electrons are excited to higher energy levels. When the electrons transition from higher energy levels back to lower energy levels, they will release energy in the form of light, producing a luminescence phenomenon. Luminescence characteristics include parameters such as luminescence intensity and spectral distribution, which are determined by the energy level structure of the material and the absorbed laser energy.

[0065] With different laser powers, the energy absorbed by the material is different, and the degree of electron excitation and transition is also different, resulting in changes in the luminescence intensity and spectral characteristics. High-power lasers will cause the material to have a stronger luminescence intensity, and the spectral distribution may also change.

[0066] By detecting the luminescence characteristics of the material, information on the laser energy absorbed by the material can be obtained from an optical perspective, providing a basis for judging the true laser power.

[0067] 3) Surface roughness: During the processing of laser die-cutting dies, the laser interacts with the material, removing the material through melting, vaporization, etc., forming cutting or engraving marks. These microscopic processing marks determine the roughness of the material surface.

[0068] If the laser power is too high, the material will be over-melted and vaporized, resulting in a large number of irregular protrusions and depressions on the surface, increasing the surface roughness; if the power is too low, the material will not be removed sufficiently, and there will be more unprocessed parts remaining on the surface, which will also increase the surface roughness. An appropriate laser power can make the surface of the material processed evenly with a smaller surface roughness.

[0069] By measuring surface roughness parameters such as the arithmetic mean deviation of the profile Ra and the ten-point height of the micro-irregularities Rz, the influence of laser processing on the microscopic structure of the material surface can be intuitively reflected, and then it can be judged whether the laser power is within a reasonable range, providing important information on the surface topography for inferring the true laser power.

[0070] Equipment such as an infrared thermal imager and an infrared temperature sensor can be used to collect temperature change information in real time; a spectrometer and a laser reflectivity measuring instrument are used to obtain laser reflectivity and luminescence information; a white light interferometer and a laser confocal microscope are used to measure surface roughness information; a non-contact three-dimensional scanner is used to obtain surface shape images before and after processing respectively.

[0071] As an example, the thermal deformation information is obtained based on the comparison between the surface shape image after processing and the surface shape image before processing, specifically:

[0072] The first surface feature and the second surface feature are respectively extracted from the surface shape image after processing and the surface shape image before processing using a feature extraction model; the first surface feature and the second surface feature are compared as a whole to obtain the degree of deviation from the expected deviation degree, that is, the thermal deformation information is obtained;

[0073] The feature extraction model includes a convolutional network module, a shape topology analysis module, and a feature fusion module;

[0074] The convolutional network module is used to respectively extract the third surface feature and the fourth surface feature from the surface shape image after processing and the surface shape image before processing;

[0075] The shape topology analysis module is used to respectively obtain a fifth surface feature and a sixth surface feature from the processed surface shape image and the pre-processed surface shape image at the topology structure level. Both the fifth surface feature and the sixth surface feature include connected components, the number of holes, and boundary curvature.

[0076] The feature fusion module is used to fuse the third surface feature with the fifth surface feature, and fuse the fourth surface feature with the sixth surface feature, respectively obtaining the first surface feature and the second surface feature.

[0077] In this embodiment, the present invention uses a pre-constructed feature extraction model to extract geometric features in the surface shape images of the material to be processed before and after processing, and then calculates the changes in the geometric features of the material to be processed before and after processing based on the obtained first surface feature and second surface feature, and further accurately calculates the actual power of the laser head.

[0078] However, traditional feature extraction methods mostly focus on geometric features of the material surface shape, such as edges, contours, textures, etc. However, these geometric features can only reflect partial information of the change in the material surface shape. When facing materials with complex internal structures or special materials, it is difficult to comprehensively and accurately describe the thermal deformation situation only relying on geometric features, resulting in the above-mentioned deviation degree, that is, the thermal deformation information obtained is not accurate enough to accurately calculate the actual power of the laser head.

[0079] To solve this technical problem, as Figure 2 shown, the feature extraction model of the present invention includes not only a convolutional network module, but also a shape topology analysis module and a feature fusion module. Specifically:

[0080] The shape topology analysis module is constructed based on a graph neural network or topological data analysis method. It can deeply explore the changes in the material surface shape before and after laser processing from the perspective of topology. For example, for a material with a hollow pattern, traditional methods may only be able to capture the geometric changes at the edge of the hollow, but it is difficult to detect the change in the connectivity of the hollow part; for multi-layer composite materials, the topological relationship changes caused by thermal deformation between layers are even a blind spot for traditional geometric feature extraction methods. The shape topology analysis module can accurately identify these subtle topological structure changes by calculating topological features such as the connected components, the number of holes, and the boundary curvature of the shape, providing a new perspective and key data support for thermal deformation information extraction.

[0081] The running process of each functional module of the feature extraction model is generally as follows:

[0082] 1) During the operation of the model, the normalized pre - processing surface shape image and the post - processing surface shape image are transmitted to the convolutional network module. The convolutional network module performs operations such as convolution, pooling, and activation on the images, extracts basic geometric features, and outputs feature maps. These feature maps are then passed to the shape topology analysis module.

[0083] 2) The shape topology analysis module performs topology structure analysis on the feature maps and extracts topology features. The topology features are features such as the connected components of the local shape of the processed area of the material before and after processing, the number of holes, and the boundary curvature.

[0084] 3) The feature fusion module fuses the geometric features (i.e., the third surface feature and the fourth surface feature) output by the convolutional network module with the topology features (i.e., the fifth surface feature and the sixth surface feature) output by the shape topology analysis module to generate a comprehensive feature vector (i.e., the first surface feature and the second surface feature), and outputs the comprehensive feature vector for comparison with the expected deviation degree, thereby obtaining thermal deformation information.

[0085] Through the feature extraction model designed by the present invention, subtle changes in the topology level such as connectivity, hole structure, and inter - layer relationship of the material surface before and after laser processing, as well as feature differences in the geometric level such as edges and contours, can be accurately captured. The shape features of the material surface are comprehensively and deeply extracted from both geometric and topological dimensions. Compared with the method that only relies on traditional geometric feature extraction, the accuracy and integrity of thermal deformation information extraction are greatly improved. This multi - dimensional feature extraction method enables the model to have stronger adaptability and robustness when facing materials with different materials and different processing technologies. The accurate thermal deformation information extracted provides a reliable basis for subsequent judgment of whether the laser power is appropriate, helps to detect abnormal situations in the laser die - cutting process in a timely manner, ensures processing quality, and improves production efficiency.

[0086] It can be understood that the convolutional network module usually includes multiple convolutional layers, pooling layers, and activation function layers, and the specific feature extraction process will not be elaborated here. The feature fusion module can use methods such as splicing and weighted summation to fuse the aforementioned geometric features and topology into a complete comprehensive feature vector.

[0087] As an example, the shape topology analysis module includes a graph structure construction unit, a node feature update unit, a graph pooling unit, and a global feature aggregation unit;

[0088] The graph structure construction unit is used to convert the post - processing surface shape image and the pre - processing surface shape image into a graph structure; the nodes in the graph structure represent the feature regions of the material surface, the node attributes include geometric features, the edges represent the topological relationships between the nodes, and the edge attributes are used to describe the relationship strength between the nodes;

[0089] The node feature update unit is used to calculate the input message based on the geometric features of adjacent nodes and edge attributes, and update the node features by combining the input message with the original geometric features of the nodes using an update function;

[0090] The graph pooling unit is used to reduce the dimension of the graph structure by screening and aggregating nodes;

[0091] The global feature aggregation unit is used to aggregate the node geometric features of each node in the dimension-reduced graph structure into a global feature vector, and perform linear transformation and activation operations on the global feature vector through a linear layer to obtain and output topological features; wherein, the topological features are the fifth surface feature or the sixth surface feature.

[0092] In this embodiment, in order to implement the foregoing function of being able to recognize subtle topological structure changes, the present invention sets the shape topological analysis module to include a graph structure construction unit, a node feature update unit, a graph pooling unit, and a global feature aggregation unit. The specific functions of each functional unit are as follows:

[0093] The graph structure construction unit: is used to convert image information into a graph data structure.

[0094] Convert the processed surface shape image and the pre-processed surface shape image into a graph structure G=(V, E) respectively, where the node set V represents the feature regions of the material surface. Assuming that the material surface is divided into n feature regions, then the node v i ∈V, i = 1, 2,..., n. The node attribute x i contains geometric features. For example, in a two-dimensional image, it can be expressed as x i =[p i,x , p i,y , c i , where p i，x , p i，y are the central coordinates of the feature region corresponding to the node v i , and c i is the average curvature of this region. The edge set E represents the topological relationship between nodes. The edge (v i , v j ) ∈ E, and the edge attribute e ij is used to describe the relationship strength between the nodes v i and v j . For example, it can be calculated by the Euclidean distance. The closer the distance, the greater the relationship strength.

[0095] The node feature update unit: The node feature update unit calculates the input message based on the geometric features of adjacent nodes and edge attributes. For the node v i , its input message m i is calculated as follows:

[0096] Among them, σ is an activation function (such as the sigmoid function), which is used to normalize the edge attribute e ij so that its weight is in the interval [0, 1]; MLP(x j ) represents the feature transformation of the geometric feature x j of the adjacent node v j through a multi-layer perceptron.

[0097] Then, the update function is used to combine the input message with the original geometric feature of the node to update the node feature: Among them, |m i , x i | means concatenating the input message m i with the original feature x i of the node, and learning a more expressive new node feature through a multi-layer perceptron to achieve the update of the node feature.

[0098] Graph pooling unit: Reducing the dimension of the graph structure through screening and aggregation.

[0099] Nodes are retained or deleted according to the node importance score to achieve screening. The node importance score s i can be calculated based on the node feature, that is, the L2 norm of the node feature vector is used as the score, and a threshold τ is set. When s i ≥τ, the node is retained; otherwise, it is deleted.

[0100] When aggregating nodes, for multiple adjacent nodes v i1 , v i2 , …, v ik , they are aggregated into a super node v s , and the super node feature x s is calculated as follows:

[0101] Global feature aggregation unit: Aggregating the node geometric features of each node in the dimension-reduced graph structure into a global feature vector g. The summation method can be used: Among them, V pool is the set of nodes retained after graph pooling.

[0102] After obtaining the global feature vector, the global feature vector is linearly transformed and activated through the linear layer W to obtain the topological feature t = ReLU(W g + b), where ReLU is the rectified linear unit function and b is the bias term.

[0103] The finally output topological feature t is either the fifth surface feature or the sixth surface feature, which is used for subsequent fusion with other features to assist in analyzing the thermal deformation information of the material. By screening and aggregating, the number of nodes in the graph structure is reduced, while the key information is retained and the computational complexity is reduced.

[0104] As an example, the analysis and prediction model includes an initial convolutional layer, a deep residual shrinkage module, and a fully connected layer. The deep residual shrinkage module is formed by cascading multiple residual blocks, and each residual block includes a convolutional layer, a batch normalization layer, an activation function layer, and a soft thresholding layer.

[0105] Among them, a coefficient related to the processing speed is set in the threshold calculation formula of the soft thresholding layer, and this coefficient is positively correlated with the processing speed.

[0106] In this embodiment, the Deep Residual Shrinkage Network (DRSN) can effectively handle the complex non-linear relationships and noise interference problems in the laser die-cutting processing data. Therefore, the present invention preferably uses DRSN to construct the analysis and prediction model, as Figure 3 shown, the constructed analysis and prediction model mainly includes an initial convolutional layer, a deep residual shrinkage module, and a fully connected layer. Of course, it will also include an input layer and an output layer, which will not be elaborated here.

[0107] The specific functions of each functional unit of the analysis and prediction model are as follows:

[0108] (1) Initial convolutional layer: The input data (i.e., material property information, set power, and surface features) is input into the initial convolutional layer. The initial convolutional layer performs a convolution operation on the above input data through a group of convolutional kernels (such as 3×3 convolutional kernels) to extract local features in the data and convert the input data into a feature map containing preliminary features. Specifically, for the surface temperature distribution data on the material surface, the initial convolutional layer can extract the local patterns of temperature changes; for the optical feature data, the initial convolutional layer can capture the local features of the reflected light intensity changes, etc.

[0109] (2) Deep residual shrinkage module: The deep residual shrinkage module is the core part of the analysis and prediction model, which is formed by cascading multiple residual blocks. Each residual block includes a convolutional layer, a batch normalization layer, an activation function layer, and a key soft thresholding layer.

[0110] Residual connection: In the residual block, on the one hand, the data is directly passed to the next layer through the shortcut connection, and on the other hand, it undergoes feature extraction and transformation through the convolutional layer, batch normalization layer, and activation function layer to obtain a feature map. This residual connection method can solve the problems of gradient disappearance and gradient explosion that may occur when the network depth increases, enabling the network to learn deeper features.

[0111] Soft thresholding layer: After operations such as convolution and activation, the feature map is transmitted to the soft thresholding layer. This layer adaptively calculates the threshold according to the data distribution. The specific calculation method is as follows: for each element x in the feature map, the value y after the soft thresholding operation is: y = sgn(x)·(|x| - λ). + . Where sgn(x) is the sign function, λ is the calculated threshold, and (·) + represents the operation of taking the positive value.

[0112] Through soft thresholding, the noise and unimportant features in the data can be suppressed, and the key features valuable for predicting the true power of the laser head can be retained. Thus, under the cascaded processing of multiple deep residual shrinkage modules, the complex features in the feature map are gradually extracted and optimized, while the noise is effectively suppressed.

[0113] In addition, in order to match the prediction speed of the deep residual shrinkage network with the processing speed of the laser head, that is, the faster the processing speed, the faster the prediction speed of the deep residual shrinkage network, so that the abnormal situation of the laser power of the laser head can be detected more timely and intervention can be carried out in time. To this end, a further improvement of the present invention is that a coefficient k related to the processing speed is added to the threshold calculation formula of the soft thresholding layer, and this coefficient k is positively correlated with the processing speed, so that λ = k·f(X).

[0114] Where f(X) = α×std(X), α is a regulation factor used to control the size of the threshold λ, for example, α = 0.2,; X represents the set of feature data corresponding to the feature map input to the soft thresholding layer, and std(X) is the standard deviation of this data set X, which is used to measure the degree of data dispersion. The larger the standard deviation, the greater the data fluctuation and the more noise.

[0115] By adding the above coefficient k, when the processing speed increases, the coefficient k in the threshold calculation is correspondingly increased, which can accelerate the processing speed of soft thresholding, reduce the time overhead of data processing, thereby improving the prediction speed of the deep residual shrinkage network, and further improving the prediction speed of the analysis and prediction model.

[0116] It should be noted that the threshold value λ is used to measure the importance of eigenvalue and determine the processing method for the element x in the feature data set X corresponding to the feature map. When the absolute value of the element x in the feature map is less than λ, it indicates that the feature represented by this element is relatively unimportant in the overall data distribution, and it may be noise or redundant information with little contribution to predicting the true power of the laser head. At this time, after the soft thresholding operation, the value of this element will be shrunk to 0; when the absolute value of the element x is greater than or equal to λ, it shows that the feature corresponding to this element has a certain importance. After subtracting λ from its absolute value, its sign direction is retained and passed to the subsequent network layer for processing. In other words, the threshold value λ plays a role in screening features and suppressing noise.

[0117] As Figure 4 shown, an embodiment of the present invention further provides a monitoring system applied to laser die processing. The system includes a controller (101) and a storage medium (102). A computer program is stored in the storage medium (102). By invoking and executing the computer program, the controller (101) realizes the following steps:

[0118] S10, judging whether the stability of laser die processing meets the preset conditions based on several determination factors. If not, execute S20;

[0119] S20, retrieving the material property information of the material to be processed, the set power of the laser head transmitted by the power sensor, and obtaining the surface features of the material to be processed during laser die processing. The surface features include thermal features, optical features, and surface roughness features;

[0120] S30, processing according to the material property information, the set power, and the surface features using an analysis and prediction model to obtain the speculated true power of the laser head;

[0121] S40, when the deviation between the speculated true power of the laser head and the set power is too large, output an alarm message.

[0122] An embodiment of the present invention further provides an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the computer program is executed by the processor, it realizes the method described in any one of the foregoing items.

[0123] An embodiment of the present invention further provides a computer storage medium, which stores a computer program that can be executed by a processor to realize the method described in any one of the foregoing items.

[0124] An embodiment of the present invention further provides a computer program product, which includes a computer program that can be executed by a processor to realize the method described in any one of the foregoing items.

[0125] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0126] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A monitoring method applied to laser die processing, characterized in that: It includes the following method steps: S10. Determine whether the stability of laser die cutting processing meets the preset conditions based on several determination factors. If not, execute S20; S20. Retrieve the material property information of the material to be processed, the set power of the laser head transmitted by the power sensor, and obtain the surface characteristics of the material to be processed during the laser die cutting process, where the surface characteristics include thermal characteristics, optical characteristics, and surface roughness characteristics; S30. Process according to the material property information, the set power, and the surface characteristics using an analysis and prediction model to obtain the speculated actual power of the laser head; S40. When the deviation between the speculated actual power of the laser head and the set power is too large, output an alarm message.

2. The monitoring method for laser die processing according to claim 1, wherein: The obtaining of the surface characteristics of the material to be processed during the laser die cutting process includes: Detect the temperature change information, laser reflectivity, and luminescence information of the processing area of the material to be processed during the processing process, where the temperature change information includes surface temperature distribution and temperature change rate; and detect the surface roughness information and processed surface shape image of the processing area of the material to be processed after processing; Construct the thermal characteristics based on the temperature change information and thermal deformation information, construct the optical characteristics based on the laser reflectivity and luminescence information, where the thermal deformation information is obtained by comparing the processed surface shape image with the pre-processed surface shape image; and construct the surface roughness characteristics based on the surface roughness information; The thermal characteristics, the optical characteristics, and the surface roughness characteristics constitute the surface characteristics of the material to be processed.

3. The monitoring method applied to laser die - cutting processing according to claim 2, wherein: The thermal deformation information is obtained by comparing the processed surface shape image with the pre-processed surface shape image. Specifically: Use a feature extraction model to respectively extract the first surface feature and the second surface feature from the processed surface shape image and the pre-processed surface shape image; perform an overall comparison on the first surface feature and the second surface feature to obtain the deviation degree from the expected deviation degree, that is, obtain the thermal deformation information; The feature extraction model includes a convolutional network module, a shape topology analysis module, and a feature fusion module; The convolutional network module is used to respectively extract the third surface feature and the fourth surface feature from the processed surface shape image and the pre-processed surface shape image; The shape topology analysis module is used to respectively extract the fifth surface feature and the sixth surface feature from the processed surface shape image and the pre-processed surface shape image at the topological structure level, where the fifth surface feature and the sixth surface feature both include connected components, the number of holes, and boundary curvature; The feature fusion module is used to fuse the third surface feature with the fifth surface feature, and fuse the fourth surface feature with the sixth surface feature to respectively obtain the first surface feature and the second surface feature.

4. The monitoring method applied to laser die processing according to claim 3, wherein: The shape topology analysis module includes a graph structure construction unit, a node feature update unit, a graph pooling unit, and a global feature aggregation unit; The graph structure construction unit is used to convert the processed surface shape image and the surface shape image before processing into a graph structure; the nodes in the graph structure represent the characteristic regions of the material surface, the node attributes include geometric features, the edges represent the topological relationships between the nodes, and the edge attributes are used to describe the relationship strength between the nodes. The node feature update unit is used to calculate the input message based on the geometric features of adjacent nodes and the edge attributes, and update the node features by combining the input message with the original geometric features of the nodes using an update function. The graph pooling unit is used to reduce the dimension of the graph structure by screening and aggregating nodes. The global feature aggregation unit is used to aggregate the node geometric features of each node in the dimension-reduced graph structure into a global feature vector, and perform a linear transformation and activation operation on the global feature vector through a linear layer to obtain and output the topological feature; wherein, the topological feature is the fifth surface feature or the sixth surface feature.

5. A monitoring method applied to laser die processing according to claim 1, characterized in that: The analysis and prediction model includes an initial convolutional layer, a deep residual shrinkage module, and a fully connected layer. The deep residual shrinkage module is formed by cascading multiple residual blocks, and each residual block includes a convolutional layer, a batch normalization layer, an activation function layer, and a soft thresholding layer. Among them, a coefficient related to the processing speed is set in the threshold calculation formula of the soft thresholding layer, and this coefficient is positively correlated with the processing speed.

6. A monitoring system applied to laser die processing, characterized in that, The system includes a controller and a storage medium. A computer program is stored in the storage medium. The controller realizes the following steps by calling and executing the computer program: S10, judging whether the stability of the laser die cutting process meets the preset conditions based on several determination factors. If not, execute S20. S20, retrieving the material property information of the material to be processed, the set power of the laser head transmitted by the power sensor, and obtaining the surface features of the material to be processed during the laser die cutting process. The surface features include thermal features, optical features, and surface roughness features. S30, using the analysis and prediction model to process according to the material property information, the set power, and the surface features to obtain the estimated actual power of the laser head. S40, when the deviation between the estimated actual power of the laser head and the set power is too large, output an alarm message.

7. The monitoring system for laser die processing according to claim 6, wherein: The obtaining of the surface features of the material to be processed during the laser die cutting process includes: Detecting the temperature change information, laser reflectivity, and luminescence information of the processing area of the material to be processed during the processing. The temperature change information includes the surface temperature distribution and the temperature change rate; and, detecting the surface roughness information and the processed surface shape image of the processing area of the material to be processed after processing. Constructing the thermal features based on the temperature change information and the thermal deformation information, constructing the optical features based on the laser reflectivity and the luminescence information. The thermal deformation information is obtained by comparing the processed surface shape image with the surface shape image before processing; and, constructing the surface roughness features based on the surface roughness information. The thermal features, the optical features, and the surface roughness features constitute the surface features of the material to be processed.

8. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, where the computer program, when executed by the processor, implements the method according to any one of claims 1-5.

9. A computer storage medium, characterized in that: The computer storage medium stores a computer program executable by a processor to implement the method according to any one of claims 1-5.

10. A computer program product, characterized in that: The computer program product includes a computer program executable by a processor to implement the method according to any one of claims 1-5.

Images